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

[Chris Wilson]

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.

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.
 	 *
-- 
2.19.0