Re: [Intel-gfx] [PATCH 2/2] drm/i915/dmabuf: drop the flush on discrete

[Daniel Vetter <daniel@ffwll.ch>]

On Thu, Oct 21, 2021 at 01:53:32PM +0100, Matthew Auld wrote:
> We were overzealous here; even though discrete is non-LLC, it should
> still be always coherent.
> 
> Signed-off-by: Matthew Auld <matthew.auld@intel.com>
> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> ---
>  drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c
> index a45d0ec2c5b6..848e81368043 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c
> @@ -251,7 +251,8 @@ static int i915_gem_object_get_pages_dmabuf(struct drm_i915_gem_object *obj)
>  		return PTR_ERR(pages);
>  
>  	/* XXX: consider doing a vmap flush or something */
> -	if (!HAS_LLC(i915) || i915_gem_object_can_bypass_llc(obj))
> +	if ((!HAS_LLC(i915) && !IS_DGFX(i915)) ||
> +	    i915_gem_object_can_bypass_llc(obj))
>  		wbinvd_on_all_cpus();

I think a comment here why we do wbinvd and not clflush like everywhere
else here would be good. There's the XXX, but it's kinda not explaining
why vmap isn't a good idea.

Just an aside.
-Daneil
>  
>  	sg_page_sizes = i915_sg_dma_sizes(pages->sgl);
> -- 
> 2.26.3
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch