[Intel-gfx] [PATCH v2 09/14] drm/i915: Remove the preempted request from the execution queue

[Chris Wilson]

After the request is cancelled, we then need to remove it from the
global execution timeline and return it to the context timeline, the
inverse of submit_request().

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_request.c  | 58 +++++++++++++++++++++++++++++++-
 drivers/gpu/drm/i915/i915_gem_request.h  |  3 ++
 drivers/gpu/drm/i915/intel_breadcrumbs.c | 19 +++++++++--
 3 files changed, 77 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
index 5db4fd1eabcd..56ebb653406d 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.c
+++ b/drivers/gpu/drm/i915/i915_gem_request.c
@@ -435,6 +435,55 @@ void i915_gem_request_submit(struct drm_i915_gem_request *request)
 	spin_unlock_irqrestore(&engine->timeline->lock, flags);
 }
 
+void __i915_gem_request_unsubmit(struct drm_i915_gem_request *request)
+{
+	struct intel_engine_cs *engine = request->engine;
+	struct intel_timeline *timeline;
+
+	assert_spin_locked(&engine->timeline->lock);
+
+	/* Only unwind in reverse order, required so that the per-context list
+	 * is kept in seqno/ring order.
+	 */
+	GEM_BUG_ON(request->global_seqno != engine->timeline->seqno);
+	engine->timeline->seqno--;
+
+	/* We may be recursing from the signal callback of another i915 fence */
+	spin_lock_nested(&request->lock, SINGLE_DEPTH_NESTING);
+	request->global_seqno = 0;
+	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags))
+		intel_engine_cancel_signaling(request);
+	spin_unlock(&request->lock);
+
+	/* Transfer back from the global per-engine timeline to per-context */
+	timeline = request->timeline;
+	GEM_BUG_ON(timeline == engine->timeline);
+
+	spin_lock(&timeline->lock);
+	list_move(&request->link, &timeline->requests);
+	spin_unlock(&timeline->lock);
+
+	/* We don't need to wake_up any waiters on request->execute, they
+	 * will get woken by any other event or us re-adding this request
+	 * to the engine timeline (__i915_gem_request_submit()). The waiters
+	 * should be quite adapt at finding that the request now has a new
+	 * global_seqno to the one they went to sleep on.
+	 */
+}
+
+void i915_gem_request_unsubmit(struct drm_i915_gem_request *request)
+{
+	struct intel_engine_cs *engine = request->engine;
+	unsigned long flags;
+
+	/* Will be called from irq-context when using foreign fences. */
+	spin_lock_irqsave(&engine->timeline->lock, flags);
+
+	__i915_gem_request_unsubmit(request);
+
+	spin_unlock_irqrestore(&engine->timeline->lock, flags);
+}
+
 static int __i915_sw_fence_call
 submit_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state)
 {
@@ -1016,9 +1065,11 @@ long i915_wait_request(struct drm_i915_gem_request *req,
 	if (flags & I915_WAIT_LOCKED)
 		add_wait_queue(errq, &reset);
 
-	intel_wait_init(&wait, i915_gem_request_global_seqno(req));
+	wait.tsk = current;
 
+restart:
 	reset_wait_queue(&req->execute, &exec);
+	wait.seqno = i915_gem_request_global_seqno(req);
 	if (!wait.seqno) {
 		do {
 			set_current_state(state);
@@ -1117,6 +1168,11 @@ long i915_wait_request(struct drm_i915_gem_request *req,
 		/* Only spin if we know the GPU is processing this request */
 		if (i915_spin_request(req, state, 2))
 			break;
+
+		if (i915_gem_request_global_seqno(req) != wait.seqno) {
+			intel_engine_remove_wait(req->engine, &wait);
+			goto restart;
+		}
 	}
 
 	intel_engine_remove_wait(req->engine, &wait);
diff --git a/drivers/gpu/drm/i915/i915_gem_request.h b/drivers/gpu/drm/i915/i915_gem_request.h
index b81f6709905c..5f73d8c0a38a 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.h
+++ b/drivers/gpu/drm/i915/i915_gem_request.h
@@ -274,6 +274,9 @@ void __i915_add_request(struct drm_i915_gem_request *req, bool flush_caches);
 void __i915_gem_request_submit(struct drm_i915_gem_request *request);
 void i915_gem_request_submit(struct drm_i915_gem_request *request);
 
+void __i915_gem_request_unsubmit(struct drm_i915_gem_request *request);
+void i915_gem_request_unsubmit(struct drm_i915_gem_request *request);
+
 struct intel_rps_client;
 #define NO_WAITBOOST ERR_PTR(-1)
 #define IS_RPS_CLIENT(p) (!IS_ERR(p))
diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
index 420bbaf1f1c0..39a0f87a32d0 100644
--- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -431,7 +431,14 @@ void intel_engine_remove_wait(struct intel_engine_cs *engine,
 	spin_unlock_irq(&b->lock);
 }
 
-static bool signal_complete(struct drm_i915_gem_request *request)
+static bool signal_valid(const struct drm_i915_gem_request *request)
+{
+	u32 seqno = READ_ONCE(request->global_seqno);
+
+	return seqno == request->signaling.wait.seqno;
+}
+
+static bool signal_complete(const struct drm_i915_gem_request *request)
 {
 	if (!request)
 		return false;
@@ -440,7 +447,7 @@ static bool signal_complete(struct drm_i915_gem_request *request)
 	 * signalled that this wait is already completed.
 	 */
 	if (intel_wait_complete(&request->signaling.wait))
-		return true;
+		return signal_valid(request);
 
 	/* Carefully check if the request is complete, giving time for the
 	 * seqno to be visible or if the GPU hung.
@@ -520,13 +527,21 @@ static int intel_breadcrumbs_signaler(void *arg)
 
 			i915_gem_request_put(request);
 		} else {
+			DEFINE_WAIT(exec);
+
 			if (kthread_should_stop()) {
 				GEM_BUG_ON(request);
 				break;
 			}
 
+			if (request)
+				add_wait_queue(&request->execute, &exec);
+
 			schedule();
 
+			if (request)
+				remove_wait_queue(&request->execute, &exec);
+
 			if (kthread_should_park())
 				kthread_parkme();
 		}
-- 
2.11.0