Re: [Intel-gfx] [PATCH 2/3] drm/i915: Fix some invalid requests cancellations

2016-02-19 Thread Tvrtko Ursulin
On 15/02/16 09:47, Daniel Vetter wrote: On Thu, Feb 11, 2016 at 01:01:34PM +, Tvrtko Ursulin wrote: On 29/01/16 16:49, Chris Wilson wrote: As we add the VMA to the request early, it may be cancelled during execbuf reservation. This will leave the context object pointing to a I don't get

Re: [Intel-gfx] [PATCH 2/3] drm/i915: Fix some invalid requests cancellations

2016-02-15 Thread Daniel Vetter
On Thu, Feb 11, 2016 at 01:01:34PM +, Tvrtko Ursulin wrote: > On 29/01/16 16:49, Chris Wilson wrote: > >As we add the VMA to the request early, it may be cancelled during > >execbuf reservation. This will leave the context object pointing to a > > I don't get it, request is created after the r

Re: [Intel-gfx] [PATCH 2/3] drm/i915: Fix some invalid requests cancellations

2016-02-11 Thread Tvrtko Ursulin
On 29/01/16 16:49, Chris Wilson wrote: As we add the VMA to the request early, it may be cancelled during execbuf reservation. This will leave the context object pointing to a I don't get it, request is created after the reservation. dangling request; i915_wait_request() simply skips the wa

[Intel-gfx] [PATCH 2/3] drm/i915: Fix some invalid requests cancellations

2016-01-29 Thread Chris Wilson
As we add the VMA to the request early, it may be cancelled during execbuf reservation. This will leave the context object pointing to a dangling request; i915_wait_request() simply skips the wait and so we may unbind the object whilst it is still active. However, if at any point we make a change