drm/i915: Reduce i915_request_alloc retirement to local context

In the continual quest to reduce the amount of global work required when
submitting requests, replace i915_retire_requests() after allocation
failure to retiring just our ring.

v2: Don't forget the list iteration included an early break, so we would
never throttle on the last request in the ring/timeline.
v3: Use the common ring_retire_requests()

References: 11abf0c5a021 ("drm/i915: Limit the backpressure for i915_request allocation")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190109215932.26454-1-chris@chris-wilson.co.uk

diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index 346418c942a25b30104b685ef145daeaae0b043d..f3c3593362ecc83940b621d665f82598d279e8e2 100644 (file)
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -477,6 +477,38 @@ submit_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state)
        return NOTIFY_DONE;
 }
 
+static void ring_retire_requests(struct intel_ring *ring)
+{
+       struct i915_request *rq, *rn;
+
+       list_for_each_entry_safe(rq, rn, &ring->request_list, ring_link) {
+               if (!i915_request_completed(rq))
+                       break;
+
+               i915_request_retire(rq);
+       }
+}
+
+static noinline struct i915_request *
+i915_request_alloc_slow(struct intel_context *ce)
+{
+       struct intel_ring *ring = ce->ring;
+       struct i915_request *rq;
+
+       if (list_empty(&ring->request_list))
+               goto out;
+
+       /* Ratelimit ourselves to prevent oom from malicious clients */
+       rq = list_last_entry(&ring->request_list, typeof(*rq), ring_link);
+       cond_synchronize_rcu(rq->rcustate);
+
+       /* Retire our old requests in the hope that we free some */
+       ring_retire_requests(ring);
+
+out:
+       return kmem_cache_alloc(ce->gem_context->i915->requests, GFP_KERNEL);
+}
+
 /**
  * i915_request_alloc - allocate a request structure
  *
@@ -559,15 +591,7 @@ 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)
-                       cond_synchronize_rcu(rq->rcustate);
-
-               rq = kmem_cache_alloc(i915->requests, GFP_KERNEL);
+               rq = i915_request_alloc_slow(ce);
                if (!rq) {
                        ret = -ENOMEM;
                        goto err_unreserve;
@@ -1218,19 +1242,6 @@ complete:
        return timeout;
 }
 
-static void ring_retire_requests(struct intel_ring *ring)
-{
-       struct i915_request *request, *next;
-
-       list_for_each_entry_safe(request, next,
-                                &ring->request_list, ring_link) {
-               if (!i915_request_completed(request))
-                       break;
-
-               i915_request_retire(request);
-       }
-}
-
 void i915_retire_requests(struct drm_i915_private *i915)
 {
        struct intel_ring *ring, *tmp;