Re: [Qemu-devel] [PATCH v10 19/24] migration: Create multifd channels

[Peter Xu <peterx@redhat.com>]

On Thu, Mar 15, 2018 at 01:57:54PM +0100, Juan Quintela wrote:
> Peter Xu <peterx@redhat.com> wrote:
> > On Wed, Mar 07, 2018 at 12:00:05PM +0100, Juan Quintela wrote:
> >> In both sides.  We still don't transmit anything through them.
> >
> > s/In/On/?
> >
> >> 
> >> Signed-off-by: Juan Quintela <quintela@redhat.com>
> >> ---
> >>  migration/ram.c | 52 ++++++++++++++++++++++++++++++++++++++++++----------
> >>  1 file changed, 42 insertions(+), 10 deletions(-)
> >> 
> >> diff --git a/migration/ram.c b/migration/ram.c
> >> index b57d9fd667..7ef0c2b7e2 100644
> >> --- a/migration/ram.c
> >> +++ b/migration/ram.c
> >> @@ -399,6 +399,7 @@ struct MultiFDSendParams {
> >>      uint8_t id;
> >>      char *name;
> >>      QemuThread thread;
> >> +    QIOChannel *c;
> >>      QemuSemaphore sem;
> >>      QemuMutex mutex;
> >>      bool running;
> >> @@ -455,6 +456,8 @@ int multifd_save_cleanup(Error **errp)
> >>              qemu_thread_join(&p->thread);
> >>              p->running = false;
> >>          }
> >> +        socket_send_channel_destroy(p->c);
> >> +        p->c = NULL;
> >>          qemu_mutex_destroy(&p->mutex);
> >>          qemu_sem_destroy(&p->sem);
> >>          g_free(p->name);
> >> @@ -514,6 +517,27 @@ static void *multifd_send_thread(void *opaque)
> >>      return NULL;
> >>  }
> >>  
> >> +static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque)
> >> +{
> >> +    MultiFDSendParams *p = opaque;
> >> +    QIOChannel *sioc = QIO_CHANNEL(qio_task_get_source(task));
> >> +    Error *local_err = NULL;
> >> +
> >> +    if (qio_task_propagate_error(task, &local_err)) {
> >> +        if (multifd_save_cleanup(&local_err) != 0) {
> >
> > Do we need to call multifd_save_cleanup() explicitly here?
> 
> Is the easiest way of stoping all multifd threads, no?

Yeah, but again, I thought it would be called later too, since...

> 
> > Asked since I saw that it would also be called in migrate_fd_cleanup(),
> > and it seems that we should call migrate_fd_cleanup() soon too when
> > this happens?
> 
> We need to stop migraiton. thtat migrate_set_error() is only used for
> reporting in info migrate, it is not acted upon.
> 
> Yes, perhaps it should, but as it is, it is not.  So, I think it is
> right O:-)

... after Dave's 688a3dcba9 ("migration: Route errors down through
migration_channel_connect", 2018-02-06), all these channel errors
should finally be routed to migrate_fd_connect(), and in that we have:

void migrate_fd_connect(MigrationState *s, Error *error_in)
{
    s->expected_downtime = s->parameters.downtime_limit;
    s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
    if (error_in) {
        migrate_fd_error(s, error_in);
        migrate_fd_cleanup(s);
        return;
    }
    ...
}

Then, in migrate_fd_cleanup() we have multifd_save_cleanup().  That's
why I thought we can skip the cleanup here since after all we'll do it
in other places (and we can keep the cleanup code unified).

Thanks,

> 
> Later, Juan.
> 
> > Otherwise it looks fine to me.  Thanks,
> >
> >> +            migrate_set_error(migrate_get_current(), local_err);
> >> +        }
> >> +    } else {
> >> +        p->c = QIO_CHANNEL(sioc);
> >> +        qio_channel_set_delay(p->c, false);
> >> +        p->running = true;
> >> +        qemu_thread_create(&p->thread, p->name, multifd_send_thread, p,
> >> +                           QEMU_THREAD_JOINABLE);
> >> +
> >> +        atomic_inc(&multifd_send_state->count);
> >> +    }
> >> +}
> >
> > [...]

-- 
Peter Xu