On 20/04/16 15:22, Chris Wilson wrote:
On Wed, Apr 20, 2016 at 03:08:19PM +0100, Tvrtko Ursulin wrote:
Otherwise execlists_check_remove_request when accessing
head_req->ctx is use after free. And I can demonstrate that easily
via gem-close-race. Put a
WARN_ON(atomic_read(&head_req->ctx->ref.refcount) == 0); and see. :)
More to the point, could we do a 10s burst of close race for BAT. What's
the likelihood of that capturing such faults?
With lockdep/kmemcheck etc enabled?
No idea, but on a lean kernel it takes a lot less than 10s. So maybe
time limited gem-close-race-basic, resend your series and see if it
catches it?
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
