Skip to contentSkip to conversation

io_uring: Add support for napi_busy_poll

PATCH Superseded Olivier Langlois 2 participants 12 messages Mar 1, 2022 – Mar 3, 2022

Subsystems (2)

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 cover io_uring: Add support for napi_busy_poll

The sqpoll thread can be used for performing the napi busy poll in a similar way that it does io polling for file systems supporting direct access bypassing the page cache.

The other way that io_uring can be used for napi busy poll is by calling io_uring_enter() to get events.

If the user specify a timeout value, it is distributed between polling and sleeping by using the systemwide setting /proc/sys/net/core/busy_poll.

The changes have been tested with this program:
https://github.com/lano1106/io_uring_udp_ping

and the result is:
Without sqpoll:
NAPI busy loop disabled:
rtt min/avg/max/mdev = 40.631/42.050/58.667/1.547 us NAPI busy loop enabled:
rtt min/avg/max/mdev = 30.619/31.753/61.433/1.456 us

With sqpoll:
NAPI busy loop disabled:
rtt min/avg/max/mdev = 42.087/44.438/59.508/1.533 us NAPI busy loop enabled:
rtt min/avg/max/mdev = 35.779/37.347/52.201/0.924 us

v2:
* Evaluate list_empty(&ctx->napi_list) outside io_napi_busy_loop() to keep __io_sq_thread() execution as fast as possible
* In io_cqring_wait(), move up the sig block to avoid needless computation if the block exits the function
* In io_cqring_wait(), protect ctx->napi_list from race condition by splicing it into a local list
* In io_cqring_wait(), allow busy polling when uts is missing
* Fix kernel test robot issues
v3:
* Fix do_div() type mismatch warning
* Reduce uring_lock contention by creating a spinlock for protecting napi_list
* Support correctly MULTISHOT poll requests
v4:
* Put back benchmark result in commit text

Olivier Langlois (2):
io_uring: minor io_cqring_wait() optimization
io_uring: Add support for napi_busy_poll

 fs/io_uring.c | 246 ++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 237 insertions(+), 9 deletions(-)

base-commit: 719fce7539cd3e186598e2aed36325fe892150cf

Signature · 1 line
2.35.1
#1 lore raw

PATCH v4 1/2 io_uring: minor io_cqring_wait() optimization

Move up the block manipulating the sig variable to execute code that may encounter an error and exit first before continuing exectuing the rest of the function and avoid useless computations

---
 fs/io_uring.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)
fs/io_uring.c +8 -8
index 4715980e9015..f7b8df79a02b 100644
@@ -7732,14 +7732,6 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 			break;
 	} while (1);
 
-	if (uts) {
-		struct timespec64 ts;
-
-		if (get_timespec64(&ts, uts))
-			return -EFAULT;
-		timeout = ktime_add_ns(timespec64_to_ktime(ts), ktime_get_ns());
-	}
-
 	if (sig) {
 #ifdef CONFIG_COMPAT
 		if (in_compat_syscall())
@@ -7753,6 +7745,14 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 			return ret;
 	}
 
+	if (uts) {
+		struct timespec64 ts;
+
+		if (get_timespec64(&ts, uts))
+			return -EFAULT;
+		timeout = ktime_add_ns(timespec64_to_ktime(ts), ktime_get_ns());
+	}
+
 	init_waitqueue_func_entry(&iowq.wq, io_wake_function);
 	iowq.wq.private = current;
 	INIT_LIST_HEAD(&iowq.wq.entry);
Signature · 1 line
2.35.1
#2 lore raw

PATCH v4 2/2 io_uring: Add support for napi_busy_poll

In reply to Olivier Langlois

The sqpoll thread can be used for performing the napi busy poll in a similar way that it does io polling for file systems supporting direct access bypassing the page cache.

The other way that io_uring can be used for napi busy poll is by calling io_uring_enter() to get events.

If the user specify a timeout value, it is distributed between polling and sleeping by using the systemwide setting /proc/sys/net/core/busy_poll.

The changes have been tested with this program:
https://github.com/lano1106/io_uring_udp_ping

and the result is:
Without sqpoll:
NAPI busy loop disabled:
rtt min/avg/max/mdev = 40.631/42.050/58.667/1.547 us NAPI busy loop enabled:
rtt min/avg/max/mdev = 30.619/31.753/61.433/1.456 us

With sqpoll:
NAPI busy loop disabled:
rtt min/avg/max/mdev = 42.087/44.438/59.508/1.533 us NAPI busy loop enabled:
rtt min/avg/max/mdev = 35.779/37.347/52.201/0.924 us

---
 fs/io_uring.c | 230 +++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 229 insertions(+), 1 deletion(-)
fs/io_uring.c +229 -1
index f7b8df79a02b..37c065786e4b 100644
@@ -63,6 +63,7 @@
 #include <net/sock.h>
 #include <net/af_unix.h>
 #include <net/scm.h>
+#include <net/busy_poll.h>
 #include <linux/anon_inodes.h>
 #include <linux/sched/mm.h>
 #include <linux/uaccess.h>
@@ -395,6 +396,11 @@ struct io_ring_ctx {
 	struct list_head	sqd_list;
 
 	unsigned long		check_cq_overflow;
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	/* used to track busy poll napi_id */
+	struct list_head	napi_list;
+	spinlock_t		napi_lock;	/* napi_list lock */
+#endif
 
 	struct {
 		unsigned		cached_cq_tail;
@@ -1464,6 +1470,10 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 	INIT_WQ_LIST(&ctx->locked_free_list);
 	INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func);
 	INIT_WQ_LIST(&ctx->submit_state.compl_reqs);
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	INIT_LIST_HEAD(&ctx->napi_list);
+	spin_lock_init(&ctx->napi_lock);
+#endif
 	return ctx;
 err:
 	kfree(ctx->dummy_ubuf);
@@ -5399,6 +5409,108 @@ IO_NETOP_FN(send);
 IO_NETOP_FN(recv);
 #endif /* CONFIG_NET */
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+
+#define NAPI_TIMEOUT			(60 * SEC_CONVERSION)
+
+struct napi_entry {
+	struct list_head	list;
+	unsigned int		napi_id;
+	unsigned long		timeout;
+};
+
+/*
+ * Add busy poll NAPI ID from sk.
+ */
+static void io_add_napi(struct file *file, struct io_ring_ctx *ctx)
+{
+	unsigned int napi_id;
+	struct socket *sock;
+	struct sock *sk;
+	struct napi_entry *ne;
+
+	if (!net_busy_loop_on())
+		return;
+
+	sock = sock_from_file(file);
+	if (!sock)
+		return;
+
+	sk = sock->sk;
+	if (!sk)
+		return;
+
+	napi_id = READ_ONCE(sk->sk_napi_id);
+
+	/* Non-NAPI IDs can be rejected */
+	if (napi_id < MIN_NAPI_ID)
+		return;
+
+	spin_lock(&ctx->napi_lock);
+	list_for_each_entry(ne, &ctx->napi_list, list) {
+		if (ne->napi_id == napi_id) {
+			ne->timeout = jiffies + NAPI_TIMEOUT;
+			goto out;
+		}
+	}
+
+	ne = kmalloc(sizeof(*ne), GFP_NOWAIT);
+	if (!ne)
+		goto out;
+
+	ne->napi_id = napi_id;
+	ne->timeout = jiffies + NAPI_TIMEOUT;
+	list_add_tail(&ne->list, &ctx->napi_list);
+out:
+	spin_unlock(&ctx->napi_lock);
+}
+
+static inline void io_check_napi_entry_timeout(struct napi_entry *ne)
+{
+	if (time_after(jiffies, ne->timeout)) {
+		list_del(&ne->list);
+		kfree(ne);
+	}
+}
+
+/*
+ * Busy poll if globally on and supporting sockets found
+ */
+static bool io_napi_busy_loop(struct list_head *napi_list)
+{
+	struct napi_entry *ne, *n;
+
+	list_for_each_entry_safe(ne, n, napi_list, list) {
+		napi_busy_loop(ne->napi_id, NULL, NULL, true,
+			       BUSY_POLL_BUDGET);
+		io_check_napi_entry_timeout(ne);
+	}
+	return !list_empty(napi_list);
+}
+
+static void io_free_napi_list(struct io_ring_ctx *ctx)
+{
+	spin_lock(&ctx->napi_lock);
+	while (!list_empty(&ctx->napi_list)) {
+		struct napi_entry *ne =
+			list_first_entry(&ctx->napi_list, struct napi_entry,
+					 list);
+
+		list_del(&ne->list);
+		kfree(ne);
+	}
+	spin_unlock(&ctx->napi_lock);
+}
+#else
+static inline void io_add_napi(struct file *file, struct io_ring_ctx *ctx)
+{
+}
+
+static inline void io_free_napi_list(struct io_ring_ctx *ctx)
+{
+}
+#endif /* CONFIG_NET_RX_BUSY_POLL */
+
 struct io_poll_table {
 	struct poll_table_struct pt;
 	struct io_kiocb *req;
@@ -5545,6 +5657,7 @@ static int io_poll_check_events(struct io_kiocb *req)
 			if (unlikely(!filled))
 				return -ECANCELED;
 			io_cqring_ev_posted(ctx);
+			io_add_napi(req->file, ctx);
 		} else if (req->result) {
 			return 0;
 		}
@@ -5777,6 +5890,7 @@ static int __io_arm_poll_handler(struct io_kiocb *req,
 		__io_poll_execute(req, mask);
 		return 0;
 	}
+	io_add_napi(req->file, req->ctx);
 
 	/*
 	 * Release ownership. If someone tried to queue a tw while it was
@@ -7519,7 +7633,11 @@ static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
 		    !(ctx->flags & IORING_SETUP_R_DISABLED))
 			ret = io_submit_sqes(ctx, to_submit);
 		mutex_unlock(&ctx->uring_lock);
-
+#ifdef CONFIG_NET_RX_BUSY_POLL
+		if (!list_empty(&ctx->napi_list) &&
+		    io_napi_busy_loop(&ctx->napi_list))
+			++ret;
+#endif
 		if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait))
 			wake_up(&ctx->sqo_sq_wait);
 		if (creds)
@@ -7650,6 +7768,9 @@ struct io_wait_queue {
 	struct io_ring_ctx *ctx;
 	unsigned cq_tail;
 	unsigned nr_timeouts;
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	unsigned busy_poll_to;
+#endif
 };
 
 static inline bool io_should_wake(struct io_wait_queue *iowq)
@@ -7711,6 +7832,87 @@ static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	return 1;
 }
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+static void io_adjust_busy_loop_timeout(struct timespec64 *ts,
+					struct io_wait_queue *iowq)
+{
+	unsigned busy_poll_to = READ_ONCE(sysctl_net_busy_poll);
+	struct timespec64 pollto = ns_to_timespec64(1000 * (s64)busy_poll_to);
+
+	if (timespec64_compare(ts, &pollto) > 0) {
+		*ts = timespec64_sub(*ts, pollto);
+		iowq->busy_poll_to = busy_poll_to;
+	} else {
+		u64 to = timespec64_to_ns(ts);
+
+		do_div(to, 1000);
+		iowq->busy_poll_to = to;
+		ts->tv_sec = 0;
+		ts->tv_nsec = 0;
+	}
+}
+
+static inline bool io_busy_loop_timeout(unsigned long start_time,
+					unsigned long bp_usec)
+{
+	if (bp_usec) {
+		unsigned long end_time = start_time + bp_usec;
+		unsigned long now = busy_loop_current_time();
+
+		return time_after(now, end_time);
+	}
+	return true;
+}
+
+static bool io_busy_loop_end(void *p, unsigned long start_time)
+{
+	struct io_wait_queue *iowq = p;
+
+	return signal_pending(current) ||
+	       io_should_wake(iowq) ||
+	       io_busy_loop_timeout(start_time, iowq->busy_poll_to);
+}
+
+static void io_blocking_napi_busy_loop(struct list_head *napi_list,
+				       struct io_wait_queue *iowq)
+{
+	unsigned long start_time =
+		list_is_singular(napi_list) ? 0 :
+		busy_loop_current_time();
+
+	do {
+		if (list_is_singular(napi_list)) {
+			struct napi_entry *ne =
+				list_first_entry(napi_list,
+						 struct napi_entry, list);
+
+			napi_busy_loop(ne->napi_id, io_busy_loop_end, iowq,
+				       true, BUSY_POLL_BUDGET);
+			io_check_napi_entry_timeout(ne);
+			break;
+		}
+	} while (io_napi_busy_loop(napi_list) &&
+		 !io_busy_loop_end(iowq, start_time));
+}
+
+static void io_putback_napi_list(struct io_ring_ctx *ctx,
+				 struct list_head *napi_list)
+{
+	struct napi_entry *cne, *lne;
+
+	spin_lock(&ctx->napi_lock);
+	list_for_each_entry(cne, &ctx->napi_list, list)
+		list_for_each_entry(lne, napi_list, list)
+			if (cne->napi_id == lne->napi_id) {
+				list_del(&lne->list);
+				kfree(lne);
+				break;
+			}
+	list_splice(napi_list, &ctx->napi_list);
+	spin_unlock(&ctx->napi_lock);
+}
+#endif /* CONFIG_NET_RX_BUSY_POLL */
+
 /*
  * Wait until events become available, if we don't already have some. The
  * application must reap them itself, as they reside on the shared cq ring.
@@ -7723,6 +7925,9 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 	struct io_rings *rings = ctx->rings;
 	ktime_t timeout = KTIME_MAX;
 	int ret;
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	LIST_HEAD(local_napi_list);
+#endif
 
 	do {
 		io_cqring_overflow_flush(ctx);
@@ -7745,13 +7950,29 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 			return ret;
 	}
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	iowq.busy_poll_to = 0;
+	if (!(ctx->flags & IORING_SETUP_SQPOLL)) {
+		spin_lock(&ctx->napi_lock);
+		list_splice_init(&ctx->napi_list, &local_napi_list);
+		spin_unlock(&ctx->napi_lock);
+	}
+#endif
 	if (uts) {
 		struct timespec64 ts;
 
 		if (get_timespec64(&ts, uts))
 			return -EFAULT;
+#ifdef CONFIG_NET_RX_BUSY_POLL
+		if (!list_empty(&local_napi_list))
+			io_adjust_busy_loop_timeout(&ts, &iowq);
+#endif
 		timeout = ktime_add_ns(timespec64_to_ktime(ts), ktime_get_ns());
 	}
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	else if (!list_empty(&local_napi_list))
+		iowq.busy_poll_to = READ_ONCE(sysctl_net_busy_poll);
+#endif
 
 	init_waitqueue_func_entry(&iowq.wq, io_wake_function);
 	iowq.wq.private = current;
@@ -7761,6 +7982,12 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 	iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events;
 
 	trace_io_uring_cqring_wait(ctx, min_events);
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	if (iowq.busy_poll_to)
+		io_blocking_napi_busy_loop(&local_napi_list, &iowq);
+	if (!list_empty(&local_napi_list))
+		io_putback_napi_list(ctx, &local_napi_list);
+#endif
 	do {
 		/* if we can't even flush overflow, don't wait for more */
 		if (!io_cqring_overflow_flush(ctx)) {
@@ -9483,6 +9710,7 @@ static __cold void io_ring_ctx_free(struct io_ring_ctx *ctx)
 	io_req_caches_free(ctx);
 	if (ctx->hash_map)
 		io_wq_put_hash(ctx->hash_map);
+	io_free_napi_list(ctx);
 	kfree(ctx->cancel_hash);
 	kfree(ctx->dummy_ubuf);
 	kfree(ctx);
Signature · 1 line
2.35.1
#3 lore raw

Hi Olivier,

Quoted 111 lines · Olivier Langlois

The sqpoll thread can be used for performing the napi busy poll in a similar way that it does io polling for file systems supporting direct access bypassing the page cache.

The other way that io_uring can be used for napi busy poll is by calling io_uring_enter() to get events.

If the user specify a timeout value, it is distributed between polling and sleeping by using the systemwide setting /proc/sys/net/core/busy_poll.

The changes have been tested with this program:
https://github.com/lano1106/io_uring_udp_ping

and the result is:
Without sqpoll:
NAPI busy loop disabled:
rtt min/avg/max/mdev = 40.631/42.050/58.667/1.547 us NAPI busy loop enabled:
rtt min/avg/max/mdev = 30.619/31.753/61.433/1.456 us

With sqpoll:
NAPI busy loop disabled:
rtt min/avg/max/mdev = 42.087/44.438/59.508/1.533 us NAPI busy loop enabled:
rtt min/avg/max/mdev = 35.779/37.347/52.201/0.924 us

---
  fs/io_uring.c | 230 +++++++++++++++++++++++++++++++++++++++++++++++++-
  1 file changed, 229 insertions(+), 1 deletion(-)
fs/io_uring.c +48 -0
index f7b8df79a02b..37c065786e4b 100644
@@ -63,6 +63,7 @@
  #include <net/sock.h>
  #include <net/af_unix.h>
  #include <net/scm.h>
+#include <net/busy_poll.h>
  #include <linux/anon_inodes.h>
  #include <linux/sched/mm.h>
  #include <linux/uaccess.h>
@@ -395,6 +396,11 @@ struct io_ring_ctx {
  	struct list_head	sqd_list;
 
  	unsigned long		check_cq_overflow;
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	/* used to track busy poll napi_id */
+	struct list_head	napi_list;
+	spinlock_t		napi_lock;	/* napi_list lock */
+#endif
 
  	struct {
  		unsigned		cached_cq_tail;
@@ -1464,6 +1470,10 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
  	INIT_WQ_LIST(&ctx->locked_free_list);
  	INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func);
  	INIT_WQ_LIST(&ctx->submit_state.compl_reqs);
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	INIT_LIST_HEAD(&ctx->napi_list);
+	spin_lock_init(&ctx->napi_lock);
+#endif
  	return ctx;
  err:
  	kfree(ctx->dummy_ubuf);
@@ -5399,6 +5409,108 @@ IO_NETOP_FN(send);
  IO_NETOP_FN(recv);
  #endif /* CONFIG_NET */
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+
+#define NAPI_TIMEOUT			(60 * SEC_CONVERSION)
+
+struct napi_entry {
+	struct list_head	list;
+	unsigned int		napi_id;
+	unsigned long		timeout;
+};
+
+/*
+ * Add busy poll NAPI ID from sk.
+ */
+static void io_add_napi(struct file *file, struct io_ring_ctx *ctx)
+{
+	unsigned int napi_id;
+	struct socket *sock;
+	struct sock *sk;
+	struct napi_entry *ne;
+
+	if (!net_busy_loop_on())
+		return;
+
+	sock = sock_from_file(file);
+	if (!sock)
+		return;
+
+	sk = sock->sk;
+	if (!sk)
+		return;
+
+	napi_id = READ_ONCE(sk->sk_napi_id);
+
+	/* Non-NAPI IDs can be rejected */
+	if (napi_id < MIN_NAPI_ID)
+		return;
+
+	spin_lock(&ctx->napi_lock);

fs/io_uring.c IO_NETOP_FN(send); · line 5459

Quoted 10 lines
fs/io_uring.c +10 -0
+	list_for_each_entry(ne, &ctx->napi_list, list) {
+		if (ne->napi_id == napi_id) {
+			ne->timeout = jiffies + NAPI_TIMEOUT;
+			goto out;
+		}
+	}
+
+	ne = kmalloc(sizeof(*ne), GFP_NOWAIT);
+	if (!ne)
+		goto out;

IMHO, we need to handle -ENOMEM here, I cut off the error handling when

I did the quick coding. Sorry for misleading.

Quoted 74 lines
fs/io_uring.c +56 -0
+
+	ne->napi_id = napi_id;
+	ne->timeout = jiffies + NAPI_TIMEOUT;
+	list_add_tail(&ne->list, &ctx->napi_list);
+out:
+	spin_unlock(&ctx->napi_lock);
+}
+
+static inline void io_check_napi_entry_timeout(struct napi_entry *ne)
+{
+	if (time_after(jiffies, ne->timeout)) {
+		list_del(&ne->list);
+		kfree(ne);
+	}
+}
+
+/*
+ * Busy poll if globally on and supporting sockets found
+ */
+static bool io_napi_busy_loop(struct list_head *napi_list)
+{
+	struct napi_entry *ne, *n;
+
+	list_for_each_entry_safe(ne, n, napi_list, list) {
+		napi_busy_loop(ne->napi_id, NULL, NULL, true,
+			       BUSY_POLL_BUDGET);
+		io_check_napi_entry_timeout(ne);
+	}
+	return !list_empty(napi_list);
+}
+
+static void io_free_napi_list(struct io_ring_ctx *ctx)
+{
+	spin_lock(&ctx->napi_lock);
+	while (!list_empty(&ctx->napi_list)) {
+		struct napi_entry *ne =
+			list_first_entry(&ctx->napi_list, struct napi_entry,
+					 list);
+
+		list_del(&ne->list);
+		kfree(ne);
+	}
+	spin_unlock(&ctx->napi_lock);
+}
+#else
+static inline void io_add_napi(struct file *file, struct io_ring_ctx *ctx)
+{
+}
+
+static inline void io_free_napi_list(struct io_ring_ctx *ctx)
+{
+}
+#endif /* CONFIG_NET_RX_BUSY_POLL */
+
  struct io_poll_table {
  	struct poll_table_struct pt;
  	struct io_kiocb *req;
@@ -5545,6 +5657,7 @@ static int io_poll_check_events(struct io_kiocb *req)
  			if (unlikely(!filled))
  				return -ECANCELED;
  			io_cqring_ev_posted(ctx);
+			io_add_napi(req->file, ctx);
  		} else if (req->result) {
  			return 0;
  		}
@@ -5777,6 +5890,7 @@ static int __io_arm_poll_handler(struct io_kiocb *req,
  		__io_poll_execute(req, mask);
  		return 0;
  	}
+	io_add_napi(req->file, req->ctx);
 
  	/*
  	 * Release ownership. If someone tried to queue a tw while it was

fs/io_uring.c static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) · line 7638

Quoted 7 lines
fs/io_uring.c +3 -1
@@ -7519,7 +7633,11 @@ static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
  		    !(ctx->flags & IORING_SETUP_R_DISABLED))
  			ret = io_submit_sqes(ctx, to_submit);
  		mutex_unlock(&ctx->uring_lock);
-
+#ifdef CONFIG_NET_RX_BUSY_POLL
+		if (!list_empty(&ctx->napi_list) &&
+		    io_napi_busy_loop(&ctx->napi_list))

I'm afraid we may need lock for sqpoll too, since io_add_napi() could be in iowq context.

I'll take a look at the lock stuff of this patch tomorrow, too late now in my timezone.

Quoted 72 lines
fs/io_uring.c +58 -0
+			++ret;
+#endif
  		if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait))
  			wake_up(&ctx->sqo_sq_wait);
  		if (creds)
@@ -7650,6 +7768,9 @@ struct io_wait_queue {
  	struct io_ring_ctx *ctx;
  	unsigned cq_tail;
  	unsigned nr_timeouts;
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	unsigned busy_poll_to;
+#endif
  };
 
  static inline bool io_should_wake(struct io_wait_queue *iowq)
@@ -7711,6 +7832,87 @@ static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
  	return 1;
  }
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+static void io_adjust_busy_loop_timeout(struct timespec64 *ts,
+					struct io_wait_queue *iowq)
+{
+	unsigned busy_poll_to = READ_ONCE(sysctl_net_busy_poll);
+	struct timespec64 pollto = ns_to_timespec64(1000 * (s64)busy_poll_to);
+
+	if (timespec64_compare(ts, &pollto) > 0) {
+		*ts = timespec64_sub(*ts, pollto);
+		iowq->busy_poll_to = busy_poll_to;
+	} else {
+		u64 to = timespec64_to_ns(ts);
+
+		do_div(to, 1000);
+		iowq->busy_poll_to = to;
+		ts->tv_sec = 0;
+		ts->tv_nsec = 0;
+	}
+}
+
+static inline bool io_busy_loop_timeout(unsigned long start_time,
+					unsigned long bp_usec)
+{
+	if (bp_usec) {
+		unsigned long end_time = start_time + bp_usec;
+		unsigned long now = busy_loop_current_time();
+
+		return time_after(now, end_time);
+	}
+	return true;
+}
+
+static bool io_busy_loop_end(void *p, unsigned long start_time)
+{
+	struct io_wait_queue *iowq = p;
+
+	return signal_pending(current) ||
+	       io_should_wake(iowq) ||
+	       io_busy_loop_timeout(start_time, iowq->busy_poll_to);
+}
+
+static void io_blocking_napi_busy_loop(struct list_head *napi_list,
+				       struct io_wait_queue *iowq)
+{
+	unsigned long start_time =
+		list_is_singular(napi_list) ? 0 :
+		busy_loop_current_time();
+
+	do {
+		if (list_is_singular(napi_list)) {
+			struct napi_entry *ne =
+				list_first_entry(napi_list,
+						 struct napi_entry, list);

fs/io_uring.c static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, · line 7897

Quoted 10 lines
fs/io_uring.c +10 -0
+
+			napi_busy_loop(ne->napi_id, io_busy_loop_end, iowq,
+				       true, BUSY_POLL_BUDGET);
+			io_check_napi_entry_timeout(ne);
+			break;
+		}
+	} while (io_napi_busy_loop(napi_list) &&
+		 !io_busy_loop_end(iowq, start_time));
+}
+

How about:

if (list is singular) {

    do something;

    return;

}

while (!io_busy_loop_end() && io_napi_busy_loop())

    ;

Btw, start_time seems not used in singular branch.

Regards,

Hao

Quoted 82 lines
fs/io_uring.c +44 -0
+static void io_putback_napi_list(struct io_ring_ctx *ctx,
+				 struct list_head *napi_list)
+{
+	struct napi_entry *cne, *lne;
+
+	spin_lock(&ctx->napi_lock);
+	list_for_each_entry(cne, &ctx->napi_list, list)
+		list_for_each_entry(lne, napi_list, list)
+			if (cne->napi_id == lne->napi_id) {
+				list_del(&lne->list);
+				kfree(lne);
+				break;
+			}
+	list_splice(napi_list, &ctx->napi_list);
+	spin_unlock(&ctx->napi_lock);
+}
+#endif /* CONFIG_NET_RX_BUSY_POLL */
+
  /*
   * Wait until events become available, if we don't already have some. The
   * application must reap them itself, as they reside on the shared cq ring.
@@ -7723,6 +7925,9 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
  	struct io_rings *rings = ctx->rings;
  	ktime_t timeout = KTIME_MAX;
  	int ret;
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	LIST_HEAD(local_napi_list);
+#endif
 
  	do {
  		io_cqring_overflow_flush(ctx);
@@ -7745,13 +7950,29 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
  			return ret;
  	}
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	iowq.busy_poll_to = 0;
+	if (!(ctx->flags & IORING_SETUP_SQPOLL)) {
+		spin_lock(&ctx->napi_lock);
+		list_splice_init(&ctx->napi_list, &local_napi_list);
+		spin_unlock(&ctx->napi_lock);
+	}
+#endif
  	if (uts) {
  		struct timespec64 ts;
 
  		if (get_timespec64(&ts, uts))
  			return -EFAULT;
+#ifdef CONFIG_NET_RX_BUSY_POLL
+		if (!list_empty(&local_napi_list))
+			io_adjust_busy_loop_timeout(&ts, &iowq);
+#endif
  		timeout = ktime_add_ns(timespec64_to_ktime(ts), ktime_get_ns());
  	}
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	else if (!list_empty(&local_napi_list))
+		iowq.busy_poll_to = READ_ONCE(sysctl_net_busy_poll);
+#endif
 
  	init_waitqueue_func_entry(&iowq.wq, io_wake_function);
  	iowq.wq.private = current;
@@ -7761,6 +7982,12 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
  	iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events;
 
  	trace_io_uring_cqring_wait(ctx, min_events);
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	if (iowq.busy_poll_to)
+		io_blocking_napi_busy_loop(&local_napi_list, &iowq);
+	if (!list_empty(&local_napi_list))
+		io_putback_napi_list(ctx, &local_napi_list);
+#endif
  	do {
  		/* if we can't even flush overflow, don't wait for more */
  		if (!io_cqring_overflow_flush(ctx)) {
@@ -9483,6 +9710,7 @@ static __cold void io_ring_ctx_free(struct io_ring_ctx *ctx)
  	io_req_caches_free(ctx);
  	if (ctx->hash_map)
  		io_wq_put_hash(ctx->hash_map);
+	io_free_napi_list(ctx);
  	kfree(ctx->cancel_hash);
  	kfree(ctx->dummy_ubuf);
  	kfree(ctx);
#4 lore raw
Quoted 9 lines · Hao Xu
Quoted 4 lines

+       ne = kmalloc(sizeof(*ne), GFP_NOWAIT);
+       if (!ne)
+               goto out;

IMHO, we need to handle -ENOMEM here, I cut off the error handling when

I did the quick coding. Sorry for misleading.

If you are correct, I would be shocked about this.

I did return in my 'Linux Device Drivers' book and nowhere it is mentionned that the kmalloc() can return something else than a pointer

No mention at all about the return value

in man page:
https://www.kernel.org/doc/htmldocs/kernel-api/API-kmalloc.html
API doc:

https://www.kernel.org/doc/html/latest/core-api/mm-api.html?highlight=kmalloc#c.kmalloc

header file:
https://elixir.bootlin.com/linux/latest/source/include/linux/slab.h#L522

I did browse into the kmalloc code. There is a lot of paths to cover but from preliminary reading, it pretty much seems that kmalloc only returns a valid pointer or NULL...

/**
* kmem_cache_alloc - Allocate an object
* @cachep: The cache to allocate from.
* @flags: See kmalloc().
*
* Allocate an object from this cache. The flags are only relevant
* if the cache has no available objects.
*
* Return: pointer to the new object or %NULL in case of error */

/**
* __do_kmalloc - allocate memory
* @size: how many bytes of memory are required.
* @flags: the type of memory to allocate (see kmalloc).
* @caller: function caller for debug tracking of the caller *
* Return: pointer to the allocated memory or %NULL in case of error */

I'll need someone else to confirm about possible kmalloc() return values with perhaps an example

I am a bit skeptic that something special needs to be done here...

Or perhaps you are suggesting that io_add_napi() returns an error code when allocation fails.

as done here:
https://elixir.bootlin.com/linux/latest/source/arch/alpha/kernel/core_marvel.c#L867

If that is what you suggest, what would this info do for the caller?

IMHO, it wouldn't help in any way...

Quoted 16 lines

fs/io_uring.c static int __io_sq_thread(struct io_ring_ctx · line 7639

Quoted 11 lines
fs/io_uring.c +3 -1
@@ -7519,7 +7633,11 @@ static int __io_sq_thread(struct io_ring_ctx
 *ctx, bool cap_entries)
                     !(ctx->flags & IORING_SETUP_R_DISABLED))
                         ret = io_submit_sqes(ctx, to_submit);
                 mutex_unlock(&ctx->uring_lock);
-
+#ifdef CONFIG_NET_RX_BUSY_POLL
+               if (!list_empty(&ctx->napi_list) &&
+                   io_napi_busy_loop(&ctx->napi_list))

I'm afraid we may need lock for sqpoll too, since io_add_napi() could be
in iowq context.

Quoted 3 lines

I'll take a look at the lock stuff of this patch tomorrow, too late now
in my timezone.

Ok, please do. I'm not a big user of io workers. I may have omitted to consider this possibility.

If that is the case, I think that this would be very easy to fix by locking the spinlock while __io_sq_thread() is using napi_list.

Quoted 7 lines

How about:

if (list is singular) {

     do something;

Quoted 8 lines

     return;

}

while (!io_busy_loop_end() && io_napi_busy_loop())

     ;

is there a concern with the current code?
What would be the benefit of your suggestion over current code?

To me, it seems that if io_blocking_napi_busy_loop() is called, a reasonable expectation would be that some busy looping is done or else you could return the function without doing anything which would, IMHO, be misleading.

By definition, napi_busy_loop() is not blocking and if you desire the device to be in busy poll mode, you need to do it once in a while or else, after a certain time, the device will return back to its interrupt mode.

IOW, io_blocking_napi_busy_loop() follows the same logic used by napi_busy_loop() that does not call loop_end() before having perform 1 loop iteration.

Quoted 1 line

Btw, start_time seems not used in singular branch.

I know. This is why it is conditionally initialized.

Greetings,

#5 lore raw
Quoted 38 lines · Olivier Langlois
Quoted 8 lines · Hao Xu

How about:

if (list is singular) {

     do something;

Quoted 8 lines

     return;

}

while (!io_busy_loop_end() && io_napi_busy_loop())

     ;

is there a concern with the current code?
What would be the benefit of your suggestion over current code?

To me, it seems that if io_blocking_napi_busy_loop() is called, a reasonable expectation would be that some busy looping is done or else
you could return the function without doing anything which would, IMHO,
be misleading.

By definition, napi_busy_loop() is not blocking and if you desire the device to be in busy poll mode, you need to do it once in a while or else, after a certain time, the device will return back to its interrupt mode.

IOW, io_blocking_napi_busy_loop() follows the same logic used by napi_busy_loop() that does not call loop_end() before having perform 1
loop iteration.

Quoted 6 lines
Quoted 1 line

Btw, start_time seems not used in singular branch.

I know. This is why it is conditionally initialized.

Greetings,

Another argument for not touching the code the way that it is:
io_napi_busy_loop() has another function on top of iterating the napi_list and calling napi_busy_loop() for each of them.

The function also check the list entries validity and frees them when they time out. Not calling io_napi_busy_loop() you would bypass this check and that could result in timed out entries to never be disposed.

#10 lore raw
In reply to Hao Xu
Quoted 34 lines · Hao Xu
Quoted 23 lines

+static void io_blocking_napi_busy_loop(struct list_head *napi_list,
+                                      struct io_wait_queue *iowq)

fs/io_uring.c +10 -0
+{
+       unsigned long start_time =
+               list_is_singular(napi_list) ? 0 :
+               busy_loop_current_time();
+
+       do {
+               if (list_is_singular(napi_list)) {
+                       struct napi_entry *ne =
+                               list_first_entry(napi_list,
+                                                struct napi_entry,

list);
+
+                       napi_busy_loop(ne->napi_id, io_busy_loop_end, iowq,
+                                      true, BUSY_POLL_BUDGET); +                       io_check_napi_entry_timeout(ne); +                       break;
+               }
+       } while (io_napi_busy_loop(napi_list) &&
+                !io_busy_loop_end(iowq, start_time));

Review

Quoted 2 lines
(quoted diff) +2 -0
+}
+

How about:

if (list is singular) {

     do something;

     return;

Quoted 8 lines

}

while (!io_busy_loop_end() && io_napi_busy_loop())

     ;

Btw, start_time seems not used in singular branch.

Hao,

it takes me few readings before being able to figure out the idea behind your suggestions. Sorry about that!

So, if I get it correctly, you are proposing extract out the singular block out of the while loop...

IMHO, this is not a good idea because you could start iterating the do/while loop with a multiple entries list that ends up becoming a singular list after one or few iterations.

Check what io_napi_busy_loop() is doing...

It does not look like that but a lot thoughts have been put into writing io_blocking_napi_busy_loop()...

#6 lore raw
In reply to Olivier Langlois
Quoted 48 lines · Olivier Langlois
Quoted 7 lines · Hao Xu
Quoted 3 lines

+       ne = kmalloc(sizeof(*ne), GFP_NOWAIT);
+       if (!ne)
+               goto out;

IMHO, we need to handle -ENOMEM here, I cut off the error handling when

I did the quick coding. Sorry for misleading.

If you are correct, I would be shocked about this.

I did return in my 'Linux Device Drivers' book and nowhere it is mentionned that the kmalloc() can return something else than a pointer

No mention at all about the return value

in man page:
https://www.kernel.org/doc/htmldocs/kernel-api/API-kmalloc.html
API doc:

https://www.kernel.org/doc/html/latest/core-api/mm-api.html?highlight=kmalloc#c.kmalloc

header file:
https://elixir.bootlin.com/linux/latest/source/include/linux/slab.h#L522

I did browse into the kmalloc code. There is a lot of paths to cover but from preliminary reading, it pretty much seems that kmalloc only returns a valid pointer or NULL...

/**
* kmem_cache_alloc - Allocate an object
* @cachep: The cache to allocate from.
* @flags: See kmalloc().
*
* Allocate an object from this cache. The flags are only relevant
* if the cache has no available objects.
*
* Return: pointer to the new object or %NULL in case of error */

/**
* __do_kmalloc - allocate memory
* @size: how many bytes of memory are required.
* @flags: the type of memory to allocate (see kmalloc).
* @caller: function caller for debug tracking of the caller *
* Return: pointer to the allocated memory or %NULL in case of error */

Quoted 7 lines

I'll need someone else to confirm about possible kmalloc() return values with perhaps an example

I am a bit skeptic that something special needs to be done here...

Or perhaps you are suggesting that io_add_napi() returns an error code when allocation fails.

This is what I mean.

Quoted 7 lines

as done here:
https://elixir.bootlin.com/linux/latest/source/arch/alpha/kernel/core_marvel.c#L867

If that is what you suggest, what would this info do for the caller?

IMHO, it wouldn't help in any way...

Hmm, I'm not sure, you're probably right based on that ENOMEM here shouldn't

fail the arm poll, but we wanna do it, we can do something like what we do for

kmalloc() in io_arm_poll_handler()). I'll leave it to others.

Quoted 29 lines
Quoted 13 lines

fs/io_uring.c static int __io_sq_thread(struct io_ring_ctx · line 7639

Quoted 9 lines
fs/io_uring.c +3 -1
@@ -7519,7 +7633,11 @@ static int __io_sq_thread(struct io_ring_ctx
 *ctx, bool cap_entries)
                     !(ctx->flags & IORING_SETUP_R_DISABLED))
                         ret = io_submit_sqes(ctx, to_submit);
                 mutex_unlock(&ctx->uring_lock);
-
+#ifdef CONFIG_NET_RX_BUSY_POLL
+               if (!list_empty(&ctx->napi_list) &&
+                   io_napi_busy_loop(&ctx->napi_list))

I'm afraid we may need lock for sqpoll too, since io_add_napi() could be
in iowq context.

Quoted 3 lines

I'll take a look at the lock stuff of this patch tomorrow, too late now
in my timezone.

Ok, please do. I'm not a big user of io workers. I may have omitted to consider this possibility.

If that is the case, I think that this would be very easy to fix by locking the spinlock while __io_sq_thread() is using napi_list.

Quoted 8 lines

How about:

if (list is singular) {

     do something;

     return;

Quoted 8 lines
Quoted 6 lines

}

while (!io_busy_loop_end() && io_napi_busy_loop())

     ;

is there a concern with the current code?
What would be the benefit of your suggestion over current code?

No, it's just coding style concern, since I see

do {

    if() {

        break;

    }

} while();

which means the if statement is actually not int the loop. Anyway, it's just

personal taste.

Quoted 6 lines

To me, it seems that if io_blocking_napi_busy_loop() is called, a reasonable expectation would be that some busy looping is done or else you could return the function without doing anything which would, IMHO, be misleading.

Quoted 8 lines

By definition, napi_busy_loop() is not blocking and if you desire the device to be in busy poll mode, you need to do it once in a while or else, after a certain time, the device will return back to its interrupt mode.

IOW, io_blocking_napi_busy_loop() follows the same logic used by napi_busy_loop() that does not call loop_end() before having perform 1 loop iteration.

I see, thanks for explanation. I'm ok with this.

Quoted 3 lines
Quoted 2 lines

Btw, start_time seems not used in singular branch.

I know. This is why it is conditionally initialized.

like what I said, just personal taste.

+static void io_blocking_napi_busy_loop(struct list_head *napi_list,

+                                      struct io_wait_queue *iowq) +{
+       if (list_is_singular(napi_list)) {
+               struct napi_entry *ne =
+                       list_first_entry(napi_list,
+                                        struct napi_entry, list); +
+               napi_busy_loop(ne->napi_id, io_busy_loop_end, iowq, +                              true, BUSY_POLL_BUDGET);
+               io_check_napi_entry_timeout(ne);
+               break;
+       }
+
+       while (io_napi_busy_loop(napi_list)) {
+               if(io_busy_loop_end(iowq, busy_loop_current_time())) +                       break;
+       }
+}

Quoted 2 lines

Greetings,

#11 lore raw
In reply to Olivier Langlois
Quoted 47 lines · Olivier Langlois
Quoted 33 lines · Hao Xu
Quoted 23 lines

+static void io_blocking_napi_busy_loop(struct list_head *napi_list,
+                                      struct io_wait_queue *iowq)

fs/io_uring.c +10 -0
+{
+       unsigned long start_time =
+               list_is_singular(napi_list) ? 0 :
+               busy_loop_current_time();
+
+       do {
+               if (list_is_singular(napi_list)) {
+                       struct napi_entry *ne =
+                               list_first_entry(napi_list,
+                                                struct napi_entry,

list);
+
+                       napi_busy_loop(ne->napi_id, io_busy_loop_end, iowq,
+                                      true, BUSY_POLL_BUDGET); +                       io_check_napi_entry_timeout(ne); +                       break;
+               }
+       } while (io_napi_busy_loop(napi_list) &&
+                !io_busy_loop_end(iowq, start_time));

Review

Quoted 2 lines
(quoted diff) +2 -0
+}
+

How about:

if (list is singular) {

     do something;

     return;

Quoted 8 lines

}

while (!io_busy_loop_end() && io_napi_busy_loop())

     ;

Btw, start_time seems not used in singular branch.

Hao,

it takes me few readings before being able to figure out the idea behind your suggestions. Sorry about that!

Quoted 6 lines

So, if I get it correctly, you are proposing extract out the singular block out of the while loop...

IMHO, this is not a good idea because you could start iterating the do/while loop with a multiple entries list that ends up becoming a singular list after one or few iterations.

My bad, I get your point now.

Quoted 5 lines

Check what io_napi_busy_loop() is doing...

It does not look like that but a lot thoughts have been put into writing io_blocking_napi_busy_loop()...

True.

#7 lore raw
In reply to Hao Xu
Quoted 63 lines · Hao Xu
Quoted 48 lines · Olivier Langlois
Quoted 7 lines · Hao Xu
Quoted 3 lines

+       ne = kmalloc(sizeof(*ne), GFP_NOWAIT);
+       if (!ne)
+               goto out;

IMHO, we need to handle -ENOMEM here, I cut off the error handling when

I did the quick coding. Sorry for misleading.

If you are correct, I would be shocked about this.

I did return in my 'Linux Device Drivers' book and nowhere it is mentionned that the kmalloc() can return something else than a pointer

No mention at all about the return value

in man page:
https://www.kernel.org/doc/htmldocs/kernel-api/API-kmalloc.html
API doc:

https://www.kernel.org/doc/html/latest/core-api/mm-api.html?highlight=kmalloc#c.kmalloc

header file:
https://elixir.bootlin.com/linux/latest/source/include/linux/slab.h#L522

I did browse into the kmalloc code. There is a lot of paths to cover but from preliminary reading, it pretty much seems that kmalloc only returns a valid pointer or NULL...

/**
  * kmem_cache_alloc - Allocate an object
  * @cachep: The cache to allocate from.
  * @flags: See kmalloc().
  *
  * Allocate an object from this cache.  The flags are only relevant
  * if the cache has no available objects.
  *
  * Return: pointer to the new object or %NULL in case of error */
    /**
  * __do_kmalloc - allocate memory
  * @size: how many bytes of memory are required.
  * @flags: the type of memory to allocate (see kmalloc).
  * @caller: function caller for debug tracking of the caller *
  * Return: pointer to the allocated memory or %NULL in case of error */

Quoted 7 lines

I'll need someone else to confirm about possible kmalloc() return values with perhaps an example

I am a bit skeptic that something special needs to be done here...

Or perhaps you are suggesting that io_add_napi() returns an error code when allocation fails.

This is what I mean.

Quoted 5 lines

as done here:
https://elixir.bootlin.com/linux/latest/source/arch/alpha/kernel/core_marvel.c#L867

Quoted 7 lines
Quoted 3 lines

If that is what you suggest, what would this info do for the caller?

IMHO, it wouldn't help in any way...

Hmm, I'm not sure, you're probably right based on that ENOMEM here shouldn't

fail the arm poll, but we wanna do it, we can do something like what we do for

                            ^---but if we wanna do it

#8 lore raw
Quoted 7 lines · Hao Xu
Quoted 7 lines
Quoted 6 lines

If that is what you suggest, what would this info do for the caller?

IMHO, it wouldn't help in any way...

Quoted 7 lines
Quoted 6 lines

Hmm, I'm not sure, you're probably right based on that ENOMEM here shouldn't

fail the arm poll, but we wanna do it, we can do something like what
we do for

                             ^---but if we wanna do it

My position is that being able to perform busy poll is a nice to have feature if the necessary resources are available. If not the request will still be handled correctly so nothing special should be done in case of mem alloc problem.

but fair enough, lets wait for Jens and Pavel to chime him if they would like to see something to be done here.

Beside that, all I need to know is if napi_list needs to be protected in __io_sq_thread with regards to io worket threads to start working on a v5.

I'll look into this question too...

#9 lore raw
Quoted 9 lines · Olivier Langlois
Quoted 8 lines · Hao Xu
Quoted 8 lines
Quoted 5 lines

If that is what you suggest, what would this info do for the caller?

IMHO, it wouldn't help in any way...

Hmm, I'm not sure, you're probably right based on that ENOMEM here shouldn't

Quoted 8 lines
Quoted 4 lines
Quoted 3 lines

fail the arm poll, but we wanna do it, we can do something like what
we do for

                             ^---but if we wanna do it

My position is that being able to perform busy poll is a nice to have feature if the necessary resources are available. If not the request will still be handled correctly so nothing special should be done in case of mem alloc problem.

Exactly what I meant.

Quoted 3 lines

but fair enough, lets wait for Jens and Pavel to chime him if they would like to see something to be done here.

Agree.

Quoted 4 lines

Beside that, all I need to know is if napi_list needs to be protected in __io_sq_thread with regards to io worket threads to start working on a v5.

Sorry for the delay, was stuck in other things. We definitely need

lock in this case too. It should be several lines code, super appreciate

if you could add it.

Thanks,

Hao

Quoted 2 lines

I'll look into this question too...