Use critical sections as mutex implementation on Windows

Windows native critical sections are now used internally in the
runtime system as mutex implementation. This since they perform
better under extreme contention than our own implementation.

4 files changed, 31 insertions, 3 deletions

diff --git a/erts/emulator/beam/erl_drv_thread.c b/erts/emulator/beam/erl_drv_thread.c
index dc578f6d2a..a49a155701 100644
--- a/erts/emulator/beam/erl_drv_thread.c
+++ b/erts/emulator/beam/erl_drv_thread.c
@@ -158,7 +158,9 @@ erl_drv_mutex_create(char *name)
 				       (sizeof(ErlDrvMutex)
 					+ (name ? sys_strlen(name) + 1 : 0)));
     if (dmtx) {
-	if (ethr_mutex_init(&dmtx->mtx) != 0) {
+	ethr_mutex_opt opt = ETHR_MUTEX_OPT_DEFAULT_INITER;
+	opt.posix_compliant = 1;
+	if (ethr_mutex_init_opt(&dmtx->mtx, &opt) != 0) {
 	    erts_free(ERTS_ALC_T_DRV_MTX, (void *) dmtx);
 	    dmtx = NULL;
 	}
@@ -226,7 +228,9 @@ erl_drv_cond_create(char *name)
 				      (sizeof(ErlDrvCond)
 				       + (name ? sys_strlen(name) + 1 : 0)));
     if (dcnd) {
-	if (ethr_cond_init(&dcnd->cnd) != 0) {
+	ethr_cond_opt opt = ETHR_COND_OPT_DEFAULT_INITER;
+	opt.posix_compliant = 1;
+	if (ethr_cond_init_opt(&dcnd->cnd, &opt) != 0) {
 	    erts_free(ERTS_ALC_T_DRV_CND, (void *) dcnd);
 	    dcnd = NULL;
 	}
diff --git a/erts/emulator/beam/erl_smp.h b/erts/emulator/beam/erl_smp.h
index a89ddfbcc1..63179dfad4 100644
--- a/erts/emulator/beam/erl_smp.h
+++ b/erts/emulator/beam/erl_smp.h
@@ -822,6 +822,16 @@ erts_smp_cnd_wait(erts_smp_cnd_t *cnd, erts_smp_mtx_t *mtx)
 #endif
 }
 
+/*
+ * IMPORTANT note about erts_smp_cnd_signal() and erts_smp_cnd_broadcast()
+ *
+ * POSIX allow a call to `pthread_cond_signal' or `pthread_cond_broadcast'
+ * even though the associated mutex/mutexes isn't/aren't locked by the
+ * caller. Our implementation do not allow that in order to avoid a
+ * performance penalty. That is, all associated mutexes *need* to be
+ * locked by the caller of erts_smp_cnd_signal()/erts_smp_cnd_broadcast()!
+ */
+
 ERTS_GLB_INLINE void
 erts_smp_cnd_signal(erts_smp_cnd_t *cnd)
 {
diff --git a/erts/emulator/beam/erl_threads.h b/erts/emulator/beam/erl_threads.h
index b4b6d0dfd5..065e7077c0 100644
--- a/erts/emulator/beam/erl_threads.h
+++ b/erts/emulator/beam/erl_threads.h
@@ -92,6 +92,8 @@ typedef struct {
 #endif
 } erts_rwmtx_t;
 
+#define ERTS_MTX_OPT_DEFAULT_INITER ETHR_MUTEX_OPT_DEFAULT_INITER
+#define ERTS_CND_OPT_DEFAULT_INITER ETHR_COND_OPT_DEFAULT_INITER
 #define ERTS_RWMTX_OPT_DEFAULT_INITER ETHR_RWMUTEX_OPT_DEFAULT_INITER
 #define ERTS_RWMTX_TYPE_NORMAL ETHR_RWMUTEX_TYPE_NORMAL
 #define ERTS_RWMTX_TYPE_FREQUENT_READ ETHR_RWMUTEX_TYPE_FREQUENT_READ
@@ -1130,6 +1132,16 @@ erts_cnd_wait(erts_cnd_t *cnd, erts_mtx_t *mtx)
 #endif
 }
 
+/*
+ * IMPORTANT note about erts_cnd_signal() and erts_cnd_broadcast()
+ *
+ * POSIX allow a call to `pthread_cond_signal' or `pthread_cond_broadcast'
+ * even though the associated mutex/mutexes isn't/aren't locked by the
+ * caller. Our implementation do not allow that in order to avoid a
+ * performance penalty. That is, all associated mutexes *need* to be
+ * locked by the caller of erts_cnd_signal()/erts_cnd_broadcast()!
+ */
+
 ERTS_GLB_INLINE void
 erts_cnd_signal(erts_cnd_t *cnd)
 {
diff --git a/erts/emulator/sys/win32/sys.c b/erts/emulator/sys/win32/sys.c
index 02d16b83a2..6f33ef7ad6 100644
--- a/erts/emulator/sys/win32/sys.c
+++ b/erts/emulator/sys/win32/sys.c
@@ -216,6 +216,9 @@ void sys_tty_reset(int exit_code)
 void erl_sys_args(int* argc, char** argv)
 {
     char *event_name;
+
+    erts_sys_env_init();
+
     nohup = get_and_remove_option(argc, argv, "-nohup");
 
 #ifdef DEBUG
@@ -3214,7 +3217,6 @@ erts_sys_pre_init(void)
     }
 #endif
     erts_smp_atomic_init_nob(&sys_misc_mem_sz, 0);
-    erts_sys_env_init();
 }
 
 void noinherit_std_handle(DWORD type)