Re: [Qemu-devel] [PATCH 15/17] migration: Test new fd infrastructure

[Dr. David Alan Gilbert]

* Juan Quintela (quint...@redhat.com) wrote:
> "Dr. David Alan Gilbert" <dgilb...@redhat.com> wrote:
> > * Juan Quintela (quint...@redhat.com) wrote:
> >> We just send the address through the alternate channels and test that it
> >> is ok.
> >> 
> >> Signed-off-by: Juan Quintela <quint...@redhat.com>
> >> ---
> >>  migration/ram.c | 36 ++++++++++++++++++++++++++++++++++++
> >>  1 file changed, 36 insertions(+)
> >> 
> >> diff --git a/migration/ram.c b/migration/ram.c
> >> index 4e530ea..95af694 100644
> >> --- a/migration/ram.c
> >> +++ b/migration/ram.c
> >> @@ -432,8 +432,22 @@ static void *multifd_send_thread(void *opaque)
> >>      qemu_mutex_lock(&params->mutex);
> >>      while (!params->quit){
> >>          if (params->pages.num) {
> >> +            int i;
> >> +            int num;
> >> +
> >> +            num = params->pages.num;
> >>              params->pages.num = 0;
> >>              qemu_mutex_unlock(&params->mutex);
> >> +
> >> +            for(i=0; i < num; i++) {
> >> +                if (qio_channel_write(params->c,
> >> +                                      (const char 
> >> *)&params->pages.address[i],
> >> +                                      sizeof(uint8_t *), &error_abort)
> >> +                    != sizeof(uint8_t*)) {
> >> +                    /* Shuoudn't ever happen */
> >> +                    exit(-1);
> >> +                }
> >
> > Nope, need to find a way to cleanly find the migration; that
> > might actually be tricky from one of these threads?
> 
> It is tricky, but the error handling is wrong in the callers of this
> already.  Will try to improve it on next series, but the problem is
> already there.

Well we should never kill the source because of a failed migration;
especially just due to a fairly boring IO error.
You could just exit the thread and return a marker value that gets
delivered to the pthread_join.
There is qemu_file_set_error() that you could call - although you'd
have to figure out thread safety.
(Which might be as simple as cmpxchg the error values and only
change the error value if it was previously 0).

Dave
--
Dr. David Alan Gilbert / dgilb...@redhat.com / Manchester, UK