On 11/4/20 11:32 AM, Jens Axboe wrote:
> On 11/4/20 10:50 AM, Jens Axboe wrote:
>> +struct io_uring_getevents_arg {
>> + sigset_t *sigmask;
>> + struct __kernel_timespec *ts;
>> +};
>> +
>
> I missed that this is still not right, I did bring it up in your last
> posting though - you can't have pointers as a user API, since the size
> of the pointer will vary depending on whether this is a 32-bit or 64-bit
> arch (or 32-bit app running on 64-bit kernel).
You also made the sigmask size go away if we're using getevent_arg, we
need to include that. It'll break right now if you give both the sigmask
and a timeout, as you're passing in the total arg size for 'ts'.
Here's my (hopefully) final fixed version:
commit 1fda0f709ac2a51c7baa9899501dbf08883fa92c
Author: Hao Xu <haoxu@xxxxxxxxxxxxxxxxx>
Date: Tue Nov 3 10:54:37 2020 +0800
io_uring: add timeout support for io_uring_enter()
Now users who want to get woken when waiting for events should submit a
timeout command first. It is not safe for applications that split SQ and
CQ handling between two threads, such as mysql. Users should synchronize
the two threads explicitly to protect SQ and that will impact the
performance.
This patch adds support for timeout to existing io_uring_enter(). To
avoid overloading arguments, it introduces a new parameter structure
which contains sigmask and timeout.
I have tested the workloads with one thread submiting nop requests
while the other reaping the cqe with timeout. It shows 1.8~2x faster
when the iodepth is 16.
Signed-off-by: Jiufei Xue <jiufei.xue@xxxxxxxxxxxxxxxxx>
Signed-off-by: Hao Xu <haoxu@xxxxxxxxxxxxxxxxx>
Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 864751d64097..8439cda54e21 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -7110,7 +7110,8 @@ static int io_run_task_work_sig(void)
* 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,
- const sigset_t __user *sig, size_t sigsz)
+ const sigset_t __user *sig, size_t sigsz,
+ struct __kernel_timespec __user *uts)
{
struct io_wait_queue iowq = {
.wq = {
@@ -7122,6 +7123,8 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
.to_wait = min_events,
};
struct io_rings *rings = ctx->rings;
+ struct timespec64 ts;
+ signed long timeout = 0;
int ret = 0;
do {
@@ -7144,6 +7147,12 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
return ret;
}
+ if (uts) {
+ if (get_timespec64(&ts, uts))
+ return -EFAULT;
+ timeout = timespec64_to_jiffies(&ts);
+ }
+
iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
trace_io_uring_cqring_wait(ctx, min_events);
do {
@@ -7157,7 +7166,15 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
break;
if (io_should_wake(&iowq, false))
break;
- schedule();
+ if (uts) {
+ timeout = schedule_timeout(timeout);
+ if (timeout == 0) {
+ ret = -ETIME;
+ break;
+ }
+ } else {
+ schedule();
+ }
} while (1);
finish_wait(&ctx->wait, &iowq.wq);
@@ -9130,20 +9147,39 @@ static void io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
}
SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
- u32, min_complete, u32, flags, const sigset_t __user *, sig,
+ u32, min_complete, u32, flags, const void __user *, argp,
size_t, sigsz)
{
struct io_ring_ctx *ctx;
long ret = -EBADF;
int submitted = 0;
struct fd f;
+ const sigset_t __user *sig;
+ struct __kernel_timespec __user *ts;
+ struct io_uring_getevents_arg arg;
io_run_task_work();
if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
- IORING_ENTER_SQ_WAIT))
+ IORING_ENTER_SQ_WAIT | IORING_ENTER_GETEVENTS_TIMEOUT))
return -EINVAL;
+ /* deal with IORING_ENTER_GETEVENTS_TIMEOUT */
+ if (flags & IORING_ENTER_GETEVENTS_TIMEOUT) {
+ if (!(flags & IORING_ENTER_GETEVENTS))
+ return -EINVAL;
+ if (sigsz != sizeof(arg))
+ return -EINVAL;
+ if (copy_from_user(&arg, argp, sizeof(arg)))
+ return -EFAULT;
+ sig = u64_to_user_ptr(arg.sigmask);
+ sigsz = arg.sigmask_sz;
+ ts = u64_to_user_ptr(arg.ts);
+ } else {
+ sig = (const sigset_t __user *)argp;
+ ts = NULL;
+ }
+
f = fdget(fd);
if (!f.file)
return -EBADF;
@@ -9199,7 +9235,7 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
!(ctx->flags & IORING_SETUP_SQPOLL)) {
ret = io_iopoll_check(ctx, min_complete);
} else {
- ret = io_cqring_wait(ctx, min_complete, sig, sigsz);
+ ret = io_cqring_wait(ctx, min_complete, sig, sigsz, ts);
}
}
@@ -9561,7 +9597,8 @@ static int io_uring_create(unsigned entries, struct io_uring_params *p,
p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
- IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED;
+ IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
+ IORING_FEAT_GETEVENTS_TIMEOUT;
if (copy_to_user(params, p, sizeof(*p))) {
ret = -EFAULT;
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 557e7eae497f..1a92985a9ee8 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -231,6 +231,7 @@ struct io_cqring_offsets {
#define IORING_ENTER_GETEVENTS (1U << 0)
#define IORING_ENTER_SQ_WAKEUP (1U << 1)
#define IORING_ENTER_SQ_WAIT (1U << 2)
+#define IORING_ENTER_GETEVENTS_TIMEOUT (1U << 3)
/*
* Passed in for io_uring_setup(2). Copied back with updated info on success
@@ -259,6 +260,7 @@ struct io_uring_params {
#define IORING_FEAT_FAST_POLL (1U << 5)
#define IORING_FEAT_POLL_32BITS (1U << 6)
#define IORING_FEAT_SQPOLL_NONFIXED (1U << 7)
+#define IORING_FEAT_GETEVENTS_TIMEOUT (1U << 8)
/*
* io_uring_register(2) opcodes and arguments
@@ -335,4 +337,11 @@ enum {
IORING_RESTRICTION_LAST
};
+struct io_uring_getevents_arg {
+ __u64 sigmask;
+ __u32 sigmask_sz;
+ __u32 pad;
+ __u64 ts;
+};
+
#endif
--
Jens Axboe
