Bug#561733: libgl1-mesa-dri: Applications die due to failed DMA assertion

[Dave Witbrodt <dawitbro@sbcglobal.net>]

Package: libgl1-mesa-dri
Version: 7.7~rc2-1
Severity: important


There has been a bug in Mesa hitting some people since August.  As an
example, here is a dropping left by Extreme Tux Racer in
~/.xsession-errors after its window suddenly disappears while you are
playing:

    Extreme TuxRacer SVN Development --  http://www.extremetuxracer.com 
    (c) 2007 The ETRacer team
    (c) 2004-2005 The PPRacer team
    (c) 1999-2001 Jasmin F. Patry<jfpatry@sunspirestudios.com>
    ETRacer comes with ABSOLUTELY NO WARRANTY. This is free software,
    and you are welcome to redistribute it under certain conditions.
    See http://www.gnu.org/copyleft/gpl.html for details.

    IRQ's not enabled, falling back to busy waits: 2 0
    %%% etracer warning: Attempt to bind to Texture unloaded texture: `b-herring_run_icon'

    bo(0x1fee820, 4431872) is mapped (-1) can't valide it.
    invalid bo(0x1fee820) [0xE0357000, 0xE0358000]
    etracer: radeon_dma.c:210: radeonRefillCurrentDmaRegion: Assertion `dma_bo->bo->cref == 1' failed.

Since November, I have been experimenting with 2D/3D acceleration using
open source drivers for my Radeon HD 4850.  I have been trying to help
track down and provide info for all bugs which I am experiencing before
I switch to experimenting with KMS.  This bug is... was... the last one
I was trying to solve, but it looks like it is fixed upstream now.

My experiments have involved enabling "r600" support in Mesa, so I was
unable to report this bug on the Debian BTS because Debian's Mesa
packages had such support disabled.  With the recent upload of Mesa
7.7~rc2-1 (which enabled "r600" support) this bug can now be reported
here.

All versions of Mesa I have used with "r600" -- either self-built or
from Debian servers -- have exhibited this bug, including 7.7~rc2.  Last
night I built my own DEBs of 7.7-rc3:  they failed as well, with ETRacer
dying the same way.

Today, I updated my Mesa git repo and found this commit:

    commit 1c28073fdfb56a241424c739b57845f47fa05002
    Author: Dave Airlie <airlied@redhat.com>
    Date:   Thu Dec 17 14:19:27 2009 +1000

        radeon: drop assert accessing cref which is meant to be hidden

There had been some earlier commits on Nov. 24 which looked promising,
so I have been patching my local Mesa builds with those as well.  I
rebuilt 7.7-rc3 patched with two commits -- bd13e6e5... and
1c28073... (was trying for 2176b3e... as well, but other recent changes
significantly altered the code it would have applied to) -- and built
new DEBs from that.  Now ETRacer runs as much as I want without dying!

I believe that some (all?) of the Debian X Strike Force also works on
Ubuntu.  If that is correct -- and if this commit works as well for
everyone else as it does for me -- then you will want to pass the news
along to the readers of this Launchpad bug report,

    https://bugs.launchpad.net/ubuntu/+source/mesa/+bug/459961

and the fellow who posted comment #24 here:

    https://bugs.launchpad.net/ubuntu/+source/celestia/+bug/30970


I emailed David Airlie to thank him for this commit, and provided him
with, more or less, the same info.  This included reference to this
upstream freedesktop.org bug report which also may be solved by the new
commit:

    https://bugs.freedesktop.org/show_bug.cgi?id=23781


Please note this Debian 'extremetuxracer' bug report may be involved
with this bug:

    http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=557750

Compare the OP's error message warning about the "unloaded texture" to
the error messages I quoted at the top.  I will post a reply to that bug
report after I send this report, but there will be nothing that user can
test (for now) unless provided with a patched version of Mesa.


HTH,
Dave Witbrodt