**** BEGIN LOGGING AT Wed Sep 12 02:59:59 2012 Sep 12 09:44:34 got a well running preempt on gta02 ... had to replace or mangle calls to gpio that spinlocked (ie as fiq is in interrupt context) Sep 12 09:49:11 I believe one does not need the fiq code I made for to get a working preempt , only those gpio calls changes (ie the only bonus of my changed fiq is that one get a backtrace of where it breaks Sep 12 09:55:03 one of those seems to be a bug ... ie the s3c handler for gio_set_value add spinlocks (when the kernel doc tells gpio_set/gat_value : "using these calls for GPIOs that can't safely be accessed without sleeping (see below) is an error" and "not to need spinlocks" Sep 12 10:07:06 http://prahal.homelinux.net/~prahal/gta02/linux-openmoko-preempt.diff.txt Sep 12 10:11:54 http://prahal.homelinux.net/~prahal/gta02/linux-openmoko-preempt_debugconfig.txt Sep 12 10:21:36 hi GNUtoo Sep 12 10:22:14 on gta02 I sorted out the preempt issue with kernel. back on enjoy . Sep 12 10:24:27 I will try to sort out why Xorg get a nice "-1" while its applications get "1" Sep 12 10:28:08 ok Sep 12 10:30:06 is all that merged already? Sep 12 10:30:41 no I bet I should send the patch to the tracker first :) I just sent the link to the patch there 10 min ago Sep 12 10:30:51 ok Sep 12 10:31:45 in short we use gpio calls that spinlocks into the fiq handler (interrupt context) Sep 12 10:32:59 which triggers a warning ... which printk ... which fiq cannot process (I wonder why it even worked without preempt ... probably spinlock are dumb without preempt) Sep 12 10:34:53 spinlock can change with reguard to preemt, yes Sep 12 10:35:10 or something like that Sep 12 10:35:23 for instance with real time stuff some locking stuff changed Sep 12 10:35:30 I don't remember the details tough Sep 12 10:56:39 I would like to split and make git format the patch before submit Sep 12 10:58:20 I have to git clone the repo ... or is there a mean to do that from the shr chroot ? Sep 12 11:00:45 I believe I found . This is a snapshot in shr chroot , so no go. I will clone the repo Sep 12 11:42:21 should I track branch "om-gta02/2.6.39/X" where X is om-stable or master ? Sep 12 11:42:40 prahal2: read wiki Sep 12 11:43:04 gee thanks and sorry Sep 12 11:43:36 no problem :) Sep 12 11:43:58 I meant this one https://github.com/shr-distribution/linux/wiki/branches Sep 12 11:45:09 :) was looking in vain into the shr-project wiki Sep 12 11:46:07 I figured out that you probably haven't seen that email about moving repos to github so you'll be in shr wiki :) Sep 12 11:54:18 well the fact is I did as you already told me the link to this email ... but even with this wiki page of github ... looking at the history of both branches helped : om-stable is with "openmoko.patch" and master is "openmoko.patch + shr.patch" Sep 12 11:55:28 yes, but vice versa... openmoko.patch is created from staboe..om-stable and shr.patch from om-stable..master Sep 12 11:57:30 btw ... now that I saw the om-gta02/3.5/master exists ... I might do a crazy thing . Try it :) Sep 12 11:57:49 I'm using 3.2 for quite long time Sep 12 11:58:07 but it's true I also don't use enjoy or even call with that :) Sep 12 11:59:46 ~spell auxillary Sep 12 11:59:48 possible spellings for auxillary: axillary auxiliary Sep 12 12:40:11 prahal2, do you have a debug board? Sep 12 12:46:11 GNUtoo: no :/ Sep 12 12:48:13 hmmm Sep 12 12:48:22 yes that might be tricky ... I did a local port of a new kernel for a101 gen8 without even a serial console . took ages Sep 12 12:48:42 prahal2, maybe we could do that : Sep 12 12:49:09 or you request a debug board, or we devide the tasks and I get back to the kernel work for gta02 and you do the rest? Sep 12 12:49:43 I could request a boad . What is the rest ? shr stack ? Sep 12 12:49:49 i have some debug boards left Sep 12 12:50:11 I live in France Sep 12 12:56:57 prahal2, test stuff and see if it's broken Sep 12 12:57:16 prahal2, ok, I'm currently in France too Sep 12 13:01:59 gosh you are the guy next door that play music so loud ! Sep 12 13:03:21 I found a debug board for sale via h:d , I believe they are in France too so it should not take too long Sep 12 13:36:24 lindi-: are you still interested into getting that fiq assembly changes from scratch ? I start to wonder if there is really a copyright issue (after all using variables to store stack and registries might not be so peculiar to the sample I took it from Sep 12 13:42:41 prahal2, what's all that FIQ stuff you're doing? Sep 12 13:42:53 there is a C FIQ version that I ported forward Sep 12 13:43:09 I also know more or less what FIQ is Sep 12 13:43:20 it's like interrupts but more privileged Sep 12 13:43:30 which leads to the FIQ debugger Sep 12 13:43:42 which can debug stuff when the phone has IRQ issues Sep 12 13:43:58 which we don't have on the freerunner (it's for some android devices) Sep 12 13:44:06 but there is a patch for mainline.... Sep 12 13:44:14 for adding FIQ to kgdb Sep 12 14:32:22 GNUtoo: I modified the way the c fiq version load the c fiq handler Sep 12 14:33:28 as a side effect I found out that when this new version is used , I get a backtrace of the c fiq handler when a warn slowpath is triggered Sep 12 14:34:01 ok Sep 12 14:34:08 I was doing all this because the c fiq handler was crashing with preempt enabled kernel Sep 12 14:35:47 it turns out that with the side effect backtrace I could find what triggered the printk into the fiq context (where printk is invalid) , ie spinlocks+preempt are scheduling and gpio calls used into the c fiq handler hdq_fiq_handler do have spinlocks Sep 12 14:36:18 so the main issue is the gpio calls . The c fiq code fix is added bonus Sep 12 14:37:52 this is a small single function thing that replace __jump_to_isr and __actual_isr Sep 12 14:38:41 but to do it I used a trick from stackoverflow sample and I am uneasy about the policy for snippets and copyright Sep 12 14:39:34 debug board ordered (I showed the slow delivery as I will be away for a few days) Sep 12 14:39:48 no "git gui" command in shr-chroot :/ Sep 12 14:40:16 I am so accustomed to this tool to split patches ... Sep 12 14:46:52 nvm found how to split line by line ... again on stackov Sep 12 15:13:57 prahal2, hmmm linux-next still fails the same way..... Sep 12 15:14:02 *current linux-next Sep 12 15:14:09 that is to say: ultra slow boot Sep 12 15:14:17 in userspace.... Sep 12 15:18:42 master-next ? Sep 12 15:23:34 xserver-nodm.service should be symlinked to "/etc/systemd/system/display-manager.service" not installed as a wantedby=multi-user.target Sep 12 15:24:17 hm... systemd is used in openmoko? awesome. :)) Sep 12 15:24:29 ie [Install]\nAlias=display-manager.service Sep 12 15:25:10 TAsn: troll land :) Sep 12 15:25:17 ? Sep 12 15:26:07 I think it's my first non-trolling message in IRC this week. Sep 12 15:29:35 JaMa|Off, hi Sep 12 15:30:04 my boot is way too slow, I've 2 solutions Sep 12 15:30:27 1) debug it on kernel space, that requires bisecting in an unbisectable zone I guess Sep 12 15:30:45 2) bootchart it Sep 12 15:30:51 how can I do 2) Sep 12 15:35:19 TAsn: I have only suggested you opened the pandorabox of all current trolls in linux area .) Sep 12 15:35:58 not that you intervene directly Sep 12 15:53:31 http://shr-project.org/trac/ticket/2065 <- the fix for preempt kernel Sep 12 15:54:40 GNUtoo: hackable1 had a bootchart-lite tool . Sep 12 15:57:38 ok Sep 12 15:57:42 busybox has one Sep 12 15:57:47 systemd maybe has something Sep 12 15:57:49 etc... Sep 12 15:57:52 we have the tools Sep 12 15:57:55 but I wonder how to use it Sep 12 15:58:03 can it work if the boot never finishes? Sep 12 15:58:21 because for now it's still not booted Sep 12 16:06:53 prahal2, ah. :) Sep 12 16:18:23 GNUtoo: I will try ... from a first look at http://code.google.com/p/bootchart-lite/source/browse/trunk/src/bootchart-lite.c ... one should be missing the header file if interrupting the boot . http://blog.hackable1.org/2009/09/study-boot-process-to-improve-boot-time.html is usage Sep 12 16:19:49 I will modify the above source file to write the header first and it should do (well afterwards we will have incomplete write thus I test to see how bad the bootchart-view desktop program handles incomplete writes Sep 12 16:20:05 since ok Sep 12 16:44:47 JaMa|Off: http://lists.linuxtogo.org/pipermail/openembedded-devel/2010-April/019299.html <- you used a modified qi to pass init= . Though the link is dead Sep 12 16:49:58 oh I bet I use an old qi , sorry Sep 12 17:01:17 ok one has to use bitbake qi-bootchart and get a qi which boot /usr/sbin/bootchartd (which is build via "bootchart" recipes done while making the qi-bootchart Sep 12 17:01:33 but I looked at "bootchart" code and it only write at the very end Sep 12 17:03:23 so I suggest a mix of both (ie mv bootchart-lite from /usr/bin/bootchart-lite to /usr/sbin/bootchard) and use qi-bootchart Sep 12 17:03:53 ie one need to make /etc/bootchart-lite to get the samples written (this is the destination dir Sep 12 17:07:29 though I wonder ... is there no better way to get qi to process init= than to flash it back and forth ? Sep 12 17:08:46 wait there might be if one tweak bootchart-lite sources Sep 12 17:14:49 prahal2, I think QI is in some git Sep 12 17:14:51 let me look Sep 12 17:17:54 in fact there is a recipe in shr to make qi-bootchart ... only I wonder if it is worth the trouble Sep 12 17:20:23 so I changed bootchart-lite to call /lib/systemd/systemd instead of /sbin/init , and replace /sbin/init link to /lib/systemd/systemd by a link to /usr/bin/bootchart-lite Sep 12 17:56:31 nope no good if broken before end ... I get a proper bootchart if so , else I get no files . Sep 12 17:56:53 by proper ... It is somewhat lite compared to a plain bootchartd output Sep 12 17:57:46 so I bet best option on shr is systemd-analyze . Might even be able to script it to output at steps in the boot process Sep 12 17:58:22 via new services files inserted at critical points in the boot Sep 12 18:14:57 JaMa|Off, hi Sep 12 20:47:43 <[Rui]> radekp, ping are you there? Sep 12 21:05:55 morphis: are you still online ? Sep 12 23:16:29 https://launchpadlibrarian.net/110061689/buildlog_ubuntu-quantal-armel.qi_20111113-1_FAILEDTOBUILD.txt.gz Sep 12 23:16:42 in case somebody has extra time to figure out this one :) Sep 13 00:09:02 lindi-: a debian post suggest that this means that libc6-dev is not installed or its version does not match libc6 Sep 13 00:10:04 all I see about libc6 is : Toolchain package versions: libc6-dev_2.15-0ubuntu15 . nothing about libc6 per se , so hard to tell if they match Sep 13 00:11:29 might well be a breakage in the toolchain setup, I bet it is not emdebian. linaro ? Sep 13 00:12:19 ie NULL not defined is quite telling Sep 13 00:14:09 well I now guess my clue was wrong (ie there a re like 10 files that compile fine beforehand ... Sep 13 00:21:38 gotcha: https://github.com/shr-distribution/meta-smartphone/blob/shr/meta-openmoko/recipes-bsp/qi/files/0001-start_qi-add-include-for-stddef.h-with-gcc-4.7.1-it-.patch Sep 13 00:22:14 lindi-: ^ **** ENDING LOGGING AT Thu Sep 13 03:00:00 2012