Bug#616301: xserver-xorg-video-radeon:screen goes black, system hangs after 2sec:[youtube(FF/Opera)-reset req.]
On Mon, Mar 7, 2011 at 4:36 AM, Ben Hutchings <ben@decadent.org.uk> wrote:
> On Sun, 2011-03-06 at 13:08 -0500, Alex Deucher wrote:
>> On Fri, Mar 4, 2011 at 8:50 PM, Ben Hutchings <ben@decadent.org.uk> wrote:
>> > On Fri, 2011-03-04 at 21:01 +0200, Faidon Liambotis wrote:
>> >> severity 616301 critical
>> >> thanks
>> >
>> > No, not unless it will affect a large proportion of users.
>> >
>> >> My system locks up whenever I click on a YouTube video link since
>> >> yesterday. I can probably live without YouTube :), but in any case this
>> >> shouldn't happen.
>> >>
>> >> This isn't a singled out case nor in exotic, possibly faulty, hardware.
>> >> It's on a standard 1½-year old Dell OptiPlex 780 desktop with a Radeon
>> >> HD card (one of the standard configurations) and this is on a stock
>> >> squeeze system.
>> >>
>> >> The findings so far seem to suggest this is a Mesa issue; I'd probably
>> >> file it under "Linux kernel bugs" (or even DoS bugs) but I'm not sure
>> >> where to properly file such bugs in the post-KMS stack world.
>> >
>> > If there is a kernel driver involved then it should be assigned to the
>> > kernel. Even without KMS, a Mesa driver should be considered untrusted
>> > and should not be able to trigger a crash or hang. With KMS, this
>> > applies to the X driver too.
>> >
>>
>> With or without KMS, the userspace acceleration drivers can certainly
>> cause GPU hangs if the 3D engine is programmed with some combination
>> of commands it doesn't like.
>
> You can't solve the halting problem but you can implement a watchdog,
> can't you?
>
We do, we reset the GPU 10s after it hangs, but this depends on a lot
of things going our way.
Occasionally we do reset the GPU when we shouldn't as well.
However if there is an issue in the kernel, ddx or mesa driver,
constants resets will pretty much DoS the GPU.
Dave.
Reply to: