selftests/epoll: add a regression test for pipe->poll_usage

pipe->poll_usage was added to ensure that edge-triggered epoll consumers
get a wakeup on every write, even if the pipe was already non-empty.
However, none of the existing epoll_wakeup_test cases cover this; the
test suite passes even with WRITE_ONCE(pipe->poll_usage, true) removed.

Add a test that writes twice to a pipe and verifies that epoll_wait with
EPOLLET reports data each time. This covers the pipe-specific per-write
wakeup behavior that edge-triggered consumers depend on.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Link: https://patch.msgid.link/amnlGZesXu-SUK2H@redhat.com
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>

diff --git a/tools/testing/selftests/filesystems/epoll/epoll_wakeup_test.c b/tools/testing/selftests/filesystems/epoll/epoll_wakeup_test.c
index f6f1a7ff01b0568090715f2feda890a59fb2470f..81a994943e121bed7b88950b9fe940e8efc88042 100644 (file)
--- a/tools/testing/selftests/filesystems/epoll/epoll_wakeup_test.c
+++ b/tools/testing/selftests/filesystems/epoll/epoll_wakeup_test.c
@@ -3538,4 +3538,27 @@ TEST(epoll65)
        close(ctx.efd[1]);
 }
 
+TEST(epoll66)
+{
+       struct epoll_event event;
+       int pfd[2], efd;
+
+       ASSERT_EQ(pipe(pfd), 0);
+
+       efd = epoll_create1(0);
+       ASSERT_GE(efd, 0);
+
+       event.events = EPOLLIN | EPOLLET;
+       ASSERT_EQ(epoll_ctl(efd, EPOLL_CTL_ADD, pfd[0], &event), 0);
+
+       for (int i = 0; i < 2; ++i) {
+               ASSERT_EQ(write(pfd[1], "", 1), 1);
+               EXPECT_EQ(epoll_wait(efd, &event, 1, 0), 1);
+       }
+
+       close(pfd[0]);
+       close(pfd[1]);
+       close(efd);
+}
+
 TEST_HARNESS_MAIN