Re: [PATCH v3 3/5] migration: Don't wait in semaphore for thread we know has finished

[Dr. David Alan Gilbert]

* Juan Quintela (address@hidden) wrote:
> If p->quit is true for any channel, we know that it has finished for
> any reason.  So don't wait for it, just continue.
> 
> Signed-off-by: Juan Quintela <address@hidden>
> 
> ---
> 
> I could be convinced that the right thing to do in that case is to
> just do a break instead of a continue.  Each option has its own
> advantages/disadvantanges.
> ---
>  migration/ram.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/migration/ram.c b/migration/ram.c
> index 44ca56e1ea..bc918ef28d 100644
> --- a/migration/ram.c
> +++ b/migration/ram.c
> @@ -1118,6 +1118,12 @@ static void multifd_send_sync_main(RAMState *rs)
>          MultiFDSendParams *p = &multifd_send_state->params[i];
>  
>          trace_multifd_send_sync_main_wait(p->id);
> +        qemu_mutex_lock(&p->mutex);
> +        if (p->quit) {
> +            qemu_mutex_unlock(&p->mutex);
> +            continue;
> +        }
> +        qemu_mutex_unlock(&p->mutex);

Why is this needed/helps?
You can't depend on the p->quit happening before the 
sem_wait, so the main thread still has to do a post on sem_sync before
the join, even with the addition of the check for p->quit.

Dave

>          qemu_sem_wait(&p->sem_sync);
>      }
>      trace_multifd_send_sync_main(multifd_send_state->packet_num);
> -- 
> 2.24.1
> 
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK