drm/i915: Stop tracking timeline->inflight_seqnos

In commit 9b6586ae9f6b ("drm/i915: Keep a global seqno per-engine"), we
moved from a global inflight counter to per-engine counters in the
hope that will be easy to run concurrently in future. However, with the
advent of the desire to move requests between engines, we do need a
global counter to preserve the semantics that no engine wraps in the
middle of a submit. (Although this semantic is now only required for gen7
semaphore support, which only supports greater-then comparisons!)

v2: Keep a global counter of all requests ever submitted and force the
reset when it wraps.

References: 9b6586ae9f6b ("drm/i915: Keep a global seqno per-engine")
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/20180430131503.5375-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/i915_debugfs.c

@@ -1340,10 +1340,9 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
 		struct rb_node *rb;
 
 		seq_printf(m, "%s:\n", engine->name);
-		seq_printf(m, "\tseqno = %x [current %x, last %x], inflight %d\n",
+		seq_printf(m, "\tseqno = %x [current %x, last %x]\n",
 			   engine->hangcheck.seqno, seqno[id],
-			   intel_engine_last_submit(engine),
-			   engine->timeline->inflight_seqnos);
+			   intel_engine_last_submit(engine));
 		seq_printf(m, "\twaiters? %s, fake irq active? %s, stalled? %s\n",
 			   yesno(intel_engine_has_waiter(engine)),
 			   yesno(test_bit(engine->id,

drivers/gpu/drm/i915/i915_drv.h

@@ -2062,6 +2062,7 @@ struct drm_i915_private {
 		struct list_head timelines;
 		struct i915_gem_timeline global_timeline;
 		u32 active_requests;
+		u32 request_serial;
 
 		/**
 		 * Is the GPU currently considered idle, or busy executing

drivers/gpu/drm/i915/i915_gem_timeline.h

@@ -37,12 +37,6 @@ struct intel_timeline {
 	u64 fence_context;
 	u32 seqno;
 
-	/**
-	 * Count of outstanding requests, from the time they are constructed
-	 * to the moment they are retired. Loosely coupled to hardware.
-	 */
-	u32 inflight_seqnos;
-
 	spinlock_t lock;
 
 	/**

drivers/gpu/drm/i915/i915_request.c

@@ -241,6 +241,7 @@ static int reset_all_global_seqno(struct drm_i915_private *i915, u32 seqno)
 			       sizeof(timeline->engine[id].global_sync));
 	}
 
+	i915->gt.request_serial = seqno;
 	return 0;
 }
 
@@ -257,18 +258,22 @@ int i915_gem_set_global_seqno(struct drm_device *dev, u32 seqno)
 	return reset_all_global_seqno(i915, seqno - 1);
 }
 
-static int reserve_engine(struct intel_engine_cs *engine)
+static int reserve_gt(struct drm_i915_private *i915)
 {
-	struct drm_i915_private *i915 = engine->i915;
-	u32 active = ++engine->timeline->inflight_seqnos;
-	u32 seqno = engine->timeline->seqno;
 	int ret;
 
-	/* Reservation is fine until we need to wrap around */
-	if (unlikely(add_overflows(seqno, active))) {
+	/*
+	 * Reservation is fine until we may need to wrap around
+	 *
+	 * By incrementing the serial for every request, we know that no
+	 * individual engine may exceed that serial (as each is reset to 0
+	 * on any wrap). This protects even the most pessimistic of migrations
+	 * of every request from all engines onto just one.
+	 */
+	while (unlikely(++i915->gt.request_serial == 0)) {
 		ret = reset_all_global_seqno(i915, 0);
 		if (ret) {
-			engine->timeline->inflight_seqnos--;
+			i915->gt.request_serial--;
 			return ret;
 		}
 	}
@@ -279,15 +284,10 @@ static int reserve_engine(struct intel_engine_cs *engine)
 	return 0;
 }
 
-static void unreserve_engine(struct intel_engine_cs *engine)
+static void unreserve_gt(struct drm_i915_private *i915)
 {
-	struct drm_i915_private *i915 = engine->i915;
-
 	if (!--i915->gt.active_requests)
 		i915_gem_park(i915);
-
-	GEM_BUG_ON(!engine->timeline->inflight_seqnos);
-	engine->timeline->inflight_seqnos--;
 }
 
 void i915_gem_retire_noop(struct i915_gem_active *active,
@@ -362,7 +362,6 @@ static void i915_request_retire(struct i915_request *request)
 	list_del_init(&request->link);
 	spin_unlock_irq(&engine->timeline->lock);
 
-	unreserve_engine(request->engine);
 	advance_ring(request);
 
 	free_capture_list(request);
@@ -424,6 +423,8 @@ static void i915_request_retire(struct i915_request *request)
 	}
 	spin_unlock_irq(&request->lock);
 
+	unreserve_gt(request->i915);
+
 	i915_sched_node_fini(request->i915, &request->sched);
 	i915_request_put(request);
 }
@@ -642,7 +643,7 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 		return ERR_CAST(ring);
 	GEM_BUG_ON(!ring);
 
-	ret = reserve_engine(engine);
+	ret = reserve_gt(i915);
 	if (ret)
 		goto err_unpin;
 
@@ -784,7 +785,7 @@ err_unwind:
 
 	kmem_cache_free(i915->requests, rq);
 err_unreserve:
-	unreserve_engine(engine);
+	unreserve_gt(i915);
 err_unpin:
 	engine->context_unpin(engine, ctx);
 	return ERR_PTR(ret);

drivers/gpu/drm/i915/intel_engine_cs.c

@@ -1321,12 +1321,11 @@ void intel_engine_dump(struct intel_engine_cs *engine,
 	if (i915_terminally_wedged(&engine->i915->gpu_error))
 		drm_printf(m, "*** WEDGED ***\n");
 
-	drm_printf(m, "\tcurrent seqno %x, last %x, hangcheck %x [%d ms], inflight %d\n",
+	drm_printf(m, "\tcurrent seqno %x, last %x, hangcheck %x [%d ms]\n",
 		   intel_engine_get_seqno(engine),
 		   intel_engine_last_submit(engine),
 		   engine->hangcheck.seqno,
-		   jiffies_to_msecs(jiffies - engine->hangcheck.action_timestamp),
-		   engine->timeline->inflight_seqnos);
+		   jiffies_to_msecs(jiffies - engine->hangcheck.action_timestamp));
 	drm_printf(m, "\tReset count: %d (global %d)\n",
 		   i915_reset_engine_count(error, engine),
 		   i915_reset_count(error));