graphics lockups with radeon DRI (radeon 9000/9100/9200)

[Michal Suchanek]

With radeon 9100 or 9250 OpenGL applications lock up the card if DRI enabled.

On Ubuntu tested with xorg 7.0 with both cards, xorg 7.1 with 9100.

The X server locks up (no crash), the X applications keep running, the sytem is accessible over the net, the onscreen picture does not change until reboot, keboard input does not work.

Originally seen on Gentoo but Ubuntu is affected as well.

https://bugs.freedesktop.org/show_bug.cgi?id=5986

[Michal Suchanek (hramrach)]

Disabled RenderAccer and got two lockups in moebius very fast.
Disabling dri now.

It looks like xserver is not the right component although it does happen with
the modular X server. Perhaps dri is for the driver bugs?

[Michal Suchanek (hramrach)]

With dri disabled I get no lockups.

Looks like a dri bug.

Is there anything the X server produces that would make it clearer why it locks up?

[Felix Kuehling (fxkuehl)]

(In reply to comment #2)
> With dri disabled I get no lockups.
>
> Looks like a dri bug.
>
> Is there anything the X server produces that would make it clearer why it
locks up?

Not really. But there have been recent changes in CVS that address lockup
problems on Radeons. You can try installing a binary snapshot on top of Xorg 7:
http://dri.freedesktop.org/wiki/Download#head-55420c59a1c2e9a70f07a6fa02f0d228ffb87b76
Or build the radeon driver from CVS.

[Michal Suchanek (hramrach)]

(In reply to comment #3)

> Not really. But there have been recent changes in CVS that address lockup
> problems on Radeons. You can try installing a binary snapshot on top of Xorg 7:
>
http://dri.freedesktop.org/wiki/Download#head-55420c59a1c2e9a70f07a6fa02f0d228ffb87b76
> Or build the radeon driver from CVS.

The binary snapshots do not work for me, and as far as I can tell there are only
build instructios for the dri cvs for use with the monolithic X server, not the
modular one.

Perhaps they will fix the snapshots eventually.

[Felix Kuehling (fxkuehl)]

(In reply to comment #4)
>
> The binary snapshots do not work for me, and as far as I can tell there are only
> build instructios for the dri cvs for use with the monolithic X server, not the
> modular one.
>
> Perhaps they will fix the snapshots eventually.
>

Binary snapshots should work with modular Xorg. I fixed that about two weeks
ago. If the snapshot installation is giving you trouble, please let me know, I
am the maintainer. And last time I checked someone updated the build
instructions for modular Xorg as well.

[Michal Suchanek (hramrach)]

With the new 20060322 driver and the 1.0.2 server X still locks up.

[Michal Suchanek (hramrach)]

Testing with 7.1 and Radeon 9250 (rv280) results in the other bug about X
crashing without DRI.

7.1 with Radeon 9100 (r200) still locks up for me.

ii libdrm2 2.0.2+git20060809-0ubuntu1 Userspace interface to
kernel DRM services -
ii libgl1-mesa-dri 6.5.1~20060817-0ubuntu2 A free implementation of
the OpenGL API -- D
ii libgl1-mesa-glx 6.5.1~20060817-0ubuntu2 A free implementation of
the OpenGL API -- G
ii xserver-xorg-core 1.1.1-0ubuntu11 X.Org X server -- core server
ii xserver-xorg-video-ati 6.6.2-0ubuntu2 X.Org X server -- ATI
display driver

The lockup can be quickly reproduced by running StepMania - I haven't seen this
run for more than a few minutes. The same could be probably achieved by some of
the xscreensaver GL hacks. I had to disable xscreensaver to prevent computer
lockups.

The symptoms are always the same: the X server locks up, the picture on the
monitor remains unchanged until reboot. The X applications keep running, and at
least some still preform functions that do not require the X server.

[Michel-tungstengraphics (michel-tungstengraphics)]

(In reply to comment #7)
>
> 7.1 with Radeon 9100 (r200) still locks up for me.

Can you try Mesa CVS (or at least the final 6.5.1 release) and possibly DRM and
xf86-video-ati git?

[Michal Suchanek (hramrach)]

With radeon 9100 or 9250 OpenGL applications lock up the card if DRI enabled.

On Ubuntu tested with xorg 7.0 with both cards, xorg 7.1 with 9100.

The X server locks up (no crash), the X applications keep running, the sytem is accessible over the net, the onscreen picture does not change until reboot, keboard input does not work.

Originally seen on Gentoo but Ubuntu is affected as well.

https://bugs.freedesktop.org/show_bug.cgi?id=5986

[Michal Suchanek (hramrach)]

Linux 2.6.18, DRM git, Mesa cvs, and ati driver git:

1st try: glxgears very slow, StepMania produces black screen only
(hard reboot)
2nd try: Xorg.0.log says DRI initialized fine but produces lots of warning about
visuals unsupported by dri driver.
glxinfo says rendering is indirect
glxgears locks up X even before it's window is shown. It looks like the same
thing as previosly - I can normally shut down the machine using the power button
(acpid), the picture displayed on the screen does not change until poweroff.

[Michal Suchanek (hramrach)]

I tried again with drm/mesa/ati as of monday/tuesday. I could run an OpenGL
application for a few hours which is certainly an improvement in stability over
the versions that crashed within minutes. This is good enough for games.

However, I would have to test for much longer to see if the original problem is
solved - if the driver is stable enough for desktop use.

[Michal Suchanek (hramrach)]

With xorg 7.2 rc3 and mesa git I got a lockup in one of the hyper- (hyperball?)
xscreensacer hacks. This one locked up the machine completely, I could not log
in over network.

It took about a day to lock up but running that hack from the start would
probably give faster result.

I am using 2.6.17 kernel and libdrm 2.3.0.
(II) RADEON(0): [dri] Found DRI library version 1.3.0 and kernel module version
1.24.0

There are lots of
(EE) RADEON(0): GetBuffer timed out, resetting engine...
(**) RADEON(0): EngineRestore (32/32)
at the end of the log but I have no idea if that is related since there are no
timestamps.

[Michel-tungstengraphics (michel-tungstengraphics)]

Both hypercube and hyperball use only lines, so it sounds like there's an issue
with line acceleration. Can you try disabling it with

    Option "XaaNoSolidBresenhamLine"
    Option "XaaNoSolidTwoPointLine"

(check the log file for which one(s) of these disable line acceleration) to
verify this?

[Michel-tungstengraphics (michel-tungstengraphics)]

Also, does this only happen with the DRI enabled, or also with it disabled?

Please also attach the full X log file.

This was also reported to
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=301531, also with a Radeon
9200. Roland, can you reproduce this with your RV280 card?

[Timo Aaltonen (tjaalton)]

duplicates: sorry for mixing this up with bug 67487, which is a different issue.

This absolutely needs to be addressed upstream before feisty is released.

[Michal Suchanek (hramrach)]

Without DRI the system appears pretty stable, I could run for more than a week without problems.

With DRI and the options disabling line acceleration the line drawing is _very_ slow (probably even slower than without DRI but with line accel). The visual glitches are gone, only the hyperball is dispalyed, no leftover line segments are present.

I could run hyperball for a few days continuosly without lockups.

[Michal Suchanek (hramrach)]

Created an attachment (id=8533)
Xorg log

Fresh Xorg log. I did not save the log of the crashed server.

[Timo Jyrinki (timo-jyrinki)]

However, many lockups may still be caused by various AGP mode / fast write problems, even though some of the reports are not directly related to 67487. Things to do, and please eg. refer with a number always in relation to what you've tried, and tell what card you are using:

1. Some of the problems are fixed in Herd-4, so the first thing one should do is to make sure you still can hang the machine with Herd-4. Success reports plus card model are welcome. http://cdimage.ubuntu.com/releases/feisty/herd-4/
2. Next, you could try the latest GIT version of xserver-xorg-video-ati, which has even more fixes to similar problems. I put compilation instructions (typo in the apt-get command though) near the end of bug 28925.
3. The upstream bug linked to this bug (fd.o 5986) links to Debian bug report that has a recently posted workaround that fixes the problems for some. Please try that on top of the current feisty driver, maybe that could be applied for feisty too: http://bugs.debian.org/cgi-bin/bugreport.cgi/radeon-rv280-cp-twopointlines.diff?bug=301531;msg=86;att=1

These would be good things to check on at least a) a 9100/9250 AGP card / machine, and also with some b) Radeon 7000/7200/7500 or c) early Radeon Mobility class machine if possible, just to get perhaps new information. No guarantees if any of the fixes fix these random hangs, but it'd be good to find out.

So when we have either 1, 2 or 3 working for all of a, b and c, we have a lot of new information :) (the bug 67487 states that radeon 9500/9600 lockups should be fixed in herd-4 anyway)

[Michal Suchanek (hramrach)]

With xorg just after 7.2 RC release and line acceleration disabled Radeon 9250 is stable.

Looks like the line acceleration is broken and causes most of the lockups.

[Timo Jyrinki (timo-jyrinki-hut)]

The Debian bug report now has a patch (a workaround) for this. Please test if it fixes the issue for you.

[Timo Jyrinki (timo-jyrinki)]

I believe so too. Can anyone confirm if it fixes lockups also on Radeon 7000-series or the Radeon Mobility M6 series (ie. 7000-series class Mobility)? So far I think the line acceleration issue affects only 9000/9100/9200-series.

[Daniel Stone (daniels)]

Sorry about the phenomenal bug spam, guys. Adding xorg-team@ to the QA contact so bugs don't get lost in future.

[Timo Jyrinki (timo-jyrinki)]

It looks likely (https://bugs.freedesktop.org/show_bug.cgi?id=3606) that the workaround from Debian's bugzilla, ie. the line acceleration fix, does not fix problems with 7000 / Mobility M6, only 9000/9100/9200. But for the latter cards, maybe the patch could be put in for Ubuntu.

I now separated the 7000/M6-series problems to bug 15219 from this bug's duplicates.

[Timo Jyrinki (timo-jyrinki)]

To put all the URLs here, the Debian bug report with the patch and "works now"-reports is http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=301531

[Łukasz Derkacz (lderkacz)]

Did anyone made patched package for Edgy? ( I mean about 9100)?

[Timo Jyrinki (timo-jyrinki)]

The patch mentioned in the Debian bug report has now been uploaded to feisty, thanks to Timo Aaltonen. Marking as Fix Released - please note that this bug solely concerns the "RV250"-series of cards, and there are other bugs for other Radeon cards. If you own one of the duplicate bugs and think it's different from this, please reopen that bug, not this one.

Version: 1:6.6.3-2ubuntu3
Distribution: feisty
Urgency: low
Maintainer: Ubuntu Core Developers <ubuntu-devel-discuss at lists.ubuntu.com>
Changed-By: Timo Aaltonen <tepsipakki at ubuntu.com>
Description:
 xserver-xorg-video-ati - X.Org X server -- ATI display driver
Changes:
 xserver-xorg-video-ati (1:6.6.3-2ubuntu3) feisty; urgency=low
 .
   * debian/patches:
     - 104_radeon_rv280_cp_twopointlines.diff
       A patch from debian that makes RV280 chips not to lock up with
       certain 3D-apps.
       (Closes: LP #63503)
     - 105_fdo_att7409_bug5437.diff
       Fix identification with Mobility Radeon X700.
       (Closes: LP #22985)
     - 106_radeon_predownscale_to_make_hd_video_work.diff
       Xv scaling patch from upstream git for big resolutions.
       (Closes: LP #74948)
Files:
 57ccb22ba9cb3b08d4a998c77f7e34e3 1218 x11 optional xserver-xorg-video-ati_6.6.3-2ubuntu3.dsc
 90c8c444863912ccc0b3cc12f345e379 28967 x11 optional xserver-xorg-video-ati_6.6.3-2ubuntu3.diff.gz

[Michal Suchanek (hramrach)]

It looks like the patch does the same as the options. Turns off line acceleration that causes the lockups.

[Michal Suchanek (hramrach)]

I guess this is fixed. The fact that line acceleration does not in fact work is not important from this point of view. Was the workaround checked into xorg?

[Michel-tungstengraphics (michel-tungstengraphics)]

(In reply to comment #19)
> Was the workaround checked into xorg?

Yes.