npfd keeps track of how many pollfds are currently being monitored. It
must be reset to 0 when fdmon_poll_wait() returns.

When npfd reaches a treshold we switch to fdmon-epoll because it scales
better.

This patch resets npfd in the case where we switch to fdmon-epoll.
Forgetting to do so results in the following assertion failure:

  util/fdmon-poll.c:65: fdmon_poll_wait: Assertion `npfd == 0' failed.

Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=1869952
Fixes: 1f050a4690f62a1e7dabc4f44141e9f762c3769f ("aio-posix: extract ppoll(2) 
and epoll(7) fd monitoring")
Reviewed-by: Philippe Mathieu-Daudé <phi...@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefa...@redhat.com>
---
 util/fdmon-poll.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/util/fdmon-poll.c b/util/fdmon-poll.c
index 488067b679..5fe3b47865 100644
--- a/util/fdmon-poll.c
+++ b/util/fdmon-poll.c
@@ -73,6 +73,7 @@ static int fdmon_poll_wait(AioContext *ctx, AioHandlerList 
*ready_list,
 
     /* epoll(7) is faster above a certain number of fds */
     if (fdmon_epoll_try_upgrade(ctx, npfd)) {
+        npfd = 0; /* we won't need pollfds[], reset npfd */
         return ctx->fdmon_ops->wait(ctx, ready_list, timeout);
     }
 
-- 
2.26.2

Reply via email to