[Intel-gfx] [PATCH 03/24] drm/i915: Store i915_gem_object_is_coherent() as a bit next to cache-dirty

Chris Wilson chris at chris-wilson.co.uk
Fri May 19 10:22:07 UTC 2017


On Thu, May 18, 2017 at 10:46:17AM +0100, Chris Wilson wrote:
> For ease of use (i.e. avoiding a few checks and function calls), store
> the object's cache coherency next to the cache is dirty bit.
> 
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Dongwon Kim <dongwon.kim at intel.com>
> Cc: Matt Roper <matthew.d.roper at intel.com>
> ---
> diff --git a/drivers/gpu/drm/i915/i915_gem_clflush.c b/drivers/gpu/drm/i915/i915_gem_clflush.c
> index 17b207e963c2..152f16c11878 100644
> --- a/drivers/gpu/drm/i915/i915_gem_clflush.c
> +++ b/drivers/gpu/drm/i915/i915_gem_clflush.c
> @@ -139,7 +139,7 @@ void i915_gem_clflush_object(struct drm_i915_gem_object *obj,
>  	 * snooping behaviour occurs naturally as the result of our domain
>  	 * tracking.
>  	 */
> -	if (!(flags & I915_CLFLUSH_FORCE) && i915_gem_object_is_coherent(obj))
> +	if (!(flags & I915_CLFLUSH_FORCE) && obj->cache_coherent)
>  		return;
>  
>  	trace_i915_gem_object_clflush(obj);
> diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> index 0b8ae0f56675..2e5f513087a8 100644
> --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> @@ -1129,7 +1129,7 @@ i915_gem_execbuffer_move_to_gpu(struct drm_i915_gem_request *req,
>  		if (vma->exec_entry->flags & EXEC_OBJECT_ASYNC)
>  			continue;
>  
> -		if (obj->cache_dirty)
> +		if (obj->cache_dirty & ~obj->cache_coherent)
>  			i915_gem_clflush_object(obj, 0);

I was worrying whether this is in conflict with v3 of the previous
patch, i.e. we must invalidate the stale cachelines. However, we are
still dropping clflushes if cache-coherent is tagged. So that narrows
down the cache-level transitions that are relevant and in particular a
clflush on transition to snooped before the GPU execbuf is not the fix
being applied in the previous patch. Odd.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre


More information about the Intel-gfx mailing list