[PATCH] drm/ttm: fix delayed ttm_bo_cleanup_refs_and_unlock delayed handling

[Jerome Glisse]

On Wed, Jan 16, 2013 at 6:10 PM, Markus Trippelsdorf
<markus at trippelsdorf.de> wrote:
> On 2013.01.16 at 17:36 -0500, Alex Deucher wrote:
>> On Tue, Jan 15, 2013 at 12:03 PM, Markus Trippelsdorf
>> <markus at trippelsdorf.de> wrote:
>> > On 2013.01.15 at 17:32 +0100, Markus Trippelsdorf wrote:
>> >> On 2013.01.15 at 16:26 +0100, Michel Dänzer wrote:
>> >> > On Die, 2013-01-15 at 16:23 +0100, Markus Trippelsdorf wrote:
>> >> > > On 2013.01.15 at 15:43 +0100, Michel Dänzer wrote:
>> >> > > > On Sam, 2013-01-05 at 11:41 +0100, Markus Trippelsdorf wrote:
>> >> > > > > On 2012.12.20 at 14:58 +0100, Markus Trippelsdorf wrote:
>> >> > > > > >
>> >> > > > > > And just in case it got lost in the noise yesterday:
>> >> > > > > > The image corruption is caused by Dave's commit:
>> >> > > > > >
>> >> > > > > > commit dd54fee7d440c4a9756cce2c24a50c15e4c17ccb
>> >> > > > > > Author: Dave Airlie <airlied at redhat.com>
>> >> > > > > > Date:   Fri Dec 14 21:04:46 2012 +1000
>> >> > > > > >
>> >> > > > > >     radeon: fix regression with eviction since evict caching changes
>> >> > > > > >
>> >> > > > > > Reverting it 'fixes' the issue.
>> >> > > > >
>> >> > > > > Ping.
>> >> > > > > The issue still happens with todays Linus git tree.
>> >> > > >
>> >> > > > Does the corruption also occur with
>> >> > > > dd54fee7d440c4a9756cce2c24a50c15e4c17ccb applied manually on top of
>> >> > > > 0d0b3e7443bed6b49cb90fe7ddc4b5578a83a88d?
>> >> > >
>> >> > > No.
>> >> >
>> >> > So, can you bisect which change between those two actually introduced
>> >> > the corruption?
>> >
>> > The real cause of the image corruption is:
>> >
>> > d025e9e2b890db679f1246037bf65bd4be512627 is the first bad commit
>> > commit d025e9e2b890db679f1246037bf65bd4be512627
>> > Author: Jerome Glisse <jglisse at redhat.com>
>> > Date:   Thu Nov 29 10:35:41 2012 -0500
>> >
>> >     drm/radeon: do not move bo to different placement at each cs
>> >
>> >     The bo creation placement is where the bo will be. Instead of trying
>> >     to move bo at each command stream let this work to another worker
>> >     thread that will use more advance heuristic.
>> >
>> >     agd5f: remove leftover unused variable
>> >
>> >     Signed-off-by: Jerome Glisse <jglisse at redhat.com>
>> >     Reviewed-by: Alex Deucher <alexander.deucher at amd.com>
>> >
>> > Reverting d025e9e2b890d on top of Linus' tree fixes the issue.
>>
>> Can you try this patch from Jerome:
>> https://bugzilla.kernel.org/attachment.cgi?id=91421
>
> It fixes the corruption, but it degrades performance so much that it
> takes several seconds to switch virtual desktops under xmonad. And
> sometimes the website used for the scroll test is stuck for several
> seconds and unscrollable during that time.
>
> --
> Markus

What about this patch instead :
http://people.freedesktop.org/~glisse/0001-drm-radeon-exclude-system-placement-when-validating-.patch

Cheers,
Jerome