[PATCH] drm/radeon: s2disk: Leave hw powered between freeze()

[Oliver Winker]

On Thursday 18 June 2015 23:16:57 Jerome Glisse wrote:
> Just to make sure we are on same page, it is the image creation that is
> problematic for you ? Not the resume from disk part, correct ?

Correct.

> While i think the part about not suspending the display make sense, i am
> more reluctant about the part not suspending the GPU. Mostly because the
> idea of freeze is to make sure nothing move and if we do not suspend the
> various GPU rings that things can change (all the wb pointer for instance)
> which might lead to an incoherent snapshot (for instance rptr/wptr of a ring
> would be from earlier version in the snapshot than what the ring content
> the snapshot capture).

Possible, I don't know the radeon in this detail. All I can tell is, that 
s2disk works perfectly well :) for me with my patch. I did use it over a week 
with resume/hibernate cycles every evening. Resume was 100% stable for me and 
counting. 

> I agree that this is unlikely given that before suspend we wait for all
> rings to drain and at that point the userspace is frozen and thus nothing
> should happen and GPU should stay idle. But i still feel that somehow
> something might go wrong with your approach.
> 
> Anyway as it is your patch would break runtime suspend.

Indeed, s2ram I didn't test at all. I only use s2disk.

> I recently posted patch to fix the resume side of hibernation, they might
> also help in the freeze()/thaw() dance. Could you please test them without
> your patch :
> https://patchwork.kernel.org/patch/6641371/
> https://patchwork.kernel.org/patch/6641381/
> 
> If it is not enough than also try with attached patch (which i have not
> tested myself but i think the logic is right).

Ok, I can do this. But only in 3-4 weeks or so. I'll let you know.

Cheers, Oliver