Re: [Qemu-devel] [PULL 2/5] migration: move bdrv_invalidate_cache_all of of coroutine context

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Paolo Bonzini (pbonzini@redhat.com) wrote:
> 
> 
> On 07/03/2016 13:49, Dr. David Alan Gilbert wrote:
> >    b) The harder problem is that there's a race where qemu_bh_delete
> >       segs, and I'm not 100% sure why yet - it only does it sometime
> >       (i.e. run virt-test and leave it and it occasionally does it).
> >       From the core it looks like qemu->bh is corrupt (0x10101010...)
> >       so maybe mis has been freed at that point?
> >       I'm suspecting this is the postcopy_ram_listen_thread freeing
> >       mis at the end of it, but I don't know yet.
> 
> That should be it.  Maybe the patch can simply be reverted, because
> loadvm_postcopy_handle_run runs from a thread and not a coroutine.  Is
> this correct?

That's still in the main thread, the 'run' comes from the packaged postcopy
state, but is after the 'listener' thread has been started.

I need to understand this anyway; the way it's supposed to work is that
if postcopy is being used then not much cleanup happens in process_incoming_migration_co
instead it exits and lets postcopy_ram_listen_thread do the cleanup
at the end; I've not quite figured out what's going on here
but it almost looks like both of them are cleaning up - that shouldn't
happen.

> However I have a bug or two for you to fix, too:
> 
> 1) as far as I can see, postcopy_ram_listen_thread is not holding the
> mutex during the call to qemu_loadvm_state_main.  Is that a bug?

No; the guest is running, the only thing that gets loaded by that
listen thread is data that's postcopied - i.e. currently just ram pages
that are loaded atomically.

> 2) no one is currently joining mis->listen_thread, I suspect it actually
> should be QEMU_THREAD_DETACHED.

OK, that looks like the easier one.

Dave

> 
> :)
> 
> Paolo
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK