[Intel-gfx] [PATCH 03/11] drm/i915: reuse i915_gem_object_put_fence_reg for fence stealing code

[Chris Wilson]

On Fri, 15 Jan 2010 13:24:10 +0100, Daniel Vetter <daniel.vetter at ffwll.ch> wrote:
> This has a few functional changes against the old code:
> 
> * a few more unnecessary loads and stores to the drm_i915_fence_reg
>   objects. Also an unnecessary store to the hw fence register.
> 
> * zaps any userspace mappings before doing other flushes. Only changes
>   anything when userspace does racy stuff against itself.
> 
> * also flush GTT domain. Now that the gtt flushing code actually
>   does something, we should put it to some good use.
> 
> Signed-off-by: Daniel Vetter <daniel.vetter at ffwll.ch>
> ---
>  drivers/gpu/drm/i915/i915_gem.c |   29 +++--------------------------
>  1 files changed, 3 insertions(+), 26 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index a092025..b11dbeb 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -2448,39 +2448,16 @@ i915_gem_object_get_fence_reg(struct drm_gem_object *obj)
>  			 */
>  			drm_gem_object_reference(old_obj);
>  
> -			/* i915 uses fences for GPU access to tiled buffers */
> -			if (IS_I965G(dev) || !old_obj_priv->active)
> -				break;
> -
> -			/* This brings the object to the head of the LRU if it
> -			 * had been written to.  The only way this should
> -			 * result in us waiting longer than the expected
> -			 * optimal amount of time is if there was a
> -			 * fence-using buffer later that was read-only.
> -			 */
> -			i915_gem_object_flush_gpu_write_domain(old_obj);
> -			ret = i915_gem_object_wait_rendering(old_obj);
> -			if (ret != 0) {
> -				drm_gem_object_unreference(old_obj);
> -				return ret;
> -			}
> -
>  			break;
>  		}
>  
> -		/*
> -		 * Zap this virtual mapping so we can set up a fence again
> -		 * for this object next time we need it.
> -		 */
> -		i915_gem_release_mmap(old_obj);
> -
>  		i = old_obj_priv->fence_reg;
>  		reg = &dev_priv->fence_regs[i];
>  
> -		old_obj_priv->fence_reg = I915_FENCE_REG_NONE;
> -		list_del_init(&old_obj_priv->fence_list);
> -
> +		ret = i915_gem_object_put_fence_reg(old_obj);
>  		drm_gem_object_unreference(old_obj);
> +		if (ret != 0) 
> +			return ret;
>  	}
>  
>  	obj_priv->fence_reg = i;

This leaves the code slightly ugly, I'd prefer the reference/unreference
to be tighter given the rearrangement.
-ickle

-- 
Chris Wilson, Intel Open Source Technology Centre