Re: [PATCH 1/6] block: don't acquire AioContext lock in bdrv_drain_all()

[Kevin Wolf <kwolf@redhat.com>]

Am 01.03.2023 um 21:57 hat Stefan Hajnoczi geschrieben:
> There is no need for the AioContext lock in bdrv_drain_all() because
> nothing in AIO_WAIT_WHILE() needs the lock and the condition is atomic.
> 
> Note that the NULL AioContext argument to AIO_WAIT_WHILE() is odd. In
> the future it can be removed.

It can be removed for all callers that run in the main loop context. For
code running in an iothread, it's still important to pass a non-NULL
context. This makes me doubt that the ctx parameter can really be
removed without changing more.

Is your plan to remove the if from AIO_WAIT_WHILE_INTERNAL(), too, and
to poll qemu_get_current_aio_context() instead of ctx_ or the main
context?

> There is an assertion in
> AIO_WAIT_WHILE() that checks that we're in the main loop AioContext and
> we would lose that check by dropping the argument. However, that was a
> precursor to the GLOBAL_STATE_CODE()/IO_CODE() macros and is now a
> duplicate check. So I think we won't lose much by dropping it, but let's
> do a few more AIO_WAIT_WHILE_UNLOCKED() coversions of this sort to
> confirm this is the case.
> 
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

Yes, it seems that we don't lose much, except maybe some consistency in
the intermediate state. The commit message could state a bit more
directly what we gain, though. Since you mention removing the parameter
as a future possibility, I assume that's the goal with it, but I
wouldn't be sure just from reading the commit message.

Kevin