drm/i915: Apply a mb between emitting the request and hangcheck
authorChris Wilson <chris@chris-wilson.co.uk>
Thu, 7 Apr 2016 06:29:17 +0000 (07:29 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Fri, 8 Apr 2016 10:44:48 +0000 (11:44 +0100)
Seal the request and mark it as pending execution before we submit it to
hardware. We assume that the actual submission cannot fail (that
guarantee is provided by preallocating space in the request for the
submission). As we may inspect this state without holding any locks
during hangcheck we should apply a barrier to ensure that we do
not see a more recent value in the HWS than we are tracking.

Based on a patch by Mika Kuoppala.

Suggested-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1460010558-10705-8-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
drivers/gpu/drm/i915/i915_gem.c
drivers/gpu/drm/i915/i915_irq.c

index 65f18f583ae1e14b549fcee66587da553088988e..339d183f05b0a0eb49f383a2d3940bacccb6d697 100644 (file)
@@ -2575,6 +2575,28 @@ void __i915_add_request(struct drm_i915_gem_request *request,
                WARN(ret, "*_ring_flush_all_caches failed: %d!\n", ret);
        }
 
+       trace_i915_gem_request_add(request);
+
+       request->head = request_start;
+
+       /* 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
+        * inactive_list and lose its active reference. Hence we do not need
+        * to explicitly hold another reference here.
+        */
+       request->batch_obj = obj;
+
+       /* Seal the request and mark it as pending execution. Note that
+        * we may inspect this state, without holding any locks, during
+        * hangcheck. Hence we apply the barrier to ensure that we do not
+        * see a more recent value in the hws than we are tracking.
+        */
+       request->emitted_jiffies = jiffies;
+       request->previous_seqno = engine->last_submitted_seqno;
+       smp_store_mb(engine->last_submitted_seqno, request->seqno);
+       list_add_tail(&request->list, &engine->request_list);
+
        /* Record the position of the start of the request so that
         * should we detect the updated seqno part-way through the
         * GPU processing the request, we never over-estimate the
@@ -2592,23 +2614,6 @@ void __i915_add_request(struct drm_i915_gem_request *request,
        /* Not allowed to fail! */
        WARN(ret, "emit|add_request failed: %d!\n", ret);
 
-       request->head = request_start;
-
-       /* 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
-        * inactive_list and lose its active reference. Hence we do not need
-        * to explicitly hold another reference here.
-        */
-       request->batch_obj = obj;
-
-       request->emitted_jiffies = jiffies;
-       request->previous_seqno = engine->last_submitted_seqno;
-       engine->last_submitted_seqno = request->seqno;
-       list_add_tail(&request->list, &engine->request_list);
-
-       trace_i915_gem_request_add(request);
-
        i915_queue_hangcheck(engine->dev);
 
        queue_delayed_work(dev_priv->wq,
index c36aa64bb1072c922b71cc11291bb3d570720a2e..cc908124983681a9b3b83c1aac6919945219f21a 100644 (file)
@@ -2806,7 +2806,8 @@ static bool
 ring_idle(struct intel_engine_cs *engine, u32 seqno)
 {
        return (list_empty(&engine->request_list) ||
-               i915_seqno_passed(seqno, engine->last_submitted_seqno));
+               i915_seqno_passed(seqno,
+                                 READ_ONCE(engine->last_submitted_seqno)));
 }
 
 static bool