[Intel-gfx] [PATCH] drm/i915: Push mm switch immediately to ring

[Chris Wilson]

On Mon, Mar 16, 2015 at 05:58:12PM +0200, Mika Kuoppala wrote:
> Sometimes when first batch is run with blitter ring,
> @@ -931,7 +931,6 @@ static int gen7_mm_switch(struct i915_hw_ppgtt *ppgtt,
>  	intel_ring_emit(ring, RING_PP_DIR_BASE(ring));
>  	intel_ring_emit(ring, get_pd_offset(ppgtt));
>  	intel_ring_emit(ring, MI_NOOP);
> -	intel_ring_advance(ring);
>  
>  	/* XXX: RCS is the only one to auto invalidate the TLBs? */
>  	if (ring->id != RCS) {
> @@ -940,6 +939,8 @@ static int gen7_mm_switch(struct i915_hw_ppgtt *ppgtt,
>  			return ret;
>  	}
>  
> +	__intel_ring_advance(ring);
> +

I really would like a comment before each of these, something like
/* XXX This papers over a bug loading PD prior to batch execution */

This last chunk is buggy. The earlier unadvanced return will break the
ringbuffer state tracking. Do you have equal success just with the plain
s/intel_ring_advance/__intel_ring_advance/ ?
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre