drm/i915: Priority boost for waiting clients

Latency is in the eye of the beholder. In the case where a client stops
and waits for the gpu, give that request chain a small priority boost
(not so that it overtakes higher priority clients, to preserve the
external ordering) so that ideally the wait completes earlier.

v2: Tvrtko recommends to keep the boost-from-user-stall as small as
possible and to allow new client flows to be preferred for interactivity
over stalls.

Testcase: igt/gem_sync/switch-default
Signed-off-by: Chris Wilson <[e-mail]>
Cc: Tvrtko Ursulin <[e-mail]>
Cc: Joonas Lahtinen <[e-mail]>
Cc: Dmitry Rogozhkin <[e-mail]>
Reviewed-by: Tvrtko Ursulin <[e-mail]>
Link: https://patchwork.freedesktop.org/patch/msgid/[e-mail]

drivers/gpu/drm/i915/i915_gem.c

···
 	 */
 	err = i915_gem_object_wait(obj,
 				   I915_WAIT_INTERRUPTIBLE |
+				   I915_WAIT_PRIORITY |
 				   (write_domain ? I915_WAIT_ALL : 0),
 				   MAX_SCHEDULE_TIMEOUT,
 				   to_rps_client(file));
···
 	start = ktime_get();
 
 	ret = i915_gem_object_wait(obj,
-				   I915_WAIT_INTERRUPTIBLE | I915_WAIT_ALL,
+				   I915_WAIT_INTERRUPTIBLE |
+				   I915_WAIT_PRIORITY |
+				   I915_WAIT_ALL,
 				   to_wait_timeout(args->timeout_ns),
 				   to_rps_client(file));
 

drivers/gpu/drm/i915/i915_request.c

···
 		add_wait_queue(errq, &reset);
 
 	intel_wait_init(&wait);
+	if (flags & I915_WAIT_PRIORITY)
+		i915_schedule_bump_priority(rq, I915_PRIORITY_WAIT);
 
 restart:
 	do {

drivers/gpu/drm/i915/i915_request.h

···
 	__attribute__((nonnull(1)));
 #define I915_WAIT_INTERRUPTIBLE	BIT(0)
 #define I915_WAIT_LOCKED	BIT(1) /* struct_mutex held, handle GPU reset */
-#define I915_WAIT_ALL		BIT(2) /* used by i915_gem_object_wait() */
-#define I915_WAIT_FOR_IDLE_BOOST BIT(3)
+#define I915_WAIT_PRIORITY	BIT(2) /* small priority bump for the request */
+#define I915_WAIT_ALL		BIT(3) /* used by i915_gem_object_wait() */
+#define I915_WAIT_FOR_IDLE_BOOST BIT(4)
 
 static inline bool intel_engine_has_started(struct intel_engine_cs *engine,
 					    u32 seqno);

drivers/gpu/drm/i915/i915_scheduler.c

···
 	return engine;
 }
 
-void i915_schedule(struct i915_request *rq, const struct i915_sched_attr *attr)
+static void __i915_schedule(struct i915_request *rq,
+			    const struct i915_sched_attr *attr)
 {
 	struct list_head *uninitialized_var(pl);
 	struct intel_engine_cs *engine, *last;
···
 	const int prio = attr->priority;
 	LIST_HEAD(dfs);
 
+	/* Needed in order to use the temporary link inside i915_dependency */
+	lockdep_assert_held(&schedule_lock);
 	GEM_BUG_ON(prio == I915_PRIORITY_INVALID);
 
 	if (i915_request_completed(rq))
···
 
 	if (prio <= READ_ONCE(rq->sched.attr.priority))
 		return;
-
-	/* Needed in order to use the temporary link inside i915_dependency */
-	spin_lock(&schedule_lock);
 
 	stack.signaler = &rq->sched;
 	list_add(&stack.dfs_link, &dfs);
···
 		rq->sched.attr = *attr;
 
 		if (stack.dfs_link.next == stack.dfs_link.prev)
-			goto out_unlock;
+			return;
 
 		__list_del_entry(&stack.dfs_link);
 	}
···
 	}
 
 	spin_unlock_irq(&engine->timeline.lock);
+}
 
-out_unlock:
+void i915_schedule(struct i915_request *rq, const struct i915_sched_attr *attr)
+{
+	spin_lock(&schedule_lock);
+	__i915_schedule(rq, attr);
 	spin_unlock(&schedule_lock);
+}
+
+void i915_schedule_bump_priority(struct i915_request *rq, unsigned int bump)
+{
+	struct i915_sched_attr attr;
+
+	GEM_BUG_ON(bump & ~I915_PRIORITY_MASK);
+
+	if (READ_ONCE(rq->sched.attr.priority) == I915_PRIORITY_INVALID)
+		return;
+
+	spin_lock_bh(&schedule_lock);
+
+	attr = rq->sched.attr;
+	attr.priority |= bump;
+	__i915_schedule(rq, &attr);
+
+	spin_unlock_bh(&schedule_lock);
 }

drivers/gpu/drm/i915/i915_scheduler.h

···
 	I915_PRIORITY_INVALID = INT_MIN
 };
 
-#define I915_USER_PRIORITY_SHIFT 1
+#define I915_USER_PRIORITY_SHIFT 2
 #define I915_USER_PRIORITY(x) ((x) << I915_USER_PRIORITY_SHIFT)
 
 #define I915_PRIORITY_COUNT BIT(I915_USER_PRIORITY_SHIFT)
 #define I915_PRIORITY_MASK (I915_PRIORITY_COUNT - 1)
 
-#define I915_PRIORITY_NEWCLIENT	((u8)BIT(0))
+#define I915_PRIORITY_WAIT	((u8)BIT(0))
+#define I915_PRIORITY_NEWCLIENT	((u8)BIT(1))
 
 struct i915_sched_attr {
 	/**
···
 
 void i915_schedule(struct i915_request *request,
 		   const struct i915_sched_attr *attr);
+
+void i915_schedule_bump_priority(struct i915_request *rq, unsigned int bump);
 
 struct list_head *
 i915_sched_lookup_priolist(struct intel_engine_cs *engine, int prio);