[RV280] runaway Xorg in drmDMA()

[Akkana Peck]

Binary package hint: xserver-xorg-video-ati

New after upgrading to jaunty (no problem with intrepid on this machine):

X will work for a while, but eventually (could be minutes or hours after booting) it suddenly jumps to 90+% CPU. I can still move the mouse, sort of (it's laggy and jumpy) but mouse clicks don't work, keyboard events don't work, and apps apparently stop redrawing.

This doesn't seem to be in response to anything in particular -- nothing intensive going on in Firefox, no flash or animations, all other windows are pretty simple (xterm, xchat, emacs).

Video card (from lspci):
01:00.0 VGA compatible controller: ATI Technologies Inc RV280 [Radeon 9200 SE] (rev 01)
01:00.1 Display controller: ATI Technologies Inc RV280 [Radeon 9200 SE] (Secondary) (rev 01)

I can still ssh in from another machine, but I can't kill (even with -9) the window manager or Xorg -- killing has no effect. All I can do is reboot.

If I attach to the Xorg process and check every few seconds, the trace each time looks like:
(gdb) where
#0 0xffffe430 in __kernel_vsyscall ()
#1 0xb7d64d59 in ioctl () from /lib/tls/i686/cmov/libc.so.6
#2 0xb79e4a6d in drmDMA () from /usr/lib/libdrm.so.2
#3 0xb795c907 in ?? () from /usr/lib/xorg/modules/drivers//radeon_drv.so
#4 0xb795cc39 in ?? () from /usr/lib/xorg/modules/drivers//radeon_drv.so
#5 0xb79b2084 in ?? () from /usr/lib/xorg/modules/drivers//radeon_drv.so
#6 0xb7798998 in exaCompositeRects () from /usr/lib/xorg/modules//libexa.so
#7 0xb779418c in exaGlyphs () from /usr/lib/xorg/modules//libexa.so
#8 0x08180465 in ?? ()
#9 0x0816ea42 in CompositeGlyphs ()
#10 0x0817bb8e in ?? ()
#11 0x08174fd5 in ?? ()
#12 0x0808d5af in Dispatch ()
#13 0x0807231d in main ()

I see xserver-xorg-video-ati-dbg in the package list, and would be happy to install it if it would help.

[Philipp Kießler (pocytac)]

Same Problem with my Laptop too.
System is Dell Latitude D600 with ATI Mobility Radeon 9000 (RV250) run by radeon. Though Xorg consumes over 90% CPU-time when in use the system doesn't freeze. Windows are reacting very, very slow (with artefacts, blank screens and so on) but keyboard and mouse are generally usable.

By now I didn't install xserver-xorg-video-ati-dbg and I'm not experienced with bugtracking. But if there's anything I can contribute, let me know.

[Philipp Kießler (pocytac)]

I did a bit of testing and discovered an interesting point: only Qt4-applications are extremly slow, but not from the start but after about three quarters of a minute to one minute. When I'm using e.g. yakuake or opera, which are both still Qt3-applications, I do not have any problems with GUI-speed, artefacts or anything else.

I also had a look into my X.org-Logs. I could not find anything extraordinary...

[Rolf Leggewie (r0lf)]

I'm sorry, I may have jumped the gun a bit. But your description sounds very much like that of bug 363238. Please test if XAA improves your situation. If not, let's undupe.

[Martin Olsson (mnemo)]

@Akkana Peck, can you please install these packages:
xserver-xorg-video-ati-dbg xserver-xorg-core-dbg libgl1-mesa-dri-dbg libgl1-mesa-dri-glx
...and then sample 2-3 stacks from xorg while it's spinning the CPU and attach that to this bug.

Can you also try separately if:
Option "AccelMethod" "XAA"
...makes the problems go away.

And then remove the XAA option and add this one instead:
Option "DRI" "off"
...and see if that makes the issue go away (checking this helps us narrow down in what part of xorg the bug is hiding).

[Akkana Peck (akkzilla)]

Unfortunately, with the released version of jaunty, the character of the problem has changed -- now what I'm seeing (which might not be the same thing at all) is that every now and then I'll do something seemingly innocuous in firefox, the screen will go black and the monitor lose signal, and the machine is dead -- not responding to pings or keyboard or anything else. No flashing keyboard LEDs to show a kernel panic either.

I'm trying XAA now and will get back to you in a day or two about that and DRI off, but I may not be able to contribute stack traces since I can't ssh in any more.

[Akkana Peck (akkzilla)]

XAA does seem to eliminate the freezes -- I've been running for several days without a problem.

Unfortunately it also seems to eliminate opengl -- at least, google earth warns at startup that it's running in opengl software emulation. And glxinfo now says
OpenGL vendor string: Mesa Project
OpenGL renderer string: Software Rasterizer

[Philipp Kießler (pocytac)]

Can't completely confirm that. On my System (RV250) run with radeon-driver DRI seems to be working. "glxinfo | grep -i opengl" says:
OpenGL vendor string: Tungsten Graphics, Inc.
OpenGL renderer string: Mesa DRI R200 20060602 AGP 4x x86/MMX/SSE2 TCL
OpenGL version string: 1.3 Mesa 7.4

[Rolf Leggewie (r0lf)]

Philipp, that is not odd at all. In fact, upstream suspects that all these issues are in fact rooted in DRI, because apparently EXA does not work correctly without it.

https://bugs.freedesktop.org/show_bug.cgi?id=21611