Uses 100% CPU with latest mesa/libdrm update

Appears to be a combination of compiz+intel. Rick Spencer had this occur since dist-upgrading today. I am not reproducing on my upgraded compiz+amd system.

Still occurs using compiz 0.8.3 in PPA:
https://launchpad.net/~compiz/+archive/ppa

Can you install the dbgsym packages for compiz then attach gdb to the running compiz and use 'where' to see what it is doing? Also, is metacity still running at this time? There was a bug where metacity kept getting restarted by the session manager and trying to take over which makes compiz use lots of CPU, perhaps it is related.

This CPU hang trigers as easily as pressing Alt+Tab

Travis -
1. I did not see metacity running
2. I tried to install dbgsym, but the packages were out of sync

I agree with Travis that having some gdb data would help pinpoint where the failure is occurring.

Aside from that, there's a hypothesis that the recent mesa upgrade caused this issue. That can be tested by downgrading to the earlier version of mesa (7.5) and verifying that the issue goes away. You may still have the old mesa deps in your /var/cache/apt/archives/ ; if not, they can also be obtained from https://edge.launchpad.net/~ubuntu-x-swat/+archive/x-retro .

If it can be confirmed that the issue goes away when downgrading to mesa 7.5, then the next steps would be to move the bug report from compiz to mesa, and I will send it upstream to get priority attention.

I am not able to reproduce this on my 965 hardware updated to current karmic. (Also did not see the error on this hardware with karmic updated as of late last week.)

I reproduced it finally

I'm also having some trouble figuring out how to get the dbgsyms for compiz installed - Robert or Travis please give guidance on this.

After downgrading libdrm and mesa to the versions in the x-retro PPA, the issue seems to have gone away. Previously, compiz would freeze reliably after <5 min usage. With the x-retro ppa I've been running now 20 minutes with nary a problem.

I'm gathering that either there's a bug in mesa or libdrm, or else compiz needs to be updated in order to work properly with the new mesa.

I've reviewed the list of changes subsequent to the snapshot date for both our libdrm and mesa, and do not see an obvious commit that fixes this issue.

I've opened bug 420321 to add a -dbg package to compiz. In the meantime you can use the build from my PPA:
https://launchpad.net/~robert-ancell/+archive/ppa

I'm still not certain whether this is going to need a fix in compiz or in something mesa/libdrm/intel-ish, but I've gone ahead and filed the bug upstream with X.org at https://bugs.freedesktop.org/show_bug.cgi?id=23566 and flagged it to our Intel rep. The fact that it showed up after we upped mesa, and goes away after downgrading to 7.5 seems too suspicious for coincidence. (I'd like to hear if this holds true for others.)

FYI I don't see this issue on my mini 10v, which hasn't been upgraded since 25 August (so mesa 7.5)

I can reliable reproduce this on 965 as well. One thing that I've found that triggers it instantly is opening up gwibber (if that's of any help).

Downgrading to libgl1-mesa-dri 7.5-1ubuntu1 makes the issue go away for me as well (sorry for the noise, computer hung the in middle!)

On Fri, Aug 28, 2009 at 08:16:35PM -0000, Martin Albisetti wrote:
> I can reliable reproduce this on 965 as well. One thing that I've found
> that triggers it instantly is opening up gwibber (if that's of any
> help).

I wonder if the issue is particular to i965. Anyone reproduced it on
945?

Bryce

My netbook is 945, and I can't reproduce it.
It's using mesa 7.6 with everything up-to-date.

Interesting, I've reproduced the compiz 100% cpu freeze with mesa 7.1 as well.

Typo in previous comment. "7.1" should be "7.5-1ubuntu1"

From the upstream bug it appears to solve this issue will need a kernel fix. Leann and Andy, mind taking a look at including this patch for Karmic?

http://lists.freedesktop.org/archives/intel-gfx/2009-August/003981.html

I get this issue every time I use alt-tab since I've upgraded

Hi Guys,

Just a quick note that I'll build a Karmic test kernel with the patch Bryce mentioned from comment #22. I'll post a link to the build when it's ready. Thanks.

Sebastien, downgrade to mesa 7.5 (https://edge.launchpad.net/~ubuntu-x-swat/+archive/x-retro) and see if that solves it for you.

downgrading libgl1-mesa-dri=7.5-1ubuntu1 and restarting compiz fixes the issue

As noted in comment #24, I've built and posted a test kernel with the patch referenced in comment #22. Please test and let us know your results. Thanks.

http://people.canonical.com/~ogasawara/lp419264/

Thanks, I've installed and booted the lp419264 kernel with mesa 7.6 and so far am not able to reproduce the hang. I'll keep running it for a bit, since downgrading mesa seemed to fix it but the issue reappeared after a few hours heavy use.

Sebastien, it would be helpful if you could also test leann's kernel with the mesa 7.6 package and see if it still solves the hang you're seeing as well, because if it is, this will be the preferred fix for this bug.

As in comment 5, I trigger the problem by pressing alt+tab. I am using the xorg stack from xorg-edgers ppa. After installing Leann's kernel, I can not reproduce the problem.

the issue is fixed too there using the karmic libgl1-mesa-dri and Leann's linux version

to be clear the karmic libgl1-mesa-dri = current 7.6

Thanks everyone for the quick testing. Ogasawara said she'll get the patch queued for the kernel, so I think this bug is nearly in the bag now.

I've posted the request to the Ubuntu kernel-team mailing list that this be included in Karmic:

https://lists.ubuntu.com/archives/kernel-team/2009-September/007128.html

Tim just applied this patch to Karmic. It appears this patch also just hit upstream so we'll only need to carry this until the next rebase with upstream.

ogasawara@emiko:~/linux-2.6$ git log -p a09ba7faf75fa4b21980d81de8e5f3d5c0785ccf
commit a09ba7faf75fa4b21980d81de8e5f3d5c0785ccf
Author: Eric Anholt <email address hidden>
Date: Sat Aug 29 12:49:51 2009 -0700

    drm/i915: Fix CPU-spinning hangs related to fence usage by using an LRU.

Hi,

Can anyone please advice us on how to apply this patch in our kernel?

00:00.0 Host bridge: Intel Corporation Mobile PM965/GM965/GL960 Memory Controller Hub (rev 03)
00:02.0 VGA compatible controller: Intel Corporation Mobile GM965/GL960 Integrated Graphics Controller (rev 03)
00:02.1 Display controller: Intel Corporation Mobile GM965/GL960 Integrated Graphics Controller (rev 03)

prageeth@prageeth-laptop:~$ lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description: Ubuntu karmic (development branch)
Release: 9.10
Codename: karmic
prageeth@prageeth-laptop:~$ uname -a
Linux prageeth-laptop 2.6.31-9-generic #29-Ubuntu SMP Sun Aug 30 17:39:26 UTC 2009 x86_64 GNU/Linux

Every time I went back from Suspend compiz went crazy , no with the kernel above it is working normal.
I have some other problem as in xournal if I hold mouse pointer on a button it couldn't show the text about it and the hole screen just flickered as going on and off very fast , now it is working normal , so the problem was bigger than just some hangs.

In the latest Kernel update (2.6.31-10.30), Compiz seems to work without issues.

That's a negat. For me mesa 7.6.0~git20090817.7c422387-0ubuntu2 is still no go. Even after kernel upgrade to 2.6.31-10 I got kernel panic when moved mouse while glMatrix screensaver was running. Next time glMatrix just froze and it was better I could do SysRq + RSIUB kungfu.
Mine bug was #419264 marked as dupe of this one maybe it's another issue then.

Indeed the patch noted in comment #35 should be in the latest 2.6.31-10.30 kernel. I'm marking the linux kernel task as Fix Released.

@Łukasz Kuryło, I'm going to undup your bug from this as it does seem you are experiencing a different issue which was not resolved by this patch. Thanks.

This was fixed in the 2.6.31-10.30 kernel and we've got 2.6.31-10.32 now so this is Fix Released.