On 04.11.25 14:48, Daniel P. Berrangé wrote:
On Tue, Nov 04, 2025 at 02:44:49PM +0300, Vladimir Sementsov-Ogievskiy wrote:
On 04.11.25 14:41, Daniel P. Berrangé wrote:
On Tue, Nov 04, 2025 at 03:35:17PM +0400, Marc-André Lureau wrote:
Hi Vladimir
On Tue, Nov 4, 2025 at 2:17 PM Vladimir Sementsov-Ogievskiy <
[email protected]> wrote:
Hi all!
That's a preparation for backend-transfer migration of
vhost-user-blk, and introduced DEFINE_PROP_CHR_NO_CONNECT()
is unused now.
v3 of "vhost-user-blk: live-backend local migration" is coming
soon, and will be based on this series (and will use
DEFINE_PROP_CHR_NO_CONNECT of course).
changes in v6:
05: move connect() call into "if (s)"
07: - drop assertion
- improve doc comment, to cover @s==NULL relations with @connect
- add r-b by Marc-André
Vladimir Sementsov-Ogievskiy (7):
chardev/char-socket: simplify reconnect-ms handling
chardev/char: split chardev_init_common() out of qemu_char_open()
chardev/char: qemu_char_open(): add return value
chardev/char: move filename and be_opened handling to qemu_char_open()
chardev/char: introduce .init() + .connect() initialization interface
chardev/char-socket: move to .init + .connect api
chardev: introduce DEFINE_PROP_CHR_NO_CONNECT
Do you need this series in 10.2? We are at soft-freeze today, this is
closer to a feature than a simple refactoring, we may just wait for the
next dev phase.
Back in v2, I had a request to convert the other chardev backends
to the new model too, as IMHO it is undesirable to introduce the
technical debt by only touching 1 backend:
https://lists.nongnu.org/archive/html/qemu-devel/2025-10/msg03272.html
Right. I remember it, and have a draft converting series. It turned out to be
more than expected,
about 24 commits, and personally I'm not sure, that it worth doing.. I'll send
an RFC too look at.
Yep, I'd be interested to see what it looks like, even if it is not
finished / not functional.
Hmm. Looking through my draft, and in parallel, moving char-socket vmsd
into correct layer (into char-socket), I now see the following:
1. splitting open() logic into init() and connect() make sense only
if it done for and together with implementing vmsd for that chardev
(as we do the split exactly in a manner which makes implementing of
migration possible)
2. implementing backend-transfer for all chardevs is too much for me.
So, seems, correct way is to allow chardevs have only .init, without
.connect when no support for migration.
And, converting all chardevs would be simply rename .open() to .init().
And I'll keep .open() semantics for .init() (i.e. additional handling
.filename and CHR_EVENT_OPENED for chardevs that don't care of it)
So, I can do the rename as first patch of v7 of this series, and we'll
keep common interaface for all chadevs (except for that most of them
will not support migration, and therefore don't implement .connect()
and .vmsd)
--
Best regards,
Vladimir
