On Wed, Jun 15, 2016 at 10:57:49AM +0200, Paolo Bonzini wrote:
> On 14/06/2016 20:17, Stefan Hajnoczi wrote:
> > +/* There is no matching mirror_resume() because mirror_run() will begin
> > + * iterating again when the job is resumed.
> > + */
> > +static void mirror_pause(BlockJob *job)
> > +{
> > +    MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
> > +
> > +    while (s->in_flight > 0) {
> > +        aio_poll(blk_get_aio_context(job->blk), true);
> > +    }
> 
> This is calling aio_poll from a coroutine, which is ugly -- see Fam's
> recent introduction of bdrv_co_drain.  I think this should call
> mirror_drain instead.

This is not called from coroutine context, so I couldn't use
mirror_drain().

BlockJobDriver->pause() and attached_aio_context() are both not
coroutine_fn.  That's because bdrv_attach_aio_context() is not
coroutine_fn and can be called from a monitor command or device
emulation code.

Stefan