Re: [PATCH 3/5] drm/vmwgfx: add a move callback.

2020-10-08 Thread Roland Scheidegger
Both vmwgxf patches look good to me too.
Reviewed-by: Roland Scheidegger 

Am 06.10.20 um 02:06 schrieb Dave Airlie:
> From: Dave Airlie 
> 
> This just copies the fallback to vmwgfx, I'm going to iterate on this
> a bit until it's not the same as the fallback path.
> 
> Signed-off-by: Dave Airlie 
> ---
>  drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c | 19 ++-
>  1 file changed, 18 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c 
> b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c
> index 2f88d2d79f9a..6e36fc932aeb 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c
> @@ -725,6 +725,23 @@ static void vmw_swap_notify(struct ttm_buffer_object *bo)
>   (void) ttm_bo_wait(bo, false, false);
>  }
>  
> +static int vmw_move(struct ttm_buffer_object *bo,
> +  bool evict,
> +  struct ttm_operation_ctx *ctx,
> +  struct ttm_resource *new_mem)
> +{
> + struct ttm_resource_manager *old_man = ttm_manager_type(bo->bdev, 
> bo->mem.mem_type);
> + struct ttm_resource_manager *new_man = ttm_manager_type(bo->bdev, 
> new_mem->mem_type);
> +
> + if (old_man->use_tt && new_man->use_tt) {
> + if (bo->mem.mem_type == TTM_PL_SYSTEM) {
> + ttm_bo_assign_mem(bo, new_mem);
> + return 0;
> + }
> + return ttm_bo_move_ttm(bo, ctx, new_mem);
> + } else
> + return ttm_bo_move_memcpy(bo, ctx, new_mem);
> +}
>  
>  struct ttm_bo_driver vmw_bo_driver = {
>   .ttm_tt_create = _ttm_tt_create,
> @@ -735,7 +752,7 @@ struct ttm_bo_driver vmw_bo_driver = {
>   .ttm_tt_destroy = _ttm_destroy,
>   .eviction_valuable = ttm_bo_eviction_valuable,
>   .evict_flags = vmw_evict_flags,
> - .move = NULL,
> + .move = vmw_move,
>   .verify_access = vmw_verify_access,
>   .move_notify = vmw_move_notify,
>   .swap_notify = vmw_swap_notify,
> 

___
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel


Re: [PATCH 3/5] drm/vmwgfx: add a move callback.

2020-10-07 Thread Dave Airlie
On Thu, 8 Oct 2020 at 13:41, Zack Rusin  wrote:
>
>
> > On Oct 5, 2020, at 20:06, Dave Airlie  wrote:
> >
> > From: Dave Airlie 
> >
> > This just copies the fallback to vmwgfx, I'm going to iterate on this
> > a bit until it's not the same as the fallback path.
> >
> > Signed-off-by: Dave Airlie 
>
> What are your plans for it? i.e. how is it going to be different?

Initial plan is to put move_notify inside the move callback, then
eventually get rid of the ttm bind/unbind callback and let the driver
do that itself if needed.

I've got most of it in a branch (and I posted a 45 patch series a week
or two ago), but I need to rebase and clean it up for reposting.

Dave.
___
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel


Re: [PATCH 3/5] drm/vmwgfx: add a move callback.

2020-10-07 Thread Zack Rusin


> On Oct 5, 2020, at 20:06, Dave Airlie  wrote:
> 
> From: Dave Airlie 
> 
> This just copies the fallback to vmwgfx, I'm going to iterate on this
> a bit until it's not the same as the fallback path.
> 
> Signed-off-by: Dave Airlie 

What are your plans for it? i.e. how is it going to be different?

Reviewed-by: Zack Rusin 

___
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel