2020-02-07 12:19:46

by Stefano Garzarella

Subject: [PATCH] io_uring: flush overflowed CQ events in the io_uring_poll()

In io_uring_poll() we must flush overflowed CQ events before to
check if there are CQ events available, to avoid missing events.

We call the io_cqring_events() that checks and flushes any overflow
and returns the number of CQ events available.

We can avoid taking the 'uring_lock' since the flush is already
protected by 'completion_lock'.

Signed-off-by: Stefano Garzarella <[email protected]>
---
fs/io_uring.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 77f22c3da30f..02e77e86abaf 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -6301,7 +6301,7 @@ static __poll_t io_uring_poll(struct file *file, poll_table *wait)
if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head !=
ctx->rings->sq_ring_entries)
mask |= EPOLLOUT | EPOLLWRNORM;
- if (READ_ONCE(ctx->rings->cq.head) != ctx->cached_cq_tail)
+ if (io_cqring_events(ctx, false))
mask |= EPOLLIN | EPOLLRDNORM;

return mask;
--
2.24.1

2020-02-07 16:15:18

by Jens Axboe

Subject: Re: [PATCH] io_uring: flush overflowed CQ events in the io_uring_poll()

On 2/7/20 5:18 AM, Stefano Garzarella wrote:
> In io_uring_poll() we must flush overflowed CQ events before to
> check if there are CQ events available, to avoid missing events.
>
> We call the io_cqring_events() that checks and flushes any overflow
> and returns the number of CQ events available.
>
> We can avoid taking the 'uring_lock' since the flush is already
> protected by 'completion_lock'.

Thanks, applied. I dropped that last sentence, as a) it doesn't
really matter, and b) we may very well already have it held here
if someone is doing a poll on the io_uring fd itself.

--
Jens Axboe

2020-02-07 16:40:12

by Stefano Garzarella

Subject: Re: [PATCH] io_uring: flush overflowed CQ events in the io_uring_poll()

On Fri, Feb 07, 2020 at 09:12:39AM -0700, Jens Axboe wrote:
> On 2/7/20 5:18 AM, Stefano Garzarella wrote:
> > In io_uring_poll() we must flush overflowed CQ events before to
> > check if there are CQ events available, to avoid missing events.
> >
> > We call the io_cqring_events() that checks and flushes any overflow
> > and returns the number of CQ events available.
> >
> > We can avoid taking the 'uring_lock' since the flush is already
> > protected by 'completion_lock'.
>
> Thanks, applied. I dropped that last sentence, as a) it doesn't
> really matter, and b) we may very well already have it held here
> if someone is doing a poll on the io_uring fd itself.

Sure, indeed I was undecided whether to put it after the three dashes
as a response to your yesterday's request.

Thanks,
Stefano