Re: [Qemu-devel] [PATCH] Check return values from g_poll and select

[Paolo Bonzini <pbonzini@redhat.com>]

Il 08/01/2013 16:30, Fabien Chouteau ha scritto:
> The current implementation of os_host_main_loop_wait() on Windows,
> returns 1 only when a g_poll() event occurs because the return value of
> select() is overridden. This is wrong as we may skip a socket event, as
> shown in this example:
> 
> 1. select() returns 0
> 2. g_poll() returns 1  (socket event occurs)
> 3. os_host_main_loop_wait() returns 1
> 4. qemu_iohandler_poll() sees no socket event because select() has
>    return before the event occurs
> 5. select() returns 1
> 6. g_poll() returns 0 (g_poll overrides select's return value)
> 7. os_host_main_loop_wait() returns 0
> 8. qemu_iohandler_poll() doesn't check for socket events because the
>    return value of os_host_main_loop_wait() is zero.
> 9. goto 5
> 
> This patch use one variable for each of these return values, so we don't
> miss a select() event anymore.
> 
> Also move the call to select() after g_poll(), this will improve latency
> as we don't have to go through two os_host_main_loop_wait() calls to
> detect a socket event.
> 
> Signed-off-by: Fabien Chouteau <chouteau@adacore.com>

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>

Paolo

> ---
>  main-loop.c |   27 +++++++++++++--------------
>  1 file changed, 13 insertions(+), 14 deletions(-)
> 
> diff --git a/main-loop.c b/main-loop.c
> index 54f38ae..6f52ac3 100644
> --- a/main-loop.c
> +++ b/main-loop.c
> @@ -330,7 +330,7 @@ void qemu_fd_register(int fd)
>  static int os_host_main_loop_wait(uint32_t timeout)
>  {
>      GMainContext *context = g_main_context_default();
> -    int ret, i;
> +    int select_ret, g_poll_ret, ret, i;
>      PollingEntry *pe;
>      WaitObjects *w = &wait_objects;
>      gint poll_timeout;
> @@ -345,13 +345,6 @@ static int os_host_main_loop_wait(uint32_t timeout)
>          return ret;
>      }
>  
> -    if (nfds >= 0) {
> -        ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
> -        if (ret != 0) {
> -            timeout = 0;
> -        }
> -    }
> -
>      g_main_context_prepare(context, &max_priority);
>      n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
>                                        poll_fds, ARRAY_SIZE(poll_fds));
> @@ -367,9 +360,9 @@ static int os_host_main_loop_wait(uint32_t timeout)
>      }
>  
>      qemu_mutex_unlock_iothread();
> -    ret = g_poll(poll_fds, n_poll_fds + w->num, poll_timeout);
> +    g_poll_ret = g_poll(poll_fds, n_poll_fds + w->num, poll_timeout);
>      qemu_mutex_lock_iothread();
> -    if (ret > 0) {
> +    if (g_poll_ret > 0) {
>          for (i = 0; i < w->num; i++) {
>              w->revents[i] = poll_fds[n_poll_fds + i].revents;
>          }
> @@ -384,12 +377,18 @@ static int os_host_main_loop_wait(uint32_t timeout)
>          g_main_context_dispatch(context);
>      }
>  
> -    /* If an edge-triggered socket event occurred, select will return a
> -     * positive result on the next iteration.  We do not need to do anything
> -     * here.
> +    /* Call select after g_poll to avoid a useless iteration and therefore
> +     * improve socket latency.
>       */
>  
> -    return ret;
> +    if (nfds >= 0) {
> +        select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
> +        if (select_ret != 0) {
> +            timeout = 0;
> +        }
> +    }
> +
> +    return select_ret || g_poll_ret;
>  }
>  #endif
>  
>