Bug#620858: xserver-xorg-video-radeon: slow overall performance upon wake-up from suspend

[Andres Cimmarusti]

2011/4/5 Michel Dänzer <daenzer at debian.org>:
> On Die, 2011-04-05 at 10:02 -0400, Andres Cimmarusti wrote:
>> >>> [ 2183.658014] [drm] radeon: 1 quad pipes, 1 z pipes initialized.
>> >>
>> >> The kernel radeon driver isn't consistently reporting the number of quad
>> >> pipes. Does the slowness correspond to the number being reported as 1?
>> >
>> > I've just checked this. Yes, it's slow when the number is reported as
>> > 1 (which happens both waking up from suspend, and when using
>> > vblank_mod=0).
>
> The latter can't really be true like that, as the number of pipe quads
> is only probed and printed at initialization and resume time.

Let me clarify. The slowness is present after waking up from suspend
OR if I boot up using vblank_mod=0.

>
>> So, this is rather strange. With a vanilla 2.6.37.6 kernel (using
>> debian config file to compile + two cpu customizations) the quad
>> number is generally reported as 3, but goes to 1 upon resuming from
>> suspend.
>>
>> With kernel 2.6.38-2 from debian main, quad seems to be 1, but after
>> resuming it goes to 3.
>
> So in the latter case, is it slow initially but becomes fast after
> resume? :)

It's always slow on resume...in any case. In fact, right now, I've
been probing the quad pipe message and it seems to vary. The last two
boots with 2.6.38-2 kernel lead to a reported number of 3. After the
resume the number remains 3:

$ dmesg | grep quad
[    9.554083] [drm] radeon: 3 quad pipes, 1 z pipes initialized.
[  173.492317] [drm] radeon: 3 quad pipes, 1 z pipes initialized.

But the slowness is still present and almost crippling...