From c1a2cf47ee7f18f08c8f8cd353eef60531f3753b Mon Sep 17 00:00:00 2001 From: Mike Crowe Date: Thu, 15 Sep 2016 10:17:53 +0100 Subject: [PATCH] epoll_wait.2: Clarify that the timeout is measured against CLOCK_MONOTONIC The existing page left the choice of clock as ambiguous. My reading of the kernel implementation is that CLOCK_MONOTONIC is always used since ep_poll() calls schedule_hrtimeout_range(), which calls schedule_hrtimeout_range_clock() passing CLOCK_MONOTONIC. References: http://lxr.free-electrons.com/source/fs/eventpoll.c?v=4.7#L1614 http://lxr.free-electrons.com/source/kernel/time/hrtimer.c?v=4.7#L1785 The "measured against" terminology was borrowed from nanosleep.2. I've checked all the way back to Linux v2.6.12 where the calculation was done using jiffies so I think that this has effectively always been true. Signed-off-by: Mike Crowe --- man2/epoll_wait.2 | 3 +++ 1 file changed, 3 insertions(+) diff --git a/man2/epoll_wait.2 b/man2/epoll_wait.2 index 963790644..48bb28134 100644 --- a/man2/epoll_wait.2 +++ b/man2/epoll_wait.2 @@ -55,6 +55,9 @@ The argument specifies the number of milliseconds that .BR epoll_wait () will block. +Time is measured against the +.B CLOCK_MONOTONIC +clock. The call will block until either: .IP * 3 a file descriptor delivers an event;