[10:45] * Ng tries to keep up with the slowmotion compiz bug [10:45] bryce: should I open a new bug about the problems I see? [10:46] I guess [10:47] there seems to be several different issues causing performance regressions [10:47] yeah, it sucks when various vague issues get conflated :/ [10:54] performance issues are the worst, they're not measurable and not differentiable since they're summation effects [10:55] freeze issues are nearly as bad since you can't tell one from the next, but at least it's pretty definite when you have the issue [10:56] Ng: so the problem you are seeing is with many windows open? [10:58] tjaalton: I have three maximised windows on separate workspaces (terminator, then thunderbird, then firefox) and flipping workspaces stutters. possibly an easier symptom to work with is the background changer - instead of being a smooth fade it's 2 or 3 frames [10:58] right, I get that too [10:59] but it's snappy when there aren't much to draw [10:59] *isn't [10:59] yeah, and often the first workspace flip or two will be ok, but the performance degrades if I keep flipping [11:00] probably the reason why they invented UXA [11:00] it feels like it's being overly affected by the system working harder, and CPU usage is higher than with UXA, which stays smooth no matter how aggressively I flip workspaces [11:00] perhaps, but EXA in Intrepid didn't do this [11:00] intrepid didn't have gem [11:00] * Ng nods [11:00] what was there before gem? is there an easy way to disable gem? [11:01] I'm a bit lost in intel acronym soup ;) [11:01] dunno if it can be disabled easily [11:01] there is a kernel patch to add a flag you can use to disable it [11:01] there was no proper memory management before, I guess :) [11:02] there's also a patch that forcefully disables it in the driver, without requiring kernel changes [11:02] ok [11:02] I can dig either up if anyone's interested. [11:03] they're on -intel bugs [11:03] I'm too chicken to try ripping out GEM at this late stage, at least, not without more testing [11:04] Ng might want to try, I'm a bit busy with packing etc :P [11:05] I'll have a go at the -intel level one. I'm not up for even thinking about kernel changes at this point ;) [11:06] standby [11:08] Ng, bug 349992 has an A17 swizzling patch for -intel [11:08] Launchpad bug 349992 in xserver-xorg-video-intel "[945 tiling] Low performance due to no A17 workaround" [High,Confirmed] https://launchpad.net/bugs/349992 [11:08] the kernel patch is also attached to that bug [11:08] there is another patch, hang on [11:10] http://paste.ubuntu.com/142337/ [11:11] all this stuff, plus the MCHBAR patch, is documented at https://wiki.ubuntu.com/X/Troubleshooting/IntelPerformance [11:12] ta [11:12] hmm that second patch looks familiar, I think I read it after reading that IntelPerformance page and dismissed it because it was talking about 945 [11:13] but I didn't think to check if 965 matches IS_945* or if there's an IS_965 I could use [11:22] about to pass out... bedtime... night [11:24] cya :) === Ampelbein is now known as Ampelbein_ === Ampelbein_ is now known as Ampelbein [19:35] anybody heard of any bugs involving the screen flickering black a few times then staying black [19:35] my music pauses momentarily, then sometimes resumes after a few seconds or others just stays dead [19:35] and keyboard does not respond [19:36] compiz + nvidia 180 [19:36] only recently started happening in jaunty :( [19:37] and with nvidia 177 compiz refuses to start so i can't rule it out [19:51] alex-weej_: I had the same problem but it was 8.04 64 bit Gnome it was because I hadn't installed the drivers right [19:51] i've reinstalled using Jockey like 3 times [19:51] 64-bit also, though [22:06] hi [22:07] can any X guru help out on #ubuntu+1 ? [22:07] cwillu isn't in today [22:08] Hey, how to I start X in failsafe mode? [22:08] I bet I broke my driver. [22:08] SnoFox: not here, please [22:09] this is not a support channel [22:09] when grub screen comes up hit the Esc key on your keyboard [22:09] O.o [22:11] bryce, sorry about pinging in the wrong channel, do you mind talking about intel and xorg for a couple minutes? [22:12] topic? [22:12] hey guys, guud evening [22:12] heya BUGabundo [22:12] why is it eating my CPU alive when I scroll? [22:12] I have an intel GMA 4500... its a decent card at least in intrepid [22:12] I got that a few times too [22:13] eats 100% of one core [22:13] causes my sound to start skipping [22:13] DBO, check lp, think there's a bug open on that [22:13] can you help me find it? [22:13] freezes are higher prio for me atm but if you have a patch I can look [22:13] I do not, I just want to understand why intel is regressing like its its job [22:14] see the IntelPerformance page I put in the X wiki to explain that [22:14] * DBO goes to look [22:15] bryce, one last question, is this going to be fixed or are intel users out to drift for a release? [22:16] bryce, http://wiki.x.org/wiki/FindPage?action=fullsearch&titlesearch=1&value=Intel am I missing something, I dont see an IntelPerformance page? [22:17] oh, its on the dri wiki =) [22:17] the Ubuntu-X wiki [22:17] http://wiki.ubuntu.com/X/Troubleshooting/ [22:17] oh, okay [22:19] as to your last question, is that just sarcasm or are you literally asking? [22:19] I am literally asking. Is it going to be fixed for jaunty [22:19] I am not trying to be insulting [22:19] make me +1 [22:20] heh, that question is sort of on par with "are we there yet??" ;-) [22:20] I always knew intel drivers a *good* place [22:20] this cycle it looks like a big washout [22:20] IF a fix is found, it will be fixed [22:20] so it's a bad bug, not drop of support? [22:20] IF no fix is found, it won't [22:20] I guess part of the frustration here is, why did we upgrade the X stack with all these regressions then? [22:21] DBO, read the IntelPerformance page [22:21] I am reading [22:22] oh I see, you addressed it [22:23] it is true the -intel drivers we've gotten have had rather poor QA. We're fixing up what we can in the distro, but there's only so much we can do at this level [22:23] of course [22:23] * jbarnes hides [22:23] heh [22:24] jbarnes has been helping tons. :-) [22:24] "I just want to understand why intel is regressing like its its job" [22:25] it's its [22:25] hehe [22:25] sorry =) [22:25] I'm adding regressions as fast as I can [22:25] a more sinister explaination would be that intel sells hardware ;) [22:25] jbarnes: damn cworth is trying to spoil all your good work [22:25] heh [22:25] GEM is a big source of regressions [22:26] since it hasn't seen much development on 8xx systems [22:26] we're hoping to fix a lot of those issues after the 2.7.0 release (i.e. for 2.7.1) [22:26] I think the movement of code into the kernel has also either caused regressions or at least made them a bit harder to get fixes for [22:26] (though one of the biggest issues was fixed recently in the kernel) [22:27] s/get fixes for/get fixes into Ubuntu for/ [22:27] is there any clear picture about whats causing the scrolling performance issues on the 4500 series? [22:27] GEM was insidious because now we have a few code paths to deal with in the 2D driver [22:27] pre-GEM, GEM, GEM+GTT, GEM+KMS [22:27] and then of course XAA, EXA and UXA on all of those [22:27] so we've been making our job a lot harder recently to add the features people want [22:27] DBO, EXA would be my guess, but hard to say [22:28] it exhibits the behavior on EXA and UXA [22:28] try XAA [22:28] I will do that [22:29] also if the memory map dump in the X log doesn't show "X tiled" for your front buffer, perf will be really bad [22:29] I didn't know XAA was viable for the 4500 series [22:29] heh, I'd doubt it could be called "viable" [22:29] but on other hw I've seen scrolling issues go away when EXA->XAA [22:30] XAA is pretty buggy on -intel (which is why we moved to EXA in the first place) [22:30] but some like it [22:31] jbarnes: btw, it seems we still have a scattering of freeze issues. I don't know if they're different bugs, or instances of the same one. At least some seem to have cropped up recently (perhaps due to patches added or reverted), but others have been going on a while [22:31] i wonder if the community could come up with a ppa with the intrepid xorg stack... [22:31] jbarnes: I'm going to try to get bug reports on freezes organized and tabulated, to see if I can categorize them better [22:31] bryce: I think cworth is hot on the trail of the biggest one, affecting 945 [22:32] jbarnes: meanwhile if you have tips... [22:32] DBO: xaa with compositing is fail, so it's not really viable given nobody's willing to fix it [22:33] jcristau, fair enough [22:33] when is 2.7.0 due? [22:33] DBO: can I see your X log from when it's slow? [22:33] DBO: supposedly tomorrow, but I'm not sure if we'll make it (still have a couple of blockers open) [22:34] jbarnes, yeah, I have to boot back over to jaunty to do it first (can't well develop GNOME Do without working X) [22:35] brb [22:36] heya RAOF! [22:38] http://paste2.org/p/181093 [22:38] jbarnes: side question... I've noticed with newer intel chips, lspci just says "Integrated Graphics Controller" - is there a better way to get the card's name? (GMA X4500HD, etc.) [22:38] Howbie [22:38] I forgot who it was that wanted my xorg log, but thats it [22:38] hey RAOF [22:38] shhh, I am pretending to be a clueless user like always [22:38] bryce: not sure... lspci gets the name info from the pci ids database [22:39] dunno if a newer one is available with friendlier names [22:40] DBO: hm looks reasonable... you're even getting uxa + dri2 [22:40] firefox scrolling is slow? [22:40] very [22:40] like if I flick my finger on the trackpad it will scroll for 3 or 4 seconds [22:40] DBO, do you see this only in firefox, or any other apps? [22:41] things that have pure text in the scroll box seem to be okay [22:41] it looks to be related to "complex" content or pixmaps [22:41] how about inkscape ? [22:41] a treeview scrolls mostly okay [22:41] will install [22:42] DBO: can you post dmesg too? [22:42] one second [22:42] you want since boot? [22:43] yeah something capturing the boot process [22:43] we've seen badness in the kernel's PAT/mtrr tracking recently [22:43] that definitely slows things down if it's wrong [22:44] oh the width of the window being scroll has a very definite impact on the performance [22:44] perhaps its more a function of the overall area [22:45] jbarnes, http://paste2.org/p/181100 [22:45] sorry for the delay again [22:46] that one doesn't have the early boot stuff [22:46] * jbarnes checks whether that's important [22:47] erm, I will double check [22:47] but I think thats all of it [22:47] you could try booting with the 'nopat' option [22:47] you are right [22:48] also /proc/mtrr [22:48] jbarnes, /proc/mtrr http://paste2.org/p/181108 [22:48] that's with X running? [22:49] yes [22:49] jbarnes, http://paste2.org/p/181109 thats a better kernel log [22:52] jbarnes, still wanting nopat? [22:53] it *might* help [22:54] i think debian disables PAT by default [22:54] hm [22:55] if that's the case I don't think you're getting write combining anywhere [22:55] since your /proc/mtrr was missing a WC entry for the framebuffer [22:55] and PAT isn't enabled so the kernel may not have done it for you [22:56] can you restate that in english? :D [22:56] X really wants write combined mappings of the framebuffer or everything gets really slow [22:57] not everything is rendered by the GPU; if X doesn't have write combining for what it does in software (includes a lot of the composite stuff that ff uses) it'll be super slow, every write will be uncached to memory [22:57] so how do I get WC? [22:58] pre-PAT you need to set up the MTRR regs to cover your graphics aperture [22:58] however if you have PAT, you can do it on a page by page basis [22:58] (when PAT isn't broken that is) [22:58] gimme a few minutes [22:59] * DBO waits patiently =) [23:03] checking libpciaccess & X for the mtrr setting bits [23:03] http://cateee.net/lkddb/web-lkddb/DRM_I915.html - huh, seems 2e02, 2e12, and 2e22 have the same generic name. [23:04] * bryce needs a decoder ring. 2e22 I think may be G45? [23:04] G43 and G41 the others maybe? [23:04] heh I'd have to look at the header files [23:05] ah good call [23:06] yeah 2e22 seems to be g45 [23:06] common.h:#define PCI_CHIP_IGD_E_G 0x2E02 [23:06] common.h:#define PCI_CHIP_G45_G 0x2E22 [23:06] common.h:#define PCI_CHIP_Q45_G 0x2E12 [23:06] common.h:#define PCI_CHIP_G41_G 0x2E32 [23:14] jbarnes, any luck? [23:14] DBO: well it looks like the kernel lies about having being able to write combine resources in sysfs w/o PAT [23:14] at least it doesn't go out of its way to make it happen [23:15] I've got a patch to libpciaccess which might help [23:15] okay [23:15] but only on systems with non-crappy mtrr layouts [23:15] is mine crappy? [23:15] no looks like it might work on yours [23:15] sweet [23:15] some systems ship these days assuming XP, which uses PAT [23:16] * DBO starts downloading the libpciaccess source [23:16] so they use all the MTRRs in weird ways [23:16] so PAT really is the right solution for many machines [23:16] but if it's disabled... :) [23:16] im not sure if it is, I just said that I read debian does that... [23:16] it looks like it's disabled [23:17] ah [23:17] is there a boot option to enable it? [23:17] bug 357908 [23:17] Launchpad bug 357908 in xserver-xorg-video-intel "[i965] X hang with EXA while scrolling in firefox" [High,Triaged] https://launchpad.net/bugs/357908 [23:17] thank you bryce, your launchpad fu is superior to mine [23:17] DBO: heh. i didn't know that. [23:18] $ grep X86_PAT /boot/config-2.6.29-1-amd64 [23:18] # CONFIG_X86_PAT is not set [23:18] indeed seems disabled on debian. [23:18] DBO: not if it's not built in to your kernel [23:18] DBO: nah just came across that randomly working on other stuff [23:19] jbarnes / jcristau I am willing to try a patch or a kernel rebuild or both or whatever you think will help [23:19] I'll sell me soul on ebay and donate the proceeds to the FSF if need be [23:20] DBO: ooh looks like my simple patch might help [23:20] ready to try it? [23:20] ready [23:20] got the source downloaded, the coolers are cool, the hot dogs are hot, its hammer time [23:20] mail addr for the patch? [23:21] jassmith@gmail.com [23:21] erm, or just paste2.org [23:22] DBO: sent [23:22] got it, applying [23:23] http://paste2.org:80/p/181137 for the curious [23:25] building package now [23:25] had to install some depends [23:26] do I need to reboot? [23:26] or just restart X? [23:26] for platforms with crappy mtrr layouts (kernel complains about "no more mtrrs") there's an enable_mtrr_cleanup boot option === jbarnes is now known as j [23:26] DBO: just X === j is now known as jbarnes [23:26] ugg [23:26] thank you [23:26] I will be right back [23:26] * jbarnes types in the wrong text widget [23:27] need to unload driver modules? [23:27] shouldn't need to [23:27] meh, I'll do it just in case [23:28] jbarnes, reg05: base=0x0d0000000 ( 3328MB), size= 256MB, count=1: write-combining [23:28] so far so good [23:29] i dont see any really improvement though [23:30] i'll see if I can do a bit of informal testing [23:30] sigh [23:30] bryce: which git commit is DBO running? [23:30] of xf86-video-intel I mean [23:30] no, its still killing my CPU on scroll [23:31] hm [23:31] 2:2.6.99.1+git20090402.fad714c4-0ubuntu0tormod2 0 [23:31] so git as of april 2 [23:32] still on UXA? [23:32] yeah [23:32] in fact thats pretty much the only thing of interest in my xorg.conf [23:32] thats the only Option [23:33] libdrm 2.4.7 has fixes, as does git after apr 2 [23:33] 2.4.6~git20090403.51d6346f-0ubuntu0tormod2 <--- libdrm [23:33] 6cd914ef315036ce8e91c7b6492994353e8ed2d8 in particular might help [23:34] I could patch that in if I knew how to get the diff [23:34] ah ok that libdrm should be fine [23:34] jbarnes: assuming he has stock ubuntu installed, that would be the 2.6.3 release (+patches) [23:34] i dont, I have xorg edgers ppa installed, that was the first thing I installed [23:34] oh, nevermind, xorg edgers [23:35] heh [23:35] plus I figure its not too useful to you guys if I dont at least try to get pretty close to upstream release [23:35] http://paste2.org:80/p/181147 [23:35] I know I hate when I get bug reports from GNOME Do 0.6 series now [23:35] that might make a big difference [23:36] DBO: yeah, right thing to do when working with jbarnes [23:36] making a package with that diff now jbarnes [23:36] bryce: unless you're a few days behind, that's ancient history to me :) [23:37] jbarnes, that diff is already in my current version [23:37] ok someone walk me through building a new xf86-video-intel deb [23:37] I guess the original package maintainer cherry picked it [23:38] DBO: arg [23:38] where can I get current source for this tree [23:38] I can build a deb [23:39] git://git.freedesktop.org/git/xorg/driver/xf86-video-intel [23:39] mmm git [23:39] git co? [23:39] git clone [23:39] thank you [23:40] * jbarnes makes room on his desk for the gm45 [23:40] jbarnes, I appreciate you debugging this with me [23:41] DBO: it's a report that keeps coming up [23:41] I'd like to nail it [23:42] well I'll stay with you until I have to go give this talk tonight [23:42] and then I'll be back for more [23:42] cool [23:43] building [23:43] please excuse if the name of the package is wrong [23:43] i forgot to rename the date [23:44] i'll fix that the second time I build it [23:45] ok let me try that again [23:46] ah damnit [23:46] the structure changed somewhere [23:46] np [23:46] I'm updating the gm45 to jaunty now [23:46] usr/share/xserver-xorg/pci/ [23:47] any idea what goes in there? [23:47] hm no [23:47] don't bother with that. [23:47] k [23:47] if it no matter, I'll just drop it from the package [23:47] (we install there a list of pci ids to help the server choose the right driver when none is configured; that's hopefully going away soon) [23:48] DBO: yeah you can do that [23:49] trying again [23:49] you must excuse me, my packaging fu is somewhat weak [23:49] RAOF only taught be enough to slaughter them terribly =P [23:49] for future reference, here are my notes for building debs of drivers (both for releases and git snapshots): https://wiki.ubuntu.com/X/DriverBuilding [23:49] ooh, nice =) [23:50] wewt [23:50] it built [23:51] jbarnes, ok, I have a package, its terrible but it should do the trick [23:51] really all you really need is './autogen.sh; make; sudo cp --remove-destination src/.libs/intel_drv.so /usr/lib/xorg/modules/drivers/; sudo killall X' :) [23:51] with one less really [23:51] restarting X [23:51] brb [23:52] hehe, somehow i killed composite [23:53] what happened? [23:53] oh, whitelisted driver check fail [23:53] i can override that [23:56] * jbarnes wishes he had a faster jaunty mirror to pull from [23:57] guess not... [23:58] its using the vesa driver now in xorg... [23:58] wonder why... [23:58] oh yuck [23:58] * DBO tries to remember how to force intel in xorg.conf [23:58] Driver "intel" [23:58] right? [23:58] yeah [23:58] brb [23:59] there we go [23:59] now its working :)