[Intel-gfx] [PATCH 1/2] dma-fence: Clear fence->status during dma_fence_init()

Chris Wilson chris at chris-wilson.co.uk
Wed Jan 4 09:43:51 UTC 2017


On Wed, Jan 04, 2017 at 10:37:32AM +0100, Daniel Vetter wrote:
> On Wed, Jan 04, 2017 at 09:24:27AM +0000, Chris Wilson wrote:
> > On Wed, Jan 04, 2017 at 10:15:01AM +0100, Daniel Vetter wrote:
> > > On Tue, Jan 03, 2017 at 02:04:44PM +0000, Tvrtko Ursulin wrote:
> > > > 
> > > > On 03/01/2017 11:05, Chris Wilson wrote:
> > > > > As the fence->status is an optional field that may be set before
> > > > > dma_fence_signal() is called to convey that the fence completed with an
> > > > > error, we have to ensure that it is always set to zero on initialisation
> > > > > so that the typical use (i.e. unset) always flags a successful completion.
> > > > > 
> > > > > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > > > > ---
> > > > >  drivers/dma-buf/dma-fence.c | 1 +
> > > > >  1 file changed, 1 insertion(+)
> > > > > 
> > > > > diff --git a/drivers/dma-buf/dma-fence.c b/drivers/dma-buf/dma-fence.c
> > > > > index 3444f293ad4a..9130f790ebf3 100644
> > > > > --- a/drivers/dma-buf/dma-fence.c
> > > > > +++ b/drivers/dma-buf/dma-fence.c
> > > > > @@ -534,6 +534,7 @@ dma_fence_init(struct dma_fence *fence, const struct dma_fence_ops *ops,
> > > > >  	fence->context = context;
> > > > >  	fence->seqno = seqno;
> > > > >  	fence->flags = 0UL;
> > > > > +	fence->status = 0;
> > > > > 
> > > > >  	trace_dma_fence_init(fence);
> > > > >  }
> > > > > 
> > > > 
> > > > Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> > > 
> > > Looking at existing users there's only the sync_file stuff. And that gets
> > > it wrong, because afaics no one ever sets fence->status to anything
> > > useful. But sync_file blindly assumes it's correct.
> > 
> > In terms of doc, sync_file is using it correctly, and dma-fence isn't
> > living up to its doc. The documented behaviour (sync_file) seems useful.
> 
> Yeah, it looks like an oversight in merging sync_file and dma_fence
> together, but instead of piecemeal fixing (like this patch does), fixing
> it all&properly might be better.

What's missing?

  * @status: Optional, only valid if < 0, must be set before calling
  * dma_fence_signal, indicates that the fence has completed with an
  * error.

dma-fence must then guarantee that is zeroed during init, then the
drivers can supply the optional information prior to calling
dma_fence_signal()

A dma_fence_set_status(fence, status) {
	BUG_ON(test_bit(SIGNALED, &fence->flags);
	BUG_ON(!IS_ERR_VALUE(status));
	BUG_ON(fence->status);
	fence->status = status;
} ?
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre


More information about the Intel-gfx mailing list