@@ -903,6 +903,17 @@ static int captureBufs(struct node *node, struct node *node_m2m_cap, const cv4l_
epollfd = epoll_create1(0);
fail_on_test(epollfd < 0);
+
+ /*
+ * Many older versions of the vb2 and m2m have a bug where
+ * EPOLLIN and EPOLLOUT events are never signaled unless they
+ * are part of the initial EPOLL_CTL_ADD. We set an initial
+ * empty set of events, which we then modify with EPOLL_CTL_MOD,
+ * in order to detect that condition.
+ */
+ ev.events = 0;
+ fail_on_test(epoll_ctl(epollfd, EPOLL_CTL_ADD, node->g_fd(), &ev));
+
if (node->is_m2m)
ev.events = EPOLLIN | EPOLLOUT | EPOLLPRI;
else if (v4l_type_is_output(q.g_type()))
@@ -910,7 +921,7 @@ static int captureBufs(struct node *node, struct node *node_m2m_cap, const cv4l_
else
ev.events = EPOLLIN;
ev.data.fd = node->g_fd();
- fail_on_test(epoll_ctl(epollfd, EPOLL_CTL_ADD, node->g_fd(), &ev));
+ fail_on_test(epoll_ctl(epollfd, EPOLL_CTL_MOD, node->g_fd(), &ev));
}
if (pollmode)
@@ -944,6 +955,10 @@ static int captureBufs(struct node *node, struct node *node_m2m_cap, const cv4l_
can_read = FD_ISSET(node->g_fd(), &rfds);
have_event = FD_ISSET(node->g_fd(), &efds);
} else if (pollmode == POLL_MODE_EPOLL) {
+ /*
+ * This can fail with a timeout on older kernels for
+ * drivers using vb2_core_poll() v4l2_m2m_poll().
+ */
ret = epoll_wait(epollfd, &ev, 1, 2000);
fail_on_test(ret == 0);
fail_on_test(ret < 0);
A bug has been discovered in the vb2 and m2m implementations of poll where EPOLLIN and EPOLLOUT events are never signaled to epoll_wait() unless at least one of them is included in the initial epoll_ctl() call. The presence of this bug can be detected by setting an initial empty set of events with EPOLL_CTL_ADD, then setting the events we are interested in using EPOLL_CTL_MOD. epoll_wait() will then invariably timeout even if there is a queue-related event pending. Signed-off-by: Alexandre Courbot <gnurou@gmail.com> --- This is the follow-up patch for the discussion on https://www.spinics.net/lists/linux-media/msg179618.html. I have tested it with vivid and vicodec, and got the expected result with both (epoll streaming tests fails without the kernel fix, passes with it). utils/v4l2-compliance/v4l2-test-buffers.cpp | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-)