Linux Virtualization - Nov 2018 - [PATCH v4 3/4] drm/virtio: add in/out fence support for explicit synchronization

On Mon, 5 Nov 2018 at 11:42, Robert Foss <robert.foss at collabora.com>
wrote:
>
> When the execbuf call receives an in-fence it will get the dma_fence
> related to that fence fd and wait on it before submitting the draw call.
>
> On the out-fence side we get fence returned by the submitted draw call
> and attach it to a sync_file and send the sync_file fd to userspace. On
> error -1 is returned to userspace.
>
> Signed-off-by: Gustavo Padovan <gustavo.padovan at collabora.com>
> Signed-off-by: Robert Foss <robert.foss at collabora.com>
> Suggested-by: Rob Herring <robh at kernel.org>
> Reviewed-by: Emil Velikov <emil.velikov at collabora.com>
> ---
>
> Changes since v3:
>  - Move all in_fence handling to the same VIRTGPU_EXECBUF_FENCE_FD_IN block
Fwiw my suggestion was to explicitly document whether the IOCTL can
support, simultaneously, IN and OUT fence.
Merging the two patches makes things a bit meh. But as before - it's
for Gerd to make the final call.

-Emil

On Mon, Nov 05, 2018 at 05:25:05PM +0000, Emil Velikov
wrote:
> On Mon, 5 Nov 2018 at 11:42, Robert Foss <robert.foss at
collabora.com> wrote:
> >
> > When the execbuf call receives an in-fence it will get the dma_fence
> > related to that fence fd and wait on it before submitting the draw
call.
> >
> > On the out-fence side we get fence returned by the submitted draw call
> > and attach it to a sync_file and send the sync_file fd to userspace.
On
> > error -1 is returned to userspace.
> >
> > Signed-off-by: Gustavo Padovan <gustavo.padovan at
collabora.com>
> > Signed-off-by: Robert Foss <robert.foss at collabora.com>
> > Suggested-by: Rob Herring <robh at kernel.org>
> > Reviewed-by: Emil Velikov <emil.velikov at collabora.com>
> > ---
> >
> > Changes since v3:
> >  - Move all in_fence handling to the same VIRTGPU_EXECBUF_FENCE_FD_IN
block
> Fwiw my suggestion was to explicitly document whether the IOCTL can
> support, simultaneously, IN and OUT fence.
Yes, that would be good.  Code looks like it is supposed to work, but
explicitly saying so in the commit message would be nice.

Also: should we use separate fields for in/out fds?

cheers,
  Gerd

On Fri, Nov 09, 2018 at 06:13:52PM +0100, Robert Foss
wrote:
> Hey Gerd,
> 
> On 2018-11-09 11:13, Gerd Hoffmann wrote:
> > On Mon, Nov 05, 2018 at 05:25:05PM +0000, Emil Velikov wrote:
> > > On Mon, 5 Nov 2018 at 11:42, Robert Foss <robert.foss at
collabora.com> wrote:
> > > > 
> > > > When the execbuf call receives an in-fence it will get the
dma_fence
> > > > related to that fence fd and wait on it before submitting
the draw call.
> > > > 
> > > > On the out-fence side we get fence returned by the submitted
draw call
> > > > and attach it to a sync_file and send the sync_file fd to
userspace. On
> > > > error -1 is returned to userspace.
> > > > 
> > > > Signed-off-by: Gustavo Padovan <gustavo.padovan at
collabora.com>
> > > > Signed-off-by: Robert Foss <robert.foss at
collabora.com>
> > > > Suggested-by: Rob Herring <robh at kernel.org>
> > > > Reviewed-by: Emil Velikov <emil.velikov at
collabora.com>
> > > > ---
> > > > 
> > > > Changes since v3:
> > > >   - Move all in_fence handling to the same
VIRTGPU_EXECBUF_FENCE_FD_IN block
> > > Fwiw my suggestion was to explicitly document whether the IOCTL
can
> > > support, simultaneously, IN and OUT fence.
> > 
> > Yes, that would be good.  Code looks like it is supposed to work, but
> > explicitly saying so in the commit message would be nice.
> 
> On it! Will send out a v5.
> 
> > 
> > Also: should we use separate fields for in/out fds?
> 
> I'm not sure I understand which fields you're referring to.
fence_in_fd & fence_out_fd in the ioctl struct (patch #2).

cheers,
  Gerd