2014-02-28 10:36:16

by Gu Zheng

Subject: [PATCH 2/2] aio: make aio_read_events_ring be aware of aio_complete

Commit 5ffac122dbda8(aio: Don't use ctx->tail unnecessarily) uses
ring->tail rather than the ctx->tail, but with this change, we fetch 'tail'
only once at the start, so that we can not be aware of adding event by aio_complete
when reading events. It seems a regression.
So here we fetch the ring->tail in start of the loop each time to make it be
aware of adding event from aio_complete.

Signed-off-by: Gu Zheng <[email protected]>
---
fs/aio.c | 6 +++++-
1 files changed, 5 insertions(+), 1 deletions(-)

diff --git a/fs/aio.c b/fs/aio.c
index 7eaa631..f5b8551 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -1029,10 +1029,14 @@ static long aio_read_events_ring(struct kioctx *ctx,
struct io_event *ev;
struct page *page;

- avail = (head <= tail ? tail : ctx->nr_events) - head;
+ ring = kmap_atomic(ctx->ring_pages[0]);
+ tail = ring->tail;
+ kunmap_atomic(ring);
+
if (head == tail)
break;

+ avail = (head <= tail ? tail : ctx->nr_events) - head;
avail = min(avail, nr - ret);
avail = min_t(long, avail, AIO_EVENTS_PER_PAGE -
((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE));
--
1.7.7

2014-02-28 20:50:39

by Kent Overstreet

Subject: Re: [PATCH 2/2] aio: make aio_read_events_ring be aware of aio_complete

On Fri, Feb 28, 2014 at 06:27:18PM +0800, Gu Zheng wrote:
> Commit 5ffac122dbda8(aio: Don't use ctx->tail unnecessarily) uses
> ring->tail rather than the ctx->tail, but with this change, we fetch 'tail'
> only once at the start, so that we can not be aware of adding event by aio_complete
> when reading events. It seems a regression.

"Seems a regression" is not good enough. What breaks?

> So here we fetch the ring->tail in start of the loop each time to make it be
> aware of adding event from aio_complete.
>
> Signed-off-by: Gu Zheng <[email protected]>
> ---
> fs/aio.c | 6 +++++-
> 1 files changed, 5 insertions(+), 1 deletions(-)
>
> diff --git a/fs/aio.c b/fs/aio.c
> index 7eaa631..f5b8551 100644
> --- a/fs/aio.c
> +++ b/fs/aio.c
> @@ -1029,10 +1029,14 @@ static long aio_read_events_ring(struct kioctx *ctx,
> struct io_event *ev;
> struct page *page;
>
> - avail = (head <= tail ? tail : ctx->nr_events) - head;
> + ring = kmap_atomic(ctx->ring_pages[0]);
> + tail = ring->tail;
> + kunmap_atomic(ring);
> +
> if (head == tail)
> break;
>
> + avail = (head <= tail ? tail : ctx->nr_events) - head;
> avail = min(avail, nr - ret);
> avail = min_t(long, avail, AIO_EVENTS_PER_PAGE -
> ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE));
> --
> 1.7.7
>