[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#524558: Occasional X hangs where drmCommandWrite() never returns



On Sat, Jul 11, 2009 at 02:04:16PM +0200, Michel Dänzer wrote:
> On Fri, 2009-07-10 at 23:05 +0200, Fabian Köster wrote:
> > 
> > I have a similar problem using Ubuntu Jaunty with the radeon-driver.
> > 
> > See https://bugs.launchpad.net/ubuntu/+source/linux/+bug/348332
> 
> Backtraces aren't useful for classifying GPU lockup problems because
> they just show one of a few places where the drivers wait for the GPU to
> catch up, which never happens because it's locked up. This is usually
> unrelated to the place where the commands causing the GPU to lock up
> were generated though. So this kind of problem needs to be classified by
> what triggers it instead.

Ok. Any idea how I can debug that? I have ssh access to the hanging
machine, and I can use gdb, strace, whatever is needed. (My card is an
intel one, not a radon.) And I'd be willing to spend time debugging
this, since having my computer freeze twice a day is annoying as hell.

There doesn't seem to be any reliable way to trigger the hang, but it
seems that more often than not the hang happens when redrawing a
Firefox window. Generally it seems to happen roughly once per five
hours of normal use.

BTW there's a "size" parameter to drmWrite() or whatever (don't have
the computer on right now), and it's often something really absurd, in
the order of 10^15 on a 64-bit computer. Any chance that could be used
to catch where such a command is generated?

> That said, a few people above mention that kernel 2.6.29 or newer solved
> their problem. Have you tried that?

I'm running on 2.6.30.1. Last happened yesterday to me, twice. No
change since 2.6.29 at least.

In the meantime I had my computer working after I switched from EXA to
UXA - UXA does not hang. That worked fine for a while. However since
recently - unfortunately I don't know if after a kernel or X update -
UXA is _really_ slow. For example, when viewing a man page in an X
terminal (konsole), pressing space to get to the next page hangs X for
around one second at 100% CPU. Same thing with gnome-terminal. And
this is without anything fancy like KDE's OpenGL-based "desktop
effects" turned on. A nethack-style game is slow to the point of being
unplayable because X can't keep up :-)

(I can also try to look at why UXA is suddenly so slow if you wish.
Actually I already filed a bug report: http://bugs.debian.org/534991
and would appreciate pointers at where to look further.)

	Sami

Attachment: signature.asc
Description: Digital signature


Reply to: