#qi-hardware IRC log for Saturday, 2012-10-27

kristianpaulhow long until a zynq based nettop/netbook?00:23
kristianpaulwell basically a zedboard could be used as a nettop..00:31
wolfspraulkristianpaul: yes, that'd be nice, no? (a zynq-based computer...)00:58
wolfspraulkeep an eye open for it and if you see something let us know...00:58
wolfspraulxilinx recently acquired petalogix (petalogix.com), presumably to offer linux/embedded solutions around the zynq01:02
xiangfuwolfspraul, good morning.01:08
wolfspraulgood morning!01:11
qi-bot[commit] Xiangfu: uboot-xburst: split spl and usbboot patch (master) http://qi-hw.com/p/openwrt-xburst/cb2b6b501:21
kristianpaulwolfspraul: nice yes :D01:37
wolfspraulmaybe I should try to experiment with zynq instead of xc6/xc7a ?01:37
kristianpaulhmm 01:38
wolfspraulwith a xc7z7020 we could run a full Linux on the arm side, and milky or some pieces of it on the fpga side01:38
kristianpaulhaving linux running is a nice entry :-)01:44
kristianpauland the open thing that about reconfigure the fpga from the same os well :)01:44
kristianpaulbut01:44
kristianpauli think still some more bits to experiement no?01:45
kristianpaulat least at the point more resources  (bram, dcm) are freed01:47
kristianpaulbut of course if you are ina hurry for chip SoC solution zynq seems way go01:48
wolfspra1lgot disconnected01:48
kristianpaulouch01:48
wolfspra1lI saw "but" and then "at least at the point"01:48
wolfspra1lwhat was in between? :-)01:48
kristianpaulbut -> i think still some more bits to experiement no? -> at least at the point more resources  (bram, dcm) are freed01:49
wolfspra1lI need to check what petalogix has01:49
kristianpauli was sewing xilinx wiki last night01:50
kristianpaulthey seems to mention upstream...01:50
wolfspra1lI think everybody by now learned that 'upstream' is the good thing01:52
wolfspra1llike "iso 9001" in china01:52
wolfspra1lwhat it means is another thing01:53
wolfspra1ljust yesterday I pealed off a sticker of a newly bought product that read "ISO 9001 QUATITY CONTROL"01:53
wolfspra1lvery convincing :-)01:53
wolfspra1lso upstream, yep. great.01:53
wolfspra1lsounds right01:53
kristianpaulnow about sdr, i remenber larsc pointed out some interesting sdr combo using zynq 01:54
wolfspra1lnice01:55
wolfspra1lhere in the channel? url/01:55
wolfspra1l?01:55
kristianpaulor in milkymist01:55
kristianpauldont remenber link.. i just googled for analog instrument zynq ...01:56
kristianpaulbut as my initial question i want to see a product that uses zynq currently01:57
kristianpauli guess now is avaliable (xc7z7020) to buy ?01:58
wolfspra1lengineering samples - yes01:58
kristianpaulah samples..01:59
wolfspra1lxiangfu: ok I'm just looking at packages...03:16
wolfspra1lthe slx9 is available in qfp144 and ftg256 as we know03:17
wolfspra1lthen also csg325 (0.8mm spacing) and csg225 and csg48403:18
wolfspra1land fgg484 at 1.0 mm spacing03:18
wolfspra1lah no, wait. not the slx9... :-)03:18
wolfspra1lthen a small cpg196 package (bga with 0.5mm spacing)03:18
wolfspra1lartix-7 engineering samples on digikey right now in csg324 (0.8mm bga) and fgg676 (1.0 mm bga)03:21
wolfspra1l(typo above - for the slx9 I meant csg324, not 325)03:21
wolfspra1llater on the xc7a100 is also planned to be available in ftg256 and fgg48403:22
qi-bot[commit] David Kühling: qemu-host: workaround for compile problem with latest openwrt (master) http://qi-hw.com/p/openwrt-packages/bafe65803:23
wolfspra1lfinally the xc7z7020 is available on digikey right now as engineering samples in clg400 and clg484 packages, both 0.8mm bga03:23
wolfspra1lxiangfu: so we could try a csg324 slx9 next? :-)03:25
qi-bot[commit] David Kühling: qemu-host: fix compilation when ccache enabled (master) http://qi-hw.com/p/openwrt-packages/bf115de04:23
qi-bot[commit] David Kühling: emacs: upgrade to version 23.4 plus various minor fixes (master) http://qi-hw.com/p/openwrt-packages/cc4792004:23
wpwrak(6 vs. 7) discipline, discipline ... :)04:32
xiangfucheckout this: http://www.kickstarter.com/projects/adapteva/parallella-a-supercomputer-for-everyone05:58
xiangfuit's using zynq 7010, running Ubuntu OS.05:58
wolfspra1lxiangfu: oh, nice06:06
wolfspra1lI think I saw this before but didn't pay attention, have to check again06:06
wolfspra1land it seems they got their 750k usd that they were aiming for :-)06:06
wolfspra1lthey promised to "open everything" after getting those 750k usd, so what stops them now? we will find out soon :-)06:07
wolfspra1lmaybe that's a new open source business model - open blackmail06:08
wolfspra1l"I will open these unbelievably valuable 'things' after I receive XXX USD"06:08
wolfspra1lbut anyway, looks like a good project - thanks!06:08
wolfspra1lQ: "will you open source the epiphany chips" A: not initially, but considered later06:17
wolfspra1lI was thinking a bit more about artix vs. zynq, and realized the cost issue06:17
wolfspra1lan xc7a100 costs 140 USD on digikey now, vs. 240 USD for the zynq (with 85k fabric cells compared to 100k for the xc7a100)06:18
wolfspra1lthat's 100 USD more for the ARM core06:18
wolfspra1lwe can divide digikey prices roughly in half, meaning xc7a100 = 70 USD, xc7z7020 = 120 USD06:18
wolfspra1lstill 50 USD more06:18
viricwolfspra1l: I remember a game writer that promised to give the games open source, if he got donetions over some limit. Didn't succeed, and then continued selling closed source.07:38
viric- he wrote that video game about making linux distributions07:39
viricIsn't the M1 running uclinux? Why the uclinux web page doesn't cite M1?07:42
larscit'srunning linux08:00
whitequarkah, was just going to write about parallella08:35
whitequarkwolfspra1l: if the zynq chip costs $240, how are they able to sell the complete boards for $99?08:37
whitequarkwell, they at least have the complete reference manual published08:47
whitequarkthat chip has an interesting topology. inter-core write transactions are 16x more efficient than read ones.08:47
lekernelwhitequark: "PSoC Creator is a Windows-based IDE. However, I exclusively use Macs"09:30
lekernel*grin*09:30
lekernelwhitequark: are you coming to EHSM?09:30
whitequarklekernel: as I've said already, I considered that a toy, so wine would be fine. but it turned out to be a completely un-fun toy.09:35
whitequarklekernel: EHSM?09:35
whitequarkhm, Dec 28, Berlin...09:35
whitequarkseems interesting. I wonder if I could get a visa soon enough.09:37
lekernelif you need an invitation letter or such, I can produce one...09:38
whitequarkthanks! I'll consider the options I have.09:40
whitequark... and if adapteva just published the sources, I could load it to my M1 and use it right now. sigh.09:54
viricthis EHSM looks very interesting10:04
wpwrakwhitequark: (un-fun toy) psoc or their IDE ?10:09
whitequarkwpwrak: the psoc-specific parts of toolchain10:10
whitequarkapparently there's no way to compile the PLD bitstream except for that .NET IDE10:10
viricmono can run many .net things10:12
wpwrakwell, the chip is completely documented at the register level. so you can just roll your own ;-)10:12
whitequarkviric: it requires native components10:13
whitequarkthe ide won't run under either mono@linux, mono@wine or dotnet@wine10:13
viricok10:14
whitequarkand I'm not going to waste my time by developing the whole application under windows. it's just unproductive.10:14
wpwrakthey help you to stay pure ;-)10:14
whitequarkwpwrak: roll my own... that's at least a year of work to make a good toolchain10:14
whitequarkand the chip is too expensive for real projects anyway10:15
wolfspra1lwhitequark: i don't fully understand the parallella thing either, need to read a bit more10:15
wolfspra1lmaybe 99 USD is just for the board containing their newly made asic10:15
wolfspra1lwhich is what they need the 750k usd for10:15
whitequarkwolfspra1l: yeah, that's for the board with ASIC10:15
wolfspra1l(and which won't be opened either)10:15
wolfspra1lso they found a way to offload the risk of an asic tape-out via kickstarter - great! :-)10:16
wolfspra1lthere should be more such projects, wherever the chips go in the end...10:16
wpwrakwhitequark: yeah, they made their new psocs very nice from a technical point of view, but also quite ugly from a price point10:16
whitequarkwolfspra1l: on one hand, they have a really really nice and well thought out ISA (I just read the reference manual), and I'm fairly certain for the completely open toolchain10:18
whitequarkon the other one, the chips will probably never be open10:19
wpwrakyou mean at the verilog level ?10:19
whitequarkyup10:19
wpwrakokay, that would be asking for a lot10:19
whitequarkindeed.10:20
wpwrakbut when i looked at them a few years ago, they had a very nice reference manual that explained every last bit of "FPGA" state. (among a ton of other things)10:21
whitequarkbut it's a coprocessor. it doesn't have as much value as it could unless it can be incorporated in other designs10:21
wpwrakso until wolfgang publishes his results, these psocs are probably the best-documented non-trivial programmable logic in the world :)10:21
larscwpwrak, whitequark: I think you are talking about differnt chips10:22
whitequarkwpwrak: oh, you're talking about PSoCs10:22
wpwrakah, and you about the parallela critter :)10:22
whitequarkPSoC is what should have actually been at the heart of arduino. it's a perfect prototyping platform.10:22
whitequarkmaybe when freeSoC completes their job, someone will write a good toolchain for it10:23
wpwrakyeah, good point10:23
larscsomething on the scale of PSoC might be a good starting point for a kind of opensource fpga10:32
e2580hi10:51
panda|zxiangfu: yo ;)10:52
e2580Anyone interested in a sneak peek at a hardware based security device we are about ready to release?10:52
panda|ze2580: sure, what's that?10:52
e2580www.cryptx2.com10:52
e2580details on the info page. i can answer questions if you have any10:53
larsce2580: can you give a short overview/introduction?10:53
e2580the device is a hardware encrypted storage device10:55
e2580in short10:55
e2580for details the website lists the features and functions on info page mostly10:55
e2580the video is not quite ready yet...10:55
panda|ze2580: OK, sounds like to protect your data on SD card?10:57
panda|ze2580: so data on those 2 SD card is not accessable without this device?10:58
e2580sd card is used for the storage of the data10:58
e2580the data on the sd cards is aes256 encrypted, you will not be able to decrypt it without the device10:58
panda|ze2580: OK, I see, it's cool, but probably I have no chance to use it :)11:00
e2580there will be suppliers outside the USA shortly after release11:01
e2580we are restricted on the export due to US law :(11:01
e2580but it is pen source, so the device will be built by other :)11:01
panda|ze2580: so a lot of people in U.S need such device to protect their data on SD card? or just for some professional users?11:04
e2580this device is mostly for IT or security professionals. but anyone interested in true data security, or small businesses, or anyone doing R&D that needs secure data storage will be interested11:06
e2580the average person will likely not need this device11:06
e2580you dont need a bank vault at fort knox to protect $20  lol11:07
panda|zyeah, it's a too heavy gun11:11
e2580it will have some other functions for normal users 11:11
e2580such as emulating a cdrom with iso, so you cn load iso on one of the sd cards, and it will act as cdrom11:12
e2580the other will be storage encrypted11:12
e2580firmware is upgradeable on the device, so you can make it do many other things11:12
whitequarke2580: so I need to enter the passkey via the buttons?..11:17
whitequarksounds quite tedious11:17
whitequarkbesides that, there were quite a few articles on circumventing hardware security11:17
whitequarksimplest is to cover parts of the chip with UV-impenetrable substance and then use UV to erase the, for example, firmware, or fuse bits11:18
whitequarkyou can do very interesting things with an AFM, and you can get one with $10k if you want11:20
whitequarkalso, hardware AES bitstream crypto was broken on a certain military-grade FPGA by analyzing EM emissions11:20
whitequarkstatistically11:20
whitequarkbeing open-source is nice, but I assume from the export restrictions that you use a specialized chip with a hardware crypto module, and it's pretty certainly not open-source11:21
whitequarkI also don't see the partnumber of the chip anywhere on the website11:22
e2580whitequark, link me to any articles 11:28
e2580DPA is not going to be possible on this device, it has ECM enabled11:29
e2580also, this device is not UV eeprom11:29
whitequarke2580: obviously it isn't UV eeprom11:30
whitequarkbut you can erase any eeprom with UV rays if you expose the die :)11:30
whitequarkhttp://www.cl.cam.ac.uk/~sps32/mcu_lock.html11:30
whitequarkhttp://hackaday.com/2011/06/27/bunnies-archives-unlocking-protected-microcontrollers/11:30
e2580also, the aes key is not stored on the device, so you wont be able to gain much from the mico even if you found a way to crack it (highly unlikely)11:30
whitequarkwhere is it stored?11:31
whitequarkor, rather, the salt part of the key11:31
whitequarkI guess the AES key is derived from the user input. is that correct?11:32
kyak5 button means total of 11:32
whitequarkkyak: the key is entered with four buttons, and it has arbitrary length11:33
kyakoh ok)11:33
whitequarkthink entering 2-bit characters per click11:33
e2580user input key > salt, stored on the micro controller > hash = aes key. aes key is not stored. instead a data block is decrypted with the resulting aes key and checked for a value, if the value that is decrypted is correct, they the key is considered valid, else it is not valid11:33
whitequarkwhich classes of attacks are you trying to prevent?11:34
e25805 buttons, each is 2bit, so you can have 4-128 digit passwords. but since the user data is salted, it doesnt matter much (in theory) what the length of user input is11:34
whitequarkit matters a lot11:35
e2580we are trying to defeat all kinds of attacks..11:35
whitequarkbecause if your input is short, and you can extract the salt from within the microcontroller11:35
whitequarkyou can trivially brute-force the key11:35
whitequarkall kinds of attacks? hah. what about the "connect standalone VBUS and just steal the device with the key in RAM" attack?11:36
e2580removing the data from the micro will not be as easy as this article describes.. i would like to see someone try it on this at32uc3a3256s. 11:36
whitequarkI would like to see an analysis of that attack channel, too.11:37
e2580however, i would suggest a long password just in case salt is some how obtained11:37
kyakbtw, i just bought a soldering iron, literally half an hour ago :) it this is used for attack, it will be very hard to prevent :)11:37
e2580the key is only in the device memory when it is unlocked11:39
e2580and then, only in secure memory11:39
e2580there isnt a ram dump to get the key from memoryon the MCU11:39
whitequarke2580: that is correct. but I don't see any hardware countermeasures against selective erasing mentioned in the datasheet11:40
whitequarkwhich means that no, the need to erase complete chip could be easily circumvented as it has been already shown11:40
whitequarkit _is_ circumvented even when the EEPROMs are covered by metal areas, and here they're probably bare.11:41
e2580there are 6 user accessable gpio on the device, you can add your own solutions11:41
e2580the countermeasure for a UV attack like you suggested, if it is even possible for this micro is already used. only the salt is store. so the key is still missing11:42
e2580you need the user entered password, and the salt to generate the aes key for encryption/decryption11:44
whitequarkyeah, I understand that11:44
whitequarkwhich hash do you use?11:45
e2580so unless the data is in an unlocked state, the aes key is not in the device.. if the device is unlocked, why do all the work for a UV attack, just read the data 11:45
whitequarkunless it's a hard-to-compute hash function like PBKDF2, a, for example, 64-bit user input could be trivially bruteforced by an array of FPGAs11:45
e2580sha256 will likely be used for the default hash, we will offer many firmware versions, and users can ad their own code.11:46
whitequarkI would suggest to use a hash function which isn't optimized for speed11:46
e2580we may chain several hash alg. to increase compute time11:47
whitequarkthere are well-known solutions for that problem already11:47
whitequarkgoogle 'bcrypt'11:48
whitequarkthe problem is, your microcontroller is severely underpowered. what takes a hour on it could be computed in a second on an fpga.11:48
qi-botThe build was successful: http://fidelio.qi-hardware.com/~xiangfu/build-nanonote/openwrt-xburst.minimal-20121026-0410 11:49
e2580the key only needs to be calculated on unlocking, so a few seconds delay to user is not a big deal. yes, i know this is less time of fpga or gpu cluster etc.. but still adds computational cost to brute force11:50
whitequarkexactly11:50
whitequarkso you need to estimate the bruteforce times and give an advice on the size of user input11:51
panda|ze2580: it's pretty interesting, AES256 key is stored on this device, so users need to maintain it in the safe place physically such as safe box11:52
e2580we do have methods to prevent hardware attack on the MCU, which will be released on the forum later. the user can make use of the gpio to add protections. we will not add these optional methods, due to praticallity, leagality or other reasons. but it is possible to prevent any type of mcu attack11:52
e2580aes256 key is not stored on the device...11:53
panda|ze2580: crackers might anonymously install some daemon like USB monitor on host side to try to sniff the data?11:53
e2580the host side data is plain text11:53
e2580plain text goes into and comes out of the device, it is encrypted on the fly in the MCU and written encrypted to the sd cards11:54
e2580you can monitor the data lines on the usb all you want, its plain text, so why bother if you have that access ?11:54
panda|ze2580: oh, I see11:55
panda|ze2580: sorry, I'm still wondering the use case :)11:55
e2580the device is for securing data at rest11:56
e2580stored data11:56
panda|ze2580: OK, so howto generate, dispatch and maintain the aes256 key?11:56
e2580dispatch ?11:57
e2580user enter password > salt > hash = aes key. then check the aes key is correct by reading a block of encrypted data and check for expected value. if value is ok, then aes key is ok, if not, key is bad11:58
e2580aes key is stored while in use only, in secure ram of the mcu11:59
panda|ze2580: ah, I used to think all the aes256 key are under control from you guys, just in case, if end user lost their crypt2x, then you guys can help them issue a new one as replacement12:00
e2580haha12:00
e2580no12:00
e2580if the key is lost, or if you wipe the salt ie by too many bad passkey attempts the data is lost12:00
panda|zno? so if end user lost their crypt2x or it's physically been destoryed by force, there is no way to recovery any crypted data on SD card?12:00
e2580there is no recovery12:01
e2580unless you can break aes256 then no12:01
panda|zem ... it is really a secure device ...12:01
panda|zso what's the risk for you guys hold user's aes key as a truested 3rd party?12:01
e2580the device goal is security over ease of use... so no recovery functions12:01
panda|zusers still hold their 5 key code, just like automobile manufacture for holding engine code12:02
e2580that adds unneeded risk to the security of the device, you cant validate our storage methods12:02
panda|zsorry, I'm totally a newbie with those naive questions, but trust me, I just ask cause those are not included in your current FAQ12:03
e2580do you trust us... companies like RSA and ssl cert issuers can be attacked, so why trust us ?12:03
e2580the goal of open source is to make the code openly available for review12:03
e2580if you give us the key, that means somewhere out there, there is a way to break your security12:04
panda|ze2580: but it's a two factor authentication, user still hold their 5 buttons password12:04
e2580ok, so you are asking if we can backup the salt ?12:05
panda|zoh, right, that's the point!12:05
e2580you can do some kinds of data back up12:06
e2580you can do a disk image of the sd cards. but you still need the cryptx2 to decrypt it, so you must use the same device, and user passkey12:06
panda|ze2580: oops, nono ...12:07
e2580or, you can unlock the cryptx2, and copy the files like you would on a normal hard drive and save them somepalce else12:07
panda|ze2580: what is the defination of 'salt', if user hold their password, they can re-generate the salt again?12:08
e2580salt is like a second part of the password12:09
e2580so to unlcok the device you need both, user entered passkey and the salt, to then make the aes key12:09
e2580the salt is stored in the MCU12:10
e2580randomly generated when you format the device12:10
e2580in the default firmware the salt is randomly generated, so you dont know what it is, and you cant restore it12:11
panda|ze2580: so user backup salt is possible and will be able to use the same password with salt to get the same aes256 key if they lost their crypt2x?12:11
e2580on alternate firmware we will allow users to create their own, so they can back it up then in case the cryptx2 needs to be replaced, they can gain access to the data on a new one12:11
e2580in the default firmware, you CANT backup the salt. that is part f the security12:12
kyakxiangfu: hi, just checking the latest minimal build.. i think you should keep the xorg.feed patches, otherwise it will fail all the time.12:12
e2580the salt is going to be more secure if it is randomly generated, VS user input which is much less likely to be high entropy12:13
panda|ze2580: well, I'm thinking of using this device in this way, plug one at home PC, another one at home, then bring SD cards with me everyday, so if I can clone with the same salt, then it's possible?12:13
panda|zsigh, maybe using crypt2x in this way will break some basic known security regulation:(12:13
e2580with the alernate firmware we will offer, you can do that. but not with the default firmware. you can change the firmware easily yourself12:13
panda|ze2580: cool, I see, thanks for your details, so how much for such a device?12:18
e2580target price is $65-75 USD with no sd cards12:21
panda|ze2580: wow! any possible to run it on kickstarter and make it much cheap with big volume?12:23
e2580we are doing kickstarter, within about 2 weeks12:26
e2580this is the kickstarter price12:26
panda|ze2580: cool! I will think about to back your project12:28
panda|zhopefully as a early bird:)12:28
e2580the website has an email notification list, we will send an advanced email to people on that list12:30
viricdamn autotools12:41
xiangfukyak, yes. applied. it will apply those 2 patch(xorg. and your openwrt.ticket12317.patch) to both minimal and full_system build.13:09
xiangfukyak, let's see how emacs build. 13:11
xiangfuthanks to David for fixing the emacs build error.13:11
viricdamn podofo. I can't build it for mips3213:12
viric/tmp/nix-build-r9q21gmnzjd324c7ygldpj53afx5mmvi-podofo-0.9.1-mipsel-unknown-linux.drv-1/podofo-0.9.1/src/base/PdfCompilerCompatPrivate.h:148:37: error: invalid 'asm': invalid use of '%w'13:12
viricI think it has some "htons" in asm13:13
xiangfuviric, some head files problem? like gcc parameters -I etc. (I just do a quick google :)13:15
viricno idea :)13:16
viricI've to dig.13:16
--- Sun Oct 28 201200:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!