Linux Virtualization - Jun 2019 - [PATCH 1/4] drm/virtio: pass gem reservation object to ttm init

With this gem and ttm will use the same reservation object,
so mixing and matching ttm / gem reservation helpers should
work fine.

Signed-off-by: Gerd Hoffmann <kraxel at redhat.com>
---
 drivers/gpu/drm/virtio/virtgpu_object.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/virtio/virtgpu_object.c
b/drivers/gpu/drm/virtio/virtgpu_object.c
index b2da31310d24..242766d644a7 100644
--- a/drivers/gpu/drm/virtio/virtgpu_object.c
+++ b/drivers/gpu/drm/virtio/virtgpu_object.c
@@ -132,7 +132,8 @@ int virtio_gpu_object_create(struct virtio_gpu_device
*vgdev,
 	virtio_gpu_init_ttm_placement(bo);
 	ret = ttm_bo_init(&vgdev->mman.bdev, &bo->tbo, params->size,
 			  ttm_bo_type_device, &bo->placement, 0,
-			  true, acc_size, NULL, NULL,
+			  true, acc_size, NULL,
+			  bo->gem_base.resv,
 			  &virtio_gpu_ttm_bo_destroy);
 	/* ttm_bo_init failure will call the destroy */
 	if (ret != 0)
-- 
2.18.1

On Mon, Jun 17, 2019 at 01:14:03PM +0200, Gerd Hoffmann
wrote:
> With this gem and ttm will use the same reservation object,
> so mixing and matching ttm / gem reservation helpers should
> work fine.
> 
> Signed-off-by: Gerd Hoffmann <kraxel at redhat.com>
While doing my prime doc+cleanup series I wondered whether we should do
this for everyone, and perhaps even remove ttm_bo.ttm_resv. Only driver
which doesn't yet have a gem_bo embedded in the same allocation is vmwgfx,
and that would be easy to fix by adding a vmwgfx_resv somehwere.

Anyway, looks like a solid start into the convergence story.

Reviewed-by: Daniel Vetter <daniel.vetter at ffwll.ch>
> ---
>  drivers/gpu/drm/virtio/virtgpu_object.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/virtio/virtgpu_object.c
b/drivers/gpu/drm/virtio/virtgpu_object.c
> index b2da31310d24..242766d644a7 100644
> --- a/drivers/gpu/drm/virtio/virtgpu_object.c
> +++ b/drivers/gpu/drm/virtio/virtgpu_object.c
> @@ -132,7 +132,8 @@ int virtio_gpu_object_create(struct virtio_gpu_device
*vgdev,
>  	virtio_gpu_init_ttm_placement(bo);
>  	ret = ttm_bo_init(&vgdev->mman.bdev, &bo->tbo,
params->size,
>  			  ttm_bo_type_device, &bo->placement, 0,
> -			  true, acc_size, NULL, NULL,
> +			  true, acc_size, NULL,
> +			  bo->gem_base.resv,
>  			  &virtio_gpu_ttm_bo_destroy);
>  	/* ttm_bo_init failure will call the destroy */
>  	if (ret != 0)
> -- 
> 2.18.1
> 
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

On Mon, Jun 17, 2019 at 04:08:25PM +0200, Daniel Vetter
wrote:
> On Mon, Jun 17, 2019 at 01:14:03PM +0200, Gerd Hoffmann wrote:
> > With this gem and ttm will use the same reservation object,
> > so mixing and matching ttm / gem reservation helpers should
> > work fine.
> > 
> > Signed-off-by: Gerd Hoffmann <kraxel at redhat.com>
> 
> While doing my prime doc+cleanup series I wondered whether we should do
> this for everyone, and perhaps even remove ttm_bo.ttm_resv. Only driver
> which doesn't yet have a gem_bo embedded in the same allocation is
vmwgfx,
> and that would be easy to fix by adding a vmwgfx_resv somehwere.
> 
> Anyway, looks like a solid start into the convergence story.
> 
> Reviewed-by: Daniel Vetter <daniel.vetter at ffwll.ch>
Aside: if virtio ever allows dma-buf sharing with something else (or
multiple virtio-gpu instances), then together with my patch series this
will fix dma-buf import. Atm virtio ignores the reservation object of the
imported dma-buf, which for foreing objects really isn't correct.
-Daniel
> 
> > ---
> >  drivers/gpu/drm/virtio/virtgpu_object.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/gpu/drm/virtio/virtgpu_object.c
b/drivers/gpu/drm/virtio/virtgpu_object.c
> > index b2da31310d24..242766d644a7 100644
> > --- a/drivers/gpu/drm/virtio/virtgpu_object.c
> > +++ b/drivers/gpu/drm/virtio/virtgpu_object.c
> > @@ -132,7 +132,8 @@ int virtio_gpu_object_create(struct
virtio_gpu_device *vgdev,
> >  	virtio_gpu_init_ttm_placement(bo);
> >  	ret = ttm_bo_init(&vgdev->mman.bdev, &bo->tbo,
params->size,
> >  			  ttm_bo_type_device, &bo->placement, 0,
> > -			  true, acc_size, NULL, NULL,
> > +			  true, acc_size, NULL,
> > +			  bo->gem_base.resv,
> >  			  &virtio_gpu_ttm_bo_destroy);
> >  	/* ttm_bo_init failure will call the destroy */
> >  	if (ret != 0)
> > -- 
> > 2.18.1
> > 
> 
> -- 
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch