Re: [RFC PATCH v2 0/8] Removal of AioContext lock, bs->parents and ->children: new rwlock

[Paolo Bonzini]

On 5/24/22 12:20, Stefan Hajnoczi wrote:
> > Maybe it's safe to run it without a lock because it runs after
> > virtio_set_status(vdev, 0) but I'd rather play it safe and protect s->rq
> > with a lock.
>
> What does the lock protect?
>
> A lock can prevent s->rq or req->vq corruption but it cannot prevent
> request leaks. This loop's job is to free all requests so there is no
> leak. If a lock is necessary then this code is already broken in a more
> fundamental way because it can leak.

Yes, you're right.  This particular list is always accessed in the 
iothread (if any) and blk_drain() is enough.  virtio-blk should already 
not need aio_context_{acquire,release}.

It's worth a comment, though!

Paolo