[Intel-gfx] i915 driver slowdown after 1-2 days?

[Chun-Yu Shei]

On 10/24/2010 2:36 PM, Paul Menzel wrote:
> Am Sonntag, den 24.10.2010, 13:03 -0400 schrieb Chun-Yu Shei:
>
>> I'm currently running xf86-video-intel 2.13.0 with an i5-540M, and I've
>> been experiencing a strange issue where after a day or two, I can no
>> longer play videos smoothly in Flash 10.2 beta, as well as in VMWare.
>> Even moving windows in VMWare guests becomes slow and visibly laggy.
>> It's almost as if there's no video acceleration -- the frame rate drops
>> drastically, and CPU usage seems to increase significantly.  Playing
>> video in mplayer still works perfectly, however.  Restarting X has no
>> effect, and I have to do a full reboot to get rid of the problem.  I'm
>> running 64-bit Gentoo with the following:
>>
>> libdrm 2.4.22
>> mesa 7.8.2
>> 2.6.36 kernel
>> xorg-server 1.7.7
>>
>> The problem also occurs with the 2.6.35 kernel, and I've also been able
>> to replicate it in a fresh install of Ubuntu 10.10.  There are no
>> indications of any problems in both the output of dmesg and the Xorg
>> logs, and the output of intel_gpu_top looks no different before/after
>> the problem occues.  I'm on a ThinkPad X201 with an external display
>> connected via DisplayPort, and I do suspend it with the hibernate-ram
>> script.  Is this a known problem by any chance?  If not, how might I go
>> about troubleshooting this?
>
> reading your report it is not clear for me, if hibernating is necessary
> to trigger the problem or not.

I cannot say for sure if hibernating is necessary to trigger the problem 
or not... every time I've seen the problem, it's been after using 
hibernate-ram a couple of times, but I also haven't left the laptop 
running continuously for a couple of days before.  I usually carry my 
laptop with me during the day, so it's not really practical to leave it 
on continuously.  It definitely doesn't happen immediately after 
resuming from suspend, though.  I can resume from suspend, watch some 
YouTube videos with no problem, and have the problem pop up a couple of 
hours later.

This bug is the only issue preventing me from switching exclusively to 
linux, so I really hope to find a fix for it -- I just don't know where 
to start. (I also hope for the ability to independently load separate 
color profiles for each monitor using xcalib, but that's minor)

Chun-Yu