[01:06] rickspencer3, you abouts? [01:07] rickspencer3, we've got a bit of a conundrum with one of our bugs [01:07] bryceh, yeah [01:07] 'sup? [01:07] I should have a good connection between here and home [01:08] rickspencer3, the background is that glx 1.4 was backported by debian from xserver 1.8, however we've found it is buggy [01:08] And seriously - bug #565981 - buggy. [01:08] Launchpad bug 565981 in xorg-server "[KMS] gem objects not deallocated" [Critical,Confirmed] https://launchpad.net/bugs/565981 [01:08] * rickspencer3 looks [01:08] rickspencer3, you might remember the issue of Clutter not working, that needed an xserver fix a few weeks ago [01:09] rickspencer3, it turns out the fix we incorporated causes the memory leak issue that RAOF just mentioned [01:09] yup [01:09] Sarvatt's been banging his head all over this issue. We're not finding a suitable solution [01:09] fudge [01:09] so we revert the change, and UNE breaks [01:09] we don't revert the change, everything breaks? [01:09] what we're considering is dropping both that fix, and the Debian glx backport stuff [01:10] what is the impact of the leak? [01:10] well, that's why I wanted to raise this with you before we did anything [01:10] the leak seems to be encountered pretty widespreadedly [01:11] so the clutter bug was caused by backported code? [01:11] yes [01:11] so you are consider going all the way back to what, glx 1.3? [01:11] fuuuuuuuudge [01:11] No; 1.2. [01:11] 1.2 [01:11] ack [01:11] 1.3 would be ok, because that's what clutter seems to use. [01:12] whatever [01:12] the point is, roll way back [01:12] ? [01:12] To the glx version (but not the code) we had in Karmic, yes. [01:12] bryceh, why did we go to glx 1.4 to begin with? [01:13] rickspencer3, debian had incorporated it into their packaging [01:13] rickspencer3, so we essentially just inherited it [01:13] ok [01:13] rickspencer3, however jcristau just mentioned that they're going to drop it in Debian due to these problems [01:13] so did any new and important functionality or bug fixes come with it? [01:13] ah [01:14] I'm not sure. What I'm wondering mainly is if it'd negatively impact clutter [01:14] okay, so the 1.2 version has been around a while, and is generally considered good? [01:14] Yes. [01:14] i.e. has clutter come to depend on having glx 1.3 / 1.4, and if so will we cause more problems if we remove it than we're solving? [01:14] ah [01:14] I don't know [01:14] I would only be able to ask njpatel that question [01:15] Apparently clutter only started using glx 1.3/1.4 in clutter 1.2, which we pulled in relatively recently. [01:15] so, to the best of our knowledge, backing out these patches seems to be the right thing to do, but we don't want to decide this unilaterally and fsck up the clutter folks [01:15] will this make my saurbraughten run slower? [01:15] yeah, so njpatel and didrocks need to be consulted [01:16] can you guys set up something that can be installed and tested in the meantime? [01:16] Sarvatt's already got a package in the x-testing PPA dropping the glx 1.4 backport. [01:16] bryceh, I agree with your concern, that there is some little thing in clutter that depends on 1.3 functionality [01:16] RAOF, can you please run UNE on it [01:16] and see how the launcher works? [01:16] also try some of the clutter based games? [01:17] robert_ancell, do you know which gnome-games rock clutter the mostest? [01:21] rickspencer3, quadrapassel [01:21] lots of bug reports on that one... [01:21] RAOF, can you run that as well? [01:21] Yup. [01:22] robert_ancell, well, we're looking for apps that we can run to validate that clutter still works [01:22] rickspencer3, it does like to break :) [01:26] clutter apps shouldn't be affected, it picks which mode to use.. I just looked at netbook-launcher and it's using glx 1.2 functions for the only things not abstracted behind clutter so it's fine, its the apps that call glx 1.3 functions directly that would be a problem (but it's an application bug if it does anyway as the warning will say) [01:29] Sarvatt: But we haven't tested clutter's glx 1.2 pathways as much recently. So while clutter apps *shouldn't* be affected… :) [01:36] UNE & quadrapassel seem to work fine with nvidia-current; I'll test moovida too. [01:37] * RAOF kills X to check on intel, too. [01:46] RAOF: nvidia doesn't use the server's GLX anyway... [01:51] Sarvatt, where would you predict to see any regressions? [01:53] Dou you know any PPA with unstable Glib/GTK+ packages? [02:01] RAOF, Sarvatt, bryce says that code that calls glx uses /usr/lib/libclutter-glx-0.8.so [02:01] is it possible to generate a list of code that calls this? [02:01] maybe packages that have a dependency? [02:02] on lucid - /usr/lib/libclutter-glx-1.0.so.0 [02:02] and /usr/lib/libglitz-glx.so.1 [02:02] shall I ask slangasek in #ubuntu-devel? [02:02] (for some reason it seems like he would know) [02:03] http://pastebin.ca/1870751 is the rdepends of libclutter-1.0-0, which contains libclutter-glx [02:03] he'd probably know [02:03] $ nm /usr/lib/libclutter-glx-1.0.so.0.200.4 [02:03] nm: /usr/lib/libclutter-glx-1.0.so.0.200.4: no symbols [02:03] hmm, I know there's a way to view the symbols of a library, but old age has made me forget [02:04] From memory it's a magic objdump invocation. [02:04] RAOF that looks like a tractable list [02:04] ahaa /usr/lib/xorg/modules/extensions/libglx.so [02:08] RAOF, can you please start to tick these off and make sure they don't break if we roll back? [02:08] So; UNE, moovida, quadrapassel all work fine on nvidia-current & -intel. [02:08] good start [02:09] so long as qudrapassel works, SHIP! [02:09] hehe [02:10] Incidentally, moovida is pretty cool :) [02:10] yes it is [02:17] RAOF, don't bother with moblin stuff [02:18] Ok. [02:19] RAOF, is there way to see what out of this list has not been updated since we were on 1.2? [02:20] Checking the changelogs would be possible. [02:22] That's not necessarily enough, though; the libclutter codepaths will be different for the GLX 1.2 case, so we could hit bugs there without any client code changes. [02:23] RAOF, right, there are two classes of problems: [02:23] 1. bugs caused by 1.2 api calls, that is, bugs in glx 1.2 [02:23] 2. apps that depend on version 1.3 api calls [02:24] just looking for apps that haven't changed since 1.2 is insufficient for case #1 [02:24] But sufficient for case 2, yeah. [02:24] and just smoke testing isn't sufficient for case #2 (as perhaps you won't exercise the code paths that make the 1.3 calls [02:25] because we are so short of time, we can't use the "upload and see who complains" QA message for this :/ [02:25] :) [02:27] A large part of the libclutter-1.0-0 rdepends list is language bindings with no in-archive users. [02:27] interesting [02:27] just sitting there waiting to be used [02:32] training pulling in soon, see you guys tomorrow [02:33] Have fun! [02:37] RAOF, do you overlap with didrocks at all? [02:37] Yes, in the late afternoon. [02:38] RAOF, great [02:38] thanks [02:38] ahh sorry, went to dinner there and just got back. clutter really is the only place I'd expect any kind of regressions (not that 1.2 isn't a regression already because it expects features from xserver 1.8). RAOF did you try running these apps with LIBGL_ALWAYS_INDIRECT=1 when you checked? also dont know if you caught my message but nvidia bypasses the server libglx anyway so it wouldnt be any different there [02:39] Sarvatt: Of course! I should have expected nvidia to use its own stack :) [02:40] Sarvatt: I haven't been testing with LIBGL_ALWAYS_INDIRECT=1; should I have? [02:40] you want to run through the server glx since KMS drivers supported glx 1.3+ before anyway [02:44] * RAOF should obviously have read the patch a little more carefully [02:45] So this is going to affect a significantly smaller fraction of our users then. [02:50] Ok. gnome-shell is an unhappy camper. [02:51] …as is netbook-launcher. [03:07] yeah netbook-launcher is all kinds of messed up here [03:08] It looks like font rendering is broken in the GLX 1.2 pathway. [03:08] it uses -efi [03:08] doesn't it? [03:09] Sarvatt: Which one? netbook-launcher, or netbook-launcher-efl? [03:09] it uses low graphics mode when i launch it with LIBGL_ALWAYS_INDIRECT=1 here [03:10] Nah; loads fine here. [03:10] But text is rendered as small white squares. [03:10] oh ok theres a seperate low quality mode for 3D? [03:10] For non-3D? Yes. [03:10] yeah its the same here, white box fonts [03:10] netbook-launcher-efl [03:11] i dont have -efl installed [03:11] (netbook-launcher:2653): libnetbook-launcher-DEBUG: CONFIG: Low graphics mode: True [03:11] (netbook-launcher:2653): libnetbook-launcher-DEBUG: CONFIG: Low graphics mode: False [03:11] first with indirect [03:11] it plain doesn't work in direct mode here [03:12] screens all garbled [03:12] RAOF, You should try: swell foop (It uses clutter+seed(js)) [03:13] LIBGL_ALWAYS_INDIRECT=1 swell-foop [03:13] Segmentation fault [03:13] :) [03:15] So, this looks both better and worse than we thought it was. [03:18] Sarvatt: What crazy X server are you using? LIBGL_ALWAYS_INDIRECT=1 swell-foop doesn't segfault here (although it also doesn't render anything, either) [03:18] RAOF, it's past EOD for me, anything else I can help with on this bug before I go? [03:20] I don't think so. [03:21] ok, I'll try checking in again in several hours, so drop me a note if something needs sponsored or whatever [03:21] Thanks. [03:26] Sarvatt: So I think this puts the kybosh on just removing the GLX 1.4 backport. [03:27] without downgrading to a stable clutter you mean? [03:27] Yes. [03:29] That would mean *at least* a bunch of rebuilds; clutter 1.2 has new API, and netbook-launcher at least hits that when you try to downgrade clutter. [03:34] i'm using lucid + x-updates xserver on intel btw, it does segfault every time [03:36] I'm using lucid + x-updates xserver on intel, it segfaults none of the time :/ [03:36] 945? [03:36] oh [03:36] my driconf settings for mesa were retained on the downgrade, I forgot about that [03:36] GM45 [03:37] It crashes on quit, if that's any consolation :) [03:37] yep GLXBadContextTag [03:38] No GLX cleanup for you! [03:38] how are you supposed to gdb that sucker? [03:38] Run with --sync and break in gdk_x_error? [03:39] nah attached to the pid, i was trying to launch it directly from gdb [03:40] Ah. It's all “I'm running your runtime in my runtime, dawg” [03:40] darn forgot mesa symbols in the archive are messed up [03:41] Anyway. That's not exactly a critical bug. [03:42] What was wrong with your master-fixes backport patch, apart from it being a code refactor? [03:45] Sarvatt: LIBGL_ALWAYS_INDIRECT might have other side-effects besides dropping down to GLX 1.2, right? [03:46] yep [03:46] sorry, afk a few [03:46] Would booting with i915.modeset=0 exercise the glx 1.2 codepaths? [03:46] Ok. [03:47] I'll give that a try anyway. [03:47] no unless you were on ATI [03:47] doh [04:00] so whats the glx story - is there a bug I can read? [04:08] lifeless: https://bugs.launchpad.net/bugs/565981 [04:08] Launchpad bug 565981 in xorg-server "[KMS] gem objects not deallocated" [Critical,Confirmed] [04:10] Sarvatt: thanks [04:11] ati and intel are getting memory leaks on the order of 1.5gb or so a day currently [04:11] Ouch. [04:12] wow [04:12] I won't reboot for a bit then [04:14] you should reboot, often if you are using KMS :) [04:14] its been this way for almost a month now [04:14] Sarvatt: 13:14:34 up 15 days, 1:27, 11 users, load average: 0.09, 0.13, 0.09 [04:15] cat /sys/kernel/debug/dri/0/gem_objects ? [04:15] still got 1/2GB unallocated [04:15] 1190 objects [04:15] 69578752 object bytes [04:15] 5 pinned [04:15] 13725696 pin bytes [04:15] 29446144 gtt bytes [04:15] 234881024 gtt total [04:15] That's the nice thing about 8GB laptops; you can have X silently consuming 6.9 GiB of memory and not notice. [04:16] I'm confused ... how long has this leak been in the distro and why hasn't it been driving people insane? [04:16] nice, can you give me some more details on your setup? what GPU, do you view a lot of flash sites on it? [04:16] Sarvatt: sure [04:16] looks like a intel 965+? [04:16] gimme a sec, phone call [04:16] it seems a widespread bug of this magnitude would drive people crazy and we would be getting a crush of feedback [04:16] i7 64-bit [04:17] rickspencer3: We possibly are, but distributed against all sorts of other packages. [04:17] This memory is not accounted for anywhere that's particularly user-visible. [04:17] its very hard to notice [04:17] so if it's "very hard to notice" we should be careful about acting in haste [04:18] RAOF: lifeless isn't hitting it here for some reason [04:18] I would prefer to see a root cause analysis that identifies the code defect(s) causing this [04:18] and see that fixed in an SRU [04:18] instead of turning knobs all over the place at this point in the cycle hoping something works [04:18] Sarvatt: 00:02.0 VGA compatible controller [0300]: Intel Corporation Core Processor Integrated Graphics Controller [8086:0046] (rev 02) [04:19] Sarvatt, RAOF do you both agree that the roll back to 1.2 is off the table? [04:19] Yes. [04:19] Sarvatt: (II) Module intel: vendor="X.Org Foundation" [04:19] compiled for 1.7.5, module version = 2.9.1 [04:19] Linux lifeless-64 2.6.32-17-generic #26-Ubuntu SMP Sat Mar 20 02:23:45 UTC 2010 x86_64 GNU/Linux [04:20] Sarvatt: anything else you want to know ? [04:20] lifeless: oh you haven't upgraded to the broken packages yet is why [04:20] they came on march 31st [04:20] Sarvatt: thats why I said 'I won't reboot for a while' :P [04:20] RAOF, bryceh, Sarvatt, what do you guys think of release noting the problem, and getting cracking with the debugger? [04:20] (or haven't rebooted into them, sorry) yeah I see what you mean :) [04:22] It's not too bad as long as you kill X every day or so. Releasenoting might be appropriate. [04:22] RAOF, there is only so far we can unwind [04:22] I would like to see an SRU ready for this before we ship, if possible [04:23] RAOF, do you feel you have the skills to snorkly into the code and find the leak? [04:23] snorkle, even? [04:23] I'd quite like to snorkly :) [04:23] RAOF, we'll talk about that at UDS ;) [04:23] Yes; I'll ask someone if I hit something incomprehensible. [04:24] desktop-maverick-xorg-snorkly [04:24] ok [04:24] Snork! Snork! [04:24] And I'd love to do some code-diving. [04:24] RAOF, can you please respond to bryceh's email with the new information [04:24] ? [04:24] ^the late question mark reveals that I didn't really mean that as a question ;) [04:24] i'll try to dig into it more, yeah. spent all day trying to backport the upstream fix since the one we are carrying that is the cause of this is the root of the problem but haven't had any luck with that [04:25] fruedian irc [04:25] rickspencer3: :) [04:25] Sarvatt, upstream fix to the leak, you mean? [04:25] I didn't quite grock what you said there [04:27] well, there was the bug where closing clutter apps crashed the server, during the course of the bug report a patch was posted that applies to xserver 1.7 that we picked up and is causing these memory leaks, and they decided to go another route to fix the bug instead of what we used [04:27] but the fix that they used applies to xserver master which is drastically different than what we have [04:28] I see [04:28] and its not something they would bring to the xserver release we are using because upstream does not have these glx 1.4 enablement patches in xserver 1.7 branch in the first place that brought up the problem [04:28] so the fixed the bug in a fundamental way in their "tip" (of xorg-xserver)? [04:29] and due tot he delta between Ubuntu and the tip, that patch couldn't just apply cleanly, so now we're a bit on our own? [04:29] yeah, exactly [04:30] debian is affected as well and they are dropping the patches, we inherited them from them [04:30] yeah [04:31] ok, so root cause analysis next, and get an SRU ready? [04:31] Sarvatt, sound right? [04:32] the backport also touches a significant amount of code and would be risky to SRU, which is why I believed dropping the glx patches would be the better route, but indeed it's bringing up problems with clutter 1.2 [04:32] so trying to fix the patch would probably be the best route [04:32] (and SRUing it like you said) [04:33] RAOF, do you agree, root cause analysis then prepare an SRU? [04:33] Yes. [04:33] ok [04:33] of course, it's a moot decision until bryceh weighs in ;) [04:34] Again, that's the least terrible of our options :) [04:34] rickspencer3: you are using intel right? can you do a cat /sys/kernel/debug/dri/0/gem_objects and paste the first two lines? [04:34] yup [04:34] 1247 objects [04:34] 278298624 object bytes [04:35] rick@rick-desktop:~$ uptime [04:35] 20:34:56 up 40 min, 3 users, load average: 0.66, 1.11, 1.21 [04:35] how long since you rebooted last? [04:35] ah, was going to say.. [04:35] (always one step ahead) [04:35] *cough* [04:35] you didn't seem affected but yeah you are :) [04:35] Sarvatt, well, I ran my UNE netbook all day for the last two days [04:35] do you have swap? [04:35] but I guess I shut it down a bit [04:36] you mean a swap partition, then yes [04:36] Sarvatt, I'm not denying the existence of the bug [04:36] if you turn off your swap you will be frozen after a day because of the leak, I guarantee it :) [04:36] that would suck [04:36] indeed [04:36] I'm not denying the severity of the bug [04:37] I think we are all in agreement that a methodical approach to fixing it is our best bet now [04:37] watch that gem_objects size there, it wraps around to a negative number around 3GB memory used [04:37] I would very much like to see an SRU ready on day zero [04:40] doing everything I can to work on it here, just found out the cause yesterday [04:41] yeah [04:41] hopefully the code defect is not too hard to identify [04:43] Sarvatt, can people who are getting kicked hard by this bug downgrad to glx 1.2 themselves until we get it fixed? [04:44] yes I posted a PPA for that on the bug report [04:44] ok [04:44] that's good [04:44] it's in x-updates, also I am putting xserver 1.8 in xorg-edgers with the fixes [04:45] by "the fixes" you mean the rolled back glx? [04:45] no xserver 1.8 is fine with glx 1.4 [04:45] I see [04:45] so another option is for users to basically update to the tip of xserver [04:45] it's a major change that requires rebuilding all of the X related packages though [04:45] so two things to try [04:46] xserver 1.8 has refactored the way that GLX resources are disposed of; they're now handled as regular X resources and cleaned up there. [04:46] nah, I wouldn't call it the tip, master is the tip and xserver 1.8 is the most recent stable release. xserver master is too cracky even for xorg-edgers crack :) [04:46] hehe [04:46] the crack for the crack [04:47] I have to rebuild all of the video and input drivers weekly for alot of the cycle if i follow master [04:48] Sarvatt, RAOF thanks! [04:48] I [04:49] 've got to go [04:49] ttyt [04:49] no worries, take care [04:49] Have fun getting home. Hope it doesn't take too long! [04:49] Or, in fact, hope you *are* home now :) [04:50] RAOF, I am home! [04:50] * Sarvatt hopes these ash clouds go away in time for UDS [04:50] Sarvatt, yes!! [04:50] ok, g'night [04:51] they're talking about it taking 2 weeks to get everyone who is stranded now home, can't afford to get stuck [04:55] Sarvatt: For people like RAOF and myself who have to come/go home on long haul flights, this is being watched with some attention, at least by me. [04:55] how long is your flight? [04:56] In total? About 28 hours. [04:56] But that isn't all in the air. [04:56] fun! [04:56] at least your flights do go through iceland though? :) [04:57] Sarvatt: I think you meant don't. [04:57] I presume you mean “don't” there :). No. But they do go through Heathrow. [04:57] yeah, sorry [04:58] yeah thats the airport with the 2 week backlog I was reading about === ara_ is now known as ara === ara is now known as Guest45645 === Guest45645 is now known as ara_ === ara_ is now known as ara [08:17] good morning [08:18] Good morning [08:19] Guten Morgen pitti, how are you? [08:20] bonjour didrocks! I'm good, thanks! how about you? [08:20] * pitti takes a stab at bug 566046 [08:20] Launchpad bug 566046 in gnome-keyring "the login password is stored in the user's keyring" [High,Confirmed] https://launchpad.net/bugs/566046 [08:20] I'm fine too, thanks ;) [08:20] oh, that one, upstream didn't reply? [08:23] didrocks: Good morning! [08:24] didrocks: right [08:24] hey RAOF [08:24] Howdie pitti [08:26] good morning RAOF [08:28] hello there [08:29] bonjour seb128 [08:29] seb128: has there been any previous investigation on bug 566046? [08:29] Launchpad bug 566046 in gnome-keyring "the login password is stored in the user's keyring" [High,Confirmed] https://launchpad.net/bugs/566046 [08:30] seb128: I wanted to take a stab at this today [08:31] hey pitti, not that I know out of mdeslaur who said he looked a bit to the code without finding the issue [08:32] salut seb128 [08:32] lut didrocks [08:49] dear launchpad, please do not oops when I try to add a bug comment, love Michael [08:55] Ok. That's enough diving throug X for one day. See y'all tomorrow :) [08:58] good evening RAOF [09:09] seb128: do you know whether there is a CLI tool to dump the keyring? [09:12] pitti, not that I know about no === almaisan-away is now known as al-maisan [09:44] morning [09:45] lut huats [09:46] hello seb128 [09:48] huats, ca va ? [09:49] seb128, yes ! [09:50] et toi seb128 ? [09:50] huats, ca va merci [09:50] :) [09:57] huats \o/ [09:57] hello james_w ! [09:57] how are you ? [09:58] it's been a long time :) [09:59] good thanks, how are you? [09:59] great thanks ! [10:00] a really happy young father :) [10:00] glad to hear it :-) [10:28] is anyone here on i386 and using the flash plugin installed from the archive? [10:29] chrisccoulson: on my netbook [10:29] pitti - do you know if it is using nspluginwrapper? [10:29] it's a clean install from yesterday [10:29] or is that just used for the amd64 version? [10:29] * pitti boots [10:30] I thought it's just needed on amd64? [10:30] chrisccoulson: are you still working on bug 555870? [10:30] Launchpad bug 555870 in gnome-screensaver "Gamma values are not being set properly after a second fade out resulting in a black screen" [High,In progress] https://launchpad.net/bugs/555870 [10:30] pitti - yeah, i think so. i just wanted to make sure we weren't using it on all archs ;) [10:30] kklimonda, no, because i've got no idea what is going on now ;) [10:31] kklimonda, if you can recreate it, please provide a xtrace log from the actual gnome-screensaver process [10:31] damn.. ;) [10:31] chrisccoulson: nspluginwrapper isn't installed here [10:31] chrisccoulson, hey, how are you? [10:31] we're unsure if the issue with test-fade is a red herring or not [10:31] thanks pitti [10:31] hey seb128, i'm good thanks. how are you? [10:31] chrisccoulson, no nspluginwrapper on i386 here either, I'm running the security ppa version which works fine btw [10:31] chrisccoulson, I'm good thank you [10:32] seb128 - excellent. i suppose i need to actually test the flash player with nspluginwrapper on amd64 ;) [10:32] i'm using the 64-bit plugin here [10:39] chrisccoulson: I can get you all xtraces you want if you think it will help :) [10:40] chrisccoulson: would you rather get it attached to the bug report or pasted somewhere http://pastebin.com/U231nmDq ? :) [10:41] kklimonda, attached to the bug would be good [10:41] ok === istaz is now known as staz [10:44] chrisccoulson: btw, sometimes I get a slightly less faded desktop instead of the pitch black one :) [11:24] tseliot, hey, what is the package which has the nvidia display configuration tool? [11:24] seb128: nvidia-settings, why? [11:25] tseliot, thanks, looking where to reassign bug #528756 [11:25] Launchpad bug 528756 in gnome-control-center "Ubuntu forgets my screen resolution settings every time I reboot" [Undecided,New] https://launchpad.net/bugs/528756 [11:25] tseliot, where the user is using the nvidia tool to change the configuration [11:26] seb128: right, that's a bug that I fixed in Lucid [11:29] tseliot, ok, feel free to close it then ;-) [11:29] sure [11:32] can someone please confirm an odd issue with gnome-panel ? first do this steps: http://pastebin.com/SfLG5Abs then do a killall gnome-panel and after it has restarted I see that one starter has been cloned and replaced another starter. e.g. I have two canonical starters but the starter for launchpad is gone. [11:32]