!916 Optimize pthread_once performance

Merge pull request !916 from yinchuang/opt_pthread_once

musl_src.gni

@@ -2188,6 +2188,7 @@ musl_src_porting_file = [
   "src/conf/legacy.c",
   "src/conf/sysconf.c",
   "src/env/getenv.c",
+  "src/thread/pthread_once.c",
 ]
 
 musl_inc_hook_files = [

musl_template.gni

@@ -393,6 +393,15 @@ template("musl_libs") {
     }
 
     defines = []
+    if (musl_arch == "arm") {
+      defines += [ "MUSL_ARM_ARCH" ]
+    }
+    if (musl_arch == "aarch64") {
+      defines += [ "MUSL_AARCH64_ARCH" ]
+    }
+    if (musl_arch == "x86_64") {
+      defines += [ "MUSL_X86_64_ARCH" ]
+    }
     if (musl_secure_level > 0) {
       defines += [ "MALLOC_FREELIST_HARDENED" ]
     }

porting/linux/user/src/thread/pthread_once.c

+#include "pthread_impl.h"
+
+static void undo(void *control)
+{
+	/* Wake all waiters, since the waiter status is lost when
+	 * resetting control to the initial state. */
+	if (a_swap(control, 0) == 3)
+		__wake(control, -1, 1);
+}
+
+hidden int __pthread_once_full(pthread_once_t *control, void (*init)(void))
+{
+	/* Try to enter initializing state. Four possibilities:
+	 *  0 - we're the first or the other cancelled; run init
+	 *  1 - another thread is running init; wait
+	 *  2 - another thread finished running init; just return
+	 *  3 - another thread is running init, waiters present; wait */
+
+	for (;;) switch (a_cas(control, 0, 1)) {
+	case 0:
+		pthread_cleanup_push(undo, control);
+		init();
+		pthread_cleanup_pop(0);
+
+		if (a_swap(control, 2) == 3)
+			__wake(control, -1, 1);
+		return 0;
+	case 1:
+		/* If this fails, so will __wait. */
+		a_cas(control, 1, 3);
+	case 3:
+		__wait(control, 0, 3, 1);
+		continue;
+	case 2:
+		return 0;
+	}
+}
+
+#if defined(MUSL_AARCH64_ARCH) || defined(MUSL_ARM_ARCH)
+int __pthread_once(pthread_once_t *control, void (*init)(void))
+{
+	/* Return immediately if init finished before, use load aquire to ensure that
+	 * effects of the init routine are visible to the caller. */
+	if (a_ll((volatile int *)control) == 2) {
+		return 0;
+	}
+	return __pthread_once_full(control, init);
+}
+#else
+int __pthread_once(pthread_once_t *control, void (*init)(void))
+{
+	/* Return immediately if init finished before, but ensure that
+	 * effects of the init routine are visible to the caller. */
+	if (*(volatile int *)control == 2) {
+		a_barrier();
+		return 0;
+	}
+	return __pthread_once_full(control, init);
+}
+#endif
+
+weak_alias(__pthread_once, pthread_once);