Dmitry Antipov <dmantipov@xxxxxxxxx> writes:
> In 'io_cqring_schedule_timeout()', do not assume that 'ktime_t' is
> equal to nanoseconds and prefer 'ktime_add()' over 'ktime_add_ns()'
> to sum two 'ktime_t' values. Compile tested only.
>
> Fixes: 1100c4a2656d ("io_uring: add support for batch wait timeout")
> Signed-off-by: Dmitry Antipov <dmantipov@xxxxxxxxx>
> ---
> io_uring/io_uring.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
> index ceacf6230e34..7f2500aca95c 100644
> --- a/io_uring/io_uring.c
> +++ b/io_uring/io_uring.c
> @@ -2434,7 +2434,7 @@ static int io_cqring_schedule_timeout(struct io_wait_queue *iowq,
> ktime_t timeout;
>
> if (iowq->min_timeout) {
> - timeout = ktime_add_ns(iowq->min_timeout, start_time);
> + timeout = ktime_add(iowq->min_timeout, start_time);
I don't think this solves the issue stated in the commit message. Look
at where the min_timeout comes from, in io_get_ext_arg:
ext_arg->min_time = READ_ONCE(w->min_wait_usec) * NSEC_PER_USEC;
Perhaps that should be:
ext_arg->min_time = us_to_ktime(READ_ONCE(w->min_wait_usec));
I also don't know whether this warrants a fixes tag, given it doesn't
change any behavior.
Cheers,
Jeff
