Use non-blocking timerfd to prevent blocking when updating timer event sources

This implements a simple fix for the blocking problem that occurs when
updating a timer event source after the timer expires, but before its
callback is dispatched. This can happen when another event happens during the
same epoll wakeup as the timer event, and causes the read() call in
wl_event_source_timer_dispatch() to block for the updated duration of the
timer.

We never want this read() call to block, so I believe it makes sense for the
timerfd to be non-blocking, and we simply ignore the case where the read fails
with EAGAIN. We still report all other errors as before, and still ignore the
actual value read from the socket.

With this change, the event_loop_timer_updates unit test case I submitted
previously now passes, and weston appears to work as before.

diff --git a/src/event-loop.c b/src/event-loop.c
index d323601..1046600 100644 (file)
--- a/src/event-loop.c
+++ b/src/event-loop.c
@@ -173,7 +173,7 @@ wl_event_source_timer_dispatch(struct wl_event_source *source,
        int len;
 
        len = read(source->fd, &expires, sizeof expires);
-       if (len != sizeof expires)
+       if (!(len == -1 && errno == EAGAIN) && len != sizeof expires)
                /* Is there anything we can do here?  Will this ever happen? */
                fprintf(stderr, "timerfd read error: %m\n");
 
@@ -196,7 +196,8 @@ wl_event_loop_add_timer(struct wl_event_loop *loop,
                return NULL;
 
        source->base.interface = &timer_source_interface;
-       source->base.fd = timerfd_create(CLOCK_MONOTONIC, TFD_CLOEXEC);
+       source->base.fd = timerfd_create(CLOCK_MONOTONIC,
+                                                                        TFD_CLOEXEC | TFD_NONBLOCK);
        source->func = func;
 
        return add_source(loop, &source->base, WL_EVENT_READABLE, data);