Re: radeon 4830 corruption after resume

[Alex Deucher <alexdeucher@gmail.com>]

On Tue, Dec 15, 2009 at 12:40 AM, Alex Chiang <achiang@hp.com> wrote:
> Adding dri-devel to cc as I should have done in the first place.
>
> Full thread for context here: http://lkml.org/lkml/2009/12/13/321
>
> More responses below.
>
> * Xiaotian Feng <xtfeng@gmail.com>:
>> On Mon, Dec 14, 2009 at 11:11 AM, Alex Chiang <achiang@hp.com> wrote:
>> > * Dave Airlie <airlied@redhat.com>:
>> >> On Sun, 2009-12-13 at 19:30 -0700, Alex Chiang wrote:
>> >> > Hi Dave, Rafael,
>> >> >
>> >> > I can successfully suspend/resume my HP Envy, but upon resume,
>> >> > the screen is corrupted.
>> >> >
>> >> > I used the gnome screenshot utility to capture this:
>> >> >
>> >> >     http://chizang.net/alex/tmp/radeon-4830-corruption.png
>> >> >
>> >> > But that screenshot leads you to believe the corruption was
>> >> > 100%, when in reality, the text in my xterms was at least
>> >> > readable, but ugly.
>> >> >
>> >> > Kernel is latest upstream pulled today. Userspace is Ubuntu Karmic.
>> >>
>> >> Has it ever worked? can you suspend/resume without X running at all?
>> >
>> > Hm, define "worked"?
>> >
>> > The machine responds to keyboard, mouse, network input, etc. It's
>> > just that the screen is garbled.
>> >
>> > Unless I'm not understanding what you're asking?
>> >
>> >> Its quite possibly a userspace problem but its hard to know, we
>> >> haven't changed the user modesetting pieces in the kernel at all in
>> >> quite a while.
>> >
>> > Ok, I'm just looking for where/how to start debugging. Any advice
>> > on where to look next would be fine too.
>>
>> Then how about ssh to your machine, and get dmesg output?
>
> The machine is actually useable. Wireless networking even works.
> It's just that the screen is garbled.
>
> Here's a much better idea of the type of corruption that I'm
> seeing. Huge file alert, it's like a 5MB jpg.
>
>        http://chizang.net/alex/tmp/corrupt-radeon2.jpg
>
> I'm attaching full dmesg and pm-suspend.log.
>
> The suspend happens some time around time 400 and the resume is
> the huge jump in time afterwards.

As Dave asked previously, is this a regression?  I.e., did s/r work at
some point in the past and if so when?

Alex