From: Leonardo Bras Soares Passos <leobras@redhat.com>
To: Eric Blake <eblake@redhat.com>
Cc: "Elena Ufimtseva" <elena.ufimtseva@oracle.com>,
"John G Johnson" <john.g.johnson@oracle.com>,
"Jagannathan Raman" <jag.raman@oracle.com>,
qemu-block@nongnu.org, "Juan Quintela" <quintela@redhat.com>,
qemu-devel <qemu-devel@nongnu.org>,
"Daniel P. Berrangé" <berrange@redhat.com>,
"Peter Xu" <peterx@redhat.com>,
"Dr. David Alan Gilbert" <dgilbert@redhat.com>,
"Paolo Bonzini" <pbonzini@redhat.com>,
"Marc-André Lureau" <marcandre.lureau@redhat.com>,
"Fam Zheng" <fam@euphon.net>
Subject: Re: [PATCH v1 1/3] io: Enable write flags for QIOChannel
Date: Thu, 2 Sep 2021 05:26:04 -0300 [thread overview]
Message-ID: <CAJ6HWG6qNjvcV5CKMSGKjZTXenguiAwN4gt528oSxkv5BqrCDw@mail.gmail.com> (raw)
In-Reply-To: <20210901205405.f3dc4vpqinfy7gkx@redhat.com>
Thanks for the feedback Eric!
On Wed, Sep 1, 2021 at 5:54 PM Eric Blake <eblake@redhat.com> wrote:
>
> On Tue, Aug 31, 2021 at 08:02:37AM -0300, Leonardo Bras wrote:
> > Some syscalls used for writting, such as sendmsg(), accept flags that
> > can modify their behavior, even allowing the usage of features such as
> > MSG_ZEROCOPY.
> >
> > Change qio_channel_write*() interface to allow passing down flags,
> > allowing a more flexible use of IOChannel.
> >
> > At first, it's use is enabled for QIOChannelSocket, but can be easily
> > extended to any other QIOChannel implementation.
>
> As a followup to this patch, I wonder if we can also get performance
> improvements by implementing MSG_MORE, and using that in preference to
> corking/uncorking to better indicate that back-to-back short messages
> may behave better when grouped together over the wire. At least the
> NBD code could make use of it (going off of my experience with the
> libnbd project demonstrating a performance boost when we added
> MSG_MORE support there).
That's interesting!
We could use this patchset for testing that out, as I believe it's easy
to add those flags to the sendmsg() we want to have it enabled.
>
> >
> > Signed-off-by: Leonardo Bras <leobras@redhat.com>
> > ---
> > chardev/char-io.c | 2 +-
> > hw/remote/mpqemu-link.c | 2 +-
> > include/io/channel.h | 56 ++++++++++++++++++++---------
> > io/channel-buffer.c | 1 +
> > io/channel-command.c | 1 +
> > io/channel-file.c | 1 +
> > io/channel-socket.c | 4 ++-
> > io/channel-tls.c | 1 +
> > io/channel-websock.c | 1 +
> > io/channel.c | 53 ++++++++++++++-------------
> > migration/rdma.c | 1 +
> > scsi/pr-manager-helper.c | 2 +-
> > tests/unit/test-io-channel-socket.c | 1 +
> > 13 files changed, 81 insertions(+), 45 deletions(-)
> >
> > diff --git a/chardev/char-io.c b/chardev/char-io.c
> > index 8ced184160..4ea7b1ee2a 100644
> > --- a/chardev/char-io.c
> > +++ b/chardev/char-io.c
> > @@ -122,7 +122,7 @@ int io_channel_send_full(QIOChannel *ioc,
> >
> > ret = qio_channel_writev_full(
> > ioc, &iov, 1,
> > - fds, nfds, NULL);
> > + fds, 0, nfds, NULL);
>
> 0 before nfds here...
Good catch! I will fix that for v2!
>
> > if (ret == QIO_CHANNEL_ERR_BLOCK) {
> > if (offset) {
> > return offset;
> > diff --git a/hw/remote/mpqemu-link.c b/hw/remote/mpqemu-link.c
> > index 7e841820e5..0d13321ef0 100644
> > --- a/hw/remote/mpqemu-link.c
> > +++ b/hw/remote/mpqemu-link.c
> > @@ -69,7 +69,7 @@ bool mpqemu_msg_send(MPQemuMsg *msg, QIOChannel *ioc, Error **errp)
> > }
> >
> > if (!qio_channel_writev_full_all(ioc, send, G_N_ELEMENTS(send),
> > - fds, nfds, errp)) {
> > + fds, nfds, 0, errp)) {
>
> Thanks for fixing the broken indentation.
I kept questioning myself if I was breaking something here :)
>
> ...but after nfds here, so one is wrong; up to this point in a linear
> review, I can't tell which was intended...
>
> > +++ b/include/io/channel.h
> > @@ -104,6 +104,7 @@ struct QIOChannelClass {
> > size_t niov,
> > int *fds,
> > size_t nfds,
> > + int flags,
> > Error **errp);
>
> ...and finally I see that in general, you wanted to add the argument
> after. Looks like the change to char-io.c is buggy.
Yeap!
>
> (You can use scripts/git.orderfile as a way to force your patch to
> list the .h file first, to make it easier for linear review).
Nice tip! just added to my qemu gitconfig :)
>
> > ssize_t (*io_readv)(QIOChannel *ioc,
> > const struct iovec *iov,
> > @@ -260,6 +261,7 @@ ssize_t qio_channel_writev_full(QIOChannel *ioc,
> > size_t niov,
> > int *fds,
> > size_t nfds,
> > + int flags,
> > Error **errp);
> >
> > /**
> > @@ -325,6 +327,7 @@ int qio_channel_readv_all(QIOChannel *ioc,
> > * @ioc: the channel object
> > * @iov: the array of memory regions to write data from
> > * @niov: the length of the @iov array
> > + * @flags: optional sending flags
> > * @errp: pointer to a NULL-initialized error object
> > *
> > * Write data to the IO channel, reading it from the
> > @@ -339,10 +342,14 @@ int qio_channel_readv_all(QIOChannel *ioc,
> > *
> > * Returns: 0 if all bytes were written, or -1 on error
> > */
> > -int qio_channel_writev_all(QIOChannel *ioc,
> > - const struct iovec *iov,
> > - size_t niov,
> > - Error **erp);
> > +int qio_channel_writev_all_flags(QIOChannel *ioc,
> > + const struct iovec *iov,
> > + size_t niov,
> > + int flags,
> > + Error **errp);
>
> You changed the function name here, but not in the comment beforehand.
>
Will fix this for v2, thanks !
> > +
> > +#define qio_channel_writev_all(ioc, iov, niov, errp) \
> > + qio_channel_writev_all_flags(ioc, iov, niov, 0, errp)
>
> In most cases, you were merely adding a new function to minimize churn
> to existing callers while making the old name a macro,...
>
> > @@ -853,6 +876,7 @@ int qio_channel_writev_full_all(QIOChannel *ioc,
> > const struct iovec *iov,
> > size_t niov,
> > int *fds, size_t nfds,
> > + int flags,
> > Error **errp);
>
> ...but this one you changed in-place. Any reason? It might be nice
> to mention how you chose which functions to wrap (to minimize churn to
> existing clients) vs. modify signatures.
It's the first one I did change, TBH.
It just had a few uses. mostly in the same file scope, and a single
use on mpqemu-link.c,
so I thought it would be ok to just change it.
But yeah, it makes sense to also add a macro to this one as well, and
create another function to keep them all looking the same.
>
> >
> > #endif /* QIO_CHANNEL_H */
> > diff --git a/io/channel-buffer.c b/io/channel-buffer.c
> > index baa4e2b089..bf52011be2 100644
> > --- a/io/channel-buffer.c
> > +++ b/io/channel-buffer.c
> > @@ -81,6 +81,7 @@ static ssize_t qio_channel_buffer_writev(QIOChannel *ioc,
> > size_t niov,
> > int *fds,
> > size_t nfds,
> > + int flags,
> > Error **errp)
> > {
> > QIOChannelBuffer *bioc = QIO_CHANNEL_BUFFER(ioc);
>
> Would it be wise to check that flags only contains values we can honor
> in this (and all other) implementations of qio backends? Do we need
> some way for a backend to advertise to the core qio code which flags
> it is willing to accept?
That's a good idea, maybe we can do as you suggest below, choose a set of
features we are willing to support and then translate it depending on the
implementation. Then this would only be testing for a mask.
>
> > +++ b/io/channel-socket.c
> > @@ -525,6 +525,7 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,
> > size_t niov,
> > int *fds,
> > size_t nfds,
> > + int flags,
> > Error **errp)
> > {
> > QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc);
> > @@ -558,7 +559,7 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,
> > }
> >
> > retry:
> > - ret = sendmsg(sioc->fd, &msg, 0);
> > + ret = sendmsg(sioc->fd, &msg, flags);
>
> Because of this line, we are forcing our use of flags to be exactly
> the same set of MSG_* flags understood by sendmsg(), which feels a bit
> fragile. Wouldn't it be safer to define our own set of QIO_MSG_
> flags, and map those into whatever flag values the underlying backends
> can support? After all, one thing I learned on libnbd is that
> MSG_MORE is not universally portable, but the goal of qio is to
> abstract away things so that the rest of the code doesn't have to do
> #ifdef tests everywhere, but instead let the qio code deal with it
> (whether to ignore an unsupported flag because it is only an
> optimization hint, or to return an error because we depend on the
> behavior change the flag would cause if supported, or...). And that
> goes back to my question of whether backends should have a way to
> inform the qio core which flags they can support.
I think you are correct and having our own QIO_MSG_* would make
sense here. This could allow us to filter incorrect flags easily, and also
have well documented what each implementation supports, by their own
masks.
Thanks!
Leonardo Bras
>
> --
> Eric Blake, Principal Software Engineer
> Red Hat, Inc. +1-919-301-3266
> Virtualization: qemu.org | libvirt.org
>
next prev parent reply other threads:[~2021-09-02 8:27 UTC|newest]
Thread overview: 62+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-08-31 11:02 [PATCH v1 0/3] QIOChannel flags + multifd zerocopy Leonardo Bras
2021-08-31 11:02 ` [PATCH v1 1/3] io: Enable write flags for QIOChannel Leonardo Bras
2021-09-01 20:54 ` Eric Blake
2021-09-02 8:26 ` Leonardo Bras Soares Passos [this message]
2021-08-31 11:02 ` [PATCH v1 2/3] io: Add zerocopy and errqueue Leonardo Bras
2021-08-31 12:57 ` Daniel P. Berrangé
2021-08-31 20:27 ` Peter Xu
2021-09-01 8:50 ` Daniel P. Berrangé
2021-09-01 15:52 ` Peter Xu
2021-09-01 15:59 ` Daniel P. Berrangé
2021-09-02 7:07 ` Leonardo Bras Soares Passos
2021-09-02 6:59 ` Leonardo Bras Soares Passos
2021-09-07 16:44 ` Peter Xu
2021-09-08 20:13 ` Leonardo Bras Soares Passos
2021-09-08 21:04 ` Peter Xu
2021-09-02 6:38 ` Leonardo Bras Soares Passos
2021-09-02 8:47 ` Daniel P. Berrangé
2021-09-02 9:34 ` Leonardo Bras Soares Passos
2021-09-02 9:49 ` Daniel P. Berrangé
2021-09-02 10:19 ` Leonardo Bras Soares Passos
2021-09-02 10:28 ` Daniel P. Berrangé
2021-09-07 11:06 ` Dr. David Alan Gilbert
2021-09-07 18:09 ` Peter Xu
2021-09-08 8:30 ` Dr. David Alan Gilbert
2021-09-08 15:24 ` Peter Xu
2021-09-09 8:49 ` Dr. David Alan Gilbert
2021-09-08 20:25 ` Leonardo Bras Soares Passos
2021-09-08 21:09 ` Peter Xu
2021-09-08 21:57 ` Daniel P. Berrangé
2021-09-09 2:05 ` Peter Xu
2021-09-09 4:58 ` Leonardo Bras Soares Passos
2021-09-09 16:40 ` Peter Xu
2021-08-31 11:02 ` [PATCH v1 3/3] migration: multifd: Enable zerocopy Leonardo Bras
2021-08-31 13:16 ` Daniel P. Berrangé
2021-08-31 20:29 ` Peter Xu
2021-09-01 8:53 ` Daniel P. Berrangé
2021-09-01 15:35 ` Peter Xu
2021-09-01 15:44 ` Daniel P. Berrangé
2021-09-01 16:01 ` Peter Xu
2021-09-02 7:57 ` Leonardo Bras Soares Passos
2021-09-07 11:13 ` Dr. David Alan Gilbert
2021-09-08 15:26 ` Daniel P. Berrangé
2021-09-02 7:23 ` Jason Wang
2021-09-02 8:08 ` Leonardo Bras Soares Passos
2021-09-02 7:27 ` Leonardo Bras Soares Passos
2021-09-02 7:22 ` Leonardo Bras Soares Passos
2021-09-02 8:20 ` Daniel P. Berrangé
2021-09-02 8:52 ` Leonardo Bras Soares Passos
2021-09-02 9:20 ` Daniel P. Berrangé
2021-09-02 9:49 ` Leonardo Bras Soares Passos
2021-09-02 9:59 ` Daniel P. Berrangé
2021-09-02 10:25 ` Leonardo Bras Soares Passos
2021-09-07 11:17 ` Dr. David Alan Gilbert
2021-09-07 18:32 ` Peter Xu
2021-09-08 2:59 ` Jason Wang
2021-09-08 3:24 ` Peter Xu
2021-09-08 3:26 ` Jason Wang
2021-09-08 8:19 ` Dr. David Alan Gilbert
2021-09-08 15:19 ` Peter Xu
2021-09-09 1:10 ` Jason Wang
2021-08-31 21:24 ` [PATCH v1 0/3] QIOChannel flags + multifd zerocopy Peter Xu
2021-09-01 19:21 ` Leonardo Bras Soares Passos
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 \
--in-reply-to=CAJ6HWG6qNjvcV5CKMSGKjZTXenguiAwN4gt528oSxkv5BqrCDw@mail.gmail.com \
--to=leobras@redhat.com \
--cc=berrange@redhat.com \
--cc=dgilbert@redhat.com \
--cc=eblake@redhat.com \
--cc=elena.ufimtseva@oracle.com \
--cc=fam@euphon.net \
--cc=jag.raman@oracle.com \
--cc=john.g.johnson@oracle.com \
--cc=marcandre.lureau@redhat.com \
--cc=pbonzini@redhat.com \
--cc=peterx@redhat.com \
--cc=qemu-block@nongnu.org \
--cc=qemu-devel@nongnu.org \
--cc=quintela@redhat.com \
/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;
as well as URLs for NNTP newsgroup(s).