libpthread: New pthread__smt_wait to put CPU in low power for spin.

This is now distinct from pthread__smt_pause, which is for spin lock
backoff with no paired wakeup.

On Arm, there is a single-bit event register per CPU, and there are two
instructions to manage it:

- wfe, wait for event -- if event register is clear, enter low power
  mode and wait until event register is set; then exit low power mode
  and clear event register

- sev, signal event -- sets event register on all CPUs (other
  circumstances like interrupts also set the event register and cause
  wfe to wake)

These can be used to reduce the power consumption of spinning for a
lock, but only if they are actually paired -- if there's no sev, wfe
might hang indefinitely.  Currently only pthread_spin(3) actually
pairs them; the other lock primitives (internal lock, mutex, rwlock)
do not -- they have spin lock backoff loops, but no corresponding
wakeup to cancel a wfe.

It may be worthwhile to teach the other lock primitives to pair
wfe/sev, but that requires some performance measurement to verify
it's actually worthwhile.  So for now, we just make sure not to use
wfe when there's no sev, and keep everything else the same -- this
should fix severe performance degredation in libpthread on Arm
without hurting anything else.

No change in the generated code on amd64 and i386.  No change in the
generated code for pthread_spin.c on arm and aarch64 -- changes only
the generated code for pthread_lock.c, pthread_mutex.c, and
pthread_rwlock.c, as intended.

PR port-arm/57437

XXX pullup-10

lib/libpthread/arch/aarch64/pthread_md.h

@@ -1,4 +1,4 @@
-/* $NetBSD: pthread_md.h,v 1.1 2014/08/10 05:47:37 matt Exp $ */
+/* $NetBSD: pthread_md.h,v 1.2 2023/05/25 14:30:02 riastradh Exp $ */
 
 /*-
  * Copyright (c) 2014 The NetBSD Foundation, Inc.
@@ -42,7 +42,7 @@ pthread__sp(void)
 	return ret;
 }
 
-#define pthread__smt_pause()	__asm __volatile("wfe") /* wfe */
+#define pthread__smt_wait()	__asm __volatile("wfe") /* wfe */
 #define pthread__smt_wake()	__asm __volatile("sev") /* sev */
 
 #define	pthread__uc_sp(ucp)	((ucp)->uc_mcontext.__gregs[_REG_SP])

lib/libpthread/arch/arm/pthread_md.h

@@ -1,4 +1,4 @@
-/*	$NetBSD: pthread_md.h,v 1.12 2023/05/25 14:29:45 riastradh Exp $	*/
+/*	$NetBSD: pthread_md.h,v 1.13 2023/05/25 14:30:02 riastradh Exp $	*/
 
 /*
  * Copyright (c) 2001 Wasabi Systems, Inc.
@@ -50,13 +50,13 @@ pthread__sp(void)
 }
 
 #if defined(__thumb__) && defined(_ARM_ARCH_6)
-#define pthread__smt_pause()	__asm __volatile(".inst.n 0xbf20") /* wfe */
+#define pthread__smt_wait()	__asm __volatile(".inst.n 0xbf20") /* wfe */
 #define pthread__smt_wake()	__asm __volatile(".inst.n 0xbf40") /* sev */
 #elif !defined(__thumb__)
-#define pthread__smt_pause()	__asm __volatile(".inst 0xe320f002") /* wfe */
+#define pthread__smt_wait()	__asm __volatile(".inst 0xe320f002") /* wfe */
 #define pthread__smt_wake()	__asm __volatile(".inst 0xe320f004") /* sev */
 #else
-#define pthread__smt_pause()	__nothing
+#define pthread__smt_wait()	__nothing
 #define pthread__smt_wake()	__nothing
 #endif
 

lib/libpthread/arch/i386/pthread_md.h

@@ -1,4 +1,4 @@
-/*	$NetBSD: pthread_md.h,v 1.20 2012/03/02 23:19:47 joerg Exp $	*/
+/*	$NetBSD: pthread_md.h,v 1.21 2023/05/25 14:30:02 riastradh Exp $	*/
 
 /*-
  * Copyright (c) 2001, 2007, 2008 The NetBSD Foundation, Inc.
@@ -61,6 +61,7 @@ _initcontext_u_md(ucontext_t *ucp)
 #define	_INITCONTEXT_U_MD(ucp)	_initcontext_u_md(ucp);
 
 #define	pthread__smt_pause()	__asm __volatile("rep; nop" ::: "memory")
+#define	pthread__smt_wait()	__asm __volatile("rep; nop" ::: "memory")
 
 /* Don't need additional memory barriers. */
 #define	PTHREAD__ATOMIC_IS_MEMBAR

lib/libpthread/arch/x86_64/pthread_md.h

@@ -1,4 +1,4 @@
-/*	$NetBSD: pthread_md.h,v 1.12 2011/01/25 19:12:06 christos Exp $	*/
+/*	$NetBSD: pthread_md.h,v 1.13 2023/05/25 14:30:03 riastradh Exp $	*/
 
 /*-
  * Copyright (c) 2001, 2007, 2008 The NetBSD Foundation, Inc.
@@ -64,6 +64,7 @@ pthread__sp(void)
 	(ucp)->uc_mcontext.__gregs[_REG_RFL] = 0x202;
 
 #define	pthread__smt_pause()	__asm __volatile("rep; nop" ::: "memory")
+#define	pthread__smt_wait()	__asm __volatile("rep; nop" ::: "memory")
 
 /* Don't need additional memory barriers. */
 #define	PTHREAD__ATOMIC_IS_MEMBAR

lib/libpthread/pthread_int.h

@@ -1,4 +1,4 @@
-/*	$NetBSD: pthread_int.h,v 1.110 2023/05/25 14:29:45 riastradh Exp $	*/
+/*	$NetBSD: pthread_int.h,v 1.111 2023/05/25 14:30:03 riastradh Exp $	*/
 
 /*-
  * Copyright (c) 2001, 2002, 2003, 2006, 2007, 2008, 2020
@@ -303,6 +303,9 @@ int	pthread__add_specific(pthread_t, pthread_key_t, const void *) PTHREAD_HIDE;
 #ifndef pthread__smt_pause
 #define	pthread__smt_pause()	__nothing
 #endif
+#ifndef pthread__smt_wait
+#define	pthread__smt_wait()	__nothing
+#endif
 #ifndef pthread__smt_wake
 #define	pthread__smt_wake()	__nothing
 #endif

lib/libpthread/pthread_spin.c

@@ -1,4 +1,4 @@
-/*	$NetBSD: pthread_spin.c,v 1.10 2022/04/10 10:38:33 riastradh Exp $	*/
+/*	$NetBSD: pthread_spin.c,v 1.11 2023/05/25 14:30:03 riastradh Exp $	*/
 
 /*-
  * Copyright (c) 2001, 2006, 2007 The NetBSD Foundation, Inc.
@@ -34,7 +34,7 @@
  */
 
 #include <sys/cdefs.h>
-__RCSID("$NetBSD: pthread_spin.c,v 1.10 2022/04/10 10:38:33 riastradh Exp $");
+__RCSID("$NetBSD: pthread_spin.c,v 1.11 2023/05/25 14:30:03 riastradh Exp $");
 
 /* Need to use libc-private names for atomic operations. */
 #include "../../common/lib/libc/atomic/atomic_op_namespace.h"
@@ -98,7 +98,7 @@ pthread_spin_lock(pthread_spinlock_t *lock)
 
 	self = pthread__self();
 	while (pthread__spintrylock(self, &lock->pts_spin) == 0) {
-		pthread__smt_pause();
+		pthread__smt_wait();
 	}
 
 	return 0;