fs/epoll: remove unnecessary wakeups of nested epoll
authorHeiher <r@hev.cc>
Thu, 5 Dec 2019 00:52:15 +0000 (16:52 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 5 Dec 2019 03:44:13 +0000 (19:44 -0800)
commit339ddb53d373baee6e7946aec17c739c4924d6d9
treea4e439109c84284fc7ecafd87c5e60e16efa7d9d
parentf6520c520842c532059c380d7a5b30b6e0e72ced
fs/epoll: remove unnecessary wakeups of nested epoll

Take the case where we have:

        t0
         | (ew)
        e0
         | (et)
        e1
         | (lt)
        s0

t0: thread 0
e0: epoll fd 0
e1: epoll fd 1
s0: socket fd 0
ew: epoll_wait
et: edge-trigger
lt: level-trigger

We remove unnecessary wakeups to prevent the nested epoll that working in edge-
triggered mode to waking up continuously.

Test code:
 #include <unistd.h>
 #include <sys/epoll.h>
 #include <sys/socket.h>

 int main(int argc, char *argv[])
 {
  int sfd[2];
  int efd[2];
  struct epoll_event e;

  if (socketpair(AF_UNIX, SOCK_STREAM, 0, sfd) < 0)
  goto out;

  efd[0] = epoll_create(1);
  if (efd[0] < 0)
  goto out;

  efd[1] = epoll_create(1);
  if (efd[1] < 0)
  goto out;

  e.events = EPOLLIN;
  if (epoll_ctl(efd[1], EPOLL_CTL_ADD, sfd[0], &e) < 0)
  goto out;

  e.events = EPOLLIN | EPOLLET;
  if (epoll_ctl(efd[0], EPOLL_CTL_ADD, efd[1], &e) < 0)
  goto out;

  if (write(sfd[1], "w", 1) != 1)
  goto out;

  if (epoll_wait(efd[0], &e, 1, 0) != 1)
  goto out;

  if (epoll_wait(efd[0], &e, 1, 0) != 0)
  goto out;

  close(efd[0]);
  close(efd[1]);
  close(sfd[0]);
  close(sfd[1]);

  return 0;

 out:
  return -1;
 }

More tests:
 https://github.com/heiher/epoll-wakeup

Link: http://lkml.kernel.org/r/20191009060516.3577-1-r@hev.cc
Signed-off-by: hev <r@hev.cc>
Reviewed-by: Roman Penyaev <rpenyaev@suse.de>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Eric Wong <e@80x24.org>
Cc: Jason Baron <jbaron@akamai.com>
Cc: Sridhar Samudrala <sridhar.samudrala@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/eventpoll.c