[PATCH 01/13] drm/scheduler: fix fence ref counting

[Christian König]

Am 25.10.22 um 05:23 schrieb Luna Nova:
> This patch doesn't apply nicely on 6.0, seems amd-staging-drm-next is missing 9b04369b060fd4885f728b7a4ab4851ffb1abb64 drm/scheduler: Don't kill jobs in interrupt context

Yeah, so what? This is a rebase on top of amd-staging-drm-next. I've 
just pushed the original patch to drm-misc-fixes.

Regards,
Christian.

>
> On Fri, 14 Oct 2022, at 01:46, Christian König wrote:
>> We leaked dependency fences when processes were beeing killed.
>>
>> Additional to that grab a reference to the last scheduled fence.
>> diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
>> index 43d337d8b153..243ff384cde8 100644
>> --- a/drivers/gpu/drm/scheduler/sched_entity.c
>> +++ b/drivers/gpu/drm/scheduler/sched_entity.c
>> @@ -207,6 +207,7 @@ static void drm_sched_entity_kill_jobs_cb(struct dma_fence *f,
>> struct drm_sched_job *job = container_of(cb, struct drm_sched_job,
>> finish_cb);
>>   
>> + dma_fence_put(f);
> Conflict here, is INIT_WORK with that commit.
>> init_irq_work(&job->work, drm_sched_entity_kill_jobs_irq_work);
>> irq_work_queue(&job->work);