From: Zorro Lang <[email protected]>
To: Brian Foster <[email protected]>
Cc: [email protected], [email protected]
Subject: Re: [PATCH v4 5/5] fsx: add IO_URING test
Date: Wed, 9 Sep 2020 11:58:07 +0800 [thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <20200908183625.GE737175@bfoster>
On Tue, Sep 08, 2020 at 02:36:25PM -0400, Brian Foster wrote:
> On Mon, Sep 07, 2020 at 01:55:13AM +0800, Zorro Lang wrote:
> > New IO_URING test for fsx, use -U option to enable IO_URING test.
> >
> > Signed-off-by: Zorro Lang <[email protected]>
> > ---
>
> Just a couple nits...
>
> > ltp/fsx.c | 134 ++++++++++++++++++++++++++++++++++++++++++++++++++++--
> > 1 file changed, 131 insertions(+), 3 deletions(-)
> >
> > diff --git a/ltp/fsx.c b/ltp/fsx.c
> > index 92f506ba..e7f23d15 100644
> > --- a/ltp/fsx.c
> > +++ b/ltp/fsx.c
> ...
> > @@ -2429,6 +2436,113 @@ aio_rw(int rw, int fd, char *buf, unsigned len, unsigned offset)
> ...
> > +int
> > +uring_rw(int rw, int fd, char *buf, unsigned len, unsigned offset)
> > +{
> > + struct io_uring_sqe *sqe;
> > + struct io_uring_cqe *cqe;
> > + struct iovec iovec;
> > + int ret;
> > + int res, res2 = 0;
> > + char *p = buf;
> > + unsigned l = len;
> > + unsigned o = offset;
>
> It looks a little odd that some variable names are aligned with tabs
> while others use a single space.. I'm not sure we care one way or
> another for xfstests, but perhaps use one approach or the other..?
>
> > +
> > + /*
> > + * Due to io_uring tries non-blocking IOs (especially read), that
> > + * always cause 'normal' short reading. To avoid this short read
> > + * fail, try to loop read/write (escpecilly read) data.
> > + */
> > + while (l > 0) {
> > + sqe = io_uring_get_sqe(&ring);
> > + if (!sqe) {
> > + fprintf(stderr, "uring_rw: io_uring_get_sqe failed: %s\n",
> > + strerror(errno));
> > + return -1;
> > + }
> > +
> > + iovec.iov_base = p;
> > + iovec.iov_len = l;
> > + if (rw == READ) {
> > + io_uring_prep_readv(sqe, fd, &iovec, 1, o);
> > + } else {
> > + io_uring_prep_writev(sqe, fd, &iovec, 1, o);
> > + }
> > +
> > + ret = io_uring_submit_and_wait(&ring, 1);
> > + if (ret != 1) {
> > + fprintf(stderr, "errcode=%d\n", -ret);
> > + fprintf(stderr, "uring %s: io_uring_submit failed: %s\n",
> > + rw == READ ? "read":"write", strerror(-ret));
> > + goto uring_error;
> > + }
> > +
> > + ret = io_uring_wait_cqe(&ring, &cqe);
> > + if (ret != 0) {
> > + fprintf(stderr, "errcode=%d\n", -ret);
> > + fprintf(stderr, "uring %s: io_uring_wait_cqe failed: %s\n",
> > + rw == READ ? "read":"write", strerror(-ret));
> > + goto uring_error;
> > + }
> > +
> > + res = cqe->res;
>
> If we assign ret here instead of res, it looks like we could optimize
> away res entirely.
I think you're right, I'll think more about that, then send V5 patches out.
As you've ACKed this patchset, I'll add "Reviewed-by Brian" in V5, many
thanks for your review :)
>
> Nits and my limited experience with uring aside, the patch otherwise
> LGTM. Are you planning any new tests to take advantage of this,
> particularly since it looks like -A (-U disabled) is the default? In any
> event, a quick test run of fsstress/fsx doesn't seem to explode:
Yes, I'm preparing more patches to use the -U option of fsx. I'll send them
out after this patchset get merged :)
>
> Reviewed-by: Brian Foster <[email protected]>
>
> > + io_uring_cqe_seen(&ring, cqe);
> > +
> > + if (res > 0) {
> > + o += res;
> > + l -= res;
> > + p += res;
> > + res2 += res;
> > + } else if (res < 0) {
> > + ret = res;
> > + fprintf(stderr, "errcode=%d\n", -ret);
> > + fprintf(stderr, "uring %s: io_uring failed: %s\n",
> > + rw == READ ? "read":"write", strerror(-ret));
> > + goto uring_error;
> > + } else {
> > + fprintf(stderr, "uring %s bad io length: %d instead of %u\n",
> > + rw == READ ? "read":"write", res2, len);
> > + break;
> > + }
> > + }
> > + return res2;
> > +
> > + uring_error:
> > + /*
> > + * The caller expects error return in traditional libc
> > + * convention, i.e. -1 and the errno set to error.
> > + */
> > + errno = -ret;
> > + return -1;
> > +}
> > +#else
> > +int
> > +uring_rw(int rw, int fd, char *buf, unsigned len, unsigned offset)
> > +{
> > + fprintf(stderr, "io_rw: need IO_URING support!\n");
> > + exit(111);
> > +}
> > +#endif
> > +
> > int
> > fsx_rw(int rw, int fd, char *buf, unsigned len, unsigned offset)
> > {
> > @@ -2436,6 +2550,8 @@ fsx_rw(int rw, int fd, char *buf, unsigned len, unsigned offset)
> >
> > if (aio) {
> > ret = aio_rw(rw, fd, buf, len, offset);
> > + } else if (uring) {
> > + ret = uring_rw(rw, fd, buf, len, offset);
> > } else {
> > if (rw == READ)
> > ret = read(fd, buf, len);
> > @@ -2498,7 +2614,7 @@ main(int argc, char **argv)
> > setvbuf(stdout, (char *)0, _IOLBF, 0); /* line buffered stdout */
> >
> > while ((ch = getopt_long(argc, argv,
> > - "b:c:dfg:i:j:kl:m:no:p:qr:s:t:w:xyABD:EFJKHzCILN:OP:RS:WXZ",
> > + "b:c:dfg:i:j:kl:m:no:p:qr:s:t:w:xyABD:EFJKHzCILN:OP:RS:UWXZ",
> > longopts, NULL)) != EOF)
> > switch (ch) {
> > case 'b':
> > @@ -2606,6 +2722,9 @@ main(int argc, char **argv)
> > case 'A':
> > aio = 1;
> > break;
> > + case 'U':
> > + uring = 1;
> > + break;
> > case 'D':
> > debugstart = getnum(optarg, &endp);
> > if (debugstart < 1)
> > @@ -2696,6 +2815,11 @@ main(int argc, char **argv)
> > if (argc != 1)
> > usage();
> >
> > + if (aio && uring) {
> > + fprintf(stderr, "-A and -U shouldn't be used together\n");
> > + usage();
> > + }
> > +
> > if (integrity && !dirpath) {
> > fprintf(stderr, "option -i <logdev> requires -P <dirpath>\n");
> > usage();
> > @@ -2786,6 +2910,10 @@ main(int argc, char **argv)
> > if (aio)
> > aio_setup();
> > #endif
> > +#ifdef URING
> > + if (uring)
> > + uring_setup();
> > +#endif
> >
> > if (!(o_flags & O_TRUNC)) {
> > off_t ret;
> > --
> > 2.20.1
> >
>
prev parent reply other threads:[~2020-09-09 3:44 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-09-06 17:55 [PATCH v4 0/5] fsstress,fsx: add io_uring test and do some fix Zorro Lang
2020-09-06 17:55 ` [PATCH v4 1/5] fsstress: add IO_URING read and write operations Zorro Lang
2020-09-08 18:34 ` Brian Foster
2020-09-06 17:55 ` [PATCH v4 2/5] fsstress: reduce the number of events when io_setup Zorro Lang
2020-09-08 18:34 ` Brian Foster
2020-09-06 17:55 ` [PATCH v4 3/5] fsstress: fix memory leak in do_aio_rw Zorro Lang
2020-09-08 18:35 ` Brian Foster
2020-09-06 17:55 ` [PATCH v4 4/5] fsx: introduce fsx_rw to combine aio_rw with general read and write Zorro Lang
2020-09-08 18:35 ` Brian Foster
2020-09-06 17:55 ` [PATCH v4 5/5] fsx: add IO_URING test Zorro Lang
2020-09-08 18:36 ` Brian Foster
2020-09-09 3:58 ` Zorro Lang [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox