[Intel-gfx] [PATCH 3/3] drm/i915: run shmem_pread_fast() after page fault

[Daniel Vetter]

On Fri, Jul 19, 2013 at 01:51:26PM +0800, Xiong Zhang wrote:
> fault_in_multipages_writeable() will load fault page into physical
> memory, then pread can run fast path, no need to run slow path
> 
> Signed-off-by: Xiong Zhang <xiong.y.zhang at intel.com>

Hm, this avoids going through the slowpath for the first page. Does this
have an impact on micro-benchmarks (we have a few) or is this just
something you've spotted? I'm a bit vary of making the already complicated
shmem pread/pwrite paths even more complicated. Chris?
-Daniel

> ---
>  drivers/gpu/drm/i915/i915_gem.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index f194d7e..982df1e 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -459,6 +459,7 @@ i915_gem_shmem_pread(struct drm_device *dev,
>  		page_do_bit17_swizzling = obj_do_bit17_swizzling &&
>  			(page_to_phys(page) & (1 << 17)) != 0;
>  
> +read_again:
>  		ret = shmem_pread_fast(page, shmem_page_offset, page_length,
>  				       user_data, page_do_bit17_swizzling,
>  				       needs_clflush);
> @@ -475,6 +476,8 @@ i915_gem_shmem_pread(struct drm_device *dev,
>  			 * and just continue. */
>  			(void)ret;
>  			prefaulted = 1;
> +			mutex_lock(&dev->struct_mutex);
> +			goto read_again;
>  		}
>  
>  		ret = shmem_pread_slow(page, shmem_page_offset, page_length,
> -- 
> 1.8.3.2
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx

-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch