#milkymist IRC log for Wednesday, 2011-02-23

wolfspraullekernel: excellent news about the Linux Magazine article!02:27
wolfspraulis there some way to download this issue as a PDF?02:28
wolfspraulah, I just see it was published last year :-) I confused 2010 and 2011 :-)02:31
wolfspraullooks like it can be downloaded for 6.50 EUR02:31
kristianpaulwolfspraul: last year indeed, you can get a pic version for free but is not very nice looking :/03:24
kristianpaulI guess somebody here already have the Magazine :-)03:24
adamw_kristianpaul, have you ever seen M1 shows " No boot medium found" under "BIOS> "03:42
adamw_kristianpaul, this msg actully is good for me, at least no boot anymore..ha... I'm testing my Sch. 3  experiments: http://en.qi-hardware.com/wiki/Milkymist_One_Power_On_Off_Sequence#The_schematics_of_three_experiments03:46
kristianpauladamw_: you got a bios pront i guess03:49
adamw_right. :-o03:49
kristianpaulare you entering in recue mode?03:51
kristianpaulIt still happing or just hapen once?03:51
adamw_that's not my goal. he.03:51
adamw_once...only..i'm reflashing image now.03:52
adamw_with  Sch .303:52
adamw_you don't need to modify your brd, this is just on mysite first.03:53
kristianpaulNo boot medium found means that BIOS cant find a "Aplication" to be booted from flash03:53
adamw_i don't want your brd get failure suddently.03:54
kristianpaulwait, you mean it just happen?03:54
adamw_yeah...i just want to know the background of s/w about this message's meaning.03:54
adamw_yeah...from my fast power cycling...(:03:55
kristianpaulwell as i said, bios cant find flicernoise from flash03:55
kristianpauli may be wrong, i'm just doing a quick read of BIOS source code ;-)03:55
adamw_don't try to do this on your site.03:55
kristianpaulahh power cycling ;)03:55
kristianpaulwho knows... :-)03:56
adamw_ok...good..just want someone can read the s/w codes to help me the meanings.:-)03:56
adamw_yeah! Don't do this.03:56
kristianpaulbefore " No boot medium found" there is a "I: Booting from flash..." message right?03:57
adamw_sure..but it was shown it quite previously.03:58
kristianpaulSo if all was okay and it just happen after the power cycling experiment, something get wrong with 0x00920000 memory addr at flash, thats from when flicernoise binary is saved03:58
kristianpaulYou should reflash if you want it back03:58
adamw_its previous log is : I: Attempting serial firmware loading03:59
kristianpaulyes, sure03:59
kristianpaulthats for flterm03:59
adamw_then "E: Timeout" -> "E: No boot medium found"03:59
adamw_yes, reflashing.03:59
kristianpaulso flash got corrupted... after power cycling?..03:59
kristianpaulbut BIOS still Okay04:00
adamw_after fast power cycling( power on -> off ->on , within 500~600ms)04:00
adamw_like I said , pls don't try to do this on your brd.04:01
kristianpaulOKAY !!04:01
adamw_yeah..i guess BIOS works well.04:01
kristianpaulSure, i dont want bother wolfspraul for a board replacement ;-)04:01
kristianpaulahh, will be nice if you could just dump the flash before re-flash, but thats late i think04:02
kristianpaulI imagine it can be usefull at least to check what got corrupted..04:03
kristianpauladamw_: how long this test took?04:05
adamw_yeah..late, also i don't know how to dump codes inside the flash under "BIOS>"04:05
kristianpaulthats is done with urjtag04:05
adamw_CH1 is DONE pin04:06
adamw_CH2 is PROGRAM_B04:06
kristianpauladamw_: I must leave, kind of sleepy here04:07
kristianpaulhope lekernel wakeup soon :-)04:07
kristianpaulread you later then04:08
adamw_get sleep first04:08
Fallenoukristianpaul: I tried the ftp server on flickernoise last night10:08
Fallenouit works with qemu at least10:08
FallenouI was able to login and to upload a file (in /dev/ lol)10:08
FallenouI used FileZilla ftp client :)10:09
Fallenou(power cycling kills the app on the flash ? :o)10:12
lekernelkristianpaul: -lgmp in cflags is a hack10:18
lekernelfix your findgmp.cmake instead10:19
lekerneladamw_: when it says "no boot medium" you have a flash CRC error before...10:29
lekernelalso, the SDRAM calibration mode is entered when PB1 is pushed at the same time as the middle button10:29
lekernelroh: are you at the agency today?10:31
adamw_lekernel, ok. btw, I tested 200ms delay & diode schematic.10:35
adamw_just temporarily clean up a bit on http://en.qi-hardware.com/wiki/Milkymist_One_Power_On_Off_Sequence10:36
adamw_but still under investigating.10:37
adamw_no further action I can do now..just fyi then if u have new ideas..let me know..10:38
Fallenoulekernel: do you have a spare M1 RC1 for ethernet debugging ?10:38
kristianpaullekernel:  show me your findgmp.cmake please i need confirm something10:46
kristianpaulFallenou: hey, (ftp) good for upload, what about for download using ftp?10:49
Fallenouwhich way ?10:49
kristianpaulrtems have a ftp10:49
kristianpaulyou can upload, i can too :-)10:49
kristianpauli cant download files even of 1Mb in size10:50
Fallenouso I have to put a file on my sdcard10:50
kristianpauldownload = login ftp then get10:50
Fallenouand then try to download it through ftp ?10:50
kristianpaulyesh !10:50
Fallenouok ok will try that :)10:50
kristianpaulplease try md5sum on it too10:50
Fallenouoh yes sure10:50
kristianpauli mean just in case :-)10:50
Fallenougood idea10:50
Fallenouam at work for the moment10:50
Fallenouwill keep you posted10:50
kristianpaulThanks !10:51
kristianpaullekernel: i put this file on https://code.google.com/p/libpbc/source/browse/trunk/FindGMP.cmake in /usr/share/cmake-2.8/Modules (yes i know i can be else where)10:56
kristianpaulmy diff is now http://paste.debian.net/108604/10:57
kristianpaulI got this after make http://paste.debian.net/108606/10:57
lekernelFallenou: there are lots of spare MM1 waiting for you at http://www.milkymist.org/buy.html ;)11:01
kristianpaulYou are at work ! thats excelent to source a buy :-)11:02
kristianpaullekernel: Can you please show me your FindGMP.cmake or equivalent?11:03
lekernelbut please don't ask me anymore about simple llhdl/cmake problems11:08
lekernelit's experimental stuff atm and I don't want to spend time on fixing your library issues... there are way more important trouble ahead11:08
kristianpaulYou wrote the code. I jsut telling hey i dint build well on debian. but OK11:09
kristianpaulgotta go, bye11:09
lekernelwell, in that case it seems it's rather cmake that doesn't work11:09
Fallenoulekernel: I was more thinking about borrowing one :)11:19
larscmwalle: how is lm32 uclibc comming along? can you send me your patches?11:20
lekernelkristianpaul: if you have questions about how to implement best that or that llhdl feature into FPGA logic cells, or what's an appropriate abstraction level in the llhdl code for some structure, those are very welcome. but i'm sorry, trivial questions like how to get the program to link against libgmp are inappropriate at the time being11:29
kristianpaullekernel: (implement FPGA logic cells) actually that will be my next question about, but first i needed at least get hhdl to compile on my side, the question will be about adding spartan-3 support, i know i must read and find _my_self_ the information until seriously bother you ;-)14:05
lekernelif you start supporting different xilinx fpga's, a lot of code can be shared with the current spartan6 support14:12
lekernelit'll probably be a good idea to but that into a semi-generic "libxilinx"14:13
wolfspraullekernel: hi there. I'm not sure you and Adam understood each other, maybe I'm just the clueless third.14:36
wolfspraulI think what Adam wanted to say is that the board he used to verify the diode fix acted very strange, and is now in a completely (for adam) unclear state.14:36
wolfspraulif I understand this correctly, after about 9-10 cycle tests he first got the CRC error14:37
lekernelhm, maybe the root cause is a DRAM problem14:38
lekerneland hi btw :)14:38
wolfspraulthen he reflashed, and it seems to have work again (?), but later on entered the 'unbooted' mode very often, and finally stops working altogether with PROGRAM_B not coming up at all anymore (fig. 16)14:38
lekernela DRAM issue could cause both the CRC error and failed memory test symptoms14:38
wolfspraulwell, the reason I am curiously following this is that we still have those 2 boards that I 'broke' in the initial round of testing14:39
lekernelat the end it does not boot _at all_?14:39
wolfspraulyes, see Fig. 1614:39
wolfspraullast picture14:39
lekernelurgh... imo that board has some other problem14:39
wolfspraulAdam should definitely speak himself, unfortunately he's already out now so we continue tomorrow14:39
wolfspraulwe have already permanently damaged 2 other boards, _ONLY_ through power cycle testig14:39
wolfspraulso maybe there are more bugs here, even if the diode fix per se is a good fix14:40
lekernelI don't know. it has never happened to me14:40
wolfspraulalso, the fact that he got the crc error and unbooted problems already after 9-10 cycles with the diode fix is strange to me14:40
lekerneland it might be a good idea to find out what is broken on those two boards14:40
lekernelmaybe it's just a DRAM solder problem14:41
wolfsprauljust not so easy14:41
wolfsprauland the dram solder problem emerges from power cycle testing?14:41
lekernelit can emerge at any time, and according to Murphy's law, during power cycle testing14:41
wolfspraulthat will mean PROGRAM_B will not come up anymore?14:41
lekernelno, definitely not14:42
lekernelwell, it's not PROGRAM_B btw, but DONE14:42
lekernelthat probably indicates a broken/corrupt flash14:42
wolfspraulthe reason we are not looking into the old two, fully unbootable (and shelfed) boards right now is that back then we did a lot of testing with a lot of boards14:42
wolfspraulso our actions are not very tracable14:42
lekernelthis happened to me once, and went away after reflashing14:42
wolfspraulthis time it's much better/much more controlled14:42
wolfspraulso I'm sure Adam will continue to work with this one board that has the diode fix14:43
wolfspraulyou think he should just reflash and it will boot again?14:43
lekerneldump the flash first14:43
lekernelso we can see exactly what has been corrupted14:43
wolfspraulsorry yes DONE. so if DONE doesn't come up you can still flash?14:43
lekernelthen reflash14:44
lekernelyes, with fjmem14:44
lekernelor the xilinx tool14:44
lekernelDONE does not come up on un-flashed boards btw14:44
wolfsprauldoes ADam know how to dump the flash on this board? also with xilinx tool I would think?14:44
lekernelI don't know how to do it, but it's definitely technically possible14:45
lekernelmaybe impact or urjtag have an option for that, maybe that needs to be coded14:45
wolfspraulthat sounds like Adam doesn't know.14:45
lekernelmaybe it's as simple as clicking an impact menu14:46
lekernelnever did it, I dont know14:46
wolfspraulso power cycling corrupts the flash?14:47
lekernelapparently, this happens sometimes14:47
wolfspraulwas the diode fix meant to fix that?14:48
wolfspraul'cannot boot once' is different from 'flash got corrupted for good'14:48
lekernelurjtag has a "readmem" command14:48
lekernelthe diode was meant to fix both problems14:48
kristianpaulurjtag have the readmem command, you just specify address, lenght and filename14:49
lekernelat least from my understanding of them14:49
wolfspraullekernel: do you have any theory what may leave a written corruption in the flash chip?14:49
lekernelso first, take that non booting board and read back the flash so we can test the "corrupted data" theory14:50
lekernelyes, I explained it already14:50
lekernelduring power up, there might be an invalid signal on the write enable pin14:50
lekernelwhich could cause write commands to be issued that would corrupt the flash's contents14:51
lekernelthe diode was supposed to fix this by asserting the flash reset low (and disabling it) until other signal levels are stable14:51
wolfspraulroh: good morning!14:53
lekernelgood afternoon :)14:53
wolfspraulbtw, I didn't realize that the 'dhl package' we sent was actually a regular mail package14:53
wolfsprauloh well14:53
wolfspraulso we are at the mercy of Taiwan Post now to deliver it14:53
lekernelin germany, dhl = regular mail, no?14:53
rohlekernel: ack.14:54
rohdhl is what german post was before14:54
lekernelit's a bit messy, I sometimes get 2 or 3 different postmen every day14:54
wolfspraulthe good news is that tracking says it's in Taiwan already14:54
wolfspraulit's moving14:54
wolfspraullekernel: ok it sounds like more work for Adam.14:55
lekernelbetween PIN AG, DHL, German post, ... :)14:55
rohlekernel: only 3?14:56
lekernelwolfspraul: I thought I explained that to him already14:56
rohthere is also gls, dpd, fedex, ups and some more (only packages) .. for letters there are only a few besides the 'post'14:56
lekernelbut if the diode+reset IC make no improvement, it's probably another problem :(14:57
wolfspraullekernel: you mean you explained the 'invalid signal on write pin'?14:58
wolfspraulif you did then I'm sure that's clear. I am just following.14:58
wolfspraulyes, there may be another problem14:58
wolfspraul'no improvement' may be too hash14:59
wolfspraullet's see what Adam says about the latest status14:59
lekernelthe invalid signal on write pin could also cause intermittent no-boot, because even if no data write happens, the flash goes to another mode than "read array" when it receives a command14:59
lekerneland the fpga would receive flash status information instead of bitstream, DONE would not go high, and the board would not boot15:00
wolfspraulif we know that the diode fix itself makes the board meet stated IC requirements, and is proven to work and cause no side-effects, then that's a good thing in itself15:00
lekernelbtw all DONE means is "fpga has successfuly read a bitstream"15:01
wolfspraulI mean "diode+reset ic"15:01
wolfsprauldo you see anything right now that would make you want to not add the diode+reset ic fix?15:02
lekernelno, I don't15:03
wolfspraulok, good15:05
rohwolfspraul: yay (package)15:21
rohwolfspraul: what would have been 'not regular mail' ?15:21
wolfspraulroh: DHL doesn't offer that to Taiwan15:43
wolfspraulI mean the 'other' DHL15:43
wolfspraulit's confusing.15:43
rohi still hope the taiwanese people can ready my address sticker and my handwriting16:00
rohwas difficult to squash the address into the 3 lines16:00
lekernelargh, a BGA socket for the spartan6 is $56618:53
lekernelI expect to fry a few chips during bitstream reverse engineering...18:54
lekernelbut at this rate, it seems cheaper to reflow the PCB, unless I really destroy a lot of FPGAs...18:56
lekernelor even make disposable breakout boards. well, let's see what the rate of chip destruction is...19:00
lekernelI've read in some paper that Altera FPGA chips need ~20 internal short circuits to blow up19:01
wpwraklekernel: interesting metric ;-) can't you just current-limit ?20:44
lekerneldon't know... maybe not so easy, the on-chip wires are thin and perhaps easily melted20:44
wpwrakwell, i guess before too long, you'll be an expert in just how much it takes to reach that melting point :)20:46
kristianpaulhe can use some cooling20:50
lekernelhaha, yeah, put the whole thing in liquid nitrogen so the fpga doesn't get damaged so easily ;)20:53
rohhm.. i'd rather use a heatsink and temperature monitoring and fast shutdown20:54
rohas in .. monitor temp AND current. if something spikes which shouldnt, power off.20:54
lekernelthe problem there might be very localized heating20:54
rohsure. but something like 20mA or so should be noticeable nonetheless20:55
lekerneland say, 20 milliamps can be a lot for a 100nm-wide wire20:55
rohwe'll see.. if you killed the first few we know20:55
rohneed to go out into the freezer outdoors hunt food. agency later20:57
kristianpaulhmm, you'll end damaging a lot of fpga's indeed21:03
rohi dont think its bad to kill a bit of hw to get to a goal21:04
wpwrakroh: you would make a good general ;-)21:14
wpwrak100 nm ... 16 uA continuous, according to http://www.geocities.com/CapeCanaveral/Lab/9643/awg.htm21:20
wpwrakbut that's of course with poor heat transfer and probably a low temperature difference21:21
lekernelany idea about what kind of CRC a 24-bit value could be?22:42
lekernelactually a 32-bit value, but the 8 MSB are always 0 and I suspect they're padding22:43
Fallenoulekernel: I have a IEEE link if you want :p22:57
Fallenouthere are two CRC-24 polynomes on the wikipedia page22:58
Fallenoumaybe you can try if they match22:58
lekernelyeah, maybe it's one of them23:04
FallenouThe first two conditions listed above persist for more than 400 cumulative hours at 85C Tj or more than 2,000 cumulative hours at 60C Tj.23:57
--- Thu Feb 24 201100:00

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