Merge branch 'egil/erts/high_accuracy_sleep/OTP-12236'

* egil/erts/high_accuracy_sleep/OTP-12236:
  erts: Add high accuracy poll timeouts

3 files changed, 167 insertions, 9 deletions

diff --git a/erts/configure.in b/erts/configure.in
index 8a9bf30433..62515fe081 100644
--- a/erts/configure.in
+++ b/erts/configure.in
@@ -1760,6 +1760,10 @@ AC_CHECK_HEADER(sys/event.h, have_kernel_poll=kqueue)
 AC_CHECK_HEADER(sys/epoll.h, have_kernel_poll=epoll)
 AC_CHECK_HEADER(sys/devpoll.h, have_kernel_poll=/dev/poll)
 
+dnl Check if we have timerfds to be used for high accuracy
+dnl epoll_wait timeouts
+AC_CHECK_HEADERS([sys/timerfd.h])
+
 dnl Check for kernel SCTP support
 AC_SUBST(LIBSCTP)
 if test "x$enable_sctp" != "xno" ; then
@@ -2105,7 +2109,7 @@ AC_CHECK_FUNCS([ieee_handler fpsetmask finite isnan isinf res_gethostbyname dlop
 		gethrtime localtime_r gmtime_r inet_pton \
 		memcpy mallopt sbrk _sbrk __sbrk brk _brk __brk \
 		flockfile fstat strlcpy strlcat setsid posix2time time2posix \
-		setlocale nl_langinfo poll mlockall])
+		setlocale nl_langinfo poll mlockall ppoll])
 
 AC_MSG_CHECKING([for isfinite])
 AC_TRY_LINK([#include <math.h>],
diff --git a/erts/emulator/sys/common/erl_poll.c b/erts/emulator/sys/common/erl_poll.c
index 446cca77ff..71c4239902 100644
--- a/erts/emulator/sys/common/erl_poll.c
+++ b/erts/emulator/sys/common/erl_poll.c
@@ -314,6 +314,9 @@ struct ErtsPollSet_ {
 #if ERTS_POLL_USE_WAKEUP_PIPE
     int wake_fds[2];
 #endif
+#if ERTS_POLL_USE_TIMERFD
+    int timer_fd;
+#endif
 #if ERTS_POLL_USE_FALLBACK
     int fallback_used;
 #endif
@@ -580,6 +583,75 @@ create_wakeup_pipe(ErtsPollSet ps)
 #endif /* ERTS_POLL_USE_WAKEUP_PIPE */
 
 /*
+ * --- timer fd -----------------------------------------------------------
+ */
+
+#if ERTS_POLL_USE_TIMERFD
+
+/* We use the timerfd when using epoll_wait to get high accuracy
+   timeouts, i.e. we want to sleep with < ms accuracy. */
+
+static void
+create_timerfd(ErtsPollSet ps)
+{
+    int do_wake = 0;
+    int timer_fd;
+    timer_fd = timerfd_create(CLOCK_MONOTONIC,0);
+    ERTS_POLL_EXPORT(erts_poll_control)(ps,
+					timer_fd,
+					ERTS_POLL_EV_IN,
+					1, &do_wake);
+#if ERTS_POLL_USE_FALLBACK
+    /* We depend on the wakeup pipe being handled by kernel poll */
+    if (ps->fds_status[timer_fd].flags & ERTS_POLL_FD_FLG_INFLBCK)
+	fatal_error("%s:%d:create_wakeup_pipe(): Internal error\n",
+		    __FILE__, __LINE__);
+#endif
+    if (ps->internal_fd_limit <= timer_fd)
+	ps->internal_fd_limit = timer_fd + 1;
+    ps->timer_fd = timer_fd;
+}
+
+static ERTS_INLINE void
+timerfd_set(ErtsPollSet ps, struct itimerspec *its)
+{
+#ifdef DEBUG
+    struct itimerspec old_its;
+    int res;
+    res = timerfd_settime(ps->timer_fd, 0, its, &old_its);
+    ASSERT(res == 0);
+    ASSERT(old_its.it_interval.tv_sec == 0 &&
+           old_its.it_interval.tv_nsec == 0 &&
+           old_its.it_value.tv_sec == 0 &&
+           old_its.it_value.tv_nsec == 0);
+
+#else
+    timerfd_settime(ps->timer_fd, 0, its, NULL);
+#endif
+}
+
+static ERTS_INLINE int
+timerfd_clear(ErtsPollSet ps, int res, int max_res) {
+
+    struct itimerspec its;
+    /* we always have to clear the timer */
+    its.it_interval.tv_sec = 0;
+    its.it_interval.tv_nsec = 0;
+    its.it_value.tv_sec = 0;
+    its.it_value.tv_nsec = 0;
+    timerfd_settime(ps->timer_fd, 0, &its, NULL);
+
+    /* only timeout fd triggered */
+    if (res == 1 && ps->res_events[0].data.fd == ps->timer_fd)
+        return 0;
+
+    return res;
+}
+
+#endif /* ERTS_POLL_USE_TIMERFD */
+
+
+/*
  * --- Poll set update requests ----------------------------------------------
  */
 #if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE
@@ -1517,6 +1589,12 @@ poll_control(ErtsPollSet ps, int fd, ErtsPollEvents events, int on, int *do_wake
 	    goto done;
 	}
 #endif
+#if ERTS_POLL_USE_TIMERFD
+	if (fd == ps->timer_fd) {
+	    new_events = ERTS_POLL_EV_NVAL;
+	    goto done;
+	}
+#endif
     }
 
     if (fd >= ps->fds_status_len)
@@ -1664,6 +1742,9 @@ save_kp_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, int chk_fds_res)
 #if ERTS_POLL_USE_WAKEUP_PIPE
     int wake_fd = ps->wake_fds[0];
 #endif
+#if ERTS_POLL_USE_TIMERFD
+    int timer_fd = ps->timer_fd;
+#endif
 
     for (i = 0; i < n; i++) {
 
@@ -1679,6 +1760,11 @@ save_kp_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, int chk_fds_res)
 		continue;
 	    }
 #endif
+#if ERTS_POLL_USE_TIMERFD
+            if (fd == timer_fd) {
+                continue;
+            }
+#endif
 	    ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK));
 	    /* epoll_wait() can repeat the same fd in result array... */
 	    ix = (int) ps->fds_status[fd].res_ev_ix;
@@ -1753,6 +1839,11 @@ save_kp_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, int chk_fds_res)
 		continue;
 	    }
 #endif
+#if ERTS_POLL_USE_TIMERFD
+	    if (fd == timer_fd) {
+		continue;
+	    }
+#endif
 	    revents = ERTS_POLL_EV_N2E(ps->res_events[i].events);
 	    pr[res].fd = fd;
 	    pr[res].events = revents;
@@ -2097,7 +2188,7 @@ get_timeout_timeval(ErtsPollSet ps,
 
 #endif
 
-#if ERTS_POLL_USE_KQUEUE
+#if ERTS_POLL_USE_KQUEUE || (ERTS_POLL_USE_POLL && defined(HAVE_PPOLL)) || ERTS_POLL_USE_TIMERFD
 
 static ERTS_INLINE int
 get_timeout_timespec(ErtsPollSet ps,
@@ -2120,7 +2211,7 @@ get_timeout_timespec(ErtsPollSet ps,
 
 	ASSERT(tsp->tv_sec >= 0);
 	ASSERT(tsp->tv_nsec >= 0);
-	ASSERT(tsp->tv_nsec < 1000*1000);
+	ASSERT(tsp->tv_nsec < 1000*1000*1000);
 
 	return !0;
     }
@@ -2128,6 +2219,22 @@ get_timeout_timespec(ErtsPollSet ps,
 
 #endif
 
+#if ERTS_POLL_USE_TIMERFD
+
+static ERTS_INLINE int
+get_timeout_itimerspec(ErtsPollSet ps,
+                       struct itimerspec *itsp,
+                       ErtsMonotonicTime timeout_time)
+{
+
+    itsp->it_interval.tv_sec = 0;
+    itsp->it_interval.tv_nsec = 0;
+
+    return get_timeout_timespec(ps, &itsp->it_value, timeout_time);
+}
+
+#endif
+
 static ERTS_INLINE int
 check_fd_events(ErtsPollSet ps, ErtsMonotonicTime timeout_time, int max_res)
 {
@@ -2145,12 +2252,29 @@ check_fd_events(ErtsPollSet ps, ErtsMonotonicTime timeout_time, int max_res)
 #if ERTS_POLL_USE_EPOLL		/* --- epoll ------------------------------- */
 	    if (max_res > ps->res_events_len)
 		grow_res_events(ps, max_res);
+#if ERTS_POLL_USE_TIMERFD
+            {
+                struct itimerspec its;
+                timeout = get_timeout_itimerspec(ps, &its, timeout_time);
+                if (timeout) {
+#ifdef ERTS_SMP
+                    erts_thr_progress_prepare_wait(NULL);
+#endif
+                    timerfd_set(ps, &its);
+                    res = epoll_wait(ps->kp_fd, ps->res_events, max_res, -1);
+                    res = timerfd_clear(ps, res, max_res);
+                } else {
+                    res = epoll_wait(ps->kp_fd, ps->res_events, max_res, 0);
+                }
+            }
+#else /* !ERTS_POLL_USE_TIMERFD */
 	    timeout = (int) get_timeout(ps, 1000, timeout_time);
 #ifdef ERTS_SMP
 	    if (timeout)
 		erts_thr_progress_prepare_wait(NULL);
 #endif
 	    res = epoll_wait(ps->kp_fd, ps->res_events, max_res, timeout);
+#endif /* !ERTS_POLL_USE_TIMERFD */
 #elif ERTS_POLL_USE_KQUEUE	/* --- kqueue ------------------------------ */
 	    struct timespec ts;
 	    if (max_res > ps->res_events_len)
@@ -2189,7 +2313,15 @@ check_fd_events(ErtsPollSet ps, ErtsMonotonicTime timeout_time, int max_res)
 #endif
 	    poll_res.dp_timeout = timeout;
 	    res = ioctl(ps->kp_fd, DP_POLL, &poll_res);
-#elif ERTS_POLL_USE_POLL	/* --- poll -------------------------------- */
+#elif ERTS_POLL_USE_POLL && defined(HAVE_PPOLL)	/* --- ppoll ---------------- */
+            struct timespec ts;
+	    timeout = get_timeout_timespec(ps, &ts, timeout_time);
+#ifdef ERTS_SMP
+	    if (timeout)
+		erts_thr_progress_prepare_wait(NULL);
+#endif
+            res = ppoll(ps->poll_fds, ps->no_poll_fds, &ts, NULL);
+#elif ERTS_POLL_USE_POLL        /* --- poll --------------------------------- */
 	    timeout = (int) get_timeout(ps, 1000, timeout_time);
 #ifdef ERTS_SMP
 	    if (timeout)
@@ -2202,7 +2334,7 @@ check_fd_events(ErtsPollSet ps, ErtsMonotonicTime timeout_time, int max_res)
 
 	    ERTS_FD_COPY(&ps->input_fds, &ps->res_input_fds);
 	    ERTS_FD_COPY(&ps->output_fds, &ps->res_output_fds);
-	    
+
 #ifdef ERTS_SMP
 	    if (timeout)
 		erts_thr_progress_prepare_wait(NULL);
@@ -2535,6 +2667,9 @@ ERTS_POLL_EXPORT(erts_poll_create_pollset)(void)
 #if ERTS_POLL_USE_WAKEUP_PIPE
     create_wakeup_pipe(ps);
 #endif
+#if ERTS_POLL_USE_TIMERFD
+    create_timerfd(ps);
+#endif
 #if ERTS_POLL_USE_FALLBACK
     if (kp_fd >= ps->fds_status_len)
 	grow_fds_status(ps, kp_fd);
@@ -2625,6 +2760,10 @@ ERTS_POLL_EXPORT(erts_poll_destroy_pollset)(ErtsPollSet ps)
     if (ps->wake_fds[1] >= 0)
 	close(ps->wake_fds[1]);
 #endif
+#if ERTS_POLL_USE_TIMERFD
+    if (ps->timer_fd >= 0)
+        close(ps->timer_fd);
+#endif
 
     erts_smp_spin_lock(&pollsets_lock);
     if (ps == pollsets)
@@ -2729,6 +2868,9 @@ ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet ps, ErtsPollInfo *pip)
 #if ERTS_POLL_USE_WAKEUP_PIPE
     pip->poll_set_size++; /* Wakeup pipe */
 #endif
+#if ERTS_POLL_USE_TIMERFD
+    pip->poll_set_size++; /* timerfd */
+#endif
 
     pip->fallback_poll_set_size =
 #if !ERTS_POLL_USE_FALLBACK
@@ -2857,14 +2999,18 @@ ERTS_POLL_EXPORT(erts_poll_get_selected_events)(ErtsPollSet ps,
 	    ev[fd] = 0;
 	else {
 	    ev[fd] = ps->fds_status[fd].events;
+            if (
 #if ERTS_POLL_USE_WAKEUP_PIPE
-	    if (fd == ps->wake_fds[0] || fd == ps->wake_fds[1])
-		ev[fd] |= ERTS_POLL_EV_NVAL;
+                fd == ps->wake_fds[0] || fd == ps->wake_fds[1] ||
+#endif
+#if ERTS_POLL_USE_TIMERFD
+                fd == ps->timer_fd ||
 #endif
 #if ERTS_POLL_USE_KERNEL_POLL
-	    if (fd == ps->kp_fd)
-		ev[fd] |= ERTS_POLL_EV_NVAL;
+                fd == ps->kp_fd ||
 #endif
+                0)
+                ev[fd] |= ERTS_POLL_EV_NVAL;
 	}
     }
     ERTS_POLLSET_UNLOCK(ps);
diff --git a/erts/emulator/sys/common/erl_poll.h b/erts/emulator/sys/common/erl_poll.h
index d02ed2396b..ae2d063805 100644
--- a/erts/emulator/sys/common/erl_poll.h
+++ b/erts/emulator/sys/common/erl_poll.h
@@ -98,6 +98,8 @@
 #  endif
 #endif
 
+#define ERTS_POLL_USE_TIMERFD 0
+
 typedef Uint32 ErtsPollEvents;
 #undef ERTS_POLL_EV_E2N
 
@@ -130,6 +132,12 @@ struct erts_sys_fd_type {
 
 #include <sys/epoll.h>
 
+#ifdef HAVE_SYS_TIMERFD_H
+#include <sys/timerfd.h>
+#undef ERTS_POLL_USE_TIMERFD
+#define ERTS_POLL_USE_TIMERFD 1
+#endif
+
 #define ERTS_POLL_EV_E2N(EV) \
   ((__uint32_t) (EV))
 #define ERTS_POLL_EV_N2E(EV) \