[Intel-gfx] [PATCH 3/3] drm/i915: Wait for the previous RCU grace period, not request completion

[Tvrtko Ursulin]

On 12/09/2018 17:40, Chris Wilson wrote:
> Under mempressure, our goal is to allow ourselves sufficient time to
> reclaim the RCU protected slabs without overly penalizing our clients.
> Currently, we use a 1 jiffie wait if the client is still active as a
> means of throttling the allocations, but we can instead wait for the end
> of the RCU grace period of the clients previous allocation.

Why did you opt for three patches changing the same code and just squash 
to last?

Regards,

Tvrtko

> Suggested-by: Daniel Vetter <daniel.vetter at ffwll.ch>
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> ---
>   drivers/gpu/drm/i915/i915_request.c | 14 ++++++--------
>   drivers/gpu/drm/i915/i915_request.h |  8 ++++++++
>   2 files changed, 14 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> index 72bcb4ca0c45..a492385b2089 100644
> --- a/drivers/gpu/drm/i915/i915_request.c
> +++ b/drivers/gpu/drm/i915/i915_request.c
> @@ -732,17 +732,13 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
>   	rq = kmem_cache_alloc(i915->requests,
>   			      GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN);
>   	if (unlikely(!rq)) {
> +		i915_retire_requests(i915);
> +
>   		/* Ratelimit ourselves to prevent oom from malicious clients */
>   		rq = i915_gem_active_raw(&ce->ring->timeline->last_request,
>   					 &i915->drm.struct_mutex);
> -		if (rq && i915_request_wait(rq,
> -					    I915_WAIT_LOCKED |
> -					    I915_WAIT_INTERRUPTIBLE,
> -					    1) == -EINTR) {
> -			ret = -EINTR;
> -			goto err_unreserve;
> -		}
> -		i915_retire_requests(i915);
> +		if (rq)
> +			cond_synchronize_rcu(rq->rcustate);
>   
>   		/*
>   		 * We've forced the client to stall and catch up with whatever
> @@ -762,6 +758,8 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
>   		}
>   	}
>   
> +	rq->rcustate = get_state_synchronize_rcu();
> +
>   	INIT_LIST_HEAD(&rq->active_list);
>   	rq->i915 = i915;
>   	rq->engine = engine;
> diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
> index 9898301ab7ef..7fa94b024968 100644
> --- a/drivers/gpu/drm/i915/i915_request.h
> +++ b/drivers/gpu/drm/i915/i915_request.h
> @@ -100,6 +100,14 @@ struct i915_request {
>   	struct i915_timeline *timeline;
>   	struct intel_signal_node signaling;
>   
> +	/*
> +	 * The rcu epoch of when this request was allocated. Used to judiciously
> +	 * apply backpressure on future allocations to ensure that under
> +	 * mempressure there is sufficient RCU ticks for us to reclaim our
> +	 * RCU protected slabs.
> +	 */
> +	unsigned long rcustate;
> +
>   	/*
>   	 * Fences for the various phases in the request's lifetime.
>   	 *
>