Re: [PATCH] drm/syncobj: Extend EXPORT_SYNC_FILE for timeline syncobjs

[Rob Clark]

On Wed, Mar 26, 2025 at 7:46 AM Rob Clark <robdclark@xxxxxxxxx> wrote:
>
> On Wed, Mar 26, 2025 at 7:41 AM Christian König
> <christian.koenig@xxxxxxx> wrote:
> >
> > Am 26.03.25 um 15:39 schrieb Rob Clark:
> > > From: Rob Clark <robdclark@xxxxxxxxxxxx>
> > >
> > > Add support for exporting a dma_fence fd for a specific point on a
> > > timeline.
> >
> > Looks good on first glance. What's the userspace use case?
>
> Timeline syncobj support for vtest/vpipe[1][2].. since core
> virglrender and drm native ctx works in terms of fences (since in the
> VM case, everything is a fence below the guest kernel uabi), we need
> to be able to turn a point on a timeline back into a fence fd.  (Plus
> it seemed like an odd omission from the existing uabi.)
>
> BR,
> -R
>
> [1] https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33433
> [2] https://gitlab.freedesktop.org/virgl/virglrenderer/-/merge_requests/805
>
> >
> > Regards,
> > Christian.
> >
> > >
> > > Signed-off-by: Rob Clark <robdclark@xxxxxxxxxxxx>
> > > ---
> > >  drivers/gpu/drm/drm_syncobj.c | 8 ++++++--
> > >  include/uapi/drm/drm.h        | 2 ++
> > >  2 files changed, 8 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c
> > > index 4f2ab8a7b50f..eb7a2dd2e261 100644
> > > --- a/drivers/gpu/drm/drm_syncobj.c
> > > +++ b/drivers/gpu/drm/drm_syncobj.c
> > > @@ -762,7 +762,7 @@ static int drm_syncobj_import_sync_file_fence(struct drm_file *file_private,
> > >  }
> > >
> > >  static int drm_syncobj_export_sync_file(struct drm_file *file_private,
> > > -                                     int handle, int *p_fd)
> > > +                                     int handle, u64 point, int *p_fd)
> > >  {
> > >       int ret;
> > >       struct dma_fence *fence;
> > > @@ -772,7 +772,7 @@ static int drm_syncobj_export_sync_file(struct drm_file *file_private,
> > >       if (fd < 0)
> > >               return fd;
> > >
> > > -     ret = drm_syncobj_find_fence(file_private, handle, 0, 0, &fence);
> > > +     ret = drm_syncobj_find_fence(file_private, handle, point, 0, &fence);
> > >       if (ret)
> > >               goto err_put_fd;
> > >
> > > @@ -882,8 +882,12 @@ drm_syncobj_handle_to_fd_ioctl(struct drm_device *dev, void *data,
> > >
> > >       if (args->flags & DRM_SYNCOBJ_HANDLE_TO_FD_FLAGS_EXPORT_SYNC_FILE)
> > >               return drm_syncobj_export_sync_file(file_private, args->handle,
> > > +                                                 args->point,
> > >                                                   &args->fd);

Hmm, maybe I should add DRM_SYNCOBJ_HANDLE_TO_FD_FLAGS_TIMELINE so
that userspace gets a clean error on older kernels, rather than having
the point param be silently ignored..

BR,
-R

> > >
> > > +     if (args->point)
> > > +             return -EINVAL;
> > > +
> > >       return drm_syncobj_handle_to_fd(file_private, args->handle,
> > >                                       &args->fd);
> > >  }
> > > diff --git a/include/uapi/drm/drm.h b/include/uapi/drm/drm.h
> > > index 7fba37b94401..c71a8f4439f2 100644
> > > --- a/include/uapi/drm/drm.h
> > > +++ b/include/uapi/drm/drm.h
> > > @@ -912,6 +912,8 @@ struct drm_syncobj_handle {
> > >
> > >       __s32 fd;
> > >       __u32 pad;
> > > +
> > > +     __u64 point;
> > >  };
> > >
> > >  struct drm_syncobj_transfer {
> >