Re: [Qemu-devel] [PATCH 11/20] mirror: support more than one in-flight AIO operation

[Paolo Bonzini <pbonzini@redhat.com>]

Il 14/01/2013 13:56, Stefan Hajnoczi ha scritto:
>> > +    /* Wait for I/O to this cluster (from a previous iteration) to be done.  */
>> > +    while (test_bit(cluster_num, s->in_flight_bitmap)) {
>> > +        trace_mirror_yield_in_flight(s, sector_num, s->in_flight);
>> > +        qemu_coroutine_yield();
>> >      }
> in_flight_bitmap is never set in this patch.  Either you'll set it in a
> later patch or this is a bug?

Yes, this must be a rebase bug.  Good catch.  I'll send an updated
series tomorrow.

> Why can we get away with testing only cluster_num and not all bits up to
> and including cluster_num + nb_chunks?  Is there an assumption that
> requests are issued in ascending order?

As of this patch, cluster_num is either a single granularity-sized
cluster or it is rounded to include a full image cluster.  If it is
rounded, any other concurrent operation will do the same rounding and
set the bit corresponding to cluster_num.

In the next patch, the same applies when each granularity-sized cluster
is added to the current operation (either alone, or rounded to include a
full image cluster).

I'll add a comment.

Paolo