#milkymist IRC log for Thursday, 2011-09-15

qi-botThe Openwrt build was successfull, see images here: http://fidelio.qi-hardware.com/~xiangfu/build-milkymist/milkymist-openwrt.minimal-09152011-0240/01:35
cdehi there01:40
wolfspraulcde: hi!01:40
wolfspraulso basically say nothing works, right?01:40
wolfspraulyour jtag-serial board doesn't show up at all over usb, your m1 doesn't boot01:41
wolfspraulat least we start from a place where things can only improve :-)01:41
cdewell leds blink, so that's at least something ;)01:41
wolfspraulbut what now? I don't know01:41
wolfspraulthe unit was 100% tested before shipping out, very extensively actually01:41
wolfspraulbut now, hmm, don't know01:41
cdeyes, I believe you01:42
wolfspraulleds blink is not enough01:42
wolfspraulwhy does the jtag-serial not register over usb?01:42
wolfspraultotally nothing coming in over usb?01:42
wolfspraulhow is that possible01:42
cdeI dont think it was damaged in transit01:42
wolfspraulwe never had that, ever01:42
cdenope, nothing :(01:42
wolfspraulthe USB power works I guess (led on jtag-serial comes on?), but the ftdi chip doesn't start sending stuff over usb?01:42
GitHub163[scripts] xiangfu pushed 1 new commit to master: http://git.io/K633PA01:43
GitHub163[scripts/master] scripts: ignore build when no new commit - Xiangfu Liu01:43
cdewell Linux detects no device at all01:43
wolfspraul_anything_ in dmesg?01:43
wolfspraulI'm at a loss01:43
wolfspraulwe never had a case like this01:44
wolfspraulsince the first prototype of the jtag-serial boards01:44
cdeI'm thinking ESD01:44
wolfspraulyour m1 seems also dead?01:44
cdepossibly, but I really have no tools to diagnose it01:44
xiangfucde, 1. replug the jtag-serial to your pc. 2. check 'dmesg'01:45
wolfspraulsomething pretty drastic must have happened somewhere01:45
wolfspraulxiangfu: nah, he checked a lot already01:45
cdexiangfu: I did that countless times01:45
wolfspraulwhen you power-on your m1, which LED of the three leds on m1 comes on?01:45
wolfspraulI mean when you plug in the power supply01:45
cdethe one on the most left iirc01:46
wolfspraulwatch out to only plug the 5V supply into m1 :-)01:46
wolfspraulis it possible that you plugged in the 12V supply at some point?01:46
xiangfucde, left?01:46
wolfspraulthere is protection against that, but just curious/asking01:46
wolfspraulcde: ok, and then when you press the middle button, what happens?01:46
wolfspraulxiangfu: after m1 boots, what is the ethernet IP of m1 ?01:46
cdenope, was vey careful to use the M1 brick01:46
wolfspraulno worries it has protection against accidentally plugging in 12V, though you shouldn't try that just for fun...01:47
cdehmm, iirc middle led turns on01:47
wolfsprauland then after 30-40 seconds the third one comes on?01:47
cdethen after awhile right led01:47
xiangfuwolfspraul, it try to get from DHCP01:47
wolfspraulcde: but nothing on vga?01:47
wolfspraulwhat do you connect the vga to?01:48
cdeunless the tv somehow didn't see the signal01:48
wolfspraulthe tv has a vga in?01:48
wolfspraulyes that is (remotely) possible01:48
wolfspraulI've heard a story before (from kristianpaul) who said he plugged m1 into something (vga) but it didn't work (on that particular projector or whatever it was)01:49
cdeyep my brother uses it to plug in an media center01:49
wolfspraulxiangfu: if dhcp fails, what's the fallback IP?01:49
wolfspraulcde: and that worked?01:49
xiangfuwolfspraul, then no IP address01:49
wolfspraulthat was in France?01:49
wolfspraulxiangfu: it should fall back to a fixed ip01:49
wolfspraulfor diagnostics01:49
wolfspraulotherwise for example cde cannot just connect it to his hotebook now to try to log on. or can he?01:50
cdewell to be franck I didn't test my m! until today01:50
cde(my brother lives in NY)01:50
wolfspraulI can tell01:50
wolfspraulbut last minute is most fun :-)01:50
wolfspraullast minute is when we get all the important things done, and make the important realizations (like no static IP fallback, he he :-))01:51
cdewell it must be disappointing for lekernel01:51
wolfspraulcde: so you saw the vga output working?01:51
wolfspraulwhy disappointing?01:51
wolfsprauland what?01:51
wolfspraulyou just wrote 'my brother uses it to plug in an media center' - I don't understand01:51
cdewell since I had to do a working demo :)01:51
wolfspraulif the TV doesn't pickup the signal, first we should try one other monitor or projector01:52
cdehowever let me just try the M! on another screen01:52
wolfsprauldid you have the demo already?01:52
wolfspraulor you say the demo failed because m1 didn't work?01:52
cdeit's happening tomorrow at the OHS01:52
wolfspraulok do you have another screen to test vga now?01:53
wolfspraulI guess you cannot log in over Ethernet unless you find something that responds to DHCP01:53
wolfspraulxiangfu: ?01:53
wolfspraulhow can he do that?01:53
wolfspraulfrom the LED action, it looks like m1 is booting01:53
cdesure 10mn01:54
wolfspraulat least that :-)01:54
wolfspraul1) jtag-serial seems 'dead', quite strange01:54
xiangfuwolfspraul, (DHCP) that is true.01:54
wolfspraul2) Ethernet is inaccessible because no static IP fallback and no DHCP partner available right now01:55
wolfspraul3) vga-out does not work with the TV, looking for another screen01:55
xiangfucde, we can try to reflash your jtag-serial, only if you make sure it is dead for now01:55
cdehmm wait this network has dhcp01:55
wolfspraulbut then how do you know the address?01:55
cdeoh I can nmap the range01:56
xiangfucde, even m1 have success get ip address, you have to setup the Username and Password under Flickernoise.01:56
xiangfuwhich mean the VGA have  to works.01:56
wolfspraulxiangfu: please think about a way to get access over ethernet without vga or keyboard01:56
xiangfucde, there are FTP and Telnet service in m1.01:56
wolfspraulthere should be a static IP fallback, that's easy01:57
wolfsprauland there should be some way to setup username/password over ethernet? we need to think about it01:57
wolfspraulxiangfu: reflash jtag-serial is a good idea, but how? is it easy for cde now?01:57
wolfspraulI doubt this will help though, something bad must have happened01:57
wolfspraulreflash will only touch the eeprom with some config data, but his jtag-serial seems totally dead, like the ftdi chip not doing anything01:58
xiangfucde, which release you using?02:01
xiangfucde, have you ever reflash your m1 ?02:02
xiangfucde, the old image have static ip address.
cdeso vga just worked!02:03
cdeit was indeed the tv, I think the M!02:03
cde1 is fine02:03
wolfspraulxiangfu: he just received a brand new rc302:04
wolfspraulok, m1 works. that's good.02:04
wolfsprauljtag-serial, hmm02:04
cdeabout the ftdi, it's not a big deal I has a couple at home that I don't use02:04
xiangfucde, m1 is stable :-)02:04
wolfspraulcde: I will mail you a replacement, free of course02:04
wolfspraulbut that's not in time for the ohs02:05
wolfspraulas for returning the one you have, it's probably not worth it02:05
cdenp, at least the demo is showing02:05
wolfspraulyou can do a bit more investigation yourself if you like, that would tremendously help our project02:05
wolfspraulyou are at least as good or better than the people and resources we have, expect no magic on my end :-)02:05
cdehmm with my 10mhz scope I might not be of much help ;)02:05
xiangfuthose two wiki pages will help on reflash jtag-serial: http://www.milkymist.org/wiki/index.php?title=Build_the_libftdi-1.0_and_new_ftdi_eeprom and http://www.milkymist.org/wiki/index.php?title=Working_ftdi_eeprom02:05
wolfspraulit's up to you of course. most likely if you return it though, it will just be discarded because I don't have enough resources to do a deep analysis.02:06
wolfspraulit is a strange finding though, totally no response02:06
wolfspraullet's see what others say later02:06
wolfspraulxiangfu: I cannot imagine that reflashing the eeprom on jtag-serial helps at all02:06
wolfspraulthe ftdi chip is not doing anything it seems02:06
wolfspraulcde can try to short the eeprom to rule it out as a source of the problem02:07
cdesure, i'll return it just in case so you can have a look02:07
wolfspraulno need02:07
wolfspraulI have no resources02:07
wolfspraulso you just waste time and money returning it02:07
xiangfuwolfspraul, seems ftdi chip totally dead.02:07
wolfspraulcde could try to short the eeprom02:07
cdeok I'll try flashing the eeprom02:07
wolfspraulsure you can try :-)02:07
wolfspraulshorting the eeprom also, I just forgot how to do it02:07
wolfspraulsearching backlog...02:07
wolfspraulyou just have to short something on the little thing02:08
wolfspraulplease keep it disconnected from m1 for now02:08
xiangfuput the pin 1 of the eeprom to the ground02:08
wolfspraulxiangfu: thanks!02:08
wolfspraulcde: so you can try that. pin 1 of eeprom to ground, then connect usb02:08
wolfspraulbut please DISCONNECTED from m102:08
cdeyeah got it :)02:08
wolfspraulmaybe someone at ohs has a usb protocol analyzer02:09
wolfspraulyou could see whether anything goes over the wire02:09
wolfspraulat some point there is nothing but the ftdi chip in the way. if you indeed have the same one and can rework it - go ahead just for fun to see whether you can fix it :-)02:10
wolfspraulbut it's a qfn I think, right? a bit hard to rework02:10
wolfspraulI'll send you a new daughterboard for sure (unless it somehow comes back to life now)02:10
cdeyeah I'll ask around at ohs! thanks very much guys for the help02:10
wolfspraulwait, try short eeprom02:11
wolfspraulI'm curious02:11
wolfspraulwe never had a case like this, want to understand02:11
wolfspraulit's a simple board02:11
wolfspraulpower comes on - why is the ftdi chip unresponsive?02:11
wolfsprauland what happened after shipping it out? total mystery to me...02:11
cdeah don't have the equipment here :) will try it when back in france02:11
wolfspraulesd? looks like an easy excuse02:11
wolfspraulcan you short the eeprom now to try?02:12
wolfspraulfor the m1 demo, you can iterate over the patches by pressing the right button I think (or the left one)02:12
wolfspraulthat will switch to the next patch02:12
wolfspraulsome patches use the camera, so you may want to connect the camera as well02:12
cdeah good to know thanks :)02:13
xiangfuleft -- for previous patch, right -- for next patch02:13
cdeI need to go to sleep now (have to wake up early) I'll let you know how it goes!02:14
wolfspraulyes, you absolutely need a line-in music input02:14
wolfspraulotherwise no point in m102:14
wolfspraulthat's the #102:14
wolfspraulthen the second one is the camera, real-time video synthesis02:14
wolfspraulsure, go sleep02:15
wolfspraullet us know how it went02:15
wolfspraulwe'll fix the jtag-serial issue too, no worries...02:15
wolfspraulcde: thanks for the feedback!02:15
xiangfucde, good night02:16
wolfsprauland of course, we should also find out one day why the TV didn't pick up the signal02:19
wolfspraulthere is probably something that can still be improved on m1, if the tv accepts a vga input that should work...02:19
wolfspraulcde: which tv model is this? just the brand...02:19
rohwolfspraul: there is no too hard in rework. only too bad yield ;)02:35
qi-botThe Firmware build was successfull, see images here: http://fidelio.qi-hardware.com/~xiangfu/build-milkymist/milkymist-firmware-09152011-0336/03:12
kristianpaulfor the tv i had problem with my theory was that some resolution may no be supported, also at that time my laptop dint worked either :/04:59
wpwrakkristianpaul: never abandon a good theory just because it's wrong ;-)04:59
kristianpaulalso at that time my m1 want able to do higher resolution for gui05:02
kristianpauland i dont bother in take tc specs as it came froma  local company that asseble tv's and other home appliaces05:02
wpwrakah yes, that brand name inflation game. as if there weren't enough brands already in china ...05:04
qi-botThe Openwrt build was successfull, see images here: http://fidelio.qi-hardware.com/~xiangfu/build-milkymist/milkymist-openwrt.minimal-09152011-0611/05:06
kristianpauland after i upgrade the firmware for that date, i again have problems when coming from gui to render with a vide proyector.. oh well..05:07
kristianpaulbut was a baracamp so again i had no time to take specs .. and i also have to run for  a workshop05:07
qi-botThe Firmware build was successfull, see images here: http://fidelio.qi-hardware.com/~xiangfu/build-milkymist/milkymist-firmware-09152011-0706/06:42
wpwrakjoy ! finally got a standby corruption. took only some 32-34 hours. and they seem to come in clusters.07:01
wpwrakxiangfu: does the CRC check in the test software start at the first word (0xffff ...) or at some offset ?07:02
xiangfuno offset.07:03
xiangfustart at 0x00000000 in nor flash.07:05
wpwrakgrmbl. so NOR corruptions do indeed come in clusters. i see a lot in the first few words, plus always one deeper inside07:05
xiangfuwpwrak, what is that mean of  'come in clusters' ?07:08
wpwrakbtw, where is the source of the test software ? (that boot.4e53273.bin)  i don't see it in git://github.com/milkymist/milkymist07:08
wpwrak(clusters) seems that more than one NOR corruption happens within a short time span. maybe even within the same cycle.07:09
wpwrakthanks !07:09
xiangfuthe https://github.com/milkymist is a group. all milkymist stuff is under 'https://github.com/milkymist'07:10
xiangfuwpwrak, (clusters) ok. got it thanks.07:10
wpwrak(group) oh, nice. thanks !07:12
wpwraklet's see if these clusters may be related to temperature. i seem to get a bit more corruption in the middle of the night.07:15
wpwrakor maybe it's the effect of moonlight ...07:16
xiangfuwolfbug :)07:19
wpwrakwerebug :)07:20
xiangfuoh. yes07:20
xiangfuwpwrak, if it related to temperature. we will have big problem in winter, just the next few month07:22
wpwrakindeed :) and i have to hurry, because i only have a few weeks left to reproduce the bug. well, unless i relocate my m1 to the fridge :)07:26
lekernelre. those screen detection problems, perhaps I should use the exact 25.175MHz pixel clock instead of 25MHz07:47
lekernelI thought all screens were able to tolerate this (and read the same in many places), but apparently not07:47
wpwraki would expect at lest that much tolerance too07:48
wpwrakmaybe it's something else ? e.g., sync polarity or such ?07:48
lekernelI had a similar issue on another beamer - detecting some very high resolution like 2048x1024 and displaying only a distorted portion of the 640x480 picture07:49
lekernelit rather looks like a video timing problem07:49
lekerneldid cde try with another USB cable btw?07:51
wpwrakmaybe the sync pulses have the wrong timing ? and yes, vga timing is interesting. changes quite a lot between resolutions. and polarities do, too07:51
lekernelit's only using negative polarity07:52
wolfspraulhe said he tried several cables yes07:54
wolfspraulfor the tv problem, it could also simply be that the vga-in of that tv is broken, or that a certain channel had to be selected manually07:54
wolfspraulbut of course, if we can improve something on the m1 side, even better.07:54
wolfspraulI hope his demo goes fine, a little strange with the dead jtag-serial board.07:55
wpwrak((tag board) yeah, the kernel should say something07:57
wpwrakfor 640x480, negative hsync/vsync would indeed be correct07:59
wolfspraulbut the kernel says nothing, and since the board appears powered (led on), it can only be a plain failure of the ftdi chip08:00
wolfsprauleeprom shorting still to be done08:00
wolfsprauli'm wondering what triggered this, since it definitely worked just a few days ago before shipping08:01
wolfsprauloh well08:01
wolfspraulhardware is hard08:01
wpwraki kinda wonder ... i would expect the kernel to at least announce that it found a low/full-speed device08:01
wolfspraulnot a single line in dmesg08:01
wpwrakmaybe the usb connector came a little loose08:01
wolfspraulwell, let's see08:01
wolfspraulwe have to wait08:01
wolfspraulgood idea [push down usb connector]08:02
wpwrak(which could also be on the pc side. particularly those ether-double-usb combos are more than fragile)08:02
wpwrak(dmesg) a good test would be to connect a known to be good device, paste the dmesh. then remove the device and connect the jtag on that very same port. paste dmesg again.08:03
qi-botThe Openwrt build was successfull, see images here: http://fidelio.qi-hardware.com/~xiangfu/build-milkymist/milkymist-openwrt.minimal-09152011-0941/08:36
GitHub168[scripts] xiangfu force-pushed master from 2b3a2d9 to 282c36d: http://git.io/DOsw5Q08:40
GitHub168[scripts/master] scripts: ignore build when no new commit - Xiangfu Liu08:40
lekernelif there's a open or short circuit on one of the USB data lines (which can come from a fault cable or connector), it will produce this symptom08:45
wolfspraulsounds exotic08:45
wolfspraulhe will test some more when he's back in france, I hope08:46
qi-botThe Firmware build was successfull, see images here: http://fidelio.qi-hardware.com/~xiangfu/build-milkymist/milkymist-firmware-09152011-1036/10:12
wpwrakhmmm ... i think it should be fun to profile this compiler11:42
kristianpaulwhat this?11:48
wpwrakcompiler.c may spend quite some time doing string operations. that may explain some of the slowness. the things it calls look better at a first glance, though. profiling should show if the inefficiency of compiler.c really has a significant impact on overall performance or not.11:51
kristianpaulflickernoise i guess11:54
kristianpaulat least wpwrak mean llhdl perhaps?11:54
wpwrakthings like pfv_from_name11:56
wpwrakmilkymist/software/libfpvm/ uses re2c and lemon. i don't know them, but they're probably reasonably efficient (i mean, why else depart from lex and yacc)11:59
wpwrakwell, maybe ... i still see a lot of copying12:07
lekernelwpwrak, because parsers generated by lex and yacc only work on GNU systems ...12:07
lekernelwpwrak, C programming is about copying :)12:08
lekernel(if I understand "copying" as "code duplication and reinventing the wheel" ...)12:08
larscdamm! and i always though it was about reuse12:09
wpwrakyou must mean flex and bison. lex and yacc must have been considered old before rms was even born ;)12:09
wpwrak(duplicate) you must have spent too much time in china ;-)12:10
wpwrakof course, unix-style reuse is often in the form of  new_interface | old_engine   ;-)12:17
larsc"and every year we decided to add a new layer to hide the crap we've down the year before"12:18
wpwrakuntil the day comes when the archeologists roll up their sleeves and go spelunking :)12:22
qi-botThe Firmware build was successfull, see images here: http://fidelio.qi-hardware.com/~xiangfu/build-milkymist/milkymist-firmware-09152011-1312/12:48
mwallewpwrak: have you tried cooler spray and a hot air gun?20:38
GitHub169[scripts] sbourdeauducq pushed 3 new commits to master: http://git.io/_Zt8Kg20:45
GitHub169[scripts/master] Update makefile to use rtems cvs repository instead of deleted milkymist git clone. - JP Bonn20:45
GitHub169[scripts/master] Updated for latest build script and rtems cvs tree. - JP Bonn20:45
GitHub169[scripts/master] Merge pull request #2 from jpbonn/master - S├ębastien Bourdeauducq20:45
mwallelekernel: what is tmu early ack?20:47
lekernelan experiment that lets the SDRAM controller acknowledge the transfer a fixed number of cycles before the data gets actually transferred, to allow for more request pipelining efficiency20:49
lekernelit doesn't make any difference from the LM32 point of view20:51
lekernelit's just a tad bit faster20:51
mwallemh qemu head is broken (at least for lm32?)21:29
mwallelekernel: i guess that experiment is just for fun? or do you want to squeeze the last bit performance out of the soc already? :)21:40
mwallelekernel: are u still using tiling windowmanages?21:48
wpwrak(early ack) i guess that's for trying to get things to work at higher resolutions22:04
wpwrak(cold/hot air) hmm, so far, i'm not desperate enough for that :)22:05
--- Fri Sep 16 201100:00

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