Fix ethread events with timeout

Lots of pthread platforms unnecessarily falled back on the pipe/select
solution. This since we tried to use the same monotonic clock source
for pthread_cond_timedwait() as used by OS monotonic time. This has
been fixed on most platforms by using another clock source.

Darwin can however not use pthread_cond_timedwait() with monotonic
clock source and has to use the pipe/select solution. On darwin we
now use select with _DARWIN_UNLIMITED_SELECT in order to be able to
handle a large amount of file descriptors.

8 files changed, 130 insertions, 18 deletions

diff --git a/erts/aclocal.m4 b/erts/aclocal.m4
index 0714ce6030..7550561737 100644
--- a/erts/aclocal.m4
+++ b/erts/aclocal.m4
@@ -750,8 +750,8 @@ AC_DEFUN(ERL_MONOTONIC_CLOCK,
 	prefer_resolution_clock_gettime_monotonic="$2"
 	;;
     *)
-	check_msg=""
-	prefer_resolution_clock_gettime_monotonic=
+	check_msg="custom "
+	prefer_resolution_clock_gettime_monotonic="$2"
 	;;
   esac
 
@@ -1472,7 +1472,7 @@ AC_ARG_WITH(with_sparc_memory_order,
 LM_CHECK_THR_LIB
 ERL_INTERNAL_LIBS
 
-ERL_MONOTONIC_CLOCK(high_resolution, undefined, no)
+ERL_MONOTONIC_CLOCK(try_find_pthread_compatible, CLOCK_HIGHRES CLOCK_MONOTONIC, no)
 
 case $erl_monotonic_clock_func in
   clock_gettime)
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index ee1dd36d48..58698173be 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -7938,11 +7938,16 @@ sched_thread_func(void *vesdp)
     ErtsThrPrgrCallbacks callbacks;
     ErtsSchedulerData *esdp = vesdp;
     Uint no = esdp->no;
+#ifdef ERTS_SMP
+    erts_tse_t *tse;
+#endif
 
     erts_sched_init_time_sup(esdp);
 
 #ifdef ERTS_SMP
-    ERTS_SCHED_SLEEP_INFO_IX(no - 1)->event = erts_tse_fetch();
+    tse = erts_tse_fetch();
+    erts_tse_prepare_timed(tse);
+    ERTS_SCHED_SLEEP_INFO_IX(no - 1)->event = tse;
     callbacks.arg = (void *) esdp->ssi;
     callbacks.wakeup = thr_prgr_wakeup;
     callbacks.prepare_wait = thr_prgr_prep_wait;
diff --git a/erts/emulator/beam/erl_threads.h b/erts/emulator/beam/erl_threads.h
index 5347979372..34f91e2ec8 100644
--- a/erts/emulator/beam/erl_threads.h
+++ b/erts/emulator/beam/erl_threads.h
@@ -649,6 +649,7 @@ ERTS_GLB_INLINE void erts_tsd_set(erts_tsd_key_t key, void *value);
 ERTS_GLB_INLINE void * erts_tsd_get(erts_tsd_key_t key);
 ERTS_GLB_INLINE erts_tse_t *erts_tse_fetch(void);
 ERTS_GLB_INLINE void erts_tse_return(erts_tse_t *ep);
+ERTS_GLB_INLINE void erts_tse_prepare_timed(erts_tse_t *ep);
 ERTS_GLB_INLINE void erts_tse_set(erts_tse_t *ep);
 ERTS_GLB_INLINE void erts_tse_reset(erts_tse_t *ep);
 ERTS_GLB_INLINE int erts_tse_wait(erts_tse_t *ep);
@@ -3461,6 +3462,15 @@ ERTS_GLB_INLINE void erts_tse_return(erts_tse_t *ep)
 #endif
 }
 
+ERTS_GLB_INLINE void erts_tse_prepare_timed(erts_tse_t *ep)
+{
+#ifdef USE_THREADS
+    int res = ethr_event_prepare_timed(&((ethr_ts_event *) ep)->event);
+    if (res != 0)
+	erts_thr_fatal_error(res, "prepare timed");
+#endif
+}
+
 ERTS_GLB_INLINE void erts_tse_set(erts_tse_t *ep)
 {
 #ifdef USE_THREADS
diff --git a/erts/emulator/drivers/unix/unix_efile.c b/erts/emulator/drivers/unix/unix_efile.c
index 06ba986044..46eccc6568 100644
--- a/erts/emulator/drivers/unix/unix_efile.c
+++ b/erts/emulator/drivers/unix/unix_efile.c
@@ -45,10 +45,10 @@
 #endif
 
 #if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__)
-#define DARWIN 1
+#define __DARWIN__ 1
 #endif
 
-#if defined(DARWIN) || defined(HAVE_LINUX_FALLOC_H) || defined(HAVE_POSIX_FALLOCATE)
+#if defined(__DARWIN__) || defined(HAVE_LINUX_FALLOC_H) || defined(HAVE_POSIX_FALLOCATE)
 #include <fcntl.h>
 #endif
 
@@ -476,11 +476,11 @@ efile_fsync(Efile_error *errInfo, /* Where to return error codes. */
 #ifdef NO_FSYNC
   undefined fsync /* XXX: Really? */
 #else
-#if defined(DARWIN) && defined(F_FULLFSYNC)
+#if defined(__DARWIN__) && defined(F_FULLFSYNC)
     return check_error(fcntl(fd, F_FULLFSYNC), errInfo);
 #else
     return check_error(fsync(fd), errInfo);
-#endif /* DARWIN */
+#endif /* __DARWIN__ */
 #endif /* NO_FSYNC */
 }
 
@@ -962,7 +962,7 @@ efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd,
 	  retval = len;
       }
     } while (len == SENDFILE_CHUNK_SIZE);
-#elif defined(DARWIN)
+#elif defined(__DARWIN__)
     int retval;
     off_t len;
     do {
diff --git a/erts/include/internal/pthread/ethr_event.h b/erts/include/internal/pthread/ethr_event.h
index 74cfa68e16..deb4b29686 100644
--- a/erts/include/internal/pthread/ethr_event.h
+++ b/erts/include/internal/pthread/ethr_event.h
@@ -83,12 +83,22 @@ ETHR_INLINE_FUNC_NAME_(ethr_event_reset)(ethr_event *e)
 #elif defined(ETHR_PTHREADS)
 /* --- Posix mutex/cond pipe/select implementation of events ---------------- */
 
+#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__)
+#  define __DARWIN__ 1
+#endif
+
+#ifdef __DARWIN__
+typedef struct ethr_event_fdsets___ ethr_event_fdsets__;
+#endif
 
 typedef struct {
     ethr_atomic32_t state;
     pthread_mutex_t mtx;
     pthread_cond_t cnd;
     int fd[2];
+#ifdef __DARWIN__
+    ethr_event_fdsets__ *fdsets;
+#endif
 } ethr_event;
 
 #define ETHR_EVENT_OFF_WAITER_SELECT__	((ethr_sint32_t) -2)
@@ -148,6 +158,7 @@ ETHR_INLINE_FUNC_NAME_(ethr_event_reset)(ethr_event *e)
 #endif
 
 int ethr_event_init(ethr_event *e);
+int ethr_event_prepare_timed(ethr_event *e);
 int ethr_event_destroy(ethr_event *e);
 int ethr_event_wait(ethr_event *e);
 int ethr_event_swait(ethr_event *e, int spincount);
diff --git a/erts/include/internal/win/ethr_event.h b/erts/include/internal/win/ethr_event.h
index bf110e10f9..458565b9ea 100644
--- a/erts/include/internal/win/ethr_event.h
+++ b/erts/include/internal/win/ethr_event.h
@@ -56,6 +56,7 @@ ETHR_INLINE_FUNC_NAME_(ethr_event_reset)(ethr_event *e)
 #endif
 
 int ethr_event_init(ethr_event *e);
+int ethr_event_prepare_timed(ethr_event *e);
 int ethr_event_destroy(ethr_event *e);
 int ethr_event_wait(ethr_event *e);
 int ethr_event_swait(ethr_event *e, int spincount);
diff --git a/erts/lib_src/pthread/ethr_event.c b/erts/lib_src/pthread/ethr_event.c
index ba664236f6..0629b4dfcd 100644
--- a/erts/lib_src/pthread/ethr_event.c
+++ b/erts/lib_src/pthread/ethr_event.c
@@ -29,6 +29,13 @@
 #include "config.h"
 #endif
 
+#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__)
+#  define __DARWIN__ 1
+#endif
+#ifdef __DARWIN__
+#  define _DARWIN_UNLIMITED_SELECT
+#endif
+
 #include "ethread.h"
 #undef ETHR_INCLUDE_MONOTONIC_CLOCK__
 #define ETHR_INCLUDE_MONOTONIC_CLOCK__
@@ -56,6 +63,12 @@ ethr_event_init(ethr_event *e)
 }
 
 int
+ethr_event_prepare_timed(ethr_event *e)
+{
+    return 0;
+}
+
+int
 ethr_event_destroy(ethr_event *e)
 {
     return 0;
@@ -171,6 +184,17 @@ return_event_on:
 #include <errno.h>
 #include <string.h>
 
+#ifdef __DARWIN__
+
+struct ethr_event_fdsets___ {
+    fd_set *rsetp;
+    fd_set *esetp;
+    size_t mem_size;
+    fd_mask mem[1];
+};
+
+#endif
+
 static void
 setup_nonblocking_pipe(ethr_event *e)
 {
@@ -188,6 +212,35 @@ setup_nonblocking_pipe(ethr_event *e)
     flgs = fcntl(e->fd[1], F_GETFL, 0);
     fcntl(e->fd[1], F_SETFL, flgs | O_NONBLOCK);
 
+
+#ifndef __DARWIN__
+    if (e->fd[0] >= FD_SETSIZE)
+	ETHR_FATAL_ERROR__(ENOTSUP);
+#else
+    {
+	int nmasks;
+	ethr_event_fdsets__ *fdsets;
+	size_t mem_size;
+
+	nmasks = (e->fd[0]+NFDBITS)/NFDBITS;
+	mem_size = 2*nmasks*sizeof(fd_mask);
+	if (mem_size < 2*sizeof(fd_set)) {
+	    mem_size = 2*sizeof(fd_set);
+	    nmasks = mem_size/(2*sizeof(fd_mask));
+	}
+
+	fdsets = malloc(sizeof(ethr_event_fdsets__)
+			+ mem_size
+			- sizeof(fd_mask));
+	if (!fdsets)
+	    ETHR_FATAL_ERROR__(ENOMEM);
+	fdsets->rsetp = (fd_set *) (char *) &fdsets->mem[0];
+	fdsets->esetp = (fd_set *) (char *) &fdsets->mem[nmasks];
+	fdsets->mem_size = mem_size;
+	e->fdsets = fdsets;
+    }
+#endif
+
     ETHR_MEMBAR(ETHR_StoreStore);
 }
 
@@ -252,6 +305,19 @@ ethr_event_init(ethr_event *e)
 
     e->fd[0] = e->fd[1] = ETHR_EVENT_INVALID_FD__;
 
+#ifdef __DARWIN__
+    e->fdsets = NULL;
+#endif
+
+    return 0;
+}
+
+int
+ethr_event_prepare_timed(ethr_event *e)
+{
+    if (e->fd[0] == ETHR_EVENT_INVALID_FD__)
+	setup_nonblocking_pipe(e);
+
     return 0;
 }
 
@@ -263,13 +329,14 @@ ethr_event_destroy(ethr_event *e)
 	close(e->fd[0]);
 	close(e->fd[1]);
     }
+#ifdef __DARWIN__
+    if (e->fdsets)
+	free(e->fdsets);
+#endif
     res = pthread_mutex_destroy(&e->mtx);
     if (res != 0)
 	return res;
-    res = pthread_cond_destroy(&e->cnd);
-    if (res != 0)
-	return res;
-    return 0;
+    return pthread_cond_destroy(&e->cnd);
 }
 
 static ETHR_INLINE int
@@ -403,8 +470,10 @@ wait__(ethr_event *e, int spincount, ethr_sint64_t timeout)
 	int fd;
 	int sres;
 	ssize_t rres;
-	fd_set rset;
-	fd_set eset;
+#ifndef __DARWIN__
+	fd_set rset, eset;
+#endif
+	fd_set *rsetp, *esetp;
 	struct timeval select_timeout;
 
 #ifdef ETHR_HAVE_ETHR_GET_MONOTONIC_TIME
@@ -457,12 +526,22 @@ wait__(ethr_event *e, int spincount, ethr_sint64_t timeout)
 	    ETHR_ASSERT(act == val);
 	}
 
+
+#ifdef __DARWIN__
+	rsetp = e->fdsets->rsetp;
+	esetp = e->fdsets->esetp;
+	memset((void *) &e->fdsets->mem[0], 0, e->fdsets->mem_size);
+#else
 	FD_ZERO(&rset);
-	FD_SET(fd, &rset);
 	FD_ZERO(&eset);
-	FD_SET(fd, &eset);
+	rsetp = &rset;
+	esetp = &eset;
+#endif
+
+	FD_SET(fd, rsetp);
+	FD_SET(fd, esetp);
 
-	sres = select(fd + 1, &rset, NULL, &eset, &select_timeout);
+	sres = select(fd + 1, rsetp, NULL, esetp, &select_timeout);
 	if (sres == 0)
 	    res = ETIMEDOUT;
 	else {
diff --git a/erts/lib_src/win/ethr_event.c b/erts/lib_src/win/ethr_event.c
index c88c8784a2..6951a216c5 100644
--- a/erts/lib_src/win/ethr_event.c
+++ b/erts/lib_src/win/ethr_event.c
@@ -47,6 +47,12 @@ ethr_event_init(ethr_event *e)
 }
 
 int
+ethr_event_prepare_timed(ethr_event *e)
+{
+    return 0;
+}
+
+int
 ethr_event_destroy(ethr_event *e)
 {
     BOOL res = CloseHandle(e->handle);