[r300] screensaver 'lattice' causes immediate Xorg freeze/crash

[Øyvind Stegard]

Binary package hint: libgl1-mesa-dri

When activating screensaver 'lattice', either by previewing in the Gnome screensaver preferences or executing the binary directly (/usr/lib/xscreensaver/lattice), Xorg immediately freezes/crashes every time. Machines needs to be rebooted, SSH does not work (hard hang). It does not matter whether Compiz is enabled or not (crash happens with Metacity as well).

I have so far been unsuccessful in trying to get at any crash output (or panic, if that's the case), because of the hard freeze. So I am hoping someone else might be able to reproduce this as well.

- Jaunty beta (tracking/current)
- Using radeon Xorg driver (xserver-xorg-video-radeon 6.12.1-0ubuntu2)
- Kernel linux-image-2.6.28-11-generic 2.6.28-11.41
- Mesa 7.4-0ubuntu1
- Xorg 7.4~5ubuntu18

Hardware:
- ATI X1400 Radeon Mobility, 128MB RAM
- Intel Core Duo 2GHz, 2GB RAM

Will attach Xorg.0.log, kernel log and output of lspci. The Xorg config file is uninteresting since it does not contain anything (default).

[Øyvind Stegard (oyvindstegard)]

rss-glx 0.8.2-1ubuntu3

[Øyvind Stegard (oyvindstegard)]

Not really sure if the problem is in the graphics DDX driver, could of course be general Xorg, DRI, radeon kernel module etc. The lattice screensaver has worked fine in earlier Ubuntu versions on the same hardware.

[Øyvind Stegard (oyvindstegard)]

I realize the difficulty in tracking this problem without any traces, so I'll try to get some kind of trace later (providing that I can manage to log into the machine after the crash).

[Øyvind Stegard (oyvindstegard)]

When activating screensaver 'lattice', either by previewing in the Gnome screensaver preferences or executing the binary directly (/usr/lib/xscreensaver/lattice), Xorg immediately freezes/crashes every time. Machines needs to be rebooted, SSH does not seem to work after crash (probably hard hang). It does not matter whether Compiz is enabled or not (happens with Metacity as well).

This error message is hard to produce because of the immediate crash, but I managed after some attempts. Here it is:

$ /usr/lib/xscreensaver/lattice
*********************************WARN_ONCE*********************************
File r300_state.c function r500SetupRSUnit line 1907
Don't know how to satisfy InputsRead=0x00000008
***************************************************************************

And then complete freeze.

- Jaunty beta (tracking/current)
- Using radeon Xorg driver (xserver-xorg-video-radeon 6.12.1-0ubuntu2)
- Kernel linux-image-2.6.28-11-generic 2.6.28-11.41
- Mesa 7.4-0ubuntu1
- Xorg 7.4~5ubuntu18
- libdrm2 2.4.5-0ubuntu4

Hardware:
- ATI X1400 Radeon Mobility, 128MB RAM
- Intel Core Duo 2GHz, 2GB RAM

Will attach Xorg.0.log, kernel log and output of lspci. The Xorg config file is uninteresting since it does not contain anything (default).

This bug was originally reported downstream here:
https://bugs.launchpad.net/ubuntu/+source/xserver-xorg-video-ati/+bug/361649

Kernel log:
http://launchpadlibrarian.net/25488287/dmesg.txt

lspci:
http://launchpadlibrarian.net/25488303/lspci.txt

Xorg log:
http://launchpadlibrarian.net/25488314/Xorg.0.log

Or would you prefer I attached these files directly to this report ?

[Øyvind Stegard (oyvindstegard)]

This error message is hard to produce because of the immediate crash, but I managed after some attempts. Here it is:

$ /usr/lib/xscreensaver/lattice
*********************************WARN_ONCE*********************************
File r300_state.c function r500SetupRSUnit line 1907
Don't know how to satisfy InputsRead=0x00000008
***************************************************************************

And then complete freeze.

[Øyvind Stegard (oyvindstegard)]

I reported this bug upstream:
https://bugs.freedesktop.org/show_bug.cgi?id=21223

[Øyvind Stegard (oyvindstegard)]

Upstream bug report now suggests this is a problem with Radeon/r300 Mesa DRI driver, so I set the affected package to that.

[Maciej Cencora (m-cencora)]

It should be fixed in current master, can you confirm?

[Maciej Cencora (m-cencora)]

Current mesa master of course :)

[Øyvind Stegard (oyvindstegard)]

(In reply to comment #1)
> It should be fixed in current master, can you confirm?
>

Is there any easy way of testing this without screwing up my default Ubuntu Jaunty installation :) ?

[Øyvind Stegard (oyvindstegard)]

(In reply to comment #3)
> (In reply to comment #1)
> > It should be fixed in current master, can you confirm?
> >
>
> Is there any easy way of testing this without screwing up my default Ubuntu
> Jaunty installation :) ?
>

I'm asking because the affected hardware is actually my work/coding laptop, so I cannot afford to mess too much with it timewise. Is it only necessary to install a newer set of radeon.ko/drm.ko or do I need to update libs as well ?

[Michel Dänzer (michel-daenzer)]

No need to install anything, just set LIBGL_DRIVERS_PATH to the directory containing the self-built r300_dri.so. You can also set LIBGL_DEBUG=verbose to verify the right one is getting picked up.

[Øyvind Stegard (oyvindstegard)]

(In reply to comment #5)
> No need to install anything, just set LIBGL_DRIVERS_PATH to the directory
> containing the self-built r300_dri.so. You can also set LIBGL_DEBUG=verbose to
> verify the right one is getting picked up.
>

Very nice approach indeed, I'll get this tested.

[Øyvind Stegard (oyvindstegard)]

Didn't really work too well with the r300_dri.so driver from Mesa master:

oyvind@blackelf:~/tmp/build/mesa$ cd src/mesa/drivers/dri/r300/
oyvind@blackelf:~/tmp/build/mesa/src/mesa/drivers/dri/r300$ export DISPLAY=:0
oyvind@blackelf:~/tmp/build/mesa/src/mesa/drivers/dri/r300$ export LIBGL_DRIVERS_PATH=$PWD
oyvind@blackelf:~/tmp/build/mesa/src/mesa/drivers/dri/r300$ export LIBGL_DEBUG=verbose
oyvind@blackelf:~/tmp/build/mesa/src/mesa/drivers/dri/r300$ /usr/lib/xscreensaver/lattice
libGL: XF86DRIGetClientDriverName: 5.3.0 r300 (screen 0)
libGL: OpenDriver: trying /home/oyvind/tmp/build/mesa/src/mesa/drivers/dri/r300/tls/r300_dri.so
libGL: OpenDriver: trying /home/oyvind/tmp/build/mesa/src/mesa/drivers/dri/r300/r300_dri.so
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is 4, (OK)
drmOpenByBusid: Searching for BusID pci:0000:01:00.0
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is 4, (OK)
drmOpenByBusid: drmOpenMinor returns 4
drmOpenByBusid: drmGetBusid reports pci:0000:01:00.0
libGL: Can't open configuration file /etc/drirc: No such file or directory.
libGL: Can't open configuration file /home/oyvind/.drirc: No such file or directory.
*********************************WARN_ONCE*********************************
File r300_mem.c function r300_mem_alloc line 225
Ran out of GART memory (for 1048576)!
Please consider adjusting GARTSize option.
***************************************************************************

Error: R300 timed out... exiting

Then X freeze.

I don't know how well the code from Mesa master in r300_dri.so is supposed to interact with the Mesa-installation in Ubuntu Jaunty ? Also, I got a build error when building the entire Mesa tree, but still managed to build the r300 DRI driver by itself.

[Bugzi09-fdo-tormod (bugzi09-fdo-tormod)]

Øyvind, you can also try the mesa packages from the xorg-edgers PPA https://launchpad.net/~xorg-edgers/+archive/ppa - the packages for Karmic are from current master head, and they will install fine in Jaunty. In addition you can install the Karmic kernel, which is based on 2.6.30-rc4.

[Øyvind Stegard (oyvindstegard)]

(In reply to comment #8)
> Øyvind, you can also try the mesa packages from the xorg-edgers PPA
> https://launchpad.net/~xorg-edgers/+archive/ppa - the packages for Karmic are
> from current master head, and they will install fine in Jaunty. In addition you
> can install the Karmic kernel, which is based on 2.6.30-rc4.
>

Ok, I did the Ubuntu LiveCD+xorg-edgers-live-test script procedure, the upgrade worked fine. However, the screensaver crashes exactly the same as with a standard Ubuntu installation. I don't know if the upgrade included the supposed fix. I am unfortunately thus not able to confirm at this moment, but will look more into this later. I'll see if I can fix the Mesa build error and try again with the r300_dri.so driver, it did indeed give a different error. My main problem is, again, that the affected hardware is not a testing/play-machine, but my work-laptop which I need to keep in something of a working condition.

[Øyvind Stegard (oyvindstegard)]

(In reply to comment #8)
> Øyvind, you can also try the mesa packages from the xorg-edgers PPA
> https://launchpad.net/~xorg-edgers/+archive/ppa - the packages for Karmic are
> from current master head, and they will install fine in Jaunty. In addition you
> can install the Karmic kernel, which is based on 2.6.30-rc4.
>

I booted an Ubuntu LiveCD, added the Karmic repo at https://launchpad.net/~xorg-edgers/+archive/ppa and did an apt-get dist upgrade. This updated at least Mesa with versions from xorg-edgers. Did an X restart and tested the lattice screensaver, but it still immediately freezes.

Tested with the following Mesa-versions:
libgl1-mesa 7.6.0~git20090502.986d4a9e-0ubuntu...
libgl1-mesa-dri 7.6.0~git20090502.986d4a9e-0ubuntu...
libgl1-mesa-glx 7.6.0~git20090502.986d4a9e-0ubuntu...
libglu1-mesa 7.6.0~git20090502.986d4a9e-0ubuntu...

[Michel Dänzer (michel-daenzer)]

(In reply to comment #7)
> *********************************WARN_ONCE*********************************
> File r300_mem.c function r300_mem_alloc line 225
> Ran out of GART memory (for 1048576)!
> Please consider adjusting GARTSize option.
> ***************************************************************************
>
> Error: R300 timed out... exiting
>
> Then X freeze.
>
> I don't know how well the code from Mesa master in r300_dri.so is supposed to
> interact with the Mesa-installation in Ubuntu Jaunty ?

The above isn't related to that, if anything it's the X driver which isn't setting up enough GART memory (hence the suggestion to increase it with Option "GARTSize"). Can you attach the X log file?

Did you already see the above output when you reported this bug? Are you still seeing the r500SetupRSUnit output with the current driver?

[Xorg (xorg)]

A bit more data:

* This also happens on a Radeon Mobility X700 card
  (system is Ubuntu Jaunty; all updates applied)

* Different behaviour occurs with the various --preset options:

lattice --preset 1
  => kills the X server, can recover with Alt-SysRQ-K

lattice --preset 2
  => seems to work OK

lattice --preset 3
  => kills system stone dead (need a power-cycle to recover)
     #surely an X-server bug shouldn't kill the kernel?

lattice --preset 4
  => works for a few seconds, then stone dead.

Of the other screensavers in the rss_glx set, most of them work just fine:
hyperspace, helios, euphoria, skyrocket, solarwinds, plasma, fieldlines, flocks. Two of them seem to do abolutely nothing , but just stay black: cyclone and flux.

The RSS set seem to be pretty good at exposing fatal bugs in ATI drivers - I recall my Rage128 mobility being locked solid by one particular invocation of Euphoria about 5 years ago. I would suggest that they might make quite a good
reliability test for new drivers.

[Bugzi09-fdo-tormod (bugzi09-fdo-tormod)]

Confirmed on another X700 Mobility, with latest radeon-rewrite mesa and kernel 2.6.30-rc4.

[Mike Hicks (hick0088)]

I also experienced a GUI crash with the lattice screensaver on what lspci reports is a "ATI Technologies Inc RV370 5B60 [Radeon X300 (PCIE)]".

Fortunately, my system didn't do a hard hang, and I was still able to use an existing SSH session to manage the box a little and determine which process had hung the X server. Unfortunately, neither killing the lattice process nor the X server was able to bring the GUI back to life, so I had to shut down the box.

I would reiterate a comment that has been made on the Freedesktop.org bug: Please, please, please use the xscreensaver-gl and rss-glx screensavers as test programs to ensure that video drivers and mesa libraries are functional before releasing updates. People like pretty screensavers, so they get enabled on *lots* of boxes in the real world, so any bugs that get triggered by them will affect many, many people.

Thanks.

[Mike Hicks (hick0088)]

I also had Xorg hang today with the "flux" screensaver, also from the rss-glx package. The X server (well, some remnant of it) also ended up hammering my CPU with 99.9% usage after I attempted to kill it, so any operations from my command-line SSH session went very slowly.

I would like to simply disable the rss-glx screensavers, but the gnome-screensaver applet doesn't let me disable them individually (see LP# 31979: https://bugs.launchpad.net/gnome-screensaver/+bug/31979). I am going to try to disable them via a gconf-editor workaround I added to that bug report, but I may end up removing the rss-glx package if gnome-screensaver still attempts to run these binaries.

rss-glx is a dependency of the ubuntu-desktop pseudo-package, so I'd have to remove ubuntu-desktop as well. I'd rather not do that, since it can assist in upgrades by guiding the system to load appropriate packages when things split or get replaced by alternate programs.

[Tormod Volden (tormodvolden)]

We hope to get this fixed in the graphics driver soon. Disabling a screensaver is a workaround (and maybe a needed one with the current state of affairs), but the bug might kick in somewhere else as well. Mike, I am not totally sure your bugs are the same as Øyvind's, so I would recommend you open new bug reports if you have more debugging information on them.

[Oibaf (oibaf)]

This bug may be related to https://bugs.freedesktop.org/show_bug.cgi?id=21849

[Bugzi09-fdo-tormod (bugzi09-fdo-tormod)]

Maybe related, but the "bad" commit 4247ca942a16745da3d09c58996b276d02655a72 in 21849 never made it into the 2.6.28 kernel. I think the lattice issues here are even older. Looking forward to try the new kernel though.

[Michel Dänzer (michel-daenzer)]

It would be nice to get a response to comment #11...

[Oibaf (oibaf)]

(In reply to comment #15)
> Maybe related, but the "bad" commit 4247ca942a16745da3d09c58996b276d02655a72 in
> 21849 never made it into the 2.6.28 kernel. I think the lattice issues here are
> even older. Looking forward to try the new kernel though.

I am having that issue with the Ubuntu 2.6.28 kernel which includes a backported radeon module.

A way to confirm it would be to try kernel 2.6.29.4 from http://kernel.ubuntu.com/~kernel-ppa/mainline/?C=M;O=D that is the latest kernel without the bad commit.

[Oibaf (oibaf)]

This bug may be related to https://bugs.freedesktop.org/show_bug.cgi?id=21849 .
A fix planned for 2.6.30 final is in drm-fixes: http://git.kernel.org/?p=linux/kernel/git/airlied/drm-2.6.git;a=commitdiff;h=9863871bd1bbf218b921af5e0bc48ca4f6ea9f12

[Oibaf (oibaf)]

Is this bug still reproducible with Ubuntu 9.10 (including mesa 7.6 and linux 2.6.31)?

[Oibaf (oibaf)]

Original reported of Ubuntu bug confirms this is no longer reproducible. Closing.

[Oibaf (oibaf)]

Is this still reproducible with Ubuntu 9.10?

[Øyvind Stegard (oyvindstegard)]

Fabio Pedretti <email address hidden> writes:

> Is this still reproducible with Ubuntu 9.10?

No.

[Oibaf (oibaf)]

Closing, then.