io_uring: add lazy poll_wq activation

Even though io_poll_wq_wake()'s waitqueue_active reuses a barrier we do
for another waitqueue, it's not going to be the case in the future and
so we want to have a fast path for it when the ring has never been
polled.

Move poll_wq wake ups into __io_commit_cqring_flush() using a new flag
called ->poll_activated. The idea behind the flag is to set it when the
ring was polled for the first time. This requires additional sync to not
miss events, which is done here by using task_work for ->task_complete
rings, and by default enabling the flag for all other types of rings.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/060785e8e9137a920b232c0c7f575b131af19cac.1673274244.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

include/linux/io_uring_types.h

@@ -206,6 +206,7 @@ struct io_ring_ctx {
 		unsigned int		syscall_iopoll: 1;
 		/* all CQEs should be posted only by the submitter task */
 		unsigned int		task_complete: 1;
+		unsigned int		poll_activated: 1;
 
 		enum task_work_notify_mode	notify_method;
 		struct io_rings			*rings;
@@ -357,6 +358,7 @@ struct io_ring_ctx {
 	u32				iowq_limits[2];
 	bool				iowq_limits_set;
 
+	struct callback_head		poll_wq_task_work;
 	struct list_head		defer_list;
 	unsigned			sq_thread_idle;
 	/* protected by ->completion_lock */

io_uring/io_uring.c

@@ -573,6 +573,8 @@ static void io_eventfd_flush_signal(struct io_ring_ctx *ctx)
 
 void __io_commit_cqring_flush(struct io_ring_ctx *ctx)
 {
+	if (ctx->poll_activated)
+		io_poll_wq_wake(ctx);
 	if (ctx->off_timeout_used)
 		io_flush_timeouts(ctx);
 	if (ctx->drain_active) {
@@ -2782,11 +2784,53 @@ static __cold void io_ring_ctx_free(struct io_ring_ctx *ctx)
 	kfree(ctx);
 }
 
+static __cold void io_activate_pollwq_cb(struct callback_head *cb)
+{
+	struct io_ring_ctx *ctx = container_of(cb, struct io_ring_ctx,
+					       poll_wq_task_work);
+
+	mutex_lock(&ctx->uring_lock);
+	ctx->poll_activated = true;
+	mutex_unlock(&ctx->uring_lock);
+
+	/*
+	 * Wake ups for some events between start of polling and activation
+	 * might've been lost due to loose synchronisation.
+	 */
+	wake_up_all(&ctx->poll_wq);
+	percpu_ref_put(&ctx->refs);
+}
+
+static __cold void io_activate_pollwq(struct io_ring_ctx *ctx)
+{
+	spin_lock(&ctx->completion_lock);
+	/* already activated or in progress */
+	if (ctx->poll_activated || ctx->poll_wq_task_work.func)
+		goto out;
+	if (WARN_ON_ONCE(!ctx->task_complete))
+		goto out;
+	if (!ctx->submitter_task)
+		goto out;
+	/*
+	 * with ->submitter_task only the submitter task completes requests, we
+	 * only need to sync with it, which is done by injecting a tw
+	 */
+	init_task_work(&ctx->poll_wq_task_work, io_activate_pollwq_cb);
+	percpu_ref_get(&ctx->refs);
+	if (task_work_add(ctx->submitter_task, &ctx->poll_wq_task_work, TWA_SIGNAL))
+		percpu_ref_put(&ctx->refs);
+out:
+	spin_unlock(&ctx->completion_lock);
+}
+
 static __poll_t io_uring_poll(struct file *file, poll_table *wait)
 {
 	struct io_ring_ctx *ctx = file->private_data;
 	__poll_t mask = 0;
 
+	if (unlikely(!ctx->poll_activated))
+		io_activate_pollwq(ctx);
+
 	poll_wait(file, &ctx->poll_wq, wait);
 	/*
 	 * synchronizes with barrier from wq_has_sleeper call in
@@ -3593,6 +3637,13 @@ static __cold int io_uring_create(unsigned entries, struct io_uring_params *p,
 	    !(ctx->flags & IORING_SETUP_SQPOLL))
 		ctx->task_complete = true;
 
+	/*
+	 * lazy poll_wq activation relies on ->task_complete for synchronisation
+	 * purposes, see io_activate_pollwq()
+	 */
+	if (!ctx->task_complete)
+		ctx->poll_activated = true;
+
 	/*
 	 * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
 	 * space applications don't need to do io completion events
@@ -3886,8 +3937,15 @@ static int io_register_enable_rings(struct io_ring_ctx *ctx)
 	if (!(ctx->flags & IORING_SETUP_R_DISABLED))
 		return -EBADFD;
 
-	if (ctx->flags & IORING_SETUP_SINGLE_ISSUER && !ctx->submitter_task)
+	if (ctx->flags & IORING_SETUP_SINGLE_ISSUER && !ctx->submitter_task) {
 		WRITE_ONCE(ctx->submitter_task, get_task_struct(current));
+		/*
+		 * Lazy activation attempts would fail if it was polled before
+		 * submitter_task is set.
+		 */
+		if (wq_has_sleeper(&ctx->poll_wq))
+			io_activate_pollwq(ctx);
+	}
 
 	if (ctx->restrictions.registered)
 		ctx->restricted = 1;

io_uring/io_uring.h

@@ -222,16 +222,14 @@ static inline void io_commit_cqring(struct io_ring_ctx *ctx)
 
 static inline void io_poll_wq_wake(struct io_ring_ctx *ctx)
 {
-	if (waitqueue_active(&ctx->poll_wq))
+	if (wq_has_sleeper(&ctx->poll_wq))
 		__wake_up(&ctx->poll_wq, TASK_NORMAL, 0,
 				poll_to_key(EPOLL_URING_WAKE | EPOLLIN));
 }
 
 /* requires smb_mb() prior, see wq_has_sleeper() */
 static inline void __io_cqring_wake(struct io_ring_ctx *ctx)
 {
-	io_poll_wq_wake(ctx);
-
 	/*
 	 * Trigger waitqueue handler on all waiters on our waitqueue. This
 	 * won't necessarily wake up all the tasks, io_should_wake() will make
@@ -316,7 +314,8 @@ static inline void io_req_complete_defer(struct io_kiocb *req)
 
 static inline void io_commit_cqring_flush(struct io_ring_ctx *ctx)
 {
-	if (unlikely(ctx->off_timeout_used || ctx->drain_active || ctx->has_evfd))
+	if (unlikely(ctx->off_timeout_used || ctx->drain_active ||
+		     ctx->has_evfd || ctx->poll_activated))
 		__io_commit_cqring_flush(ctx);
 }