[Intel-gfx] [PATCH] drm/i915/execlists: stall on render flush before writing seqno

Chris Wilson chris at chris-wilson.co.uk
Tue Aug 27 12:03:45 UTC 2019


Quoting Chris Wilson (2019-08-27 12:54:13)
> Quite rarely we see that the CS completion event fires before the
> breadcrumb is coherent. Try rearranging the breadcrumb write sequence
> such that the CS_STALL is on the post-sync write pipecontrol.
> 
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> ---
>  drivers/gpu/drm/i915/gt/intel_lrc.c | 17 ++++++++---------
>  1 file changed, 8 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c
> index 80a3f1dbb456..669e8bd9f830 100644
> --- a/drivers/gpu/drm/i915/gt/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
> @@ -2961,18 +2961,17 @@ static u32 *gen8_emit_fini_breadcrumb(struct i915_request *request, u32 *cs)
>  
>  static u32 *gen8_emit_fini_breadcrumb_rcs(struct i915_request *request, u32 *cs)
>  {
> -       cs = gen8_emit_ggtt_write_rcs(cs,
> -                                     request->fence.seqno,
> -                                     request->timeline->hwsp_offset,
> -                                     PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH |
> -                                     PIPE_CONTROL_DEPTH_CACHE_FLUSH |
> -                                     PIPE_CONTROL_DC_FLUSH_ENABLE);
> -
>         /* XXX flush+write+CS_STALL all in one upsets gem_concurrent_blt:kbl */
>         cs = gen8_emit_pipe_control(cs,
> -                                   PIPE_CONTROL_FLUSH_ENABLE |
> -                                   PIPE_CONTROL_CS_STALL,
> +                                   PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH |
> +                                   PIPE_CONTROL_DEPTH_CACHE_FLUSH |
> +                                   PIPE_CONTROL_DC_FLUSH_ENABLE,
>                                     0);
> +       cs = gen8_emit_ggtt_write_rcs(cs,
> +                                     request->fence.seqno,
> +                                     request->timeline->hwsp_offset,
> +                                     PIPE_CONTROL_FLUSH_ENABLE |

Or perhaps we need PIPE_CONTROL_DC_FLUSH_ENABLE here.

I think that might make more sense (replace DC_FLUSH with whatever might
flush the post-sync write).
-Chris


More information about the Intel-gfx mailing list