drm/i915: Keep a global seqno per-engine

Replace the global device seqno with one for each engine, and account
for in-flight seqno on each separately. This is consistent with
dma-fence as each timeline has separate fence-contexts for each engine
and a seqno is only ordered within a fence-context (i.e.  seqno do not
need to be ordered wrt to other engines, just ordered within a single
engine). This is required to enable request rewinding for preemption on
individual engines (we have to rewind the global seqno to avoid
overflow, and we do not have to rewind all engines just to preempt one.)

v2: Rename active_seqno to inflight_seqnos to more clearly indicate that
it is a counter and not equivalent to the existing seqno. Update
functions that operated on active_seqno similarly.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170223074422.4125-3-chris@chris-wilson.co.uk

1 files changed, 45 insertions, 38 deletions

diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
index 12b06e055490..b94ae6cf9837 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.c
+++ b/drivers/gpu/drm/i915/i915_gem_request.c
@@ -198,6 +198,12 @@ i915_priotree_init(struct i915_priotree *pt)
 	pt->priority = INT_MIN;
 }
 
+static void unreserve_seqno(struct intel_engine_cs *engine)
+{
+	GEM_BUG_ON(!engine->timeline->inflight_seqnos);
+	engine->timeline->inflight_seqnos--;
+}
+
 void i915_gem_retire_noop(struct i915_gem_active *active,
 			  struct drm_i915_gem_request *request)
 {
@@ -237,6 +243,7 @@ static void i915_gem_request_retire(struct drm_i915_gem_request *request)
 				 &request->i915->gt.idle_work,
 				 msecs_to_jiffies(100));
 	}
+	unreserve_seqno(request->engine);
 
 	/* Walk through the active list, calling retire on each. This allows
 	 * objects to track their GPU activity and mark themselves as idle
@@ -307,7 +314,7 @@ void i915_gem_request_retire_upto(struct drm_i915_gem_request *req)
 	} while (tmp != req);
 }
 
-static int i915_gem_init_global_seqno(struct drm_i915_private *i915, u32 seqno)
+static int reset_all_global_seqno(struct drm_i915_private *i915, u32 seqno)
 {
 	struct i915_gem_timeline *timeline = &i915->gt.global_timeline;
 	struct intel_engine_cs *engine;
@@ -325,15 +332,19 @@ static int i915_gem_init_global_seqno(struct drm_i915_private *i915, u32 seqno)
 	GEM_BUG_ON(i915->gt.active_requests > 1);
 
 	/* If the seqno wraps around, we need to clear the breadcrumb rbtree */
-	if (!i915_seqno_passed(seqno, atomic_read(&timeline->seqno))) {
-		while (intel_breadcrumbs_busy(i915))
-			cond_resched(); /* spin until threads are complete */
-	}
-	atomic_set(&timeline->seqno, seqno);
+	for_each_engine(engine, i915, id) {
+		struct intel_timeline *tl = &timeline->engine[id];
 
-	/* Finally reset hw state */
-	for_each_engine(engine, i915, id)
+		if (!i915_seqno_passed(seqno, tl->seqno)) {
+			/* spin until threads are complete */
+			while (intel_breadcrumbs_busy(engine))
+				cond_resched();
+		}
+
+		/* Finally reset hw state */
+		tl->seqno = seqno;
 		intel_engine_init_global_seqno(engine, seqno);
+	}
 
 	list_for_each_entry(timeline, &i915->gt.timelines, link) {
 		for_each_engine(engine, i915, id) {
@@ -358,37 +369,38 @@ int i915_gem_set_global_seqno(struct drm_device *dev, u32 seqno)
 	/* HWS page needs to be set less than what we
 	 * will inject to ring
 	 */
-	return i915_gem_init_global_seqno(dev_priv, seqno - 1);
+	return reset_all_global_seqno(dev_priv, seqno - 1);
 }
 
-static int reserve_global_seqno(struct drm_i915_private *i915)
+static int reserve_seqno(struct intel_engine_cs *engine)
 {
-	u32 active_requests = ++i915->gt.active_requests;
-	u32 seqno = atomic_read(&i915->gt.global_timeline.seqno);
+	u32 active = ++engine->timeline->inflight_seqnos;
+	u32 seqno = engine->timeline->seqno;
 	int ret;
 
 	/* Reservation is fine until we need to wrap around */
-	if (likely(seqno + active_requests > seqno))
+	if (likely(!add_overflows(seqno, active)))
 		return 0;
 
-	ret = i915_gem_init_global_seqno(i915, 0);
+	/* Even though we are tracking inflight seqno individually on each
+	 * engine, other engines may be observing us using hw semaphores and
+	 * so we need to idle all engines before wrapping around this engine.
+	 * As all engines are then idle, we can reset the seqno on all, so
+	 * we don't stall in quick succession if each engine is being
+	 * similarly utilized.
+	 */
+	ret = reset_all_global_seqno(engine->i915, 0);
 	if (ret) {
-		i915->gt.active_requests--;
+		engine->timeline->inflight_seqnos--;
 		return ret;
 	}
 
 	return 0;
 }
 
-static u32 __timeline_get_seqno(struct i915_gem_timeline *tl)
-{
-	/* seqno only incremented under a mutex */
-	return ++tl->seqno.counter;
-}
-
-static u32 timeline_get_seqno(struct i915_gem_timeline *tl)
+static u32 timeline_get_seqno(struct intel_timeline *tl)
 {
-	return atomic_inc_return(&tl->seqno);
+	return ++tl->seqno;
 }
 
 void __i915_gem_request_submit(struct drm_i915_gem_request *request)
@@ -402,14 +414,10 @@ void __i915_gem_request_submit(struct drm_i915_gem_request *request)
 	GEM_BUG_ON(timeline == request->timeline);
 	assert_spin_locked(&timeline->lock);
 
-	seqno = timeline_get_seqno(timeline->common);
+	seqno = timeline_get_seqno(timeline);
 	GEM_BUG_ON(!seqno);
 	GEM_BUG_ON(i915_seqno_passed(intel_engine_get_seqno(engine), seqno));
 
-	GEM_BUG_ON(i915_seqno_passed(timeline->last_submitted_seqno, seqno));
-	request->previous_seqno = timeline->last_submitted_seqno;
-	timeline->last_submitted_seqno = seqno;
-
 	/* We may be recursing from the signal callback of another i915 fence */
 	spin_lock_nested(&request->lock, SINGLE_DEPTH_NESTING);
 	request->global_seqno = seqno;
@@ -516,7 +524,7 @@ i915_gem_request_alloc(struct intel_engine_cs *engine,
 	if (ret)
 		return ERR_PTR(ret);
 
-	ret = reserve_global_seqno(dev_priv);
+	ret = reserve_seqno(engine);
 	if (ret)
 		goto err_unpin;
 
@@ -568,7 +576,7 @@ i915_gem_request_alloc(struct intel_engine_cs *engine,
 		       &i915_fence_ops,
 		       &req->lock,
 		       req->timeline->fence_context,
-		       __timeline_get_seqno(req->timeline->common));
+		       timeline_get_seqno(req->timeline));
 
 	/* We bump the ref for the fence chain */
 	i915_sw_fence_init(&i915_gem_request_get(req)->submit, submit_notify);
@@ -613,6 +621,8 @@ i915_gem_request_alloc(struct intel_engine_cs *engine,
 	 */
 	req->head = req->ring->tail;
 
+	/* Check that we didn't interrupt ourselves with a new request */
+	GEM_BUG_ON(req->timeline->seqno != req->fence.seqno);
 	return req;
 
 err_ctx:
@@ -623,7 +633,7 @@ err_ctx:
 
 	kmem_cache_free(dev_priv->requests, req);
 err_unreserve:
-	dev_priv->gt.active_requests--;
+	unreserve_seqno(engine);
 err_unpin:
 	engine->context_unpin(engine, ctx);
 	return ERR_PTR(ret);
@@ -836,8 +846,7 @@ void __i915_add_request(struct drm_i915_gem_request *request, bool flush_caches)
 	 * our i915_gem_request_alloc() and called __i915_add_request() before
 	 * us, the timeline will hold its seqno which is later than ours.
 	 */
-	GEM_BUG_ON(i915_seqno_passed(timeline->last_submitted_seqno,
-				     request->fence.seqno));
+	GEM_BUG_ON(timeline->seqno != request->fence.seqno);
 
 	/*
 	 * To ensure that this call will not fail, space for its emissions
@@ -891,16 +900,14 @@ void __i915_add_request(struct drm_i915_gem_request *request, bool flush_caches)
 	list_add_tail(&request->link, &timeline->requests);
 	spin_unlock_irq(&timeline->lock);
 
-	GEM_BUG_ON(i915_seqno_passed(timeline->last_submitted_seqno,
-				     request->fence.seqno));
-
-	timeline->last_submitted_seqno = request->fence.seqno;
+	GEM_BUG_ON(timeline->seqno != request->fence.seqno);
 	i915_gem_active_set(&timeline->last_request, request);
 
 	list_add_tail(&request->ring_link, &ring->request_list);
 	request->emitted_jiffies = jiffies;
 
-	i915_gem_mark_busy(engine);
+	if (!request->i915->gt.active_requests++)
+		i915_gem_mark_busy(engine);
 
 	/* Let the backend know a new request has arrived that may need
 	 * to adjust the existing execution schedule due to a high priority