drm/i915: Move the global sync optimisation to the timeline

Currently we try to reduce the number of synchronisations (now the
number of requests we need to wait upon) by noting that if we have
earlier waited upon a request, all subsequent requests in the timeline
will be after the wait. This only applies to requests in this timeline,
as other timelines will not be ordered by that waiter.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-30-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/i915_debugfs.c

@@ -3347,15 +3347,6 @@ static int i915_semaphore_status(struct seq_file *m, void *unused)
 		seq_putc(m, '\n');
 	}
 
-	seq_puts(m, "\nSync seqno:\n");
-	for_each_engine(engine, dev_priv, id) {
-		for (j = 0; j < num_rings; j++)
-			seq_printf(m, "  0x%08x ",
-				   engine->semaphore.sync_seqno[j]);
-		seq_putc(m, '\n');
-	}
-	seq_putc(m, '\n');
-
 	intel_runtime_pm_put(dev_priv);
 	mutex_unlock(&dev->struct_mutex);
 	return 0;

drivers/gpu/drm/i915/i915_drv.h

@@ -802,7 +802,6 @@ struct drm_i915_error_state {
 		u32 cpu_ring_tail;
 
 		u32 last_seqno;
-		u32 semaphore_seqno[I915_NUM_ENGINES - 1];
 
 		/* Register state */
 		u32 start;

drivers/gpu/drm/i915/i915_gem_request.c

@@ -238,35 +238,41 @@ static int i915_gem_check_wedge(struct drm_i915_private *dev_priv)
 	return 0;
 }
 
-static int i915_gem_init_global_seqno(struct drm_i915_private *dev_priv,
+static int i915_gem_init_global_seqno(struct drm_i915_private *i915, u32 seqno)
-				      u32 seqno)
 {
-	struct i915_gem_timeline *timeline = &dev_priv->gt.global_timeline;
+	struct i915_gem_timeline *timeline = &i915->gt.global_timeline;
 	struct intel_engine_cs *engine;
 	enum intel_engine_id id;
 	int ret;
 
 	/* Carefully retire all requests without writing to the rings */
-	ret = i915_gem_wait_for_idle(dev_priv,
+	ret = i915_gem_wait_for_idle(i915,
 				     I915_WAIT_INTERRUPTIBLE |
 				     I915_WAIT_LOCKED);
 	if (ret)
 		return ret;
 
-	i915_gem_retire_requests(dev_priv);
+	i915_gem_retire_requests(i915);
 
 	/* If the seqno wraps around, we need to clear the breadcrumb rbtree */
 	if (!i915_seqno_passed(seqno, timeline->next_seqno)) {
-		while (intel_kick_waiters(dev_priv) ||
+		while (intel_kick_waiters(i915) || intel_kick_signalers(i915))
-		       intel_kick_signalers(dev_priv))
 			yield();
 		yield();
 	}
 
 	/* Finally reset hw state */
-	for_each_engine(engine, dev_priv, id)
+	for_each_engine(engine, i915, id)
 		intel_engine_init_global_seqno(engine, seqno);
 
+	list_for_each_entry(timeline, &i915->gt.timelines, link) {
+		for_each_engine(engine, i915, id) {
+			struct intel_timeline *tl = &timeline->engine[id];
+
+			memset(tl->sync_seqno, 0, sizeof(tl->sync_seqno));
+		}
+	}
+
 	return 0;
 }
 
@@ -462,7 +468,7 @@ static int
 i915_gem_request_await_request(struct drm_i915_gem_request *to,
 			       struct drm_i915_gem_request *from)
 {
-	int idx, ret;
+	int ret;
 
 	GEM_BUG_ON(to == from);
 
@@ -483,8 +489,7 @@ i915_gem_request_await_request(struct drm_i915_gem_request *to,
 		return ret < 0 ? ret : 0;
 	}
 
-	idx = intel_engine_sync_index(from->engine, to->engine);
+	if (from->global_seqno <= to->timeline->sync_seqno[from->engine->id])
-	if (from->global_seqno <= from->engine->semaphore.sync_seqno[idx])
 		return 0;
 
 	trace_i915_gem_ring_sync_to(to, from);
@@ -502,7 +507,7 @@ i915_gem_request_await_request(struct drm_i915_gem_request *to,
 			return ret;
 	}
 
-	from->engine->semaphore.sync_seqno[idx] = from->global_seqno;
+	to->timeline->sync_seqno[from->engine->id] = from->global_seqno;
 	return 0;
 }
 

drivers/gpu/drm/i915/i915_gem_timeline.h

@@ -48,6 +48,7 @@ struct intel_timeline {
 	 * struct_mutex.
 	 */
 	struct i915_gem_active last_request;
+	u32 sync_seqno[I915_NUM_ENGINES];
 
 	struct i915_gem_timeline *common;
 };

drivers/gpu/drm/i915/i915_gpu_error.c

@@ -415,17 +415,13 @@ static void error_print_engine(struct drm_i915_error_state_buf *m,
 	if (INTEL_GEN(m->i915) >= 6) {
 		err_printf(m, "  RC PSMI: 0x%08x\n", ee->rc_psmi);
 		err_printf(m, "  FAULT_REG: 0x%08x\n", ee->fault_reg);
-		err_printf(m, "  SYNC_0: 0x%08x [last synced 0x%08x]\n",
+		err_printf(m, "  SYNC_0: 0x%08x\n",
-			   ee->semaphore_mboxes[0],
+			   ee->semaphore_mboxes[0]);
-			   ee->semaphore_seqno[0]);
+		err_printf(m, "  SYNC_1: 0x%08x\n",
-		err_printf(m, "  SYNC_1: 0x%08x [last synced 0x%08x]\n",
+			   ee->semaphore_mboxes[1]);
-			   ee->semaphore_mboxes[1],
+		if (HAS_VEBOX(m->i915))
-			   ee->semaphore_seqno[1]);
+			err_printf(m, "  SYNC_2: 0x%08x\n",
-		if (HAS_VEBOX(m->i915)) {
+				   ee->semaphore_mboxes[2]);
-			err_printf(m, "  SYNC_2: 0x%08x [last synced 0x%08x]\n",
-				   ee->semaphore_mboxes[2],
-				   ee->semaphore_seqno[2]);
-		}
 	}
 	if (USES_PPGTT(m->i915)) {
 		err_printf(m, "  GFX_MODE: 0x%08x\n", ee->vm_info.gfx_mode);
@@ -972,6 +968,26 @@ static void i915_gem_record_fences(struct drm_i915_private *dev_priv,
 	}
 }
 
+static inline u32
+gen8_engine_sync_index(struct intel_engine_cs *engine,
+		       struct intel_engine_cs *other)
+{
+	int idx;
+
+	/*
+	 * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
+	 * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
+	 * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
+	 * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
+	 * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
+	 */
+
+	idx = (other - engine) - 1;
+	if (idx < 0)
+		idx += I915_NUM_ENGINES;
+
+	return idx;
+}
 
 static void gen8_record_semaphore_state(struct drm_i915_error_state *error,
 					struct intel_engine_cs *engine,
@@ -995,10 +1011,9 @@ static void gen8_record_semaphore_state(struct drm_i915_error_state *error,
 		signal_offset =
 			(GEN8_SIGNAL_OFFSET(engine, id) & (PAGE_SIZE - 1)) / 4;
 		tmp = error->semaphore->pages[0];
-		idx = intel_engine_sync_index(engine, to);
+		idx = gen8_engine_sync_index(engine, to);
 
 		ee->semaphore_mboxes[idx] = tmp[signal_offset];
-		ee->semaphore_seqno[idx] = engine->semaphore.sync_seqno[idx];
 	}
 }
 
@@ -1009,14 +1024,9 @@ static void gen6_record_semaphore_state(struct intel_engine_cs *engine,
 
 	ee->semaphore_mboxes[0] = I915_READ(RING_SYNC_0(engine->mmio_base));
 	ee->semaphore_mboxes[1] = I915_READ(RING_SYNC_1(engine->mmio_base));
-	ee->semaphore_seqno[0] = engine->semaphore.sync_seqno[0];
+	if (HAS_VEBOX(dev_priv))
-	ee->semaphore_seqno[1] = engine->semaphore.sync_seqno[1];
-
-	if (HAS_VEBOX(dev_priv)) {
 		ee->semaphore_mboxes[2] =
 			I915_READ(RING_SYNC_2(engine->mmio_base));
-		ee->semaphore_seqno[2] = engine->semaphore.sync_seqno[2];
-	}
 }
 
 static void error_record_engine_waiters(struct intel_engine_cs *engine,

drivers/gpu/drm/i915/intel_engine_cs.c

@@ -204,8 +204,6 @@ void intel_engine_init_global_seqno(struct intel_engine_cs *engine, u32 seqno)
 				       I915_NUM_ENGINES * gen8_semaphore_seqno_size);
 		kunmap(page);
 	}
-	memset(engine->semaphore.sync_seqno, 0,
-	       sizeof(engine->semaphore.sync_seqno));
 
 	intel_write_status_page(engine, I915_GEM_HWS_INDEX, seqno);
 	if (engine->irq_seqno_barrier)

drivers/gpu/drm/i915/intel_ringbuffer.c

@@ -2003,9 +2003,6 @@ static int intel_init_ring_buffer(struct intel_engine_cs *engine)
 
 	intel_engine_setup_common(engine);
 
-	memset(engine->semaphore.sync_seqno, 0,
-	       sizeof(engine->semaphore.sync_seqno));
-
 	ret = intel_engine_init_common(engine);
 	if (ret)
 		goto error;

drivers/gpu/drm/i915/intel_ringbuffer.h

@@ -314,8 +314,6 @@ struct intel_engine_cs {
 	 *  ie. transpose of f(x, y)
 	 */
 	struct {
-		u32	sync_seqno[I915_NUM_ENGINES-1];
-
 		union {
 #define GEN6_SEMAPHORE_LAST	VECS_HW
 #define GEN6_NUM_SEMAPHORES	(GEN6_SEMAPHORE_LAST + 1)
@@ -385,27 +383,6 @@ intel_engine_flag(const struct intel_engine_cs *engine)
 	return 1 << engine->id;
 }
 
-static inline u32
-intel_engine_sync_index(struct intel_engine_cs *engine,
-			struct intel_engine_cs *other)
-{
-	int idx;
-
-	/*
-	 * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
-	 * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
-	 * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
-	 * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
-	 * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
-	 */
-
-	idx = (other->id - engine->id) - 1;
-	if (idx < 0)
-		idx += I915_NUM_ENGINES;
-
-	return idx;
-}
-
 static inline void
 intel_flush_status_page(struct intel_engine_cs *engine, int reg)
 {