[Intel-gfx] [PATCH] dma-buf/fence: Avoid use of uninitialised timestamp

[Chris Wilson]

On Tue, Feb 14, 2017 at 12:22:02PM -0200, Gustavo Padovan wrote:
> 2017-02-14 Chris Wilson <chris at chris-wilson.co.uk>:
> 
> > On Tue, Feb 14, 2017 at 11:40:38AM -0200, Gustavo Padovan wrote:
> > > Hi Chris,
> > > 
> > > 2017-02-14 Chris Wilson <chris at chris-wilson.co.uk>:
> > One thing that occurs to me is whether we should be setting the
> > timestamp when we set an error. The above (sync_debug though) implies
> > that it expects the error to have the timestamp. sync_fence_info could
> > go either way.
> 
> We could do it. I don't see any reason against it.

It is just uncertain as to what timestamp means, and what the user
wants. In i915 we may end up flagging an error on a fence long (many
10s) before the fence is signaled. It still feels more appropriate to
set the timestamp on when the fence is complete - certainly for the case
where we replay the request (and we flag fence->error that we did so as
there may have been side-effects that we should inform the user about).
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre