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@xxxxxxxxx>
---
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
