Am 19.09.2018 um 17:39 schrieb Michel Dänzer:
On 2018-09-19 2:30 p.m., Christian König wrote:
Am 18.09.2018 um 18:17 schrieb Nayan Deshmukh:
having a delayed work item per job is redundant as we only need one per scheduler to track the time out the currently executing job.
Well that looks simpler than I thought it would be.
But it shows the next problem that the timeout and the completion could race.
As far as I can see that can be fixed by moving the dma_fence_remove_callback()/dma_fence_add_callback() dance from drm_sched_hw_job_reset() to drm_sched_job_timedout().
BTW, while you guys are looking into this code, please keep an eye open for things that could explain https://bugs.freedesktop.org/107762 .
Yeah, since we now have only one timer that should be fixed by this as well.
Christian.