Re: [PATCH v2] block: Factor out bdrv_run_co()

[Vladimir Sementsov-Ogievskiy]

20.05.2020 17:49, Kevin Wolf wrote:
> Am 20.05.2020 um 16:05 hat Kevin Wolf geschrieben:
> > Am 19.05.2020 um 19:56 hat Vladimir Sementsov-Ogievskiy geschrieben:
> > > We have a few bdrv_*() functions that can either spawn a new coroutine
> > > and wait for it with BDRV_POLL_WHILE() or use a fastpath if they are
> > > alreeady running in a coroutine. All of them duplicate basically the
> > > same code.
> > >
> > > Factor the common code into a new function bdrv_run_co().
> > >
> > > Signed-off-by: Kevin Wolf <address@hidden>
> > > Signed-off-by: Vladimir Sementsov-Ogievskiy <address@hidden>
> > >     [Factor out bdrv_run_co_entry too]
> > > ---
> > >
> > > Hi!
> > >
> > > I'm a bit lost on rebasing "block/io: safer inc/dec in_flight sections"
> > > (is it needed or not?), so, I decided to send just this one patch:
> > >
> > > I suggest to go a bit further, and refactor that bdrv_run_co don't need
> > > additional *ret argument neither NOT_DONE logic.
> >
> > Hm, this approach adds another indirection and bdrv_pread/pwrite still
> > seems to be on some hot paths. But maybe this is just the right
> > motivation to clean up qcow2 a bit and use explicit bdrv_co_*() where it
> > is possible. I might take a look later.
>
> Still not easily possible it seems. We can add a few coroutine_fn
> markers here and there (and probably should do that), but the
> interesting I/O is in the Qcow2Cache, which is used from basically
> everywhere.

Hmm. Calling *_entry on in-coroutine path is also an extra inderection.

I've posted another solution for this think: auto generation of coroutine 
wrappers code, which produce no extra inderection.

RFC was here, a year ago: 
https://lists.gnu.org/archive/html/qemu-devel/2019-02/msg05322.html

May be, it's time to resend it, as your point gives it a point.

--
Best regards,
Vladimir