#milkymist IRC log for Saturday, 2011-07-30

kristianpaulcool, the openwrt toolchain for lm32 seems to work02:20
wolfspraulkristianpaul: can you try to download this zip file?03:12
wolfspraulit's linked on the right side03:12
wolfspraulsupposedly it's easy, but I am unable to successfully login, it keeps complaining about my password even though I successfully reset it with their own server etc. stupid.03:13
kristianpaulYou do not have permission to access this content. You can choose to register to gain access to the protected content, or you can return to your originating page.03:14
kristianpaul(i already loged in with my xilinx user)03:14
kristianpaulillregister, let see03:15
kristianpauloh, i acept a doc about XILINX CONFIDENTIAL ;-)03:15
wolfspraulif you don't want to, don't do it03:16
kristianpaulgot it :-)03:16
wolfspraulthis zip file contains some source snippet that lekernel was considering using in the Milkymist SoC03:17
wolfspraulthe problem is that the source comes with restrictions that do only allow redistribution in binary format inside a binary bitstream03:17
wolfspraulso anybody who would want to build the Milkymist SoC from source would need to download the zip file from Xilinx first, which I am currently unable to do even03:17
wolfspraulso it looks like that would create quite a high barrier of entry for contributors, in reality I would think you loose 80% of people over a step like this :-)03:18
kristianpaulcheck your mail03:19
wolfspraulbut then Sebastien has to weigh the benefits of that code to Milkymist today...03:19
kristianpaulan not yet03:19
wolfspraulbest is if more Verilog hackers join and help on the hundreds of things that can be improved :-)03:19
Action: kristianpaul still need to learn more verilog..03:20
kristianpaulbut yes03:20
kristianpauli dont see bitstreams..  actually there is source code03:21
kristianpaulat least is all blobs embeded in .v files ;-)03:21
kristianpaulhaha, lekernel want to drop its hpdmc :-), what nasa will say now ..03:22
kristianpaul(kidding of course)03:23
kristianpaulnow check email wolfspraul03:23
wolfspraulwe were just discussing the implications of using a source like this in Milkymist03:23
wolfsprauleverything has pros and cons. Sebastien is the only one with deep overview of the Milkymist SoC architecture today and how it could evolve in the future, right now.03:24
kristianpaulsure,hes is the only one as always.. :)03:24
wolfspraulthis source snippet which Xilinx created to support its customers would clearly raise the barrier of entry for new contributors, but I guess it would also improve performance03:24
wolfspraulit would make us depend on proprietary Spartan-6 specific features03:25
kristianpaulyeah,it seems (performance)03:25
kristianpaulwell, take advantage is hardware is good03:26
wolfspraulbut then I do share Sebastien's belief that we have to make great functioning products today, and not just theoretically 'free' but crappy and uncompetitive stuff03:26
wolfspraulthe best would be as I said if more developers join and help lift the free parts up, no doubt.03:26
kristianpaulmay be you can ask to Northwest Logic, Inc, for let you distribute the DDR2 controoler03:29
wolfspraulunlikely that we ever get a response03:29
wolfspraulyou can email them too :-)03:30
wolfspraulit's not even because they are bad or anything, but simply because no other of their important customers ever has a request like that03:30
kristianpaulwell, i dont complain yet about ddr controller made by sebastien :-D03:30
wolfsprauland at the volume we are buying chips at, all of these companies would be out of business tomorrow, so they are wise to ignore our strange requests :-)03:31
kristianpauli agree that take advange of our hardware is the best03:31
kristianpaulsadlt the only way is using xilinx specific libraries for their propietary software,03:32
kristianpaullike then using specific intel instructions, etc...03:33
wolfspraulit will be a middle path for a long time03:33
kristianpaulso the only practical problem, distribution for profit03:34
wolfspraulcopyleft hardware is clear to me ;-) push 100% free software and tools to the point that all remaining parts can be sourced from 2 or more independent parties03:34
wolfspraulbut that's a long term goal03:34
kristianpaulyeah, sure03:34
kristianpaulthe concern now a fast and fool prove Milkymist One, i understand :)03:35
lekernelwe're not using the NWL DDR controller, just the PHY, i.e. some non-portable spartan6-specific design they made to tame the pesky I/O and clocking system of that particular FPGA08:45
lekernelyou can read the spec document of the PHY ...08:45
AlarmI can not find "sys/ioct1.h" ?09:29
rohwolfspraul: maybe we are looking at this the wrong way around.. why is xilinx again not sponsoring sebastien?09:41
roh ;)09:41
kristianpaul(only PHY) oh, nice :-)11:49
kristianpaulhe, yeah roh :)11:49
GitHub88[flickernoise] sbourdeauducq pushed 1 new commit to master: https://github.com/milkymist/flickernoise/commit/33091c15e2e8864b09e2e0b03899155a1fb5e39e13:01
GitHub88[flickernoise/master] Do not include curl/types.h - Sebastien Bourdeauducq13:01
lekernelomg... lm32 is so so broken in rtems head13:23
lekernelsemaphores are fucked up, interrupts are probably fucked up, termios is fucked up, ... upstreaming wtg ...not13:24
lekerneland of course, everything works in QEMU or when you add printk() to track down issues13:24
larsctime to switch to linux, i guess ;)13:40
lekernelhaha, don't tell me such things do not happen all the time with linux :)13:55
larsctook me almost a day to figure out why the rootfs could not be mounted anymore with 3.014:01
larscturns out there is a function called user_strnlen, which returns the number of characters in a string including the null byte. somebody accidentally changed it to not include the null byte14:02
larscand thus the last byte was always truncated when copying the rootfs path14:03
Alarmlekernel: I can not find "sys/ioct1.h" ?14:45
mwallelekernel: components arrived14:45
mwallei guess the middle pin of the ir receiver is GND :)14:46
lekernelmwalle, just drop it instead of the existing sensor14:46
lekernelit's the same P/N14:47
lekernelyou shouldn't need to care about pinout14:47
lekernelAlarm, it works for me (and everyone else). please give more information.14:47
mwallelekernel: na because the heat is sucked up by the plane :)14:48
lekernelalso, flickernoise does use this include file, and you said you were able to recompile it... so have a look at how it's done14:48
lekernele.g. check compiler command like14:48
lekernelalso, it's <sys/ioctl.h> and not "sys/ioct1.h" (l != 1; " != <)14:50
lekernelmwalle, does this look good to you? http://pastebin.com/5q8GmNq514:52
lekernelthe internal interrupt/context switch management API was changed in RTEMS and broken for LM32, which created all sort of problems (stack corruption etc.)14:53
lekernelI have copied the current Nios code and added _exception_stack_frame14:54
mwallelekernel: whats the difference between this and the old handling?14:55
mwalle#if( CPU_HAS_SOFTWARE_INTERRUPT_STACK == TRUE) << shouldnt this be always true?14:56
lekernelyes, it should be always true14:56
lekerneland I checked it is :)14:56
lekernelhttps://github.com/milkymist/rtems/blob/mmstaging/cpukit/score/cpu/lm32/irq.c <= current broken handling14:56
lekernelhttps://github.com/milkymist/rtems-old/blob/master/cpukit/score/cpu/lm32/irq.c <= old working handling14:57
lekernel_Context_Switch_necessary was renamed to _Thread_Dispatch_necessary14:57
lekernelthey removed _ISR_Signals_to_thread_executing (I do not understand what it does atm)14:58
lekernelthe new Nios code also calls _Thread_Dispatch_in_critical_section()14:58
lekernelthe pastebin code seems to work correctly ...15:01
lekernelat least much better than the other which froze the board some 20ms after boot15:01
lekerneli get weird behaviour after trying to render in flickernoise though...15:03
lekernelnot sure if it's related15:03
lekernelbut after I have run rendering it takes several seconds to input one character on the shell, the mouse cursor lags, etc.15:04
mwallemh there seems to be two differences, _Thread_Dispatch_{in,de}crement_disable_level is called before and after the stack switch15:08
mwallei guess its just a wrapper around the former _Thread_Dispatch_disable_level--;15:09
lekernelyeah, or should be...I saw a Changelog entry about that15:09
Alarmlm32-rtems4.11-gcc -O2 -mbarrel-shift-enabled -mmultiply-enabled -mdivide-enabled -msign-extend-enabled -I $RTEMS_MAKEFILE_PATH/lib/include  -B  $RTEMS_MAKEFILE_PATH/lib -specs bsp_specs -qrtems -o VGA VGA.c15:09
AlarmVGA.c:3:23: fatal error: sys/ioct1.h: No such file or directory15:09
lekernelAlarm, ioctL.h15:10
Alarmoh sorry15:10
mwalleand the context switch was changed15:11
mwallelast time i touched that code was 2 years ago :)15:11
lekernelI have committed the 'working better' code https://github.com/milkymist/rtems/commit/e0d4d55f693886b991c1b35b53091d005cea8cf515:17
kristianpaulthere is no current way to boot linux from flash right?15:19
lekernelkristianpaul, no but should be an easy exercise for you :-)15:19
kristianpaulgood !15:19
kristianpaulwtf Kernel panic - not syncing: VFS: Unable to mount root fs on unknown-block(1,0)15:22
kristianpaullast owrt-milkymist revision..15:22
kristianpaulah, thats what initrd for..15:23
GitHub192[rtems-yaffs2] sbourdeauducq pushed 1 new commit to master: https://github.com/milkymist/rtems-yaffs2/commit/875d6f89d85cc4a4c4dd6bf649b83230cf2db14415:31
GitHub192[rtems-yaffs2/master] added missing file - Sebastien Bourdeauducq15:31
GitHub199[rtems-yaffs2] sbourdeauducq pushed 1 new commit to master: https://github.com/milkymist/rtems-yaffs2/commit/3dc866a6f0c5f286ca81924d93c242747934b61715:35
GitHub199[rtems-yaffs2/master] Remove references to non-existing bits - Sebastien Bourdeauducq15:35
kristianpaulvery easy, at least a boot from mecard load the simpleImage.milkymist_one16:25
kristianpaulnow lets fill CMDLINE.TXT :-)16:27
kristianpaulno need boot from flash after all lekernel :916:31
kristianpaulah, you already diccused about it.. well wasnt so obvious too me16:36
kristianpaullarsc: rootfstype is ext2 even when in the target image said ext4?16:54
kristianpaulhum.. E: File size larger than the blocks read (corrupted FS or IO error ?)16:59
kristianpaulagrhrhrh i formated the flashmemory , write bios.bin and such again and now the bios cant  find it ....17:52
GitHub119[rtems-yaffs2] sbourdeauducq pushed 1 new commit to master: https://github.com/milkymist/rtems-yaffs2/commit/083cf978d030a6c841203bb5390a9952b952d07a18:21
GitHub119[rtems-yaffs2/master] Compile needed yaffs_summary - Sebastien Bourdeauducq18:21
GitHub0[rtems-yaffs2] sbourdeauducq pushed 1 new commit to master: https://github.com/milkymist/rtems-yaffs2/commit/3849a105e3fe4606704609d96a6025385598cdba18:23
GitHub0[rtems-yaffs2/master] Add memory management - Sebastien Bourdeauducq18:23
GitHub74[flickernoise] sbourdeauducq pushed 1 new commit to master: https://github.com/milkymist/flickernoise/commit/8e7aa92cceb96bd6e1b3356c66b4031da80a697118:39
GitHub74[flickernoise/master] Use new YAFFS API - Sebastien Bourdeauducq18:39
kristianpaularghgh ERROR:Place:1108 - A clock ..19:11
kristianpauldamn, there are not more clock deticated input in the mm1? at least router trought the exp connector..19:56
kristianpaulhum, may be i can still it from some TP, bad thing i can search text in the pdf... oh wel..20:02
kristianpaulhum where is B9..20:20
lekernelclocks in spartan 6 are a major source of trouble. i spent two weeks or so getting them to work for milkymist soc.22:22
lekernelthe tools have improved since then, however22:22
lekernelISE 11.x had a hefty dose of bugs that made any non-trivial clocking a complete nightmare22:23
mwallei hate that fucking toolchain23:24
wpwrakmwalle: mmh, toolchain build time ? or did you find some binutils/gcc/... bugs ?23:26
mwallei just want to rebuild it23:26
mwallenow everything is broken, lol23:26
kristianpaulopenwrt toolchain?23:48
mwalleno own one23:50
kristianpaulok, best wishes :-)23:58
--- Sun Jul 31 201100:00

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