Il lun 26 set 2022, 14:21 Vladimir Sementsov-Ogievskiy <
vsement...@yandex-team.ru> ha scritto:
> On 9/18/22 20:12, Emanuele Giuseppe Esposito wrote:
> >>> --- a/qemu-img.c
> >>> +++ b/qemu-img.c
> >>> @@ -911,7 +911,6 @@ static void run_block_job(BlockJob *job, Error
> >>> **errp)
> >>> AioContext *aio_context = block_job_get_aio_context(job);
> >>> int ret = 0;
> >>> - aio_context_acquire(aio_context);
> >>> job_lock();
> >>> job_ref_locked(&job->job);
> >>> do {
> >> aio_poll() call here, doesn't require aio_context to be acquired?
> > On the contrary I think, if you see in AIO_WAIT_WHILE we explicitly
> > release it because we don't want to allow something else to run with the
> > aiocontext acquired.
> >
>
> Still, in AIO_WAIT_WHILE() we release ctx_, but do
> aio_poll(qemu_get_aio_context(), true), so we poll in other context.
>
> But here in qemu-img.c we drop aiocontext lock exactly for aio_context,
> which is an argument of aio_poll()..
>
It's the same, the acquire/release is done again in file descriptor
callback or bottom halves (typically via aio_co_wake).
Paolo
> --
> Best regards,
> Vladimir
>
>
