[PATCH 30/45] drm/ttm: add a new invalidate notify callback.

Christian König christian.koenig at amd.com
Tue Sep 29 07:05:25 UTC 2020


Am 29.09.20 um 05:23 schrieb Dave Airlie:
> On Thu, 24 Sep 2020 at 22:25, Christian König <christian.koenig at amd.com> wrote:
>> Am 24.09.20 um 07:18 schrieb Dave Airlie:
>>> From: Dave Airlie <airlied at redhat.com>
>>>
>>> Signed-off-by: Dave Airlie <airlied at redhat.com>
>> NAK, completely unnecessary.
>>
>> We should rather do the remaining accounting in the already existing
>> release_notify() callback.
>>
>> That makes much more sense and if I'm not completely mistaken could
>> actually fix a bug in amdgpu.
> release_notify is called from one path, but there is a path in
> eviction where if it gets pass 0 placements
> it tears down the memory, and allocates a tt.

You mean for the pipelined gutting? Mhm, I see. Probably better to call 
the move callback for those cases as well.

Ok, go ahead with that approach for now. But we really need to clean 
that up further.

Christian.

>
> I'm considering whether it should be acceptable to call evict with no
> placements (though maybe that just means evict to system).
>
> Dave.



More information about the dri-devel mailing list