[PATCH v2 0/3] io_uring: avoid ring quiesce in io_uring_register for eventfd opcodes

[PATCH v2 0/3] io_uring: avoid ring quiesce in io_uring_register for eventfd opcodes

[Usama Arif]

This is done by creating a new RCU data structure (io_ev_fd) as part of
io_ring_ctx that holds the eventfd_ctx, with reads to the structure protected
by rcu_read_lock and writes (register/unregister calls) protected by a mutex.

With the above approach ring quiesce can be avoided which is much more
expensive then using RCU lock. On the system tested, io_uring_reigster with
IORING_REGISTER_EVENTFD takes less than 1ms with RCU lock, compared to 15ms
before with ring quiesce.

The first patch creates the RCU protected data structure and removes ring
quiesce for IORING_REGISTER_EVENTFD and IORING_UNREGISTER_EVENTFD.

The second patch builds on top of the first patch and removes ring quiesce
for IORING_REGISTER_EVENTFD_ASYNC.

---
v1->v2:
- Added patch to remove eventfd from tracepoint (Patch 1) (Jens Axboe)
- Made the code of io_should_trigger_evfd as part of io_eventfd_signal (Jens Axboe)

Usama Arif (3):
  io_uring: remove trace for eventfd
  io_uring: avoid ring quiesce while registering/unregistering eventfd
  io_uring: avoid ring quiesce for IORING_REGISTER_EVENTFD_ASYNC

 fs/io_uring.c                   | 124 ++++++++++++++++++++++----------
 include/trace/events/io_uring.h |  13 ++--
 2 files changed, 93 insertions(+), 44 deletions(-)

-- 
2.25.1

[PATCH v2 1/3] io_uring: remove trace for eventfd

[Usama Arif]

The information on whether eventfd is registered is not very useful
and would result in the tracepoint being enclosed in an rcu_readlock
in a later patch that tries to avoid ring quiesce for registering
eventfd.

Suggested-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Usama Arif <usama.arif@bytedance.com>
---
 fs/io_uring.c                   |  3 +--
 include/trace/events/io_uring.h | 13 +++++--------
 2 files changed, 6 insertions(+), 10 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 2e04f718319d..21531609a9c6 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -11171,8 +11171,7 @@ SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
 	mutex_lock(&ctx->uring_lock);
 	ret = __io_uring_register(ctx, opcode, arg, nr_args);
 	mutex_unlock(&ctx->uring_lock);
-	trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
-							ctx->cq_ev_fd != NULL, ret);
+	trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, ret);
 out_fput:
 	fdput(f);
 	return ret;
diff --git a/include/trace/events/io_uring.h b/include/trace/events/io_uring.h
index 7346f0164cf4..098beda7601a 100644
--- a/include/trace/events/io_uring.h
+++ b/include/trace/events/io_uring.h
@@ -57,10 +57,9 @@ TRACE_EVENT(io_uring_create,
  * @opcode:		describes which operation to perform
  * @nr_user_files:	number of registered files
  * @nr_user_bufs:	number of registered buffers
- * @cq_ev_fd:		whether eventfs registered or not
  * @ret:		return code
  *
- * Allows to trace fixed files/buffers/eventfds, that could be registered to
+ * Allows to trace fixed files/buffers, that could be registered to
  * avoid an overhead of getting references to them for every operation. This
  * event, together with io_uring_file_get, can provide a full picture of how
  * much overhead one can reduce via fixing.
@@ -68,16 +67,15 @@ TRACE_EVENT(io_uring_create,
 TRACE_EVENT(io_uring_register,
 
 	TP_PROTO(void *ctx, unsigned opcode, unsigned nr_files,
-			 unsigned nr_bufs, bool eventfd, long ret),
+			 unsigned nr_bufs, long ret),
 
-	TP_ARGS(ctx, opcode, nr_files, nr_bufs, eventfd, ret),
+	TP_ARGS(ctx, opcode, nr_files, nr_bufs, ret),
 
 	TP_STRUCT__entry (
 		__field(  void *,	ctx			)
 		__field(  unsigned,	opcode		)
 		__field(  unsigned,	nr_files	)
 		__field(  unsigned,	nr_bufs		)
-		__field(  bool,		eventfd		)
 		__field(  long,		ret			)
 	),
 
@@ -86,14 +84,13 @@ TRACE_EVENT(io_uring_register,
 		__entry->opcode		= opcode;
 		__entry->nr_files	= nr_files;
 		__entry->nr_bufs	= nr_bufs;
-		__entry->eventfd	= eventfd;
 		__entry->ret		= ret;
 	),
 
 	TP_printk("ring %p, opcode %d, nr_user_files %d, nr_user_bufs %d, "
-			  "eventfd %d, ret %ld",
+			  "ret %ld",
 			  __entry->ctx, __entry->opcode, __entry->nr_files,
-			  __entry->nr_bufs, __entry->eventfd, __entry->ret)
+			  __entry->nr_bufs, __entry->ret)
 );
 
 /**
-- 
2.25.1

[PATCH v2 2/3] io_uring: avoid ring quiesce while registering/unregistering eventfd

[Usama Arif]

This is done by creating a new RCU data structure (io_ev_fd) as part of
io_ring_ctx that holds the eventfd_ctx.

The function io_eventfd_signal is executed under rcu_read_lock with a
single rcu_dereference to io_ev_fd so that if another thread unregisters
the eventfd while io_eventfd_signal is still being executed, the
eventfd_signal for which io_eventfd_signal was called completes
successfully.

The process of registering/unregistering eventfd is done under a lock
so multiple threads don't enter a race condition while
registering/unregistering eventfd.

With the above approach ring quiesce can be avoided which is much more
expensive then using RCU lock. On the system tested, io_uring_reigster with
IORING_REGISTER_EVENTFD takes less than 1ms with RCU lock, compared to 15ms
before with ring quiesce.

Signed-off-by: Usama Arif <usama.arif@bytedance.com>
---
 fs/io_uring.c | 101 +++++++++++++++++++++++++++++++++++++-------------
 1 file changed, 76 insertions(+), 25 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 21531609a9c6..9b6ee3b8b9f2 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -326,6 +326,12 @@ struct io_submit_state {
 	struct blk_plug		plug;
 };
 
+struct io_ev_fd {
+	struct eventfd_ctx	*cq_ev_fd;
+	struct io_ring_ctx	*ctx;
+	struct rcu_head		rcu;
+};
+
 struct io_ring_ctx {
 	/* const or read-mostly hot data */
 	struct {
@@ -399,7 +405,8 @@ struct io_ring_ctx {
 	struct {
 		unsigned		cached_cq_tail;
 		unsigned		cq_entries;
-		struct eventfd_ctx	*cq_ev_fd;
+		struct io_ev_fd	__rcu	*io_ev_fd;
+		struct mutex		ev_fd_lock;
 		struct wait_queue_head	cq_wait;
 		unsigned		cq_extra;
 		atomic_t		cq_timeouts;
@@ -1448,6 +1455,7 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 	xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1);
 	xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1);
 	mutex_init(&ctx->uring_lock);
+	mutex_init(&ctx->ev_fd_lock);
 	init_waitqueue_head(&ctx->cq_wait);
 	spin_lock_init(&ctx->completion_lock);
 	spin_lock_init(&ctx->timeout_lock);
@@ -1726,13 +1734,24 @@ static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx)
 	return &rings->cqes[tail & mask];
 }
 
-static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
+static void io_eventfd_signal(struct io_ring_ctx *ctx)
 {
-	if (likely(!ctx->cq_ev_fd))
-		return false;
+	struct io_ev_fd *ev_fd;
+
+	rcu_read_lock();
+	/* rcu_dereference ctx->io_ev_fd once and use it for both for checking and eventfd_signal */
+	ev_fd = rcu_dereference(ctx->io_ev_fd);
+
+	if (likely(!ev_fd))
+		goto out;
 	if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
-		return false;
-	return !ctx->eventfd_async || io_wq_current_is_worker();
+		goto out;
+
+	if (!ctx->eventfd_async || io_wq_current_is_worker())
+		eventfd_signal(ev_fd->cq_ev_fd, 1);
+
+out:
+	rcu_read_unlock();
 }
 
 /*
@@ -1751,8 +1770,7 @@ static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
 	 */
 	if (wq_has_sleeper(&ctx->cq_wait))
 		wake_up_all(&ctx->cq_wait);
-	if (io_should_trigger_evfd(ctx))
-		eventfd_signal(ctx->cq_ev_fd, 1);
+	io_eventfd_signal(ctx);
 }
 
 static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
@@ -1764,8 +1782,7 @@ static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
 		if (waitqueue_active(&ctx->cq_wait))
 			wake_up_all(&ctx->cq_wait);
 	}
-	if (io_should_trigger_evfd(ctx))
-		eventfd_signal(ctx->cq_ev_fd, 1);
+	io_eventfd_signal(ctx);
 }
 
 /* Returns true if there are no backlogged entries after the flush */
@@ -9353,35 +9370,67 @@ static int __io_sqe_buffers_update(struct io_ring_ctx *ctx,
 
 static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
 {
+	struct io_ev_fd *ev_fd;
 	__s32 __user *fds = arg;
-	int fd;
+	int fd, ret;
 
-	if (ctx->cq_ev_fd)
-		return -EBUSY;
+	mutex_lock(&ctx->ev_fd_lock);
+	ret = -EBUSY;
+	if (rcu_dereference_protected(ctx->io_ev_fd, lockdep_is_held(&ctx->ev_fd_lock)))
+		goto out;
 
+	ret = -EFAULT;
 	if (copy_from_user(&fd, fds, sizeof(*fds)))
-		return -EFAULT;
+		goto out;
 
-	ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
-	if (IS_ERR(ctx->cq_ev_fd)) {
-		int ret = PTR_ERR(ctx->cq_ev_fd);
+	ret = -ENOMEM;
+	ev_fd = kmalloc(sizeof(*ev_fd), GFP_KERNEL);
+	if (!ev_fd)
+		goto out;
 
-		ctx->cq_ev_fd = NULL;
-		return ret;
+	ev_fd->cq_ev_fd = eventfd_ctx_fdget(fd);
+	if (IS_ERR(ev_fd->cq_ev_fd)) {
+		ret = PTR_ERR(ev_fd->cq_ev_fd);
+		kfree(ev_fd);
+		goto out;
 	}
+	ev_fd->ctx = ctx;
 
-	return 0;
+	rcu_assign_pointer(ctx->io_ev_fd, ev_fd);
+	ret = 0;
+
+out:
+	mutex_unlock(&ctx->ev_fd_lock);
+	return ret;
+}
+
+static void io_eventfd_put(struct rcu_head *rcu)
+{
+	struct io_ev_fd *ev_fd = container_of(rcu, struct io_ev_fd, rcu);
+	struct io_ring_ctx *ctx = ev_fd->ctx;
+
+	eventfd_ctx_put(ev_fd->cq_ev_fd);
+	kfree(ev_fd);
+	rcu_assign_pointer(ctx->io_ev_fd, NULL);
 }
 
 static int io_eventfd_unregister(struct io_ring_ctx *ctx)
 {
-	if (ctx->cq_ev_fd) {
-		eventfd_ctx_put(ctx->cq_ev_fd);
-		ctx->cq_ev_fd = NULL;
-		return 0;
+	struct io_ev_fd *ev_fd;
+	int ret;
+
+	mutex_lock(&ctx->ev_fd_lock);
+	ev_fd = rcu_dereference_protected(ctx->io_ev_fd, lockdep_is_held(&ctx->ev_fd_lock));
+	if (ev_fd) {
+		call_rcu(&ev_fd->rcu, io_eventfd_put);
+		ret = 0;
+		goto out;
 	}
+	ret = -ENXIO;
 
-	return -ENXIO;
+out:
+	mutex_unlock(&ctx->ev_fd_lock);
+	return ret;
 }
 
 static void io_destroy_buffers(struct io_ring_ctx *ctx)
@@ -10960,6 +11009,8 @@ static bool io_register_op_must_quiesce(int op)
 	case IORING_REGISTER_FILES:
 	case IORING_UNREGISTER_FILES:
 	case IORING_REGISTER_FILES_UPDATE:
+	case IORING_REGISTER_EVENTFD:
+	case IORING_UNREGISTER_EVENTFD:
 	case IORING_REGISTER_PROBE:
 	case IORING_REGISTER_PERSONALITY:
 	case IORING_UNREGISTER_PERSONALITY:
-- 
2.25.1

[PATCH v2 3/3] io_uring: avoid ring quiesce for IORING_REGISTER_EVENTFD_ASYNC

[Usama Arif]

This is done using the RCU data structure (io_ev_fd). eventfd_async
is moved from io_ring_ctx to io_ev_fd which is RCU protected hence
avoiding ring quiesce which is much more expensive than an RCU lock.

Signed-off-by: Usama Arif <usama.arif@bytedance.com>
---
 fs/io_uring.c | 22 ++++++++++++----------
 1 file changed, 12 insertions(+), 10 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 9b6ee3b8b9f2..05fd059b3f1e 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -329,6 +329,7 @@ struct io_submit_state {
 struct io_ev_fd {
 	struct eventfd_ctx	*cq_ev_fd;
 	struct io_ring_ctx	*ctx;
+	unsigned int		eventfd_async: 1;
 	struct rcu_head		rcu;
 };
 
@@ -341,7 +342,6 @@ struct io_ring_ctx {
 		unsigned int		flags;
 		unsigned int		compat: 1;
 		unsigned int		drain_next: 1;
-		unsigned int		eventfd_async: 1;
 		unsigned int		restricted: 1;
 		unsigned int		off_timeout_used: 1;
 		unsigned int		drain_active: 1;
@@ -1747,7 +1747,7 @@ static void io_eventfd_signal(struct io_ring_ctx *ctx)
 	if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
 		goto out;
 
-	if (!ctx->eventfd_async || io_wq_current_is_worker())
+	if (!ev_fd->eventfd_async || io_wq_current_is_worker())
 		eventfd_signal(ev_fd->cq_ev_fd, 1);
 
 out:
@@ -9368,7 +9368,8 @@ static int __io_sqe_buffers_update(struct io_ring_ctx *ctx,
 	return done ? done : err;
 }
 
-static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
+static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg,
+			       unsigned int eventfd_async)
 {
 	struct io_ev_fd *ev_fd;
 	__s32 __user *fds = arg;
@@ -9395,6 +9396,7 @@ static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
 		goto out;
 	}
 	ev_fd->ctx = ctx;
+	ev_fd->eventfd_async = eventfd_async;
 
 	rcu_assign_pointer(ctx->io_ev_fd, ev_fd);
 	ret = 0;
@@ -11010,6 +11012,7 @@ static bool io_register_op_must_quiesce(int op)
 	case IORING_UNREGISTER_FILES:
 	case IORING_REGISTER_FILES_UPDATE:
 	case IORING_REGISTER_EVENTFD:
+	case IORING_REGISTER_EVENTFD_ASYNC:
 	case IORING_UNREGISTER_EVENTFD:
 	case IORING_REGISTER_PROBE:
 	case IORING_REGISTER_PERSONALITY:
@@ -11110,17 +11113,16 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
 		ret = io_register_files_update(ctx, arg, nr_args);
 		break;
 	case IORING_REGISTER_EVENTFD:
-	case IORING_REGISTER_EVENTFD_ASYNC:
 		ret = -EINVAL;
 		if (nr_args != 1)
 			break;
-		ret = io_eventfd_register(ctx, arg);
-		if (ret)
+		ret = io_eventfd_register(ctx, arg, 0);
+		break;
+	case IORING_REGISTER_EVENTFD_ASYNC:
+		ret = -EINVAL;
+		if (nr_args != 1)
 			break;
-		if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
-			ctx->eventfd_async = 1;
-		else
-			ctx->eventfd_async = 0;
+		ret = io_eventfd_register(ctx, arg, 1);
 		break;
 	case IORING_UNREGISTER_EVENTFD:
 		ret = -EINVAL;
-- 
2.25.1