/srv/irclogs.ubuntu.com/2012/07/27/#ubuntu-kernel.txt

=== Ming is now known as Guest98249
=== smb` is now known as smb
smbmorning07:36
ppisatimoin07:37
cooloneyapw: hey, andy07:59
cooloneyi'm trying to use aufs in quantal, but found it's not built as module. like modprobe aufs will fail.08:02
cooloneyeven i tried my own built kernel package from latest master-next08:02
cooloneysmb, morning, ^^^08:03
RAOFAufs has never been in mainline, right? And we've stopped patching it in - it's now overlayfs only.08:03
smbcooloney, morning, yes afaik aufs is not build to see who screams08:03
smbIts unionfs i think 08:04
smbor overlay08:04
smbbah08:04
smbRAOF, Oh btw, when I just see you around... May I express my unhappiness with X for Quantal (bug 1029582)? :)08:05
ubot2Launchpad bug 1029582 in xorg "Unusable graphics with ATI RS690M" [High,New] https://launchpad.net/bugs/102958208:05
cooloneyRAOF and smb, exactly, we are using overlayfs from upstream now. but i'm working on a bug which is on overlayfs but not on aufs08:06
cooloneyhttps://bugs.launchpad.net/ubuntu/+source/linux/+bug/95935208:06
ubot2Ubuntu bug 959352 in lxc "Ephemeral containers have "/rootfs" prefix in /proc/self/maps entries" [High,Confirmed]08:06
cooloneyit looks like there is a bug in overlayfs, but i wanna make sure it is not on aufs.08:06
cooloneyso trying to enable aufs in kernel now08:07
smbcooloney, I am not very confident that the aufs module has been kept in a buildable state... even if it is still in there08:07
cooloneysmb: hmmm. looks like i can only use old kernel 08:10
smbcooloney, Well I'd probably try to follow the "rootfs" hint. There seems to be not too many places in fs/ than have it. So fs/namespace.c and fs/ramfs/inode.c seem to be the only places. namespace.c looks like a good place to start...08:21
cooloneysmb: thanks, i'm looking at that. probably fs/overlayfs/inode.c, but i fail to find the lxc configuration in Quantal, looks like lxc changed a log08:25
RAOFsmb: Sadface.08:25
smbcooloney, Right, maybe it would help to talk to stgraber / hallyn about the differences in mounting between ephemeral and non (like whether one uses its own namepsace or not). Since one seem affected but not the other08:26
smbRAOF, Me too, I use that laptop normally for iso tests. but right now it is not very helpful08:28
smbAnd I could not even tell why... To me the logs are not very bad looking... at least not that critically bad...08:29
smbNot that I know what to look for...08:29
RAOFsmb: It looks like some acceleration snafu.08:29
RAOFYeah, all the logs look fine.08:30
smbRAOF, Yeah, well I will help as much as possible. Just need someone to add someone to add a few pointers about debugging this.08:31
smboops08:31
smbmental loop detected... abort for more coffee08:32
RAOF:)08:32
RAOFDoes it actually work, just fail to display correctly?08:32
RAOFie: if you type blindly to log in, does the screen change?08:32
smbNo, with the current state I seem to get some more broken graphics and then fall back tolightdm08:33
smbWell probably you can say that works in some way a bit08:33
RAOFAh, so it actually crashes somewhere?08:34
smbBut sooner or later (if I insist on using graphics) I also had total lockups08:34
smbLet me try again and see whther there is more in and *old.log08:35
* smb just was glad optional text mode worked to get sshd installed...08:36
cooloneysmb: i found the templates of lxc moved from /usr/lib/lxc to /usr/share/lxc/08:36
smbOk, right now there is a black screen but I heard drums...08:36
cooloneysmb: do you know why lxc should use overlayfs? is it for chroot rootfs?08:36
smbcooloney, Well since the dictionary say ephemeral means temporary I assume their purpose is to be playgrounds to revert back to what things where when you end the container...08:38
cooloneysmb: yeah, ephemeral just start the lxc container and execute some commands in it then shutdown the container08:39
smbcooloney, But also I guess your fs will be unchanged whatever you do. Thus overlay or aufs...08:40
cooloneysmb: and in /usr/bin/lxc-start-ephemeral, it use overlayfs as default unionfs08:40
* smb would prefer temporary for simplicity08:40
RAOFsmb: Could you try adding an Xorg.conf with Option "NoAccel" "true"?08:40
smbRAOF, Yes, if I get it up in a state it is usable. Seems it exploded right after the drum sound. No ssh no nothing... arg08:41
RAOFOk, that suggests that the driver is just sending garbage to the drm device. That's not wonderful.08:42
smbRAOF, :) No I think it makes the device a bit unhappy08:42
ckingsmb, http://en.wikipedia.org/wiki/Zero_insertion_force09:07
ckingso many standards09:09
* apw yawns09:10
smbapw, slacker. :)09:12
smbapw, cking and I already completed all of our morning whining...09:13
apwi can't speak yet, know i am moaning inside09:13
ckingyep, as ever, lots of stuff to whine about09:13
smbapw, We pretend we can hear you complain then09:14
ckingthe old geezers whining channel09:20
smbRAOF, Ok NoAccel allows to use things... /me goes updating the bug09:50
=== yofel_ is now known as yofel
ckingapw, smb, I've sent some slub/slab results to you for your eyeballing10:21
apwcking, ack10:21
smbok10:22
apwcking, remind me, did we decide that EFI would need FAT11:05
apwi think we did 11:05
ckingyes11:05
apwthought so11:05
cooloneycking: hey, cking, i've setup ceph server on my host and i'm going to setup a ceph client on a PandaES board11:07
ckingcooloney, cool, lemme know if you need any assistance11:11
smbcking, After finishing my netbook surgery I am looking at the slab/slub report. One note about the graphs: if possible I would use the same y-scale for left and right. But that is probably just my personal preferance. :) And about locality, that is the unsigned byte distance? Just wanting to understand better. Maybe I missed the real explanation,,,11:14
ckingsmb, it's the unsigned difference between the addresses11:14
cooloneycking: so i've mount vm'ceph fs on my host machine. if i write some file to this /mnt/cephs, can we read it out in vm?11:14
smbcking, Ah ok, thanks11:15
* cooloney is going to upgrade his old macbook pro to SSD and 8G RAM, and hope to do some testing in the future11:15
ckingcooloney, you can only make sense of the data on the client - on the host it's in CEPH's format11:15
cooloneycking: it means i need setup a ceph client somewhere and mount the directory in that ceph client, right?11:16
ckingcooloney, yes, you need to mount it with a ceph client11:17
ckingcooloney, but ceph client is built into our kernel, so it should work easily on a client11:18
* henrix -> lunch11:19
cooloneycking: ok, thanks, will bring up on my pandaES11:37
cooloneyapw and cking, enjoy your Olympic Game Openning11:38
apwcooloney, just waiting for the rain to start11:38
ckingcooloney, so you probably need to do three tests of tests:  client on x86, server on arm, and client on arm and server on arm and finally client on arm, server on arm11:38
apwcking, on arm .. fun11:47
ckingcking, gotta be thorough :-)11:47
ckingbah, /me now talking to himself11:47
njinhallo guys, can be this a linux bug ?  [drm] nouveau 0000:00:05.0: PGRAPH - ERROR nsource: FORMAT_EXCEPTION nstatus:11:53
njin[drm] nouveau 0000:00:05.0: PGRAPH - ch 3 (0x000db000) subc 6 class 0x009f mthd 0x0308 data 0x0018004211:53
njin[drm] nouveau 0000:00:05.0: PGRAPH - ERROR nsource: FORMAT_EXCEPTION nstatus: INVALID_STATE BAD_ARGUMENT PROTECTION_FAULT11:55
apwnjin, that is the kernel complaining about the format of a GPU command it was passed likely from mesa11:56
njinapw, thanks, so this is not an error ? my syslog is full of this11:57
njinandf the graphic is garbled11:58
apwnjin, its clearly an error, its not necessarily a kernel bug, it is as likely a mesa bug as mesa writes the GPU instructions11:59
njinapw thanks, so i open a bug report against mesa11:59
apwas the X folk have more knowledge of the GPU side i tend to file their side and let them decide if its kernle :)11:59
apwi would file it against the xorg drvier for your card, and they will reassign it sensibly if its wrong12:00
njinok, i do then, thanks again12:00
apwnp12:00
infinitylibdrm seems like a sane spot to assign blame.12:00
infinityIt could just as easily be an overheating video card, though.12:01
apwinfinity, by filing it against the ddx for nouveau the right logs get got automatically, then X bots it into the right place generally12:01
apwinfinity, indeed, could be the kernel not doing powermanagement or 1000 other things :/12:01
infinitynjin: File wherever apw tells you to file, but I do recommend checking your fans. :P12:01
apwif only these vendors realised they need to _support_ the crap they sell12:02
apwespecially as win8 looks like its going to be an utter disaster and is driving the games people away12:02
infinityapw: Nah, I didn't mean to imply it was software overheating it, but that 99% of "my screen is garbled" video card issures are airflow and/or loose heatsinks.12:02
infinityissues...12:02
njininfinity, ok I try to put an additional one, but the case is already opened12:02
apwinfinity, oh i cannot argue with your analysis at all.  i still file against ddx just to get right X info12:02
infinitynjin: I didn't mean "add more fans", I just meant "check that the one on the card is still spinning, check that the heatsink is hot to touch", etc.12:03
apwthose heatsinks where the paste is dried out are the worst12:03
ckingor fluff stuck in the fan outlets12:03
infinity(The heatsink should be so hot that it burns your fingers under load, if it's not, it's come loose or the thermal paste has gone bad)12:04
njininfinity, fan is spinning, and cleaned12:04
infinitynjin: Alright.  Then happy bug filing.  ;)12:04
njinthen I file a bug against xserver-xorg-video-nouveau12:04
infinityFrying video cards is actually (seriously) why I gave up on high-end gaming. :/12:05
apwwtf i am suddenly getting 40k a second instead of 500 from the DC ... wobble12:05
apwoh i wonder if we are suffering olympic internet syndrome already12:05
infinityapw: A flood of Romney-mocking tweets, no doubt.12:05
rtghenrix, I just pushed a closing entry to Precise master-next. do some test builds on that.12:47
henrixrtg: thanks. i'll check that12:48
henrixrtg: i'm building it now, it will take a while.12:51
henrixrtg: so, there was nothing wrong with your patch. the prob was that i modified the abi directory (basically, i renamed it). since there were changes done to the arm abi, these changes were lost12:52
henrixrtg: the origin of this issue was the rename, due to the unreleased version on git12:53
rtghenrix, but you are agreed that what is in master-next right now is correct ?12:55
henrixrtg: yep, it is and it should fix the prob. the ignore file should fix it12:55
rtghenrix, and you understand _why_ the ignore files are OK in this instance ?12:55
henrixrtg: yep. i guess i overcomplicated this :(12:56
henrixrtg: solution is far less complex12:56
henrixrtg: thanks12:56
rtghenrix, normally an ABI ignore on a minor update is dangerous, but since that kernel was never published...12:57
henrixrtg: ack12:57
rtgapw, drat. I've just done a 'reset --hard' on a branch that I hadn't pushed. Is there a way to figure out what the origin SHA1 was on that branch before I lose all the commits leading up to it ?13:32
rtgoriginal*13:32
rtgaha! ORIG_HEAD perhaps ?13:33
apwrtg, if you did a git reset --hard then git log -g will show the previous locations of the HEAD13:36
apwover time back for about 2 weeks13:37
apwthis drops out a list of somethig called the reflog, git log -g shows HEADs location, git log -g <branch> does just the branch tip13:37
apwrtg, did you find it ?13:39
rtgapw, yep, ORIG_HEAD had what I needed13:40
apwrtg, ok the reflogs have a longer view should you really mess up :)13:40
rtgI assume that a 'git gc' cleans up all history wrt previous HEADs ?13:40
apwrtg, the reflogs hold references to old trees based on time, so they should be there i think it is either 14 or 21 days, then the reflog entries go and release them13:41
rtgok13:41
apwthat said my reflogs for my ubuntu-quantal head goes all the way back to 18-nov-2011 wow13:42
apwrtg ok its 90 days for entries which point at things which still exist from a the tip you are on, and 30 days for others13:45
* rtg relocates...13:50
josephtI'm trying to bisect commits between mainline v3.4.6-quantal and v3.5.0-quantal.  Should I be using the Ubuntu... tags or the v3... tags?14:18
rtgjosepht, likely the v3 tags since the commits are more or less linear. the Ubuntu tags could be across a rebase.14:21
josephtrtg: thanks14:21
apwbjf, https://wiki.ubuntu.com/Kernel/KernelPPA i have created a new page which describes the KernelPPA, how to make one, how to upload, how to copy to -proposed, why debug is bad etc.  but it needs the how to copy bit filling in ... perhaps you could do the honours14:27
bjfapw, will do, thanks14:28
apwbjf i have also rejigged the bottom of stable release cadance so it tells you sh14:28
apw'who to give the package to, or where to upload it' sort of thing instead of all those details on the Kernel PPA which are now in the other page14:29
bjfnice14:29
dilekshttp://kernel.ubuntu.com/git?p=ubuntu/ubuntu-quantal.git;a=commitdiff;h=dc9bd4098f0cf9703f30fe2fa3c433fafddf435c14:34
dilekshttp://kernel.ubuntu.com/~cking/iosched/14:34
* cking suspects an impending discussion about this14:35
* dileks looks at Overall.. section14:35
* cking recommends dileks to read the entire report rather than just drawn a conclusion from the final section14:36
* dileks cross-read it14:37
ckingah14:37
* dileks ...and looked at the result sheet (ods)14:37
dileksdepends whats your focus... more modern hardware and filesystem, you might be on the right track14:40
dileksmore on*14:40
ckingdileks, like all things, some win, some lose14:41
dileksIIRC the linux-fsdevel guys have a testsuite. you tested with it?14:41
* dileks checks mailbox14:41
ckingdileks, no, but if I tested with all the fs test suites out there I wouldn't get any results out before 14.02 :-/14:42
ckingapw, so, I think the conclusion from the slub/slab data is we should stick with slub 14:45
dilekshehe. visited donald knuths blog yesterday for 1st time... referenced from a x32abi link... and while reading more articles taocp page was listing about book part V an ETA 202014:47
ckinghopefully knuth with be alive then, we can only hope14:47
ckingbjf, we got any precise SRU kernel updates coming down the pipe? I think I'm waiting for the U300 S3 bug to get verified14:48
henrixcking: on its way14:48
dileksage 7414:49
bjfcking: yup, this is prep week. 100+ are inbound14:49
ckingooh, that's a whole load of patches  in the pipe 14:49
* ogasawara back in 2014:50
dilekscking: nice editing with test-results14:50
ckingdileks, much appreciated :-)14:51
apwcking, slub> thanks for investigating14:53
ckingapw, if you think I should explore more avenues, I'm happy to do so, but I think we have enough data to make a decision14:53
apwcking, no i trust your judgement, we had what was an arbitrary choice made "this is new and cool" some two years ago, now we have numbers to say it was cool14:54
apwcking, that is good enough for me14:54
ckingand other distros are using the same config too, so that's helpful14:54
rtgapw, IIRC there was _some_ LKML data to justify the switch.14:55
apwrtg, yeah but i knew it was by someone who tended to say things were great for everyone when only had tested them on vast machines with zillions of cpus14:56
rtgwell, true14:56
apwso its good to have some proper analysis tell us we made a sane choice :)  and more importantly it remains a sane choice14:56
rtgapw, I was just pointing out that our decision to change  wasn't _completely_ frivolous.14:57
apwrtg, heh sorry, i know it wasn't quite as arbitrary as i make it sound ...14:58
ckingit wasn't a flip of a coin choice that's for sure14:58
dileksso, slub + deadline is todays choice?14:58
apwat least while cfq continues to have these hideous corner behaviours i recon14:59
ckingdileks, slub has been our choice for quite a while14:59
apwcfq must be one of the most incongiously named drivers there is14:59
rtgdileks, yes, deadline appears to behave a little better under extreme conditions.14:59
* dileks remembers discussions with fellows from grml and debian and both switched to slab... so this might have changed again15:03
dileksany link to "...LKML data to justify the switch"15:04
dilekscking: testsuite is called xfstests which you can use also for numerous other fs#s15:11
ckingdileks, yes, I'm familiar with those, I used them to exercise eCryptfs a lot15:12
* cking --> more coffee. when will this flippin' jet lag wear off?!15:17
* dileks recommends a centrifugal machine for astronauts to kill jetlag15:19
* cking recommends sleep 15:24
apwppisati, remdind me did you tell us why we have CONFIG_SATA_AHCI_PLATFORM off on ti-omap4 ?15:26
apwppisati, i sort of think you said it turned something ugly on15:26
ppisatiapw: it breaks compilation if OMAP5 is not defined (and we don't define it)15:27
apwppisati, ahh thanks thats the one15:27
=== ev_ is now known as ev
infinityapw: It's somewhat moot anyway, unless you can figure out how to staple an AHCI controller onto your Panda.16:01
apwinfinity, indeed but we are looser with things in the config review, clearly here it should be off though16:02
infinityAt least until we start flirting with the idea of building unified omap kernels, yeah.16:03
infinity(Which would be nice, if omap4 and omap5 are finally being upstreamed enough to get away with it...)16:03
apwogasawara, i think we decided we'd keep CONFIG_WAN_ROUTER on only where it might possibly have been used before, which for me is i386/amd64 and powerpc ... right ?16:03
apwinfinity, oh indeed, which is one reason we try and turn on as much common stuff as we can so any future merge is easier16:04
ogasawaraapw: yep16:04
ppisatiinfinity: omap3 + omap4 is doable for R IMO16:04
apwogasawara, and can you remember anything about GPIO_SYSFS16:04
ppisatiinfinity: while omap5 is clearly impossible16:04
infinityppisati: Well, Linaro is already doing it now, but obviously not with mainline.16:05
infinityNot sure how massive the TILT changeset still is, I haven't looked in a long time, but I trust that you're on top of it, so I try not to care. :P16:05
ogasawaraapw: i remember we asked jj and he didn't have a good recommendation for it16:05
ogasawaraapw:  so it seems we tabled it for later16:06
apwogasawara, oh hrmmm ... 16:06
ppisatiinfinity: http://people.canonical.com/~ppisati/ti-omap4-tilt-34-on-35/tilt-3.4/16:07
infinityppisati: So, still "huge", then.  Basically.16:07
infinityAlthough, they're all rather tiny changes.  Maybe they just need to upstream harder.16:08
apwheh so we can merge omap3 and 4 if we pull it out of master16:08
ppisatiyes16:09
infinityapw: Which we could do if master could either (a) build linux-libc-dev without a kernel, or (b) we added a different flavour (like, say, vexpress)16:09
infinityOh, wait, we have another flavour now.  Hi there, highbank.16:09
apwinfinity, i beleive we can do that anyway16:09
infinity(Except on armel, but we'll cross that bridge when we get to it)16:10
* smb -> EOW16:14
* herton -> lunch16:21
* ppisati -> gym16:23
apwogasawara, ok other than GPIO_SYSFS there are only three red lines in the 'common drivers' sections, they are all bryan's fault, once my test builds are done i'll push16:46
ogasawaraapw: heh, awesome thanks16:47
apwogasawara, if we ignore ti-omap4 things are very close to as good as they are going to gt for Q i recon16:51
apwget16:51
ogasawaraapw: I can be happy with that16:51
apwogasawara, i recon we let bryan do a bit more and then relegate the rest to R's torture16:53
ogasawaraapw: do we want to have a WI for cooloney/ppisati to clean up ti-omap4 or feed us annotations?16:53
ogasawaraapw: whatever they don't hammer out we can grind through at UDS-R and our next sprint16:53
apwogasawara, i think they have one for some of it anyhow... i have some open WIs for it so i may just hammer them out and use them as testing bitches16:54
ogasawaraapw: ack16:54
apws/it/general beta2 review/16:54
* apw waits impatiently for gomeisa16:56
apwogasawara, heh a 15m load average of 178 ...16:58
ckingit's almost on fire16:59
* cking --> food16:59
apwogasawara, ok it seems to build for me, highbank is an abi bumper so you'd need to slip one of those underneath when you are tieing the bow17:31
apw(as and when)17:31
ogasawaraapw: ack17:31
apwogasawara, i've close that WI so we are making progress ...17:31
apwogasawara, i think that gets us back to normalised relative to the slow even after the crap we added last week17:33
apws/slow/slope/17:33
* rtg -> lunch17:57
* cking --> EOW18:19
=== JanC_ is now known as JanC
* rtg -> EOW19:35
joshhunt_i saw usn-1515-1 for a new kernel pass by the other day on the mailing list, but then it seems to have disappeared. does anyone know the status of this?19:55
sbeattiejoshhunt_: hunh. that's odd. Will look.20:03
sbeattiejoshhunt_: I've caused the USN to be regenerated at http://www.ubuntu.com/usn/usn-1515-1/ I'm not sure why it wasn't there before20:05
joshhunt_sbeattie: thanks20:10
joshhunt_sbeattie: yeah it was weird. i saw it in the mailing list, but then it wasn't on the rss feed/web site.20:11

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!