Re: [PATCH 4/4] aio: use iovec array rather than the single one

From: Jeff Moyer
Date: Tue Jul 22 2014 - 11:21:06 EST


Gu Zheng <guz.fnst@xxxxxxxxxxxxxx> writes:

> use an iovec array rather than the single one, so that we can avoid
> to alloc more iovecs buffer in small(< 8) PREADV/PWRITEV cases.

I did some basic functional testing of this change and the change in
patch 1/4. That testing included using aio-stress to drive queue depths
of 7, 8 and 9, and verify that it didn't fall over. I also ran xfstests
'./check -g aio', and libaio's 'make partcheck'.

The change looks good to me, and passed testing, so:

Reviewed-by: Jeff Moyer <jmoyer@xxxxxxxxxx>

However, I still would like some comment on the reasoning behind it, and
whether there is some measurable performance advantage for some
workload. Additionally, it would be nice if that comment made its way
into the commit message.

Cheers,
Jeff

>
> Signed-off-by: Gu Zheng <guz.fnst@xxxxxxxxxxxxxx>
> ---
> fs/aio.c | 10 +++++-----
> 1 files changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/fs/aio.c b/fs/aio.c
> index 0cd0479..ef21efe 100644
> --- a/fs/aio.c
> +++ b/fs/aio.c
> @@ -1260,12 +1260,12 @@ static ssize_t aio_setup_vectored_rw(struct kiocb *kiocb,
> if (compat)
> ret = compat_rw_copy_check_uvector(rw,
> (struct compat_iovec __user *)buf,
> - *nr_segs, 1, *iovec, iovec);
> + *nr_segs, UIO_FASTIOV, *iovec, iovec);
> else
> #endif
> ret = rw_copy_check_uvector(rw,
> (struct iovec __user *)buf,
> - *nr_segs, 1, *iovec, iovec);
> + *nr_segs, UIO_FASTIOV, *iovec, iovec);
> if (ret < 0)
> return ret;
>
> @@ -1302,7 +1302,7 @@ static ssize_t aio_run_iocb(struct kiocb *req, unsigned opcode,
> fmode_t mode;
> aio_rw_op *rw_op;
> rw_iter_op *iter_op;
> - struct iovec inline_vec, *iovec = &inline_vec;
> + struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
> struct iov_iter iter;
>
> switch (opcode) {
> @@ -1337,7 +1337,7 @@ rw_common:
> if (!ret)
> ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes);
> if (ret < 0) {
> - if (iovec != &inline_vec)
> + if (iovec != inline_vecs)
> kfree(iovec);
> return ret;
> }
> @@ -1384,7 +1384,7 @@ rw_common:
> return -EINVAL;
> }
>
> - if (iovec != &inline_vec)
> + if (iovec != inline_vecs)
> kfree(iovec);
>
> if (ret != -EIOCBQUEUED) {
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/