=== Ursinha` is now known as Ursinha === Ursinha is now known as Guest7533 === Jack87|Away is now known as Jack87 === Guest7533 is now known as Ursula === rsalveti` is now known as rsalveti [08:50] ppisati, ddi you include the fix for bug 746137 in your last upload ? [08:50] Launchpad bug 746137 in linux-ti-omap4 "Page allocation failure on Pandaboard and Beagle XM" [High,Confirmed] https://launchpad.net/bugs/746137 [08:50] seems QA is seeing it recently on the bamboo-feeder http://pastebin.ubuntu.com/1212016/ [09:15] ogra_: AFAIK there's no fix [09:15] err, cooloney worked one out weeks agai [09:16] *ago [09:16] then i'm not aware of it [09:16] i know there's a systcl to workaround it [09:16] but IMO all the fixes proposed so far [09:16] are not working [09:16] GFP_ATOMIC vs GFP_KERNEL i think [09:16] cooloney: ^ [09:16] ogra_: i saw patches for Panda was merged by ppisati [09:17] the patch is a hack from Andy Green [09:17] oh, he merged it without knowing ? [09:17] haha [09:17] k, as long as its in ... [09:17] that's for panda, not for Beagle [09:17] ah [09:17] well, would be good to have it in beagle too [09:17] * ppisati is schizophrenic then [09:17] well guys [09:17] since we have no way to apply userspace hacks anymore like we did with preinstalled images [09:18] but if we still have that problem even with the supposed fix [09:18] ogra_: i tried a patch MingLei gave to me from upstream some maintainer. [09:18] it's clearly not working [09:18] ogra_: but it doesn't work. [09:18] ppisati, how would you know ? [09:18] your upload is in the archive but we have no new images with it yet [09:18] ogra_: but it's not there for sure [09:18] i mean [09:18] cooloney: which patch were you referring? [09:19] ogra_: in the last upload there's the video/unity flickering fix [09:19] hmpf, i had the impression that was long fixed [09:19] ogra_: and a rebase on origin/master (latest 3.5 stable) [09:19] ppisati: a patch not in public, it is for smsc95xx usb net driver [09:19] ah, if its driver specific that should work for beagle too then [09:19] ppisati: the patch from Andy is in your ti-omap4, so it should be workarounded [09:20] they use the same chip [09:20] but for beagle it is in master. [09:20] ogra_: iirc, conrad approached me weeks ago and told they have the same problem on builders [09:20] the hack will touch other flavors. [09:20] conrad ? [09:20] ogra_: adam [09:20] do you mean infinity ? [09:20] yes [09:20] infinity: ^^ [09:20] ah, conrad adam ... the evil twin of adam conrad *g* [09:21] :) [09:21] i dont think the builders run quantal, so thats not a surprise [09:22] and they dont use preinstalled images to set them up (which would get them the fix) but use ubuntu-core with manual config or some such [09:22] ogra_: yep [09:22] ogra_: what i meant was that we never really fixed it [09:22] s/get them the fix/get them the userspace fix/ [09:23] ogra_: and if we have a supposed workaround in [09:23] ogra_: but we still experience it [09:23] well, i lost the opportunity to work around it in userspace ... [09:23] ogra_: then we should a) investigate the problem b) back out the workaround [09:23] and i havent seen it in ages here in my tests [09:24] so i personally had the impression it was long fixed [09:24] ogra_: The buildds use d-i over PXE on precise, but they have the userspace hack in puppet. [09:24] until hggdh pinged me over night today [09:24] infinity: and do they suffer the problem? [09:24] ppisati: No. [09:24] uhm [09:24] and funnily the saem thing is discussed on #beagle today too [09:24] ppisati, the userspace hack always worked [09:25] ppisati: But any machine installing from an installer that doesn't put the userspace hack in place will have the issue. [09:25] the userspace stuff was the sysctl config, right? [09:25] Right. [09:25] right [09:25] worst case we should just bump it at build time for certain arches [09:25] in the kernel code ... if thats possible [09:26] so we would just move the userspace hack into the kernel [09:26] i wonder if can pass it via cmdline [09:26] well you can definitely just create the sysctl.d file ... its a one liner [09:27] not susre sysctl values are read from cmdline [09:27] right [09:27] If it should always start at a certain value, and it's not, passing it on the cmdline (or sysctl) is the wrong answer. [09:27] The initial value should be correct. [09:27] I think we've been over this a few dozen times. [09:27] but they are kernel variables [09:27] heh, yeah [09:27] ppisati: Yes, it's a variable, but variables should have sane defaults. This one clearly isn't sane for ARM. [09:30] vm.min_free_kbytes = 8192 [09:30] Anyhow, I should sleep. [09:30] uhm [09:30] If I recall, apw was deeply involved in the last round of this conversation. [09:30] ppisati: i suggest we instroduce a config option in master branch like CONFIG_VM_MIN_FREE_KBYTES [09:30] sleep well [09:30] And he doesn't seem to be in the channel. [09:30] So you might want to take this to #-kernel. [09:30] which should be 8192 by default for x86, powerpc [09:31] and 32768 for ARM [09:31] vm.min_free_kbytes = 32768 [09:31] on my Q/omap4 [09:31] and porting Andy's hacking patch to master [09:31] ppisati: yeah, it should be 32768 for ARM since it was forced to be that in mm/page_alloc.c [09:32] so we are "already "good" [09:32] ? [09:32] Oh, kay, so we HAVE worked around it in the kernel? [09:32] yeah, but beagle is omap3, which is master kernel [09:32] on omap4, but not on omap (which uses the same driver) [09:32] not on your branch, [09:32] Ahh, not in master. Check. [09:32] wait [09:32] we need put such similar patch in master [09:33] Couldn't you just wrap it in #ifdef __ARM__ and commit to master? [09:33] ogra_: didn't you say we hit this problem on omap4 tonight? [09:33] if you put the same patch in master, it will set it to 32768 for all the flavors [09:33] like x86 and powerpc [09:33] I guess highbank and armadaxp would pick it up too, then, but that's not world-ending. [09:33] ppisati, i was notified by hggdh in #ubuntu-bugs tonight with the pastebin i gave above ... havent talked to him yet [09:33] cooloney: Hence why I said wrap it in an ifdef. [09:33] ogra_: oh [09:33] cooloney: Though, obviously a CONFIG_ is better. [09:34] (since he is asleep like every sane person on that american continent atm) [09:34] so, even with the fix [09:34] we hit the problem [09:34] cooloney: And the CONFIG_ change would be upstreamable even. [09:34] infinity: yeah, that should be ok for just local hacking, heh [09:34] ppisati, well, lets see what exactly he tested there [09:34] ok [09:34] i need to talk to him first [09:34] infinity: i don't think we need such hacking for highbank and amadaxp [09:34] ok [09:34] wait a sec [09:34] in any case people in #beagle are seeing it and have discussed it today [09:35] because they don't use usb ethernet controller [09:35] cooloney: Right, hence why a CONFIG_ would be better. [09:35] qait wait [09:35] i just noticed one thing [09:35] only beagle [09:35] infinity: CONFIG_ can be set as different value for different flavor [09:35] ogra_: That pastebin is an ancient kernel. [09:36] tight [09:36] rigjt [09:36] ah [09:36] it's a 3.4.x [09:36] i have heard many reports that people on intel have seen it too btw [09:36] cooloney: Yes, I know. :P [09:36] we have the fix in 3.5 [09:36] it isnt arch specific but USB related [09:36] or just use #ifdef USB_NET SMSC95XX [09:36] like this kind of hacking [09:37] ppisati: what's kind of fix in 3.5? [09:44] 86b83747713545234c28e5347c6f0e5efb652332 [09:44] HACK force min_free_kbytes set by proc to min 32K to workaround smsc problems [09:44] Signed-off-by: Andy Green [09:45] actually even this one [09:45] 89b16e837180b215b667affdd4227220827f8bb4 [09:45] HACK force min_free_kbytes to 32K to workaround smsc problems [09:45] both are needed [09:45] the second one is a safe belt [09:45] the first one is the real one [09:53] ppisati, did you put the LP bug number for the flickering in the changelog ? (i didnt see it in teh pull request mail) [09:53] Laney, ^^^ [09:54] ppisati: i didn't see that patch in the real code. can you see that code in the mm/page_alloc.c? [09:54] ppisati: although i can git show that patches you posted [09:58] ogra_: i didn't know there was a bug opened for it [09:59] bug 1045491 [09:59] Launchpad bug 1045491 in linux-ti-omap4 "Moving mouse messes up the desktop" [High,Confirmed] https://launchpad.net/bugs/1045491 [09:59] ah ok [09:59] ok [09:59] i'll close it manually [10:00] well, let me take care [10:00] i'll close it if it has been tested with the images :) [10:02] (if i dont, kate will hunt me down anyway) [11:34] ogra_: where is the template for preEnv.txt? or, how do i make it look as i like? or how do i make flash-kernel NOT replace its content at every run? [11:35] ppisati, /etc/default/flash-kernel [11:35] ogra_: ack [12:09] As a part of a custom/proprietary first-boot system, I need to run a series of disk partitioning operations prior to mounting the real root. What would be preferred: a dash script with the required support tools (sfdisk, awk, grep, etc.) or to pull in python into initramfs? [12:10] sveinse, have a look at the jasper-initramfs code [12:11] ogra_, voila. Thanks [12:11] specifically at the jasper_growroot script [12:11] * sveinse discoveres that the wheel has already been invented [12:12] and obsoleted too :) === doko_ is now known as doko [12:12] I'm not there yet :P [12:39] Any of you people the author(s) of jasper-initramfs? [12:39] yes [12:39] Why do you reboot after resizing root? [12:39] any of me are :) [12:40] because i'm a coward :) [12:40] and want to make sure it actually worked before moving on with the install [12:40] its not actually necessary if you are sure its all fine [12:41] (i.e. if you dont have people with random SD cards as media) [12:41] its just that there any isn't more you can do with a reboot compared to just continue. But I get the point [12:42] Except that you know that the bootloader is still intact after altering the parttables [12:42] you do test if the UUID and partitioning is still right and if the bootloader still does the right thing on a fresh boot [12:43] yup [12:43] jasper_growroot also isnt the only script there ... jasper modifies other things too and sets them up [12:43] i also noticed a loop around resize2fs. Doesn't it do enough in one iteration? [12:44] (not that I've deciphered the sed and char logic in there) [12:44] thats for plymouth [12:45] ah. of course [12:45] resize2fs runs through several iterations, the loop just picks up the output and dumps it to plymouth or the text console [12:45] its not resize2fs itself thats gets looped over ... only the output [12:46] you mean resize2fs does NOT run in loop, yes, I see that now [12:46] right [12:46] the while sits after the pipe [12:48] and I also found some nice sysctl settings for having root on sd-cards. Are there any more goodies like this I should be aware of? [12:49] I need to start using that :p [12:50] I often get users of my transformer images complaining about it just stopping when resize2fs is running [12:51] sveinse, the zram-conf package is helpful if your kernel supports it (speeds up swapping a lot) [12:51] and here on my ac100 i just experiment with a journal-less ext4 ... thats classes faster than any FS i have tried yet [12:52] though with that you should force an fsck on every boot [12:53] ogra@horus:~$ cat /etc/security/limits.d/90-stack.conf [12:53] * soft stack 256 [12:53] thats also pretty helpful when being low on ram [12:54] what is required for zram anyway [12:54] (default is 8k per thread stack, makes your apps use less ram (and since its soft they can just ask for more if needed)) [12:54] lilstevie, that you enable it in your kernel config [12:54] I have no idea if I included the right bits [12:55] :p [12:55] i think there is only one option (or probably two) [12:55] config_cramfs? [12:56] I guess I should just try it lol [12:56] no, thats a filesystem [12:56] ah right [12:56] search for zram [12:57] (hit / in menuconfig) [12:57] yeah I know how to search :p [12:58] ok yeah CONFIG_ZRAM [12:58] and it is enabled [12:58] great [13:00] I have double the ram of the ac100, but IMO 1GB is far from enough [13:02] yeah, imho zram should become a default as staging swap [13:03] so that you first swap into compressed ram before actually writing to disk [13:03] yeah [13:03] I will have to give it a go, see how much of a performance increase it gets [13:05] you will only note it when swapping [13:05] so the moment I open chromium then :p [13:06] yeah, firefox uses lots less ram in its recent versions ... [13:26] What's the deal with CHS geometry on omap3 and bootloader? I haven't been able to find any other combination other than H=255, S=63 that works. But I see H=128, S=32 is used in jasper [13:26] so I guess that also works [13:26] the ROM needs to find the first stage loader at a certain point on a vfat to boot [13:27] (I have to admit I'm confused about CHS vs LBA and how they interact) [13:27] thats omap specific [13:28] i doubt you need CHS on tegra [13:30] ogra_, tegra scans until it finds the bct [13:30] right [13:30] from what I understand, there partition table contains both CHS and LBA fields. And Linux uses only LBA, right. [13:31] So given that you're able to boot, when sfdisk complains about partitions not on cylinder boundaries is just a warning, but it does not affect anything === Quintasan_ is now known as Quintasan === jkridner__ is now known as jkridner [13:49] ogra_, A proposal: You dont strictly need to use fdisk to determine the size of the disk. sfdisk can do that as well. So you can spare a binary in initramfs [13:49] sveinse, well, jasper is pretty much dead beef [13:50] ok, just saying [13:50] we only keep it in the archive in case we will actually spin preinstalleds once again [13:50] over time it will vanish from the arhcive though, i wont put any work into it [13:51] and yeah, i know that sfdisk can do this [14:40] * ogra_ looks at bug 1044709 and wonders what got infinity the honor to be the new unity slave [14:40] Launchpad bug 1044709 in unity/6.0 "unity-6.4.0 from quantal-proposed crashed with SIGSEGV on omap4" [Critical,Fix committed] https://launchpad.net/bugs/1044709 [14:41] oh you actually commited that to their tree === cmagina is now known as cmagina_away === cmagina_away is now known as cmagina === cmagina is now known as cmagina_away === orated_ is now known as orated === rickspencer3_ is now known as rickspencer3 === cmagina_away is now known as cmagina [20:40] omap4+armhf: it works. enough said. [20:44] jimerickson: Erm, when didn't it? [20:54] was have some flashing screen problems with the last kernel. but this latest update has resolved it. [21:03] Ah. [21:14] Anyone here know about performance counters? [21:46] I do, sort of. [21:46] At least from the Pentium Pro days. === cmagina_ is now known as cmagina_away [22:54] GrueMaster, For some reason on omap4 devices the performance counters are disabled and there doesn't seem to be documentation for enabling them. [22:54] So I suppose pentium pro isn't relevant. [22:55] Yea, propably not. :P [22:55] You might ask on #pandaboard though. They can probably help, if it is harder than simply rebuilding the kernel. [23:00] It should be fairly straight-forward though. Looks like it can be done by downloading the kernel source package for the kernel you are using, then modify the config to enable perf counters. [23:00] And rebuild. [23:00] Documentation is at http://help.ubuntu.com/community/Kernel/Compile [23:27] GrueMaster, I tried that. [23:28] There is a bug report on launchpad. [23:28] And the new custom kernel didn't work? [23:28] The existing kernel has it enabled. [23:28] ah, ok. [23:29] But warmcat says on the launchpad bug report that it is intentionally disabled, i cannot find what they disabled or where. [23:29] I just get errors every time I try to use the PMU. [23:32] Sounds like possibly a #linaro or #pandaboard issue. [23:32] Ok. [23:33] I seem to remember some discussion on this a while back (like 11.10 timeframe).