[PATCH v3] io_uring: enable toggle of iowait usage

[Jens Axboe <axboe@kernel.dk>]

By default, io_uring marks a waiting task as being in iowait, if it's
sleeping waiting on events and there are pending requests. This isn't
necessarily always useful, and may be confusing on non-storage setups
where iowait isn't expected. It can also cause extra power usage, by
preventing the CPU from entering lower sleep states.

Add a sysctl knob to control this, /proc/sys/kernel/io_uring_iowait. It
defaults to '1' which is the current behavior, and can be set to 0 if
iowait accounting and boosting isn't deemed suitable on that system.

Implemented as an int proc variable rather than a bool, in case there's
a need to expand this in the future to distinguish between iowait
accounting and cpufreq boosting. Bool proc entries do allow > 1 values
without erroring, let's retain those for when we may actually use them.

In the future, enter flags may be added to control this as well. For
now, a system-wide knob is enough.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

---

Changes since v2:
- Make it a sysctl knob instead, leaving per-ring enter flags as a
  future kind of thing.

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 58003fa6b327..2866ab55a739 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -156,9 +156,10 @@ static struct workqueue_struct *iou_wq __ro_after_init;
 
 static int __read_mostly sysctl_io_uring_disabled;
 static int __read_mostly sysctl_io_uring_group = -1;
+static int __read_mostly sysctl_io_uring_iowait = 1;
 
 #ifdef CONFIG_SYSCTL
-static const struct ctl_table kernel_io_uring_disabled_table[] = {
+static const struct ctl_table kernel_io_uring_sysctl_table[] = {
 	{
 		.procname	= "io_uring_disabled",
 		.data		= &sysctl_io_uring_disabled,
@@ -175,6 +176,15 @@ static const struct ctl_table kernel_io_uring_disabled_table[] = {
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec,
 	},
+	{
+		.procname	= "io_uring_iowait",
+		.data		= &sysctl_io_uring_iowait,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE,
+	},
 };
 #endif
 
@@ -2496,7 +2506,7 @@ static int __io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	 * can take into account that the task is waiting for IO - turns out
 	 * to be important for low QD IO.
 	 */
-	if (current_pending_io())
+	if (sysctl_io_uring_iowait && current_pending_io())
 		current->in_iowait = 1;
 	if (iowq->timeout != KTIME_MAX || iowq->min_timeout)
 		ret = io_cqring_schedule_timeout(iowq, ctx->clockid, start_time);
@@ -3959,7 +3969,7 @@ static int __init io_uring_init(void)
 	BUG_ON(!iou_wq);
 
 #ifdef CONFIG_SYSCTL
-	register_sysctl_init("kernel", kernel_io_uring_disabled_table);
+	register_sysctl_init("kernel", kernel_io_uring_sysctl_table);
 #endif
 
 	return 0;

-- 
Jens Axboe