[Intel-gfx] [PATCH] drm/i915: Workaround to avoid lite restore with HEAD==TAIL

Daniel Vetter daniel at ffwll.ch
Fri Apr 10 03:01:08 PDT 2015


On Fri, Apr 10, 2015 at 10:32:33AM +0100, Michel Thierry wrote:
> WaIdleLiteRestore is an execlists-only workaround, and requires the driver
> to ensure that any context always has HEAD!=TAIL when attempting lite
> restore.
> 
> Add two extra MI_NOOP instructions at the end of each request, but keep
> the requests tail pointing before the MI_NOOPs.
> 
> If we submit a context to the ELSP which has previously been submitted,
> move the tail pointer past the MI_NOOPs. This ensures HEAD!=TAIL.
> 
> Signed-off-by: Thomas Daniel <thomas.daniel at intel.com>
> Signed-off-by: Michel Thierry <michel.thierry at intel.com>

Is there an igt to provoke this? I'm thinking of submitting tiny noop
batches on 2 alternating contexts. That should be able to make this go
boom pretty reliable, we only need a slight delay in processing the
execlist completion interrupt.
-Daniel

> ---
>  drivers/gpu/drm/i915/i915_gem.c  | 11 +++++++++++
>  drivers/gpu/drm/i915/intel_lrc.c | 29 +++++++++++++++++++++++++++++
>  2 files changed, 40 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index ca522c9..a44070a 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -2382,6 +2382,17 @@ int __i915_add_request(struct intel_engine_cs *ring,
>  	request->head = request_start;
>  	request->tail = intel_ring_get_tail(ringbuf);
>  
> +	if (i915.enable_execlists &&
> +	    (IS_GEN8(ring->dev) || IS_GEN9(ring->dev))) {
> +		/*
> +		 * Here we add two extra NOOPs as padding to avoid
> +		 * lite restore of a context with HEAD==TAIL.
> +		 */
> +		intel_logical_ring_emit(ringbuf, MI_NOOP);
> +		intel_logical_ring_emit(ringbuf, MI_NOOP);
> +		intel_logical_ring_advance(ringbuf);
> +	}
> +
>  	/* Whilst this request exists, batch_obj will be on the
>  	 * active_list, and so will hold the active reference. Only when this
>  	 * request is retired will the the batch_obj be moved onto the
> diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
> index f18f96c..8acedca 100644
> --- a/drivers/gpu/drm/i915/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/intel_lrc.c
> @@ -413,6 +413,26 @@ static void execlists_context_unqueue(struct intel_engine_cs *ring)
>  		}
>  	}
>  
> +	if (IS_GEN8(ring->dev) || IS_GEN9(ring->dev)) {
> +		/*
> +		 * WaIdleLiteRestore: make sure we never cause a lite
> +		 * restore with HEAD==TAIL
> +		 */
> +		if (req0 && req0->elsp_submitted == 1) {
> +			/*
> +			 * Consume the buffer NOOPs to ensure HEAD != TAIL when
> +			 * submitting. elsp_submitted can only be >1 after
> +			 * reset, in which case we don't need the workaround as
> +			 * a lite restore will not occur.
> +			 */
> +			struct intel_ringbuffer *ringbuf;
> +
> +			ringbuf = req0->ctx->engine[ring->id].ringbuf;
> +			req0->tail += 8;
> +			req0->tail &= ringbuf->size - 1;
> +		}
> +	}
> +
>  	WARN_ON(req1 && req1->elsp_submitted);
>  
>  	execlists_submit_contexts(ring, req0->ctx, req0->tail,
> @@ -829,6 +849,15 @@ static int intel_logical_ring_begin(struct intel_ringbuffer *ringbuf,
>  	struct drm_i915_private *dev_priv = dev->dev_private;
>  	int ret;
>  
> +	if (IS_GEN8(dev) || IS_GEN9(dev)) {
> +		/*
> +		 * Reserve space for 2 NOOPs at the end of each request to be
> +		 * used as a workaround for not being allowed to do lite
> +		 * restore with HEAD==TAIL.
> +		 */
> +		num_dwords += 2;
> +	}
> +
>  	ret = i915_gem_check_wedge(&dev_priv->gpu_error,
>  				   dev_priv->mm.interruptible);
>  	if (ret)
> -- 
> 2.1.1
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx

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


More information about the Intel-gfx mailing list