force app kill patch

Liu, Monk Monk.Liu at
Wed Apr 18 07:11:07 UTC 2018

Hi Christian & Emily

I think the v4 fix for "fix force app kill hang" is still not good enough:

See that in "sched_entity_fini", we only call dma_fence_put(entity->last_scheduled" under the condition of "If (entity->fini_status)", so
This way there is memory leak for the case of "entity->fini_stats ==0"

If we move dma_fence_put(entity->last_scheduled) out of the condition of "if (entity->fini_status)", the memory leak issue can be fixed
But there will be kernel NULL pointer access, I think the time you call dma_fence_put(entity->last_scheduled") may actually executed *not*
On the last scheduled fence of this entity, because it is run without "thread_park/unpark" pair which to make sure scheduler not dealing this entity

So with certain race issue, here is the scenario:

1.        scheduler is doing the dma_fence_put() on the 1st fence,

2.        scheduler set entity->last_scheduled to 1st fence

3.        now sched_entity_fini() run, and it call dma_fence_put() on entity->last_scheduled

4.        now this 1st fence is actually put double time and the real last fence won't get put by expected

any idea?

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the amd-gfx mailing list