[PATCHSET v3 0/5] Add support for batched min timeout

[PATCHSET v3 0/5] Add support for batched min timeout

[Jens Axboe]

Hi,

After somewhat of a lengthy delay from v2, here's v3 of this patchset.
We've now validated this actually makes sense and helps for production
workloads where there's a big gap between the peak and slow hours in
terms of load. Having a min time enables the same settings to be used
for peak hours (guaranteeing minimum response time), while avoiding
excessive context switches during lulls.

For a full description, see the v2 posting:

https://lore.kernel.org/io-uring/20240215161002.3044270-1-axboe@kernel.dk/

As before, there's a liburing branch with added test cases, it can be
found here:

https://git.kernel.dk/cgit/liburing/log/?h=min-wait

The patches are on top of for-6.12/io_uring, and with David Wei's
new iowait feat and enter flag added to avoid conflicts with that.

Changes since v2:
- Rebase on current 6.12 tree, which was mostly centered around Pavel's
  rework of the timeout handling and ABS_TIME additions
- Since there's now one more argument for io_cqring_wait(), wrap those
  output arguments in a struct ext_arg.

 include/uapi/linux/io_uring.h |   3 +-
 io_uring/io_uring.c           | 187 +++++++++++++++++++++++++---------
 io_uring/io_uring.h           |   4 +
 3 files changed, 147 insertions(+), 47 deletions(-)

-- 
Jens Axboe

[PATCH 1/5] io_uring: encapsulate extraneous wait flags into a separate struct

[Jens Axboe]

Rather than need to pass in 2 or 3 separate arguments, add a struct
to encapsulate the timeout and sigset_t parts of waiting. In preparation
for adding another argument for waiting.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/io_uring.c | 45 ++++++++++++++++++++++++---------------------
 1 file changed, 24 insertions(+), 21 deletions(-)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 5e75672525df..f90a4490908b 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -2384,13 +2384,18 @@ static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	return ret;
 }
 
+struct ext_arg {
+	size_t argsz;
+	struct __kernel_timespec __user *ts;
+	const sigset_t __user *sig;
+};
+
 /*
  * 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.
  */
 static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
-			  const sigset_t __user *sig, size_t sigsz,
-			  struct __kernel_timespec __user *uts)
+			  struct ext_arg *ext_arg)
 {
 	struct io_wait_queue iowq;
 	struct io_rings *rings = ctx->rings;
@@ -2416,10 +2421,10 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 	iowq.timeout = KTIME_MAX;
 	iowq.no_iowait = flags & IORING_ENTER_NO_IOWAIT;
 
-	if (uts) {
+	if (ext_arg->ts) {
 		struct timespec64 ts;
 
-		if (get_timespec64(&ts, uts))
+		if (get_timespec64(&ts, ext_arg->ts))
 			return -EFAULT;
 
 		iowq.timeout = timespec64_to_ktime(ts);
@@ -2427,14 +2432,14 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 			iowq.timeout = ktime_add(iowq.timeout, io_get_time(ctx));
 	}
 
-	if (sig) {
+	if (ext_arg->sig) {
 #ifdef CONFIG_COMPAT
 		if (in_compat_syscall())
-			ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
-						      sigsz);
+			ret = set_compat_user_sigmask((const compat_sigset_t __user *)ext_arg->sig,
+						      ext_arg->argsz);
 		else
 #endif
-			ret = set_user_sigmask(sig, sigsz);
+			ret = set_user_sigmask(ext_arg->sig, ext_arg->argsz);
 
 		if (ret)
 			return ret;
@@ -3113,9 +3118,8 @@ static int io_validate_ext_arg(unsigned flags, const void __user *argp, size_t a
 	return 0;
 }
 
-static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
-			  struct __kernel_timespec __user **ts,
-			  const sigset_t __user **sig)
+static int io_get_ext_arg(unsigned flags, const void __user *argp,
+			  struct ext_arg *ext_arg)
 {
 	struct io_uring_getevents_arg arg;
 
@@ -3124,8 +3128,8 @@ static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz
 	 * is just a pointer to the sigset_t.
 	 */
 	if (!(flags & IORING_ENTER_EXT_ARG)) {
-		*sig = (const sigset_t __user *) argp;
-		*ts = NULL;
+		ext_arg->sig = (const sigset_t __user *) argp;
+		ext_arg->ts = NULL;
 		return 0;
 	}
 
@@ -3133,15 +3137,15 @@ static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz
 	 * EXT_ARG is set - ensure we agree on the size of it and copy in our
 	 * timespec and sigset_t pointers if good.
 	 */
-	if (*argsz != sizeof(arg))
+	if (ext_arg->argsz != sizeof(arg))
 		return -EINVAL;
 	if (copy_from_user(&arg, argp, sizeof(arg)))
 		return -EFAULT;
 	if (arg.pad)
 		return -EINVAL;
-	*sig = u64_to_user_ptr(arg.sigmask);
-	*argsz = arg.sigmask_sz;
-	*ts = u64_to_user_ptr(arg.ts);
+	ext_arg->sig = u64_to_user_ptr(arg.sigmask);
+	ext_arg->argsz = arg.sigmask_sz;
+	ext_arg->ts = u64_to_user_ptr(arg.ts);
 	return 0;
 }
 
@@ -3247,15 +3251,14 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
 			}
 			mutex_unlock(&ctx->uring_lock);
 		} else {
-			const sigset_t __user *sig;
-			struct __kernel_timespec __user *ts;
+			struct ext_arg ext_arg = { .argsz = argsz };
 
-			ret2 = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
+			ret2 = io_get_ext_arg(flags, argp, &ext_arg);
 			if (likely(!ret2)) {
 				min_complete = min(min_complete,
 						   ctx->cq_entries);
 				ret2 = io_cqring_wait(ctx, min_complete, flags,
-						      sig, argsz, ts);
+						      &ext_arg);
 			}
 		}
 
-- 
2.43.0

[PATCH 2/5] io_uring: move schedule wait logic into helper

[Jens Axboe]

In preparation for expanding how we handle waits, move the actual
schedule and schedule_timeout() handling into a helper.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/io_uring.c | 37 +++++++++++++++++++++----------------
 1 file changed, 21 insertions(+), 16 deletions(-)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index f90a4490908b..2bdb66902f58 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -2350,22 +2350,10 @@ static bool current_pending_io(void)
 	return percpu_counter_read_positive(&tctx->inflight);
 }
 
-/* when returns >0, the caller should retry */
-static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
-					  struct io_wait_queue *iowq)
+static int __io_cqring_wait_schedule(struct io_ring_ctx *ctx,
+				     struct io_wait_queue *iowq)
 {
-	int ret;
-
-	if (unlikely(READ_ONCE(ctx->check_cq)))
-		return 1;
-	if (unlikely(!llist_empty(&ctx->work_llist)))
-		return 1;
-	if (unlikely(test_thread_flag(TIF_NOTIFY_SIGNAL)))
-		return 1;
-	if (unlikely(task_sigpending(current)))
-		return -EINTR;
-	if (unlikely(io_should_wake(iowq)))
-		return 0;
+	int ret = 0;
 
 	/*
 	 * Mark us as being in io_wait if we have pending requests, so cpufreq
@@ -2374,7 +2362,6 @@ static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	 */
 	if (!iowq->no_iowait && current_pending_io())
 		current->in_iowait = 1;
-	ret = 0;
 	if (iowq->timeout == KTIME_MAX)
 		schedule();
 	else if (!schedule_hrtimeout_range_clock(&iowq->timeout, 0,
@@ -2384,6 +2371,24 @@ static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	return ret;
 }
 
+/* If this returns > 0, the caller should retry */
+static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
+					  struct io_wait_queue *iowq)
+{
+	if (unlikely(READ_ONCE(ctx->check_cq)))
+		return 1;
+	if (unlikely(!llist_empty(&ctx->work_llist)))
+		return 1;
+	if (unlikely(test_thread_flag(TIF_NOTIFY_SIGNAL)))
+		return 1;
+	if (unlikely(task_sigpending(current)))
+		return -EINTR;
+	if (unlikely(io_should_wake(iowq)))
+		return 0;
+
+	return __io_cqring_wait_schedule(ctx, iowq);
+}
+
 struct ext_arg {
 	size_t argsz;
 	struct __kernel_timespec __user *ts;
-- 
2.43.0

[PATCH 3/5] io_uring: implement our own schedule timeout handling

[Jens Axboe]

In preparation for having two distinct timeouts and avoid waking the
task if we don't need to.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/io_uring.c | 41 ++++++++++++++++++++++++++++++++++++-----
 io_uring/io_uring.h |  2 ++
 2 files changed, 38 insertions(+), 5 deletions(-)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 2bdb66902f58..6e53ebd58aab 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -2322,7 +2322,7 @@ static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
 	 * Cannot safely flush overflowed CQEs from here, ensure we wake up
 	 * the task, and the next invocation will do it.
 	 */
-	if (io_should_wake(iowq) || io_has_work(iowq->ctx))
+	if (io_should_wake(iowq) || io_has_work(iowq->ctx) || iowq->hit_timeout)
 		return autoremove_wake_function(curr, mode, wake_flags, key);
 	return -1;
 }
@@ -2350,6 +2350,38 @@ static bool current_pending_io(void)
 	return percpu_counter_read_positive(&tctx->inflight);
 }
 
+static enum hrtimer_restart io_cqring_timer_wakeup(struct hrtimer *timer)
+{
+	struct io_wait_queue *iowq = container_of(timer, struct io_wait_queue, t);
+	struct io_ring_ctx *ctx = iowq->ctx;
+
+	WRITE_ONCE(iowq->hit_timeout, 1);
+	if (ctx->flags & IORING_SETUP_DEFER_TASKRUN)
+		wake_up_process(ctx->submitter_task);
+	else
+		io_cqring_wake(ctx);
+	return HRTIMER_NORESTART;
+}
+
+static int io_cqring_schedule_timeout(struct io_wait_queue *iowq,
+				      clockid_t clock_id)
+{
+	iowq->hit_timeout = 0;
+	hrtimer_init_on_stack(&iowq->t, clock_id, HRTIMER_MODE_ABS);
+	iowq->t.function = io_cqring_timer_wakeup;
+	hrtimer_set_expires_range_ns(&iowq->t, iowq->timeout, 0);
+	hrtimer_start_expires(&iowq->t, HRTIMER_MODE_ABS);
+
+	if (!READ_ONCE(iowq->hit_timeout))
+		schedule();
+
+	hrtimer_cancel(&iowq->t);
+	destroy_hrtimer_on_stack(&iowq->t);
+	__set_current_state(TASK_RUNNING);
+
+	return READ_ONCE(iowq->hit_timeout) ? -ETIME : 0;
+}
+
 static int __io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 				     struct io_wait_queue *iowq)
 {
@@ -2362,11 +2394,10 @@ static int __io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	 */
 	if (!iowq->no_iowait && current_pending_io())
 		current->in_iowait = 1;
-	if (iowq->timeout == KTIME_MAX)
+	if (iowq->timeout != KTIME_MAX)
+		ret = io_cqring_schedule_timeout(iowq, ctx->clockid);
+	else
 		schedule();
-	else if (!schedule_hrtimeout_range_clock(&iowq->timeout, 0,
-						 HRTIMER_MODE_ABS, ctx->clockid))
-		ret = -ETIME;
 	current->in_iowait = 0;
 	return ret;
 }
diff --git a/io_uring/io_uring.h b/io_uring/io_uring.h
index e0868b79774c..bac830a2d6ec 100644
--- a/io_uring/io_uring.h
+++ b/io_uring/io_uring.h
@@ -40,7 +40,9 @@ struct io_wait_queue {
 	struct io_ring_ctx *ctx;
 	unsigned cq_tail;
 	unsigned nr_timeouts;
+	int hit_timeout;
 	ktime_t timeout;
+	struct hrtimer t;
 #ifdef CONFIG_NET_RX_BUSY_POLL
 	ktime_t napi_busy_poll_dt;
 	bool napi_prefer_busy_poll;
-- 
2.43.0

[PATCH 4/5] io_uring: add support for batch wait timeout

[Jens Axboe]

Waiting for events with io_uring has two knobs that can be set:

1) The number of events to wake for
2) The timeout associated with the event

Waiting will abort when either of those conditions are met, as expected.

This adds support for a third event, which is associated with the number
of events to wait for. Applications generally like to handle batches of
completions, and right now they'd set a number of events to wait for and
the timeout for that. If no events have been received but the timeout
triggers, control is returned to the application and it can wait again.
However, if the application doesn't have anything to do until events are
reaped, then it's possible to make this waiting more efficient.

For example, the application may have a latency time of 50 usecs and
wanting to handle a batch of 8 requests at the time. If it uses 50 usecs
as the timeout, then it'll be doing 20K context switches per second even
if nothing is happening.

This introduces the notion of min batch wait time. If the min batch wait
time expires, then we'll return to userspace if we have any events at all.
If none are available, the general wait time is applied. Any request
arriving after the min batch wait time will cause waiting to stop and
return control to the application.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/io_uring.c | 75 +++++++++++++++++++++++++++++++++++++++------
 io_uring/io_uring.h |  2 ++
 2 files changed, 67 insertions(+), 10 deletions(-)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 6e53ebd58aab..27d949ff84a3 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -2363,13 +2363,62 @@ static enum hrtimer_restart io_cqring_timer_wakeup(struct hrtimer *timer)
 	return HRTIMER_NORESTART;
 }
 
+/*
+ * Doing min_timeout portion. If we saw any timeouts, events, or have work,
+ * wake up. If not, and we have a normal timeout, switch to that and keep
+ * sleeping.
+ */
+static enum hrtimer_restart io_cqring_min_timer_wakeup(struct hrtimer *timer)
+{
+	struct io_wait_queue *iowq = container_of(timer, struct io_wait_queue, t);
+	struct io_ring_ctx *ctx = iowq->ctx;
+
+	/* no general timeout, or shorter, we are done */
+	if (iowq->timeout == KTIME_MAX ||
+	    ktime_after(iowq->min_timeout, iowq->timeout))
+		goto out_wake;
+	/* work we may need to run, wake function will see if we need to wake */
+	if (io_has_work(ctx))
+		goto out_wake;
+	/* got events since we started waiting, min timeout is done */
+	if (iowq->cq_min_tail != READ_ONCE(ctx->rings->cq.tail))
+		goto out_wake;
+	/* if we have any events and min timeout expired, we're done */
+	if (io_cqring_events(ctx))
+		goto out_wake;
+
+	/*
+	 * If using deferred task_work running and application is waiting on
+	 * more than one request, ensure we reset it now where we are switching
+	 * to normal sleeps. Any request completion post min_wait should wake
+	 * the task and return.
+	 */
+	if (ctx->flags & IORING_SETUP_DEFER_TASKRUN)
+		atomic_set(&ctx->cq_wait_nr, 1);
+
+	iowq->t.function = io_cqring_timer_wakeup;
+	hrtimer_set_expires(timer, iowq->timeout);
+	return HRTIMER_RESTART;
+out_wake:
+	return io_cqring_timer_wakeup(timer);
+}
+
 static int io_cqring_schedule_timeout(struct io_wait_queue *iowq,
-				      clockid_t clock_id)
+				      clockid_t clock_id, ktime_t start_time)
 {
+	ktime_t timeout;
+
 	iowq->hit_timeout = 0;
 	hrtimer_init_on_stack(&iowq->t, clock_id, HRTIMER_MODE_ABS);
-	iowq->t.function = io_cqring_timer_wakeup;
-	hrtimer_set_expires_range_ns(&iowq->t, iowq->timeout, 0);
+	if (iowq->min_timeout) {
+		timeout = ktime_add_ns(iowq->min_timeout, start_time);
+		iowq->t.function = io_cqring_min_timer_wakeup;
+	} else {
+		timeout = iowq->timeout;
+		iowq->t.function = io_cqring_timer_wakeup;
+	}
+
+	hrtimer_set_expires_range_ns(&iowq->t, timeout, 0);
 	hrtimer_start_expires(&iowq->t, HRTIMER_MODE_ABS);
 
 	if (!READ_ONCE(iowq->hit_timeout))
@@ -2383,7 +2432,8 @@ static int io_cqring_schedule_timeout(struct io_wait_queue *iowq,
 }
 
 static int __io_cqring_wait_schedule(struct io_ring_ctx *ctx,
-				     struct io_wait_queue *iowq)
+				     struct io_wait_queue *iowq,
+				     ktime_t start_time)
 {
 	int ret = 0;
 
@@ -2394,8 +2444,8 @@ static int __io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	 */
 	if (!iowq->no_iowait && current_pending_io())
 		current->in_iowait = 1;
-	if (iowq->timeout != KTIME_MAX)
-		ret = io_cqring_schedule_timeout(iowq, ctx->clockid);
+	if (iowq->timeout != KTIME_MAX || iowq->min_timeout != KTIME_MAX)
+		ret = io_cqring_schedule_timeout(iowq, ctx->clockid, start_time);
 	else
 		schedule();
 	current->in_iowait = 0;
@@ -2404,7 +2454,8 @@ static int __io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 
 /* If this returns > 0, the caller should retry */
 static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
-					  struct io_wait_queue *iowq)
+					  struct io_wait_queue *iowq,
+					  ktime_t start_time)
 {
 	if (unlikely(READ_ONCE(ctx->check_cq)))
 		return 1;
@@ -2417,7 +2468,7 @@ static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	if (unlikely(io_should_wake(iowq)))
 		return 0;
 
-	return __io_cqring_wait_schedule(ctx, iowq);
+	return __io_cqring_wait_schedule(ctx, iowq, start_time);
 }
 
 struct ext_arg {
@@ -2435,6 +2486,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 {
 	struct io_wait_queue iowq;
 	struct io_rings *rings = ctx->rings;
+	ktime_t start_time;
 	int ret;
 
 	if (!io_allowed_run_tw(ctx))
@@ -2453,9 +2505,12 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 	INIT_LIST_HEAD(&iowq.wq.entry);
 	iowq.ctx = ctx;
 	iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
+	iowq.cq_min_tail = READ_ONCE(ctx->rings->cq.tail);
 	iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events;
+	iowq.min_timeout = 0;
 	iowq.timeout = KTIME_MAX;
 	iowq.no_iowait = flags & IORING_ENTER_NO_IOWAIT;
+	start_time = io_get_time(ctx);
 
 	if (ext_arg->ts) {
 		struct timespec64 ts;
@@ -2465,7 +2520,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 
 		iowq.timeout = timespec64_to_ktime(ts);
 		if (!(flags & IORING_ENTER_ABS_TIMER))
-			iowq.timeout = ktime_add(iowq.timeout, io_get_time(ctx));
+			iowq.timeout = ktime_add(iowq.timeout, start_time);
 	}
 
 	if (ext_arg->sig) {
@@ -2496,7 +2551,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 							TASK_INTERRUPTIBLE);
 		}
 
-		ret = io_cqring_wait_schedule(ctx, &iowq);
+		ret = io_cqring_wait_schedule(ctx, &iowq, start_time);
 		__set_current_state(TASK_RUNNING);
 		atomic_set(&ctx->cq_wait_nr, IO_CQ_WAKE_INIT);
 
diff --git a/io_uring/io_uring.h b/io_uring/io_uring.h
index bac830a2d6ec..24ecd31c81e9 100644
--- a/io_uring/io_uring.h
+++ b/io_uring/io_uring.h
@@ -39,8 +39,10 @@ struct io_wait_queue {
 	struct wait_queue_entry wq;
 	struct io_ring_ctx *ctx;
 	unsigned cq_tail;
+	unsigned cq_min_tail;
 	unsigned nr_timeouts;
 	int hit_timeout;
+	ktime_t min_timeout;
 	ktime_t timeout;
 	struct hrtimer t;
 #ifdef CONFIG_NET_RX_BUSY_POLL
-- 
2.43.0

[PATCH 5/5] io_uring: wire up min batch wake timeout

[Jens Axboe]

Expose min_wait_usec in io_uring_getevents_arg, replacing the pad member
that is currently in there. The value is in usecs, which is explained in
the name as well.

Note that if min_wait_usec and a normal timeout is used in conjunction,
the normal timeout is still relative to the base time. For example, if
min_wait_usec is set to 100 and the normal timeout is 1000, the max
total time waited is still 1000. This also means that if the normal
timeout is shorter than min_wait_usec, then only the min_wait_usec will
take effect.

See previous commit for an explanation of how this works.

IORING_FEAT_MIN_TIMEOUT is added as a feature flag for this, as
applications doing submit_and_wait_timeout() style operations will
generally not see the -EINVAL from the wait side as they return the
number of IOs submitted. Only if no IOs are submitted will the -EINVAL
bubble back up to the application.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/uapi/linux/io_uring.h | 3 ++-
 io_uring/io_uring.c           | 9 +++++----
 2 files changed, 7 insertions(+), 5 deletions(-)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 3a94afa8665e..da0d472d5ec7 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -545,6 +545,7 @@ struct io_uring_params {
 #define IORING_FEAT_REG_REG_RING	(1U << 13)
 #define IORING_FEAT_RECVSEND_BUNDLE	(1U << 14)
 #define IORING_FEAT_IOWAIT_TOGGLE	(1U << 15)
+#define IORING_FEAT_MIN_TIMEOUT		(1U << 16)
 
 /*
  * io_uring_register(2) opcodes and arguments
@@ -768,7 +769,7 @@ enum io_uring_register_restriction_op {
 struct io_uring_getevents_arg {
 	__u64	sigmask;
 	__u32	sigmask_sz;
-	__u32	pad;
+	__u32	min_wait_usec;
 	__u64	ts;
 };
 
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 27d949ff84a3..ecebcea5cbd7 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -2475,6 +2475,7 @@ struct ext_arg {
 	size_t argsz;
 	struct __kernel_timespec __user *ts;
 	const sigset_t __user *sig;
+	ktime_t min_time;
 };
 
 /*
@@ -2507,7 +2508,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 	iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
 	iowq.cq_min_tail = READ_ONCE(ctx->rings->cq.tail);
 	iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events;
-	iowq.min_timeout = 0;
+	iowq.min_timeout = ext_arg->min_time;
 	iowq.timeout = KTIME_MAX;
 	iowq.no_iowait = flags & IORING_ENTER_NO_IOWAIT;
 	start_time = io_get_time(ctx);
@@ -3232,8 +3233,7 @@ static int io_get_ext_arg(unsigned flags, const void __user *argp,
 		return -EINVAL;
 	if (copy_from_user(&arg, argp, sizeof(arg)))
 		return -EFAULT;
-	if (arg.pad)
-		return -EINVAL;
+	ext_arg->min_time = arg.min_wait_usec * NSEC_PER_USEC;
 	ext_arg->sig = u64_to_user_ptr(arg.sigmask);
 	ext_arg->argsz = arg.sigmask_sz;
 	ext_arg->ts = u64_to_user_ptr(arg.ts);
@@ -3634,7 +3634,8 @@ static __cold int io_uring_create(unsigned entries, struct io_uring_params *p,
 			IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS |
 			IORING_FEAT_RSRC_TAGS | IORING_FEAT_CQE_SKIP |
 			IORING_FEAT_LINKED_FILE | IORING_FEAT_REG_REG_RING |
-			IORING_FEAT_RECVSEND_BUNDLE | IORING_FEAT_IOWAIT_TOGGLE;
+			IORING_FEAT_RECVSEND_BUNDLE | IORING_FEAT_IOWAIT_TOGGLE |
+			IORING_FEAT_MIN_TIMEOUT;
 
 	if (copy_to_user(params, p, sizeof(*p))) {
 		ret = -EFAULT;
-- 
2.43.0

[PATCHSET v4 0/5] Add support for batched min timeout

[Jens Axboe]

Hi,

After somewhat of a lengthy delay from v2, here's v3 of this patchset.
We've now validated this actually makes sense and helps for production
workloads where there's a big gap between the peak and slow hours in
terms of load. Having a min time enables the same settings to be used
for peak hours (guaranteeing minimum response time), while avoiding
excessive context switches during lulls.

For a full description, see the v2 posting:

https://lore.kernel.org/io-uring/20240215161002.3044270-1-axboe@kernel.dk/

As before, there's a liburing branch with added test cases, it can be
found here:

https://git.kernel.dk/cgit/liburing/log/?h=min-wait

The patches are on top of for-6.12/io_uring, and with David Wei's
new iowait feat and enter flag added to avoid conflicts with that.

Changes since v3:
- Rebase on current 6.12 tree, without the NOWAIT toggle patch. Mostly
  mechanical, and FEAT flag reordering.

 include/uapi/linux/io_uring.h |   3 +-
 io_uring/io_uring.c           | 186 +++++++++++++++++++++++++---------
 io_uring/io_uring.h           |   4 +
 3 files changed, 146 insertions(+), 47 deletions(-)

-- 
Jens Axboe

[PATCH 1/5] io_uring: encapsulate extraneous wait flags into a separate struct

[Jens Axboe]

Rather than need to pass in 2 or 3 separate arguments, add a struct
to encapsulate the timeout and sigset_t parts of waiting. In preparation
for adding another argument for waiting.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/io_uring.c | 45 ++++++++++++++++++++++++---------------------
 1 file changed, 24 insertions(+), 21 deletions(-)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 20229e72b65c..37053d32c668 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -2384,13 +2384,18 @@ static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	return ret;
 }
 
+struct ext_arg {
+	size_t argsz;
+	struct __kernel_timespec __user *ts;
+	const sigset_t __user *sig;
+};
+
 /*
  * 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.
  */
 static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
-			  const sigset_t __user *sig, size_t sigsz,
-			  struct __kernel_timespec __user *uts)
+			  struct ext_arg *ext_arg)
 {
 	struct io_wait_queue iowq;
 	struct io_rings *rings = ctx->rings;
@@ -2415,10 +2420,10 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 	iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events;
 	iowq.timeout = KTIME_MAX;
 
-	if (uts) {
+	if (ext_arg->ts) {
 		struct timespec64 ts;
 
-		if (get_timespec64(&ts, uts))
+		if (get_timespec64(&ts, ext_arg->ts))
 			return -EFAULT;
 
 		iowq.timeout = timespec64_to_ktime(ts);
@@ -2426,14 +2431,14 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 			iowq.timeout = ktime_add(iowq.timeout, io_get_time(ctx));
 	}
 
-	if (sig) {
+	if (ext_arg->sig) {
 #ifdef CONFIG_COMPAT
 		if (in_compat_syscall())
-			ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
-						      sigsz);
+			ret = set_compat_user_sigmask((const compat_sigset_t __user *)ext_arg->sig,
+						      ext_arg->argsz);
 		else
 #endif
-			ret = set_user_sigmask(sig, sigsz);
+			ret = set_user_sigmask(ext_arg->sig, ext_arg->argsz);
 
 		if (ret)
 			return ret;
@@ -3112,9 +3117,8 @@ static int io_validate_ext_arg(unsigned flags, const void __user *argp, size_t a
 	return 0;
 }
 
-static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
-			  struct __kernel_timespec __user **ts,
-			  const sigset_t __user **sig)
+static int io_get_ext_arg(unsigned flags, const void __user *argp,
+			  struct ext_arg *ext_arg)
 {
 	struct io_uring_getevents_arg arg;
 
@@ -3123,8 +3127,8 @@ static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz
 	 * is just a pointer to the sigset_t.
 	 */
 	if (!(flags & IORING_ENTER_EXT_ARG)) {
-		*sig = (const sigset_t __user *) argp;
-		*ts = NULL;
+		ext_arg->sig = (const sigset_t __user *) argp;
+		ext_arg->ts = NULL;
 		return 0;
 	}
 
@@ -3132,15 +3136,15 @@ static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz
 	 * EXT_ARG is set - ensure we agree on the size of it and copy in our
 	 * timespec and sigset_t pointers if good.
 	 */
-	if (*argsz != sizeof(arg))
+	if (ext_arg->argsz != sizeof(arg))
 		return -EINVAL;
 	if (copy_from_user(&arg, argp, sizeof(arg)))
 		return -EFAULT;
 	if (arg.pad)
 		return -EINVAL;
-	*sig = u64_to_user_ptr(arg.sigmask);
-	*argsz = arg.sigmask_sz;
-	*ts = u64_to_user_ptr(arg.ts);
+	ext_arg->sig = u64_to_user_ptr(arg.sigmask);
+	ext_arg->argsz = arg.sigmask_sz;
+	ext_arg->ts = u64_to_user_ptr(arg.ts);
 	return 0;
 }
 
@@ -3246,15 +3250,14 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
 			}
 			mutex_unlock(&ctx->uring_lock);
 		} else {
-			const sigset_t __user *sig;
-			struct __kernel_timespec __user *ts;
+			struct ext_arg ext_arg = { .argsz = argsz };
 
-			ret2 = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
+			ret2 = io_get_ext_arg(flags, argp, &ext_arg);
 			if (likely(!ret2)) {
 				min_complete = min(min_complete,
 						   ctx->cq_entries);
 				ret2 = io_cqring_wait(ctx, min_complete, flags,
-						      sig, argsz, ts);
+						      &ext_arg);
 			}
 		}
 
-- 
2.43.0

[PATCHSET v5 0/5] Add support for batched min timeout

[Jens Axboe]

Hi,

Here's v5 of the min-wait patchset. For a full description, see the v2
posting:

https://lore.kernel.org/io-uring/20240215161002.3044270-1-axboe@kernel.dk/

As before, there's a liburing branch with added test cases, it can be
found here:

https://git.kernel.dk/cgit/liburing/log/?h=min-wait

The patches are on top of master with for-6.12/io_uring pulled in.

Changes since v4:
- Use READ/WRITE_ONCE consistently with iowq->hit_timeout
- Unify how io_cqring_timer_wakeup() handles ring types by always using
  wake_up_process().
- Fix race in min timer wakeup with DEFER_TASKRUN
- Don't reset ctx->cq_wait_nr if timeout has been hit

 include/uapi/linux/io_uring.h |   3 +-
 io_uring/io_uring.c           | 191 +++++++++++++++++++++++++---------
 io_uring/io_uring.h           |   4 +
 3 files changed, 150 insertions(+), 48 deletions(-)

-- 
Jens Axboe

[PATCH 1/5] io_uring: encapsulate extraneous wait flags into a separate struct

[Jens Axboe]

Rather than need to pass in 2 or 3 separate arguments, add a struct
to encapsulate the timeout and sigset_t parts of waiting. In preparation
for adding another argument for waiting.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/io_uring.c | 45 ++++++++++++++++++++++++---------------------
 1 file changed, 24 insertions(+), 21 deletions(-)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 20229e72b65c..37053d32c668 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -2384,13 +2384,18 @@ static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	return ret;
 }
 
+struct ext_arg {
+	size_t argsz;
+	struct __kernel_timespec __user *ts;
+	const sigset_t __user *sig;
+};
+
 /*
  * 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.
  */
 static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
-			  const sigset_t __user *sig, size_t sigsz,
-			  struct __kernel_timespec __user *uts)
+			  struct ext_arg *ext_arg)
 {
 	struct io_wait_queue iowq;
 	struct io_rings *rings = ctx->rings;
@@ -2415,10 +2420,10 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 	iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events;
 	iowq.timeout = KTIME_MAX;
 
-	if (uts) {
+	if (ext_arg->ts) {
 		struct timespec64 ts;
 
-		if (get_timespec64(&ts, uts))
+		if (get_timespec64(&ts, ext_arg->ts))
 			return -EFAULT;
 
 		iowq.timeout = timespec64_to_ktime(ts);
@@ -2426,14 +2431,14 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, u32 flags,
 			iowq.timeout = ktime_add(iowq.timeout, io_get_time(ctx));
 	}
 
-	if (sig) {
+	if (ext_arg->sig) {
 #ifdef CONFIG_COMPAT
 		if (in_compat_syscall())
-			ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
-						      sigsz);
+			ret = set_compat_user_sigmask((const compat_sigset_t __user *)ext_arg->sig,
+						      ext_arg->argsz);
 		else
 #endif
-			ret = set_user_sigmask(sig, sigsz);
+			ret = set_user_sigmask(ext_arg->sig, ext_arg->argsz);
 
 		if (ret)
 			return ret;
@@ -3112,9 +3117,8 @@ static int io_validate_ext_arg(unsigned flags, const void __user *argp, size_t a
 	return 0;
 }
 
-static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
-			  struct __kernel_timespec __user **ts,
-			  const sigset_t __user **sig)
+static int io_get_ext_arg(unsigned flags, const void __user *argp,
+			  struct ext_arg *ext_arg)
 {
 	struct io_uring_getevents_arg arg;
 
@@ -3123,8 +3127,8 @@ static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz
 	 * is just a pointer to the sigset_t.
 	 */
 	if (!(flags & IORING_ENTER_EXT_ARG)) {
-		*sig = (const sigset_t __user *) argp;
-		*ts = NULL;
+		ext_arg->sig = (const sigset_t __user *) argp;
+		ext_arg->ts = NULL;
 		return 0;
 	}
 
@@ -3132,15 +3136,15 @@ static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz
 	 * EXT_ARG is set - ensure we agree on the size of it and copy in our
 	 * timespec and sigset_t pointers if good.
 	 */
-	if (*argsz != sizeof(arg))
+	if (ext_arg->argsz != sizeof(arg))
 		return -EINVAL;
 	if (copy_from_user(&arg, argp, sizeof(arg)))
 		return -EFAULT;
 	if (arg.pad)
 		return -EINVAL;
-	*sig = u64_to_user_ptr(arg.sigmask);
-	*argsz = arg.sigmask_sz;
-	*ts = u64_to_user_ptr(arg.ts);
+	ext_arg->sig = u64_to_user_ptr(arg.sigmask);
+	ext_arg->argsz = arg.sigmask_sz;
+	ext_arg->ts = u64_to_user_ptr(arg.ts);
 	return 0;
 }
 
@@ -3246,15 +3250,14 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
 			}
 			mutex_unlock(&ctx->uring_lock);
 		} else {
-			const sigset_t __user *sig;
-			struct __kernel_timespec __user *ts;
+			struct ext_arg ext_arg = { .argsz = argsz };
 
-			ret2 = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
+			ret2 = io_get_ext_arg(flags, argp, &ext_arg);
 			if (likely(!ret2)) {
 				min_complete = min(min_complete,
 						   ctx->cq_entries);
 				ret2 = io_cqring_wait(ctx, min_complete, flags,
-						      sig, argsz, ts);
+						      &ext_arg);
 			}
 		}
 
-- 
2.43.0