Reimplement mutexes, condvars, and rwlocks to eliminate bugs,

particularly the "consume the signal you just sent" hang, and putting
the wait queues in userspace.

Do cancellation handling in pthread_cond_*wait(), pthread_join(),
and sem_wait().

Add __ prefix to thr{sleep,wakeup,exit,sigdivert}() syscalls; add
'abort" argument to thrsleep to close cancellation race; make
thr{sleep,wakeup} return errno values via *retval to avoid touching
userspace errno.

34 files changed, 839 insertions, 390 deletions

diff --git a/lib/libc/arch/amd64/Makefile.inc b/lib/libc/arch/amd64/Makefile.inc
index 2cfeef61734..891bde15d16 100644
--- a/lib/libc/arch/amd64/Makefile.inc
+++ b/lib/libc/arch/amd64/Makefile.inc
@@ -1,4 +1,4 @@
-#	$OpenBSD: Makefile.inc,v 1.3 2005/06/29 14:03:25 mickey Exp $
+#	$OpenBSD: Makefile.inc,v 1.4 2012/01/17 02:34:18 guenther Exp $
 
 KMINCLUDES= arch/amd64/SYS.h
-KMSRCS=	ffs.S strlen.S htonl.S ntohl.S htons.S ntohs.S
+KMSRCS=	ffs.S htonl.S ntohl.S htons.S ntohs.S
diff --git a/lib/libc/arch/amd64/string/Makefile.inc b/lib/libc/arch/amd64/string/Makefile.inc
index a384f0f20ae..e9662393875 100644
--- a/lib/libc/arch/amd64/string/Makefile.inc
+++ b/lib/libc/arch/amd64/string/Makefile.inc
@@ -1,8 +1,8 @@
-#	$OpenBSD: Makefile.inc,v 1.2 2007/05/15 18:42:31 otto Exp $
+#	$OpenBSD: Makefile.inc,v 1.3 2012/01/17 02:34:18 guenther Exp $
 
 SRCS+=  bcmp.c ffs.S index.c memchr.c memcmp.c bcopy.c bzero.c \
-        rindex.c strcat.c strcmp.c strcpy.c strcspn.c strlen.S \
+        rindex.c strcat.c strcmp.c strcpy.c strcspn.c strlen.c \
         strncat.c strncmp.c strncpy.c strpbrk.c strsep.c \
         strspn.c strstr.c swab.c memset.c strlcpy.c strlcat.c
 
-LSRCS+=	ffs.c strlen.c
+LSRCS+=	ffs.c
diff --git a/lib/libc/shlib_version b/lib/libc/shlib_version
index 73198ebb5bc..d9553a8cf66 100644
--- a/lib/libc/shlib_version
+++ b/lib/libc/shlib_version
@@ -1,4 +1,4 @@
-major=61
+major=62
 minor=0
 # note: If changes were made to include/thread_private.h or if system
 # calls were added/changed then libpthread must also be updated.
diff --git a/lib/libc/sys/Makefile.inc b/lib/libc/sys/Makefile.inc
index 2902075c49e..affed8158c4 100644
--- a/lib/libc/sys/Makefile.inc
+++ b/lib/libc/sys/Makefile.inc
@@ -1,4 +1,4 @@
-#	$OpenBSD: Makefile.inc,v 1.96 2011/11/22 21:13:30 guenther Exp $
+#	$OpenBSD: Makefile.inc,v 1.97 2012/01/17 02:34:18 guenther Exp $
 #	$NetBSD: Makefile.inc,v 1.35 1995/10/16 23:49:07 jtc Exp $
 #	@(#)Makefile.inc	8.1 (Berkeley) 6/17/93
 
@@ -66,7 +66,7 @@ ASM=	accept.o access.o acct.o adjfreq.o adjtime.o bind.o chdir.o chflags.o \
 	symlink.o sync.o sysarch.o umask.o unlink.o unmount.o \
 	utimes.o wait4.o write.o writev.o nnpfspioctl.o __semctl.o \
 	__syscall.o __sysctl.o __getcwd.o sched_yield.o getthrid.o \
-	thrsleep.o thrwakeup.o threxit.o thrsigdivert.o \
+	__thrsleep.o __thrwakeup.o __threxit.o __thrsigdivert.o \
 	setrtable.o getrtable.o __set_tcb.o __get_tcb.o \
 	openat.o fchmodat.o fstatat.o mkdirat.o mkfifoat.o mknodat.o \
 	faccessat.o fchownat.o linkat.o readlinkat.o renameat.o symlinkat.o \
diff --git a/lib/librthread/arch/alpha/rfork_thread.S b/lib/librthread/arch/alpha/rfork_thread.S
index 3205958f6a9..3cd2ab68106 100644
--- a/lib/librthread/arch/alpha/rfork_thread.S
+++ b/lib/librthread/arch/alpha/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.3 2011/10/17 06:39:20 guenther Exp $	*/
+/*	$OpenBSD: rfork_thread.S,v 1.4 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * Copyright (c) 2005, Miodrag Vallat
@@ -52,6 +52,6 @@ LEAF(__tfork_thread,0)
 	jsr	ra, (pv)
 
 	mov	zero, a0
-	CALLSYS_NOERROR(threxit)
+	CALLSYS_NOERROR(__threxit)
 	
 END(__tfork_thread)
diff --git a/lib/librthread/arch/amd64/rfork_thread.S b/lib/librthread/arch/amd64/rfork_thread.S
index 77b62ef8005..d5bd9a428eb 100644
--- a/lib/librthread/arch/amd64/rfork_thread.S
+++ b/lib/librthread/arch/amd64/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.5 2011/10/17 06:39:20 guenther Exp $ */
+/*	$OpenBSD: rfork_thread.S,v 1.6 2012/01/17 02:34:18 guenther Exp $ */
 /*-
  * Copyright (c) 2000 Peter Wemm <peter@FreeBSD.org>
  * Copyright (c) 2003 Alan L. Cox <alc@cs.rice.edu>
@@ -68,7 +68,7 @@ ENTRY(__tfork_thread)
 	/*
 	 * If we are in the child (new thread), then
 	 * set-up the call to the internal subroutine.  If it
-	 * returns, then call threxit.
+	 * returns, then call __threxit.
 	 */
 1:
 	movq	%rsi, %rsp
@@ -78,7 +78,7 @@ ENTRY(__tfork_thread)
 	/*
 	 * Thread exit system call
 	 */
-	movl	$SYS_threxit, %eax
+	movl	$SYS___threxit, %eax
 	xorl	%edi, %edi
 	syscall
 
diff --git a/lib/librthread/arch/arm/rfork_thread.S b/lib/librthread/arch/arm/rfork_thread.S
index a631b1b8ce3..749ba0490fa 100644
--- a/lib/librthread/arch/arm/rfork_thread.S
+++ b/lib/librthread/arch/arm/rfork_thread.S
@@ -1,4 +1,4 @@
-/* $OpenBSD: rfork_thread.S,v 1.3 2011/10/17 06:39:20 guenther Exp $ */
+/* $OpenBSD: rfork_thread.S,v 1.4 2012/01/17 02:34:18 guenther Exp $ */
 /*
  * Copyright (c) 2005 Dale Rahn <drahn@openbsd.org>
  *
@@ -42,7 +42,7 @@ ENTRY(__tfork_thread)
 	mov	lr, pc
 	mov	pc, r2
 	nop
-	SYSTRAP(threxit)
+	SYSTRAP(__threxit)
 1:
 	ldmia   sp!, {r4}
 	b     PIC_SYM(CERROR, PLT)
diff --git a/lib/librthread/arch/hppa/rfork_thread.S b/lib/librthread/arch/hppa/rfork_thread.S
index c98d4ee3bac..e29488db4d1 100644
--- a/lib/librthread/arch/hppa/rfork_thread.S
+++ b/lib/librthread/arch/hppa/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.2 2011/10/17 06:39:20 guenther Exp $	*/
+/*	$OpenBSD: rfork_thread.S,v 1.3 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * Copyright (c) 2005, Miodrag Vallat
@@ -56,7 +56,7 @@ ENTRY(__tfork_thread, 0)
 	copy	r31, rp
 
 	copy	r0, arg0
-	SYSCALL(threxit)
+	SYSCALL(__threxit)
 
 1:
 	bv	r0(rp)
diff --git a/lib/librthread/arch/i386/rfork_thread.S b/lib/librthread/arch/i386/rfork_thread.S
index 1bc1d3e9e32..9fe8da9310c 100644
--- a/lib/librthread/arch/i386/rfork_thread.S
+++ b/lib/librthread/arch/i386/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.4 2011/10/17 06:39:20 guenther Exp $ */
+/*	$OpenBSD: rfork_thread.S,v 1.5 2012/01/17 02:34:18 guenther Exp $ */
 /*-
  * Copyright (c) 2000 Peter Wemm <peter@FreeBSD.org>
  * All rights reserved.
@@ -101,7 +101,7 @@ ENTRY(__tfork_thread)
 	 */
 	pushl	%eax
 	pushl	$0
-	movl	$SYS_threxit, %eax
+	movl	$SYS___threxit, %eax
 	int	$0x80
 
 	/*
diff --git a/lib/librthread/arch/m68k/rfork_thread.S b/lib/librthread/arch/m68k/rfork_thread.S
index 80c682baf50..3931a4a8eb1 100644
--- a/lib/librthread/arch/m68k/rfork_thread.S
+++ b/lib/librthread/arch/m68k/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.2 2011/10/17 06:39:20 guenther Exp $	*/
+/*	$OpenBSD: rfork_thread.S,v 1.3 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * Copyright (c) 2005, Miodrag Vallat
@@ -61,7 +61,7 @@ ENTRY(__tfork_thread)
 	jsr	a1@		/* func */
 	addq	#4, sp
 
-	__DO_SYSCALL(threxit)
+	__DO_SYSCALL(__threxit)
 
 9:
 	/*
diff --git a/lib/librthread/arch/m88k/rfork_thread.S b/lib/librthread/arch/m88k/rfork_thread.S
index 2c820232a40..d8f7657bd30 100644
--- a/lib/librthread/arch/m88k/rfork_thread.S
+++ b/lib/librthread/arch/m88k/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.2 2011/10/17 06:39:20 guenther Exp $	*/
+/*	$OpenBSD: rfork_thread.S,v 1.3 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * Copyright (c) 2005, Miodrag Vallat
@@ -51,5 +51,5 @@ ENTRY(__tfork_thread)
 	jsr.n	r4		/* func */
 	 or	r2,  r5,  r0	/* arg */
 
-	or	r13, r0,  __SYSCALLNAME(SYS_,threxit)
+	or	r13, r0,  __SYSCALLNAME(SYS_,__threxit)
 	tb0	0,   r0,  128
diff --git a/lib/librthread/arch/mips64/rfork_thread.S b/lib/librthread/arch/mips64/rfork_thread.S
index 32ee5307bee..adb73dbc24d 100644
--- a/lib/librthread/arch/mips64/rfork_thread.S
+++ b/lib/librthread/arch/mips64/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.2 2011/10/17 06:39:20 guenther Exp $	*/
+/*	$OpenBSD: rfork_thread.S,v 1.3 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * Copyright (c) 2005, Miodrag Vallat
@@ -64,7 +64,7 @@ LEAF(__tfork_thread, 32)
 	 move	v0, zero
 
 	move	a0, zero
-	__DO_SYSCALL(threxit)
+	__DO_SYSCALL(__threxit)
 
 9:
 	/*
diff --git a/lib/librthread/arch/powerpc/rfork_thread.S b/lib/librthread/arch/powerpc/rfork_thread.S
index 5731553fbe7..103ff9f10a0 100644
--- a/lib/librthread/arch/powerpc/rfork_thread.S
+++ b/lib/librthread/arch/powerpc/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.4 2011/10/17 06:39:20 guenther Exp $	*/
+/*	$OpenBSD: rfork_thread.S,v 1.5 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * Copyright (c) 2005 Tim Wiess <tim@nop.cx>
@@ -45,7 +45,7 @@ ENTRY(__tfork_thread)
 	blrl
 	
 	/* child returned, call _exit */
-	li	%r0, SYS_threxit
+	li	%r0, SYS___threxit
 	sc
 1:
 	li	%r3, -1
diff --git a/lib/librthread/arch/sh/rfork_thread.S b/lib/librthread/arch/sh/rfork_thread.S
index fb6002477fc..51100162285 100644
--- a/lib/librthread/arch/sh/rfork_thread.S
+++ b/lib/librthread/arch/sh/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.2 2011/10/17 06:39:20 guenther Exp $	*/
+/*	$OpenBSD: rfork_thread.S,v 1.3 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * Copyright (c) 2007 Miodrag Vallat.
@@ -45,7 +45,7 @@ ENTRY(__tfork_thread)
 	jsr	@r6
 	 mov	r7, r4
 
-	mov.l	.LSYS_threxit, r0
+	mov.l	.LSYS___threxit, r0
 	.word	0xc380		/* trapa #0x80 */
 
 9:
@@ -56,6 +56,6 @@ ENTRY(__tfork_thread)
 
 	.align 2
 .LSYS___tfork:	.long SYS___tfork
-.LSYS_threxit:	.long SYS_threxit
+.LSYS___threxit:	.long SYS___threxit
 
 	SET_ENTRY_SIZE(__tfork_thread)
diff --git a/lib/librthread/arch/sparc/rfork_thread.S b/lib/librthread/arch/sparc/rfork_thread.S
index fc4b4f998c7..2feac006077 100644
--- a/lib/librthread/arch/sparc/rfork_thread.S
+++ b/lib/librthread/arch/sparc/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.2 2011/10/17 06:39:20 guenther Exp $	*/
+/*	$OpenBSD: rfork_thread.S,v 1.3 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * Copyright (c) 2005, Miodrag Vallat
@@ -63,7 +63,7 @@ ENTRY(__tfork_thread)
 	call	%o2		/* func */
 	 mov	%o3, %o0	/* arg */
 
-	mov	SYS_threxit, %g1
+	mov	SYS___threxit, %g1
 	clr	%o0
 	t	ST_SYSCALL	/* will not return */
 
diff --git a/lib/librthread/arch/sparc64/rfork_thread.S b/lib/librthread/arch/sparc64/rfork_thread.S
index d35f6ba1d74..86b876db714 100644
--- a/lib/librthread/arch/sparc64/rfork_thread.S
+++ b/lib/librthread/arch/sparc64/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.3 2011/10/17 06:39:20 guenther Exp $	*/
+/*	$OpenBSD: rfork_thread.S,v 1.4 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * Copyright (c) 2005, Miodrag Vallat
@@ -62,7 +62,7 @@ ENTRY(__tfork_thread)
 	call	%o2				/* func */
 	 mov	%o3, %o0			/* arg */
 
-	mov	SYS_threxit, %g1
+	mov	SYS___threxit, %g1
 	clr	%o0
 	t	ST_SYSCALL	/* will not return */
 
diff --git a/lib/librthread/arch/vax/rfork_thread.S b/lib/librthread/arch/vax/rfork_thread.S
index a7e6259b9e0..f5cc4231c58 100644
--- a/lib/librthread/arch/vax/rfork_thread.S
+++ b/lib/librthread/arch/vax/rfork_thread.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rfork_thread.S,v 1.3 2011/10/17 06:39:20 guenther Exp $	*/
+/*	$OpenBSD: rfork_thread.S,v 1.4 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * Copyright (c) 2005, Miodrag Vallat
@@ -60,7 +60,7 @@ ENTRY(__tfork_thread, R2|R3|R4)
 	pushl	r4		/* arg */
 	calls	$1, *4(sp)	/* func */
 
-	__DO_SYSCALL(threxit)
+	__DO_SYSCALL(__threxit)
 
 9:
 	/*
diff --git a/lib/librthread/rthread.c b/lib/librthread/rthread.c
index b4016f656d5..8f761c3896d 100644
--- a/lib/librthread/rthread.c
+++ b/lib/librthread/rthread.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rthread.c,v 1.49 2011/12/28 04:59:31 guenther Exp $ */
+/*	$OpenBSD: rthread.c,v 1.50 2012/01/17 02:34:18 guenther Exp $ */
 /*
  * Copyright (c) 2004,2005 Ted Unangst <tedu@openbsd.org>
  * All Rights Reserved.
@@ -103,9 +103,31 @@ sigthr_handler(__unused int sig)
 {
 	pthread_t self = pthread_self();
 
-	if ((self->flags & (THREAD_CANCELED | THREAD_CANCEL_COND)) ==
-	    THREAD_CANCELED && (self->cancel_point ||
-	    (self->flags & THREAD_CANCEL_DEFERRED) == 0))
+	/*
+	 * Do nothing unless
+	 * 1) pthread_cancel() has been called on this thread,
+	 * 2) cancelation is enabled for it, and
+	 * 3) we're not already in cancelation processing
+	 */
+	if ((self->flags & (THREAD_CANCELED|THREAD_CANCEL_ENABLE|THREAD_DYING))
+	    != (THREAD_CANCELED|THREAD_CANCEL_ENABLE))
+		return;
+
+	/*
+	 * If delaying cancels inside complex ops (pthread_cond_wait,
+	 * pthread_join, etc), just mark that this has happened to
+	 * prevent a race with going to sleep
+	 */
+	if (self->flags & THREAD_CANCEL_DELAY) {
+		self->delayed_cancel = 1;
+		return;
+	}
+
+	/*
+	 * otherwise, if in a cancel point or async cancels are
+	 * enabled, then exit
+	 */
+	if (self->cancel_point || (self->flags & THREAD_CANCEL_DEFERRED) == 0)
 		pthread_exit(PTHREAD_CANCELED);
 }
 
@@ -123,10 +145,12 @@ _rthread_init(void)
 	strlcpy(thread->name, "Main process", sizeof(thread->name));
 	LIST_INSERT_HEAD(&_thread_list, thread, threads);
 	_rthread_debug_init();
-	_rthread_debug(1, "rthread init\n");
+
 	_threads_ready = 1;
 	__isthreaded = 1;
 
+	_rthread_debug(1, "rthread init\n");
+
 #if defined(__ELF__) && defined(PIC)
 	/*
 	 * To avoid recursion problems in ld.so, we need to trigger the
@@ -177,7 +201,7 @@ _rthread_free(pthread_t thread)
 	}
 }
 
-static void
+void
 _rthread_setflag(pthread_t thread, int flag)
 {
 	_spinlock(&thread->flags_lock);
@@ -185,7 +209,7 @@ _rthread_setflag(pthread_t thread, int flag)
 	_spinunlock(&thread->flags_lock);
 }
 
-static void
+void
 _rthread_clearflag(pthread_t thread, int flag)
 {
 	_spinlock(&thread->flags_lock);
@@ -269,34 +293,38 @@ pthread_exit(void *retval)
 		_sem_post(&thread->donesem);
 	}
 
-	threxit(&thread->tid);
+	__threxit(&thread->tid);
 	for(;;);
 }
 
 int
 pthread_join(pthread_t thread, void **retval)
 {
-	int e;
+	int e, r;
 	pthread_t self = pthread_self();
 
+	e = r = 0;
+	_enter_delayed_cancel(self);
 	if (thread == NULL)
 		e = EINVAL;
 	else if (thread == self)
 		e = EDEADLK;
 	else if (thread->flags & THREAD_DETACHED)
 		e = EINVAL;
-	else {
-		_sem_wait(&thread->donesem, 0);
+	else if ((r = _sem_wait(&thread->donesem, 0, &self->delayed_cancel))) {
 		if (retval)
 			*retval = thread->retval;
-		e = 0;
-		/* We should be the last having a ref to this thread, but
-		 * someone stupid or evil might haved detached it;
-		 * in that case the thread will cleanup itself */
+
+		/*
+		 * We should be the last having a ref to this thread,
+		 * but someone stupid or evil might haved detached it;
+		 * in that case the thread will clean up itself
+		 */
 		if ((thread->flags & THREAD_DETACHED) == 0)
 			_rthread_free(thread);
 	}
 
+	_leave_delayed_cancel(self, !r);
 	_rthread_reaper();
 	return (e);
 }
@@ -439,7 +467,6 @@ pthread_setcancelstate(int state, int *oldstatep)
 	    PTHREAD_CANCEL_ENABLE : PTHREAD_CANCEL_DISABLE;
 	if (state == PTHREAD_CANCEL_ENABLE) {
 		_rthread_setflag(self, THREAD_CANCEL_ENABLE);
-		pthread_testcancel();
 	} else if (state == PTHREAD_CANCEL_DISABLE) {
 		_rthread_clearflag(self, THREAD_CANCEL_ENABLE);
 	} else {
diff --git a/lib/librthread/rthread.h b/lib/librthread/rthread.h
index e20bf11d734..391f0746699 100644
--- a/lib/librthread/rthread.h
+++ b/lib/librthread/rthread.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rthread.h,v 1.30 2011/12/21 00:49:47 guenther Exp $ */
+/*	$OpenBSD: rthread.h,v 1.31 2012/01/17 02:34:18 guenther Exp $ */
 /*
  * Copyright (c) 2004,2005 Ted Unangst <tedu@openbsd.org>
  * All Rights Reserved.
@@ -54,7 +54,8 @@ struct sem {
 TAILQ_HEAD(pthread_queue, pthread);
 
 struct pthread_mutex {
-	struct sem sem;
+	_spinlock_lock_t lock;
+	struct pthread_queue lockers;
 	int type;
 	pthread_t owner;
 	int count;
@@ -68,7 +69,9 @@ struct pthread_mutex_attr {
 };
 
 struct pthread_cond {
-	struct sem sem;
+	_spinlock_lock_t lock;
+	struct pthread_queue waiters;
+	struct pthread_mutex *mutex;
 };
 
 struct pthread_cond_attr {
@@ -76,10 +79,10 @@ struct pthread_cond_attr {
 };
 
 struct pthread_rwlock {
-	struct sem sem;
 	_spinlock_lock_t lock;
+	pthread_t owner;
+	struct pthread_queue writers;
 	int readers;
-	int writer;
 };
 
 struct pthread_rwlockattr {
@@ -133,22 +136,32 @@ struct pthread {
 	struct stack *stack;
 	LIST_ENTRY(pthread) threads;
 	TAILQ_ENTRY(pthread) waiting;
+	pthread_cond_t blocking_cond;
 	int sched_policy;
 	struct pthread_attr attr;
 	struct sched_param sched_param;
 	struct rthread_storage *local_storage;
 	struct rthread_cleanup_fn *cleanup_fns;
 	int myerrno;
+
+	/* currently in a cancel point? */
 	int cancel_point;
+
+	/* cancel received in a delayed cancel block? */
+	int delayed_cancel;
 };
 #define	THREAD_DONE		0x001
 #define	THREAD_DETACHED		0x002
 #define THREAD_CANCELED		0x004
 #define THREAD_CANCEL_ENABLE	0x008
 #define THREAD_CANCEL_DEFERRED	0x010
-#define THREAD_CANCEL_COND	0x020
+#define THREAD_CANCEL_DELAY	0x020
 #define THREAD_DYING            0x040
 
+#define	IS_CANCELED(thread) \
+	(((thread)->flags & (THREAD_CANCELED|THREAD_DYING)) == THREAD_CANCELED)
+
+
 extern int _threads_ready;
 extern LIST_HEAD(listhead, pthread) _thread_list;
 extern struct pthread _initial_thread;
@@ -156,12 +169,11 @@ extern _spinlock_lock_t _thread_lock;
 
 void	_spinlock(_spinlock_lock_t *);
 void	_spinunlock(_spinlock_lock_t *);
-int	_sem_wait(sem_t, int);
-int	_sem_waitl(sem_t, int, clockid_t, const struct timespec *);
+int	_sem_wait(sem_t, int, int *);
 int	_sem_post(sem_t);
-int	_sem_wakeup(sem_t);
-int	_sem_wakeall(sem_t);
 
+void	_rthread_setflag(pthread_t, int);
+void	_rthread_clearflag(pthread_t, int);
 struct stack *_rthread_alloc_stack(pthread_t);
 void	_rthread_free_stack(struct stack *);
 void	_rthread_tls_destructors(pthread_t);
@@ -174,19 +186,22 @@ void	_rthread_bind_lock(int);
 #endif
 
 /* rthread_cancel.c */
-void	_leave_cancel(pthread_t);
 void	_enter_cancel(pthread_t);
+void	_leave_cancel(pthread_t);
+void	_enter_delayed_cancel(pthread_t);
+void	_leave_delayed_cancel(pthread_t, int);
 
 void	_thread_dump_info(void);
 
 int	_atomic_lock(register volatile _spinlock_lock_t *);
 
 /* syscalls */
-int getthrid(void);
-void threxit(pid_t *);
-int thrsleep(const volatile void *, clockid_t, const struct timespec *,
-    volatile void *);
-int thrwakeup(void *, int n);
-int sched_yield(void);
-int thrsigdivert(sigset_t, siginfo_t *, const struct timespec *);
-int _thread_sys_sigaction(int, const struct sigaction *, struct sigaction *);
+int	getthrid(void);
+void	__threxit(pid_t *);
+int	__thrsleep(const volatile void *, clockid_t, const struct timespec *,
+	    void *, const int *);
+int	__thrwakeup(const volatile void *, int n);
+int	__thrsigdivert(sigset_t, siginfo_t *, const struct timespec *);
+int	sched_yield(void);
+int	_thread_sys_sigaction(int, const struct sigaction *,
+	    struct sigaction *);
diff --git a/lib/librthread/rthread_cancel.c b/lib/librthread/rthread_cancel.c
index 4d57c8ee530..1db785a9a9e 100644
--- a/lib/librthread/rthread_cancel.c
+++ b/lib/librthread/rthread_cancel.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: rthread_cancel.c,v 1.3 2012/01/04 05:46:38 guenther Exp $ */
+/* $OpenBSD: rthread_cancel.c,v 1.4 2012/01/17 02:34:18 guenther Exp $ */
 /* $snafu: libc_tag.c,v 1.4 2004/11/30 07:00:06 marc Exp $ */
 
 /* PUBLIC DOMAIN: No Rights Reserved. Marco S Hyman <marc@snafu.org> */
@@ -66,8 +66,7 @@ _enter_cancel(pthread_t self)
 {
 	if (self->flags & THREAD_CANCEL_ENABLE) {
 		self->cancel_point++;
-		if ((self->flags & (THREAD_CANCELED | THREAD_DYING)) ==
-		    THREAD_CANCELED)
+		if (IS_CANCELED(self))
 			pthread_exit(PTHREAD_CANCELED);
 	}
 }
@@ -79,6 +78,31 @@ _leave_cancel(pthread_t self)
 		self->cancel_point--;
 }
 
+void
+_enter_delayed_cancel(pthread_t self)
+{
+	if (self->flags & THREAD_CANCEL_ENABLE) {
+		self->delayed_cancel = 0;
+		self->cancel_point++;
+		if (IS_CANCELED(self))
+			pthread_exit(PTHREAD_CANCELED);
+		_rthread_setflag(self, THREAD_CANCEL_DELAY);
+	}
+}
+
+void
+_leave_delayed_cancel(pthread_t self, int can_cancel)
+{
+	if (self->flags & THREAD_CANCEL_ENABLE) {
+		if (self->flags & THREAD_CANCEL_DELAY) {
+			self->cancel_point--;
+			_rthread_clearflag(self, THREAD_CANCEL_DELAY);
+		}
+		if (IS_CANCELED(self) && can_cancel)
+			pthread_exit(PTHREAD_CANCELED);
+		self->delayed_cancel = 0;
+	}
+}
 
 int
 accept(int fd, struct sockaddr *addr, socklen_t *addrlen)
@@ -419,7 +443,7 @@ select(int nfds, fd_set *readfds, fd_set *writefds, fd_set *exceptfds,
 
 #if 0
 sem_timedwait()			/* don't have yet */
-sem_wait()			/* don't have yet */
+sem_wait()			/* in rthread_sem.c */
 send()				/* built on sendto() */
 #endif
 
diff --git a/lib/librthread/rthread_file.c b/lib/librthread/rthread_file.c
index 14190662885..28bad52d0c2 100644
--- a/lib/librthread/rthread_file.c
+++ b/lib/librthread/rthread_file.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rthread_file.c,v 1.3 2011/11/06 11:48:59 guenther Exp $	*/
+/*	$OpenBSD: rthread_file.c,v 1.4 2012/01/17 02:34:18 guenther Exp $	*/
 /*
  * Copyright (c) 1995 John Birrell <jb@cimlogic.com.au>.
  * All rights reserved.
@@ -204,7 +204,7 @@ void
 		 */
 		TAILQ_INSERT_TAIL(&p->lockers,self,waiting);
 		while (p->owner != self) {
-			thrsleep(self, 0, NULL, &hash_lock);
+			__thrsleep(self, 0, NULL, &hash_lock, NULL);
 			_spinlock(&hash_lock);
 		}
 	}
@@ -292,7 +292,7 @@ void
 				 */
 				p->count = 1;
 
-				thrwakeup(p->owner, 1);
+				__thrwakeup(p->owner, 1);
 			}
 		}
 	}
diff --git a/lib/librthread/rthread_rwlock.c b/lib/librthread/rthread_rwlock.c
index 0d391fe9afd..c148089d996 100644
--- a/lib/librthread/rthread_rwlock.c
+++ b/lib/librthread/rthread_rwlock.c
@@ -1,6 +1,7 @@
-/*	$OpenBSD: rthread_rwlock.c,v 1.1 2011/12/21 23:59:03 guenther Exp $ */
+/*	$OpenBSD: rthread_rwlock.c,v 1.2 2012/01/17 02:34:18 guenther Exp $ */
 /*
  * Copyright (c) 2004,2005 Ted Unangst <tedu@openbsd.org>
+ * Copyright (c) 2012 Philip Guenther <guenther@openbsd.org>
  * All Rights Reserved.
  *
  * Permission to use, copy, modify, and distribute this software for any
@@ -20,6 +21,7 @@
  */
 
 
+#include <assert.h>
 #include <stdlib.h>
 #include <unistd.h>
 #include <errno.h>
@@ -31,8 +33,10 @@
 
 static _spinlock_lock_t rwlock_init_lock = _SPINLOCK_UNLOCKED;
 
+/* ARGSUSED1 */
 int
-pthread_rwlock_init(pthread_rwlock_t *lockp, const pthread_rwlockattr_t *attrp)
+pthread_rwlock_init(pthread_rwlock_t *lockp,
+    const pthread_rwlockattr_t *attrp __unused)
 {
 	pthread_rwlock_t lock;
 
@@ -40,7 +44,8 @@ pthread_rwlock_init(pthread_rwlock_t *lockp, const pthread_rwlockattr_t *attrp)
 	if (!lock)
 		return (errno);
 	lock->lock = _SPINLOCK_UNLOCKED;
-	lock->sem.lock = _SPINLOCK_UNLOCKED;
+	TAILQ_INIT(&lock->writers);
+
 	*lockp = lock;
 
 	return (0);
@@ -49,13 +54,19 @@ pthread_rwlock_init(pthread_rwlock_t *lockp, const pthread_rwlockattr_t *attrp)
 int
 pthread_rwlock_destroy(pthread_rwlock_t *lockp)
 {
-	if ((*lockp) && ((*lockp)->readers || (*lockp)->writer)) {
+	pthread_rwlock_t lock;
+
+	assert(lockp);
+	lock = *lockp;
+	if (lock) {
+		if (lock->readers || !TAILQ_EMPTY(&lock->writers)) {
 #define MSG "pthread_rwlock_destroy on rwlock with waiters!\n"
-		write(2, MSG, sizeof(MSG) - 1);
+			write(2, MSG, sizeof(MSG) - 1);
 #undef MSG
-		return (EBUSY);
+			return (EBUSY);
+		}
+		free(lock);
 	}
-	free(*lockp);
 	*lockp = NULL;
 
 	return (0);
@@ -81,86 +92,72 @@ _rthread_rwlock_ensure_init(pthread_rwlock_t *lockp)
 }
 
 
-int
-pthread_rwlock_rdlock(pthread_rwlock_t *lockp)
+static int
+_rthread_rwlock_rdlock(pthread_rwlock_t *lockp, const struct timespec *abstime,
+    int try)
 {
 	pthread_rwlock_t lock;
+	pthread_t thread = pthread_self();
 	int error;
 
 	if ((error = _rthread_rwlock_ensure_init(lockp)))
 		return (error);
 
 	lock = *lockp;
-again:
+	_rthread_debug(5, "%p: rwlock_rdlock %p\n", (void *)thread,
+	    (void *)lock);
 	_spinlock(&lock->lock);
-	if (lock->writer) {
-		_spinlock(&lock->sem.lock);
-		_spinunlock(&lock->lock);
-		_sem_waitl(&lock->sem, 0, 0, NULL);
-		goto again;
+
+	/* writers have precedence */
+	if (lock->owner == NULL && TAILQ_EMPTY(&lock->writers))
+		lock->readers++;
+	else if (try)
+		error = EBUSY;
+	else if (lock->owner == thread)
+		error = EDEADLK;
+	else {
+		do {
+			if (__thrsleep(lock, CLOCK_REALTIME, abstime,
+			    &lock->lock, NULL) == EWOULDBLOCK)
+				return (ETIMEDOUT);
+			_spinlock(&lock->lock);
+		} while (lock->owner != NULL || !TAILQ_EMPTY(&lock->writers));
+		lock->readers++;
 	}
-	lock->readers++;
 	_spinunlock(&lock->lock);
 
-	return (0);
+	return (error);
 }
 
 int
-pthread_rwlock_timedrdlock(pthread_rwlock_t *lockp,
-    const struct timespec *abstime)
+pthread_rwlock_rdlock(pthread_rwlock_t *lockp)
 {
-	pthread_rwlock_t lock;
-	int do_wait = 1;
-	int error;
-
-	if ((error = _rthread_rwlock_ensure_init(lockp)))
-		return (error);
-
-	lock = *lockp;
-	_spinlock(&lock->lock);
-	while (lock->writer && do_wait) {
-		_spinlock(&lock->sem.lock);
-		_spinunlock(&lock->lock);
-		do_wait = _sem_waitl(&lock->sem, 0, CLOCK_REALTIME, abstime);
-		_spinlock(&lock->lock);
-	}
-	if (lock->writer) {
-		/* do_wait must be 0, so timed out */
-		_spinunlock(&lock->lock);
-		return (ETIMEDOUT);
-	}
-	lock->readers++;
-	_spinunlock(&lock->lock);
-
-	return (0);
+	return (_rthread_rwlock_rdlock(lockp, NULL, 0));
 }
 
 int
 pthread_rwlock_tryrdlock(pthread_rwlock_t *lockp)
 {
-	pthread_rwlock_t lock;
-	int error;
-
-	if ((error = _rthread_rwlock_ensure_init(lockp)))
-		return (error);
-
-	lock = *lockp;
-
-	_spinlock(&lock->lock);
-	if (lock->writer) {
-		_spinunlock(&lock->lock);
-		return (EBUSY);
-	}
-	lock->readers++;
-	_spinunlock(&lock->lock);
-
-	return (0);
+	return (_rthread_rwlock_rdlock(lockp, NULL, 1));
 }
 
 int
-pthread_rwlock_wrlock(pthread_rwlock_t *lockp)
+pthread_rwlock_timedrdlock(pthread_rwlock_t *lockp,
+    const struct timespec *abstime)
+{
+	if (abstime == NULL || abstime->tv_sec < 0 || abstime->tv_nsec < 0 ||
+	    abstime->tv_nsec > 1000000000)
+		return (EINVAL);
+	return (_rthread_rwlock_rdlock(lockp, abstime, 0));
+}
+
+
+static int
+_rthread_rwlock_wrlock(pthread_rwlock_t *lockp, const struct timespec *abstime,
+    int try)
 {
 	pthread_rwlock_t lock;
+	pthread_t thread = pthread_self();
 	int error;
 
 	if ((error = _rthread_rwlock_ensure_init(lockp)))
@@ -168,95 +165,98 @@ pthread_rwlock_wrlock(pthread_rwlock_t *lockp)
 
 	lock = *lockp;
 
+	_rthread_debug(5, "%p: rwlock_timedwrlock %p\n", (void *)thread,
+	    (void *)lock);
 	_spinlock(&lock->lock);
-	lock->writer++;
-	while (lock->readers) {
-		_spinlock(&lock->sem.lock);
-		_spinunlock(&lock->lock);
-		_sem_waitl(&lock->sem, 0, 0, NULL);
-		_spinlock(&lock->lock);
+	if (lock->readers == 0 && lock->owner == NULL)
+		lock->owner = thread;
+	else if (try)
+		error = EBUSY;
+	else if (lock->owner == thread)
+		error = EDEADLK;
+	else {
+		int do_wait;
+
+		/* gotta block */
+		TAILQ_INSERT_TAIL(&lock->writers, thread, waiting);
+		do {
+			do_wait = __thrsleep(thread, CLOCK_REALTIME, abstime,
+			    &lock->lock, NULL) != EWOULDBLOCK;
+			_spinlock(&lock->lock);
+		} while (lock->owner != thread && do_wait);
+
+		if (lock->owner != thread) {
+			/* timed out, sigh */
+			TAILQ_REMOVE(&lock->writers, thread, waiting);
+			error = ETIMEDOUT;
+		}
 	}
-	lock->readers = -pthread_self()->tid;
 	_spinunlock(&lock->lock);
 
-	return (0);
+	return (error);
 }
 
 int
-pthread_rwlock_timedwrlock(pthread_rwlock_t *lockp,
-    const struct timespec *abstime)
+pthread_rwlock_wrlock(pthread_rwlock_t *lockp)
 {
-	pthread_rwlock_t lock;
-	int do_wait = 1;
-	int error;
-
-	if ((error = _rthread_rwlock_ensure_init(lockp)))
-		return (error);
-
-	lock = *lockp;
-
-	_spinlock(&lock->lock);
-	lock->writer++;
-	while (lock->readers && do_wait) {
-		_spinlock(&lock->sem.lock);
-		_spinunlock(&lock->lock);
-		do_wait = _sem_waitl(&lock->sem, 0, CLOCK_REALTIME, abstime);
-		_spinlock(&lock->lock);
-	}
-	if (lock->readers) {
-		/* do_wait must be 0, so timed out */
-		lock->writer--;
-		_spinunlock(&lock->lock);
-		return (ETIMEDOUT);
-	}
-	lock->readers = -pthread_self()->tid;
-	_spinunlock(&lock->lock);
-
-	return (0);
+	return (_rthread_rwlock_wrlock(lockp, NULL, 0));
 }
 
 int
 pthread_rwlock_trywrlock(pthread_rwlock_t *lockp)
 {
-	pthread_rwlock_t lock;
-	int error;
-
-	if ((error = _rthread_rwlock_ensure_init(lockp)))
-		return (error);
-
-	lock = *lockp;
-
-	_spinlock(&lock->lock);
-	if (lock->readers || lock->writer) {
-		_spinunlock(&lock->lock);
-		return (EBUSY);
-	}
-	lock->writer = 1;
-	lock->readers = -pthread_self()->tid;
-	_spinunlock(&lock->lock);
+	return (_rthread_rwlock_wrlock(lockp, NULL, 1));
+}
 
-	return (0);
+int
+pthread_rwlock_timedwrlock(pthread_rwlock_t *lockp,
+    const struct timespec *abstime)
+{
+	if (abstime == NULL || abstime->tv_sec < 0 || abstime->tv_nsec < 0 ||
+	    abstime->tv_nsec > 1000000000)
+		return (EINVAL);
+	return (_rthread_rwlock_wrlock(lockp, abstime, 0));
 }
 
+
 int
 pthread_rwlock_unlock(pthread_rwlock_t *lockp)
 {
 	pthread_rwlock_t lock;
+	pthread_t thread = pthread_self();
+	pthread_t next;
+	int was_writer;
 
 	lock = *lockp;
 
+	_rthread_debug(5, "%p: rwlock_unlock %p\n", (void *)thread,
+	    (void *)lock);
 	_spinlock(&lock->lock);
-	if (lock->readers == -pthread_self()->tid) {
-		lock->readers = 0;
-		lock->writer--;
-	} else if (lock->readers > 0) {
-		lock->readers--;
+	if (lock->owner != NULL) {
+		assert(lock->owner == thread);
+		was_writer = 1;
 	} else {
+		assert(lock->readers > 0);
+		lock->readers--;
+		if (lock->readers > 0)
+			goto out;
+		was_writer = 0;
+	}
+
+	lock->owner = next = TAILQ_FIRST(&lock->writers);
+	if (next != NULL) {
+		/* dequeue and wake first writer */
+		TAILQ_REMOVE(&lock->writers, next, waiting);
 		_spinunlock(&lock->lock);
-		return (EPERM);
+		__thrwakeup(next, 1);
+		return (0);
 	}
+
+	/* could there have been blocked readers?  wake them all */
+	if (was_writer)
+		__thrwakeup(lock, 0);
+out:
 	_spinunlock(&lock->lock);
-	_sem_wakeall(&lock->sem);
 
 	return (0);
 }
diff --git a/lib/librthread/rthread_sem.c b/lib/librthread/rthread_sem.c
index 6c065cc4b78..66c22af4b9a 100644
--- a/lib/librthread/rthread_sem.c
+++ b/lib/librthread/rthread_sem.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rthread_sem.c,v 1.3 2012/01/04 21:01:25 guenther Exp $ */
+/*	$OpenBSD: rthread_sem.c,v 1.4 2012/01/17 02:34:18 guenther Exp $ */
 /*
  * Copyright (c) 2004,2005 Ted Unangst <tedu@openbsd.org>
  * All Rights Reserved.
@@ -28,42 +28,29 @@
  * Internal implementation of semaphores
  */
 int
-_sem_wait(sem_t sem, int tryonly)
+_sem_wait(sem_t sem, int tryonly, int *delayed_cancel)
 {
+	int r;
 
 	_spinlock(&sem->lock);
-	return (_sem_waitl(sem, tryonly, 0, NULL));
-}
-
-int
-_sem_waitl(sem_t sem, int tryonly, clockid_t clock_id,
-    const struct timespec *abstime)
-{
-	int do_sleep;
-
-again:
-	if (sem->value == 0) {
-		if (tryonly) {
-			_spinunlock(&sem->lock);
-			return (0);
-		}
-		sem->waitcount++;
-		do_sleep = 1;
-	} else {
+	if (sem->value) {
 		sem->value--;
-		do_sleep = 0;
-	}
-
-	if (do_sleep) {
-		if (thrsleep(sem, clock_id, abstime, &sem->lock) == -1 &&
-		    errno == EWOULDBLOCK)
-			return (0);
-		_spinlock(&sem->lock);
+		r = 1;
+	} else if (tryonly) {
+		r = 0;
+	} else {
+		sem->waitcount++;
+		do {
+			r = __thrsleep(&sem->waitcount, 0, NULL, &sem->lock,
+			    delayed_cancel) == 0;
+			_spinlock(&sem->lock);
+		} while (r && sem->value == 0);
 		sem->waitcount--;
-		goto again;
+		if (r)
+			sem->value--;
 	}
 	_spinunlock(&sem->lock);
-	return (1);
+	return (r);
 }
 
 /* always increment count */
@@ -75,44 +62,13 @@ _sem_post(sem_t sem)
 	_spinlock(&sem->lock);
 	sem->value++;
 	if (sem->waitcount) {
-		thrwakeup(sem, 1);
+		__thrwakeup(&sem->waitcount, 1);
 		rv = 1;
 	}
 	_spinunlock(&sem->lock);
 	return (rv);
 }
 
-/* only increment count if a waiter */
-int
-_sem_wakeup(sem_t sem)
-{
-	int rv = 0;
-
-	_spinlock(&sem->lock);
-	if (sem->waitcount) {
-		sem->value++;
-		thrwakeup(sem, 1);
-		rv = 1;
-	}
-	_spinunlock(&sem->lock);
-	return (rv);
-}
-
-
-int
-_sem_wakeall(sem_t sem)
-{
-	int rv;
-
-	_spinlock(&sem->lock);
-	rv = sem->waitcount;
-	sem->value += rv;
-	thrwakeup(sem, 0);
-	_spinunlock(&sem->lock);
-
-	return (rv);
-}
-
 /*
  * exported semaphores
  */
@@ -199,13 +155,17 @@ int
 sem_wait(sem_t *semp)
 {
 	sem_t sem = *semp;
+	pthread_t self = pthread_self();
+	int r;
 
 	if (!semp || !*semp) {
 		errno = EINVAL;
 		return (-1);
 	}
 
-	_sem_wait(sem, 0);
+	_enter_delayed_cancel(self);
+	r = _sem_wait(sem, 0, &self->delayed_cancel);
+	_leave_delayed_cancel(self, !r);
 
 	return (0);
 }
@@ -221,7 +181,7 @@ sem_trywait(sem_t *semp)
 		return (-1);
 	}
 
-	rv = _sem_wait(sem, 1);
+	rv = _sem_wait(sem, 1, NULL);
 
 	if (!rv) {
 		errno = EAGAIN;
@@ -231,22 +191,25 @@ sem_trywait(sem_t *semp)
 	return (0);
 }
 
+/* ARGSUSED */
 sem_t *
-sem_open(const char *name, int oflag, ...)
+sem_open(const char *name __unused, int oflag __unused, ...)
 {
 	errno = ENOSYS;
 	return (SEM_FAILED);
 }
 
+/* ARGSUSED */
 int
-sem_close(sem_t *sem)
+sem_close(sem_t *sem __unused)
 {
 	errno = ENOSYS;
 	return (-1);
 }
 
+/* ARGSUSED */
 int
-sem_unlink(const char *name)
+sem_unlink(const char *name __unused)
 {
 	errno = ENOSYS;
 	return (-1);
diff --git a/lib/librthread/rthread_sig.c b/lib/librthread/rthread_sig.c
index 11351ddafc3..c38fed3998b 100644
--- a/lib/librthread/rthread_sig.c
+++ b/lib/librthread/rthread_sig.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rthread_sig.c,v 1.11 2011/12/27 17:36:59 guenther Exp $ */
+/*	$OpenBSD: rthread_sig.c,v 1.12 2012/01/17 02:34:18 guenther Exp $ */
 /*
  * Copyright (c) 2005 Ted Unangst <tedu@openbsd.org>
  * All Rights Reserved.
@@ -56,7 +56,7 @@ sigwait(const sigset_t *set, int *sig)
 
 	sigdelset(&s, SIGTHR);
 	_enter_cancel(self);
-	ret = thrsigdivert(s, NULL, NULL);
+	ret = __thrsigdivert(s, NULL, NULL);
 	_leave_cancel(self);
 	if (ret == -1)
 		return (errno);
diff --git a/lib/librthread/rthread_sync.c b/lib/librthread/rthread_sync.c
index cbe529a83aa..ae3bdbb40d9 100644
--- a/lib/librthread/rthread_sync.c
+++ b/lib/librthread/rthread_sync.c
@@ -1,6 +1,7 @@
-/*	$OpenBSD: rthread_sync.c,v 1.28 2012/01/04 17:43:34 mpi Exp $ */
+/*	$OpenBSD: rthread_sync.c,v 1.29 2012/01/17 02:34:18 guenther Exp $ */
 /*
  * Copyright (c) 2004,2005 Ted Unangst <tedu@openbsd.org>
+ * Copyright (c) 2012 Philip Guenther <guenther@openbsd.org>
  * All Rights Reserved.
  *
  * Permission to use, copy, modify, and distribute this software for any
@@ -20,7 +21,9 @@
  */
 
 
+#include <assert.h>
 #include <stdlib.h>
+#include <string.h>
 #include <unistd.h>
 #include <errno.h>
 
@@ -36,16 +39,16 @@ static _spinlock_lock_t static_init_lock = _SPINLOCK_UNLOCKED;
 int
 pthread_mutex_init(pthread_mutex_t *mutexp, const pthread_mutexattr_t *attr)
 {
-	pthread_mutex_t mutex;
+	struct pthread_mutex *mutex;
 
 	mutex = calloc(1, sizeof(*mutex));
 	if (!mutex)
 		return (errno);
-	mutex->sem.lock = _SPINLOCK_UNLOCKED;
-	mutex->sem.value = 1;	/* unlocked */
+	mutex->lock = _SPINLOCK_UNLOCKED;
+	TAILQ_INIT(&mutex->lockers);
 	if (attr == NULL) {
 		mutex->type = PTHREAD_MUTEX_ERRORCHECK;
-		mutex->prioceiling = PTHREAD_PRIO_NONE;
+		mutex->prioceiling = -1;
 	} else {
 		mutex->type = (*attr)->ma_type;
 		mutex->prioceiling = (*attr)->ma_protocol ==
@@ -59,23 +62,29 @@ pthread_mutex_init(pthread_mutex_t *mutexp, const pthread_mutexattr_t *attr)
 int
 pthread_mutex_destroy(pthread_mutex_t *mutexp)
 {
+	struct pthread_mutex *mutex;
 
-	if ((*mutexp) && (*mutexp)->count) {
+	assert(mutexp);
+	mutex = (struct pthread_mutex *)*mutexp;
+	if (mutex) {
+		if (mutex->count || mutex->owner != NULL ||
+		    !TAILQ_EMPTY(&mutex->lockers)) {
 #define MSG "pthread_mutex_destroy on mutex with waiters!\n"
-		write(2, MSG, sizeof(MSG) - 1);
+			write(2, MSG, sizeof(MSG) - 1);
 #undef MSG
-		return (EBUSY);
+			return (EBUSY);
+		}
+		free(mutex);
+		*mutexp = NULL;
 	}
-	free((void *)*mutexp);
-	*mutexp = NULL;
 	return (0);
 }
 
 static int
 _rthread_mutex_lock(pthread_mutex_t *mutexp, int trywait)
 {
-	pthread_mutex_t mutex;
-	pthread_t thread = pthread_self();
+	struct pthread_mutex *mutex;
+	pthread_t self = pthread_self();
 	int ret = 0;
 
 	/*
@@ -92,19 +101,42 @@ _rthread_mutex_lock(pthread_mutex_t *mutexp, int trywait)
 		if (ret != 0)
 			return (EINVAL);
 	}
-	mutex = *mutexp;
-	if (mutex->owner == thread) {
-		if (mutex->type == PTHREAD_MUTEX_RECURSIVE) {
-			mutex->count++;
-			return (0);
+	mutex = (struct pthread_mutex *)*mutexp;
+
+	_rthread_debug(5, "%p: mutex_lock %p\n", (void *)self, (void *)mutex);
+	_spinlock(&mutex->lock);
+	if (mutex->owner == NULL && TAILQ_EMPTY(&mutex->lockers)) {
+		assert(mutex->count == 0);
+		mutex->owner = self;
+	} else if (mutex->owner == self) {
+		assert(mutex->count > 0);
+
+		/* already owner?  handle recursive behavior */
+		if (mutex->type != PTHREAD_MUTEX_RECURSIVE)
+		{
+			if (trywait ||
+			    mutex->type == PTHREAD_MUTEX_ERRORCHECK) {
+				_spinunlock(&mutex->lock);
+				return (trywait ? EBUSY : EDEADLK);
+			}
+			abort();
 		}
-		if (mutex->type == PTHREAD_MUTEX_ERRORCHECK)
-			return (trywait ? EBUSY : EDEADLK);
-	}
-	if (!_sem_wait((void *)&mutex->sem, trywait))
+	} else if (trywait) {
+		/* try failed */
+		_spinunlock(&mutex->lock);
 		return (EBUSY);
-	mutex->owner = thread;
-	mutex->count = 1;
+	} else {
+		/* add to the wait queue and block until at the head */
+		TAILQ_INSERT_TAIL(&mutex->lockers, self, waiting);
+		while (mutex->owner != self) {
+			__thrsleep(self, 0, NULL, &mutex->lock, NULL);
+			_spinlock(&mutex->lock);
+			assert(mutex->owner != NULL);
+		}
+	}
+
+	mutex->count++;
+	_spinunlock(&mutex->lock);
 
 	return (0);
 }
@@ -124,15 +156,25 @@ pthread_mutex_trylock(pthread_mutex_t *p)
 int
 pthread_mutex_unlock(pthread_mutex_t *mutexp)
 {
-	pthread_t thread = pthread_self();
-	pthread_mutex_t mutex = *mutexp;
+	pthread_t self = pthread_self();
+	struct pthread_mutex *mutex = (struct pthread_mutex *)*mutexp;
+
+	_rthread_debug(5, "%p: mutex_unlock %p\n", (void *)self,
+	    (void *)mutex);
 
-	if (mutex->owner != thread)
+	if (mutex->owner != self)
 		return (EPERM);
 
 	if (--mutex->count == 0) {
-		mutex->owner = NULL;
-		_sem_post((void *)&mutex->sem);
+		pthread_t next;
+
+		_spinlock(&mutex->lock);
+		mutex->owner = next = TAILQ_FIRST(&mutex->lockers);
+		if (next != NULL)
+			TAILQ_REMOVE(&mutex->lockers, next, waiting);
+		_spinunlock(&mutex->lock);
+		if (next != NULL)
+			__thrwakeup(next, 1);
 	}
 
 	return (0);
@@ -141,15 +183,18 @@ pthread_mutex_unlock(pthread_mutex_t *mutexp)
 /*
  * condition variables
  */
+/* ARGSUSED1 */
 int
-pthread_cond_init(pthread_cond_t *condp, const pthread_condattr_t *attrp)
+pthread_cond_init(pthread_cond_t *condp,
+    const pthread_condattr_t *attrp __unused)
 {
 	pthread_cond_t cond;
 
 	cond = calloc(1, sizeof(*cond));
 	if (!cond)
 		return (errno);
-	cond->sem.lock = _SPINLOCK_UNLOCKED;
+	cond->lock = _SPINLOCK_UNLOCKED;
+	TAILQ_INIT(&cond->waiters);
 
 	*condp = cond;
 
@@ -159,8 +204,19 @@ pthread_cond_init(pthread_cond_t *condp, const pthread_condattr_t *attrp)
 int
 pthread_cond_destroy(pthread_cond_t *condp)
 {
+	pthread_cond_t cond;
 
-	free(*condp);
+	assert(condp);
+	cond = *condp;
+	if (cond) {
+		if (!TAILQ_EMPTY(&cond->waiters)) {
+#define MSG "pthread_cond_destroy on condvar with waiters!\n"
+			write(2, MSG, sizeof(MSG) - 1);
+#undef MSG
+			return (EBUSY);
+		}
+		free(cond);
+	}
 	*condp = NULL;
 
 	return (0);
@@ -170,37 +226,310 @@ int
 pthread_cond_timedwait(pthread_cond_t *condp, pthread_mutex_t *mutexp,
     const struct timespec *abstime)
 {
+	pthread_cond_t cond;
+	struct pthread_mutex *mutex = (struct pthread_mutex *)*mutexp;
+	pthread_t self = pthread_self();
+	pthread_t next;
+	int mutex_count;
+	int canceled = 0;
+	int rv = 0;
 	int error;
-	int rv;
 
 	if (!*condp)
 		if ((error = pthread_cond_init(condp, NULL)))
 			return (error);
+	cond = *condp;
+	_rthread_debug(5, "%p: cond_timed %p,%p\n", (void *)self,
+	    (void *)cond, (void *)mutex);
 
-	_spinlock(&(*condp)->sem.lock);
-	pthread_mutex_unlock(mutexp);
-	rv = _sem_waitl(&(*condp)->sem, 0, CLOCK_REALTIME, abstime);
-	error = pthread_mutex_lock(mutexp);
+	if (mutex->owner != self)
+		return (EPERM);
+	if (abstime == NULL || abstime->tv_sec < 0 || abstime->tv_nsec < 0 ||
+	    abstime->tv_nsec >= 1000000000)
+		return (EINVAL);
+
+	_enter_delayed_cancel(self);
+
+	_spinlock(&cond->lock);
+
+	/* mark the condvar as being associated with this mutex */
+	if (cond->mutex == NULL) {
+		cond->mutex = mutex;
+		assert(TAILQ_EMPTY(&cond->waiters));
+	} else if (cond->mutex != mutex) {
+		assert(cond->mutex == mutex);
+		_spinunlock(&cond->lock);
+		_leave_delayed_cancel(self, 1);
+		return (EINVAL);
+	} else
+		assert(! TAILQ_EMPTY(&cond->waiters));
+
+	/* snag the count in case this is a recursive mutex */
+	mutex_count = mutex->count;
+
+	/* transfer from the mutex queue to the condvar queue */
+	_spinlock(&mutex->lock);
+	self->blocking_cond = cond;
+	TAILQ_INSERT_TAIL(&cond->waiters, self, waiting);
+	_spinunlock(&cond->lock);
+
+	/* wake the next guy blocked on the mutex */
+	mutex->count = 0;
+	mutex->owner = next = TAILQ_FIRST(&mutex->lockers);
+	if (next != NULL) {
+		TAILQ_REMOVE(&mutex->lockers, next, waiting);
+		__thrwakeup(next, 1);
+	}
 
-	return (error ? error : rv ? 0 : ETIMEDOUT);
+	/* wait until we're the owner of the mutex again */
+	while (mutex->owner != self) {
+		error = __thrsleep(self, CLOCK_REALTIME, abstime, &mutex->lock,
+		    &self->delayed_cancel);
+
+		/*
+		 * If abstime == NULL, then we're definitely waiting
+		 * on the mutex instead of the condvar, and are
+		 * just waiting for mutex ownership, regardless of
+		 * why we woke up.
+		 */
+		if (abstime == NULL) {
+			_spinlock(&mutex->lock);
+			continue;
+		}
+
+		/*
+		 * If we took a normal signal (not from
+		 * cancellation) then we should just go back to
+		 * sleep without changing state (timeouts, etc).
+		 */
+		if (error == EINTR && !IS_CANCELED(self)) {
+			_spinlock(&mutex->lock);
+			continue;
+		}
+
+		/*
+		 * The remaining reasons for waking up (normal
+		 * wakeup, timeout, and cancellation) all mean that
+		 * we won't be staying in the condvar queue and
+		 * we'll no longer time out or be cancelable.
+		 */
+		abstime = NULL;
+		_leave_delayed_cancel(self, 0);
+
+		/*
+		 * If we're no longer in the condvar's queue then
+		 * we're just waiting for mutex ownership.  Need
+		 * cond->lock here to prevent race with cond_signal().
+		 */
+		_spinlock(&cond->lock);
+		if (self->blocking_cond == NULL) {
+			_spinunlock(&cond->lock);
+			_spinlock(&mutex->lock);
+			continue;
+		}
+		assert(self->blocking_cond == cond);
+
+		/* if timeout or canceled, make note of that */
+		if (error == EWOULDBLOCK)
+			rv = ETIMEDOUT;
+		else if (error == EINTR)
+			canceled = 1;
+
+		/* transfer between the queues */
+		TAILQ_REMOVE(&cond->waiters, self, waiting);
+		assert(mutex == cond->mutex);
+		if (TAILQ_EMPTY(&cond->waiters))
+			cond->mutex = NULL;
+		self->blocking_cond = NULL;
+		_spinunlock(&cond->lock);
+		_spinlock(&mutex->lock);
+
+		/* mutex unlocked right now? */
+		if (mutex->owner == NULL &&
+		    TAILQ_EMPTY(&mutex->lockers)) {
+			assert(mutex->count == 0);
+			mutex->owner = self;
+			break;
+		}
+		TAILQ_INSERT_TAIL(&mutex->lockers, self, waiting);
+	}
+
+	/* restore the mutex's count */
+	mutex->count = mutex_count;
+	_spinunlock(&mutex->lock);
+
+	_leave_delayed_cancel(self, canceled);
+
+	return (rv);
 }
 
 int
 pthread_cond_wait(pthread_cond_t *condp, pthread_mutex_t *mutexp)
 {
-	return (pthread_cond_timedwait(condp, mutexp, NULL));
+	pthread_cond_t cond;
+	struct pthread_mutex *mutex = (struct pthread_mutex *)*mutexp;
+	pthread_t self = pthread_self();
+	pthread_t next;
+	int mutex_count;
+	int canceled = 0;
+	int error;
+
+	if (!*condp)
+		if ((error = pthread_cond_init(condp, NULL)))
+			return (error);
+	cond = *condp;
+	_rthread_debug(5, "%p: cond_timed %p,%p\n", (void *)self,
+	    (void *)cond, (void *)mutex);
+
+	if (mutex->owner != self)
+		return (EPERM);
+
+	_enter_delayed_cancel(self);
+
+	_spinlock(&cond->lock);
+
+	/* mark the condvar as being associated with this mutex */
+	if (cond->mutex == NULL) {
+		cond->mutex = mutex;
+		assert(TAILQ_EMPTY(&cond->waiters));
+	} else if (cond->mutex != mutex) {
+		assert(cond->mutex == mutex);
+		_spinunlock(&cond->lock);
+		_leave_delayed_cancel(self, 1);
+		return (EINVAL);
+	} else
+		assert(! TAILQ_EMPTY(&cond->waiters));
+
+	/* snag the count in case this is a recursive mutex */
+	mutex_count = mutex->count;
+
+	/* transfer from the mutex queue to the condvar queue */
+	_spinlock(&mutex->lock);
+	self->blocking_cond = cond;
+	TAILQ_INSERT_TAIL(&cond->waiters, self, waiting);
+	_spinunlock(&cond->lock);
+
+	/* wake the next guy blocked on the mutex */
+	mutex->count = 0;
+	mutex->owner = next = TAILQ_FIRST(&mutex->lockers);
+	if (next != NULL) {
+		TAILQ_REMOVE(&mutex->lockers, next, waiting);
+		__thrwakeup(next, 1);
+	}
+
+	/* wait until we're the owner of the mutex again */
+	while (mutex->owner != self) {
+		error = __thrsleep(self, 0, NULL, &mutex->lock,
+		    &self->delayed_cancel);
+
+		/*
+		 * If we took a normal signal (not from
+		 * cancellation) then we should just go back to
+		 * sleep without changing state (timeouts, etc).
+		 */
+		if (error == EINTR && !IS_CANCELED(self)) {
+			_spinlock(&mutex->lock);
+			continue;
+		}
+
+		/*
+		 * The remaining reasons for waking up (normal
+		 * wakeup and cancellation) all mean that we won't
+		 * be staying in the condvar queue and we'll no
+		 * longer be cancelable.
+		 */
+		_leave_delayed_cancel(self, 0);
+
+		/*
+		 * If we're no longer in the condvar's queue then
+		 * we're just waiting for mutex ownership.  Need
+		 * cond->lock here to prevent race with cond_signal().
+		 */
+		_spinlock(&cond->lock);
+		if (self->blocking_cond == NULL) {
+			_spinunlock(&cond->lock);
+			_spinlock(&mutex->lock);
+			continue;
+		}
+		assert(self->blocking_cond == cond);
+
+		/* if canceled, make note of that */
+		if (error == EINTR)
+			canceled = 1;
+
+		/* transfer between the queues */
+		TAILQ_REMOVE(&cond->waiters, self, waiting);
+		assert(mutex == cond->mutex);
+		if (TAILQ_EMPTY(&cond->waiters))
+			cond->mutex = NULL;
+		self->blocking_cond = NULL;
+		_spinunlock(&cond->lock);
+		_spinlock(&mutex->lock);
+
+		/* mutex unlocked right now? */
+		if (mutex->owner == NULL &&
+		    TAILQ_EMPTY(&mutex->lockers)) {
+			assert(mutex->count == 0);
+			mutex->owner = self;
+			break;
+		}
+		TAILQ_INSERT_TAIL(&mutex->lockers, self, waiting);
+	}
+
+	/* restore the mutex's count */
+	mutex->count = mutex_count;
+	_spinunlock(&mutex->lock);
+
+	_leave_delayed_cancel(self, canceled);
+
+	return (0);
 }
 
+
 int
 pthread_cond_signal(pthread_cond_t *condp)
 {
-	int error;
+	pthread_cond_t cond;
+	struct pthread_mutex *mutex;
+	pthread_t thread;
+	int wakeup;
 
+	/* uninitialized?  Then there's obviously no one waiting! */
 	if (!*condp)
-		if ((error = pthread_cond_init(condp, NULL)))
-			return (error);
+		return 0;
+
+	cond = *condp;
+	_rthread_debug(5, "%p: cond_signal %p,%p\n", (void *)pthread_self(),
+	    (void *)cond, (void *)cond->mutex);
+	_spinlock(&cond->lock);
+	thread = TAILQ_FIRST(&cond->waiters);
+	if (thread == NULL) {
+		assert(cond->mutex == NULL);
+		_spinunlock(&cond->lock);
+		return (0);
+	}
 
-	_sem_wakeup(&(*condp)->sem);
+	assert(thread->blocking_cond == cond);
+	TAILQ_REMOVE(&cond->waiters, thread, waiting);
+	thread->blocking_cond = NULL;
+
+	mutex = cond->mutex;
+	assert(mutex != NULL);
+	if (TAILQ_EMPTY(&cond->waiters))
+		cond->mutex = NULL;
+
+	/* link locks to prevent race with timedwait */
+	_spinlock(&mutex->lock);
+	_spinunlock(&cond->lock);
+
+	wakeup = mutex->owner == NULL && TAILQ_EMPTY(&mutex->lockers);
+	if (wakeup)
+		mutex->owner = thread;
+	else
+		TAILQ_INSERT_TAIL(&mutex->lockers, thread, waiting);
+	_spinunlock(&mutex->lock);
+	if (wakeup)
+		__thrwakeup(thread, 1);
 
 	return (0);
 }
@@ -208,10 +537,65 @@ pthread_cond_signal(pthread_cond_t *condp)
 int
 pthread_cond_broadcast(pthread_cond_t *condp)
 {
+	pthread_cond_t cond;
+	struct pthread_mutex *mutex;
+	pthread_t thread;
+	pthread_t p;
+	int wakeup;
+
+	/* uninitialized?  Then there's obviously no one waiting! */
 	if (!*condp)
-		pthread_cond_init(condp, NULL);
+		return 0;
+
+	cond = *condp;
+	_rthread_debug(5, "%p: cond_broadcast %p,%p\n", (void *)pthread_self(),
+	    (void *)cond, (void *)cond->mutex);
+	_spinlock(&cond->lock);
+	thread = TAILQ_FIRST(&cond->waiters);
+	if (thread == NULL) {
+		assert(cond->mutex == NULL);
+		_spinunlock(&cond->lock);
+		return (0);
+	}
+
+	mutex = cond->mutex;
+	assert(mutex != NULL);
+
+	/* walk the list, clearing the "blocked on condvar" pointer */
+	p = thread;
+	do
+		p->blocking_cond = NULL;
+	while ((p = TAILQ_NEXT(p, waiting)) != NULL);
+
+	/*
+	 * We want to transfer all the threads from the condvar's list
+	 * to the mutex's list.  The TAILQ_* macros don't let us do that
+	 * efficiently, so this is direct list surgery.  Pay attention!
+	 */
 
-	_sem_wakeall(&(*condp)->sem);
+	/* 1) attach the first thread to the end of the mutex's list */
+	_spinlock(&mutex->lock);
+	wakeup = mutex->owner == NULL && TAILQ_EMPTY(&mutex->lockers);
+	thread->waiting.tqe_prev = mutex->lockers.tqh_last;
+	*(mutex->lockers.tqh_last) = thread;
+
+	/* 2) fix up the end pointer for the mutex's list */
+	mutex->lockers.tqh_last = cond->waiters.tqh_last;
+	_spinunlock(&mutex->lock);
+
+	if (wakeup) {
+		TAILQ_REMOVE(&mutex->lockers, thread, waiting);
+		mutex->owner = thread;
+		_spinunlock(&mutex->lock);
+		__thrwakeup(thread, 1);
+	} else
+		_spinunlock(&mutex->lock);
+
+	/* 3) reset the condvar's list and mutex pointer */
+	TAILQ_INIT(&cond->waiters);
+	assert(cond->mutex != NULL);
+	cond->mutex = NULL;
+	_spinunlock(&cond->lock);
 
 	return (0);
 }
diff --git a/lib/librthread/shlib_version b/lib/librthread/shlib_version
index 890c57389b5..3066b9771e7 100644
--- a/lib/librthread/shlib_version
+++ b/lib/librthread/shlib_version
@@ -1,2 +1,2 @@
-major=4
-minor=1
+major=5
+minor=0
diff --git a/sys/kern/init_sysent.c b/sys/kern/init_sysent.c
index 215f213454e..be822ad146e 100644
--- a/sys/kern/init_sysent.c
+++ b/sys/kern/init_sysent.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: init_sysent.c,v 1.131 2011/10/15 23:36:31 guenther Exp $	*/
+/*	$OpenBSD: init_sysent.c,v 1.132 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * System call switch table.
@@ -731,14 +731,14 @@ struct sysent sysent[] = {
 	    sys_sched_yield },			/* 298 = sched_yield */
 	{ 0, 0, SY_NOLOCK | 0,
 	    sys_getthrid },			/* 299 = getthrid */
-	{ 4, s(struct sys_thrsleep_args), 0,
-	    sys_thrsleep },			/* 300 = thrsleep */
-	{ 2, s(struct sys_thrwakeup_args), 0,
-	    sys_thrwakeup },			/* 301 = thrwakeup */
-	{ 1, s(struct sys_threxit_args), 0,
-	    sys_threxit },			/* 302 = threxit */
-	{ 3, s(struct sys_thrsigdivert_args), 0,
-	    sys_thrsigdivert },			/* 303 = thrsigdivert */
+	{ 5, s(struct sys___thrsleep_args), 0,
+	    sys___thrsleep },			/* 300 = __thrsleep */
+	{ 2, s(struct sys___thrwakeup_args), 0,
+	    sys___thrwakeup },			/* 301 = __thrwakeup */
+	{ 1, s(struct sys___threxit_args), 0,
+	    sys___threxit },			/* 302 = __threxit */
+	{ 3, s(struct sys___thrsigdivert_args), 0,
+	    sys___thrsigdivert },		/* 303 = __thrsigdivert */
 	{ 2, s(struct sys___getcwd_args), 0,
 	    sys___getcwd },			/* 304 = __getcwd */
 	{ 2, s(struct sys_adjfreq_args), 0,
diff --git a/sys/kern/kern_exit.c b/sys/kern/kern_exit.c
index ecd63cf0fda..50e9e0e3cf9 100644
--- a/sys/kern/kern_exit.c
+++ b/sys/kern/kern_exit.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_exit.c,v 1.105 2011/12/14 07:32:16 guenther Exp $	*/
+/*	$OpenBSD: kern_exit.c,v 1.106 2012/01/17 02:34:18 guenther Exp $	*/
 /*	$NetBSD: kern_exit.c,v 1.39 1996/04/22 01:38:25 christos Exp $	*/
 
 /*
@@ -91,9 +91,9 @@ sys_exit(struct proc *p, void *v, register_t *retval)
 }
 
 int
-sys_threxit(struct proc *p, void *v, register_t *retval)
+sys___threxit(struct proc *p, void *v, register_t *retval)
 {
-	struct sys_threxit_args /* {
+	struct sys___threxit_args /* {
 		syscallarg(pid_t *) notdead;
 	} */ *uap = v;
 
diff --git a/sys/kern/kern_sig.c b/sys/kern/kern_sig.c
index 0775a1edf52..859e1ff75aa 100644
--- a/sys/kern/kern_sig.c
+++ b/sys/kern/kern_sig.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_sig.c,v 1.131 2011/12/11 19:42:28 guenther Exp $	*/
+/*	$OpenBSD: kern_sig.c,v 1.132 2012/01/17 02:34:18 guenther Exp $	*/
 /*	$NetBSD: kern_sig.c,v 1.54 1996/04/22 01:38:32 christos Exp $	*/
 
 /*
@@ -1503,9 +1503,9 @@ sys_nosys(struct proc *p, void *v, register_t *retval)
 }
 
 int
-sys_thrsigdivert(struct proc *p, void *v, register_t *retval)
+sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
 {
-	struct sys_thrsigdivert_args /* {
+	struct sys___thrsigdivert_args /* {
 		syscallarg(sigset_t) sigmask;
 		syscallarg(siginfo_t *) info;
 		syscallarg(const struct timespec *) timeout;
diff --git a/sys/kern/kern_synch.c b/sys/kern/kern_synch.c
index 4b711711c02..6b7f80ab747 100644
--- a/sys/kern/kern_synch.c
+++ b/sys/kern/kern_synch.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_synch.c,v 1.98 2011/12/11 19:42:28 guenther Exp $	*/
+/*	$OpenBSD: kern_synch.c,v 1.99 2012/01/17 02:34:18 guenther Exp $	*/
 /*	$NetBSD: kern_synch.c,v 1.37 1996/04/22 01:38:37 christos Exp $	*/
 
 /*
@@ -404,37 +404,49 @@ sys_sched_yield(struct proc *p, void *v, register_t *retval)
 }
 
 int
-sys_thrsleep(struct proc *p, void *v, register_t *revtal)
+sys___thrsleep(struct proc *p, void *v, register_t *retval)
 {
-	struct sys_thrsleep_args /* {
-		syscallarg(void *) ident;
+	struct sys___thrsleep_args /* {
+		syscallarg(const volatile void *) ident;
 		syscallarg(clockid_t) clock_id;
 		syscallarg(struct timespec *) tp;
 		syscallarg(void *) lock;
+		syscallarg(const int *) abort;
 	} */ *uap = v;
 	long ident = (long)SCARG(uap, ident);
 	_spinlock_lock_t *lock = SCARG(uap, lock);
 	static _spinlock_lock_t unlocked = _SPINLOCK_UNLOCKED;
 	long long to_ticks = 0;
-	int error;
+	int abort, error;
 
-	if (!rthreads_enabled)
-		return (ENOTSUP);
+	if (!rthreads_enabled) {
+		*retval = ENOTSUP;
+		return (0);
+	}
+	if (ident == 0) {
+		*retval = EINVAL;
+		return (0);
+	}
 	if (SCARG(uap, tp) != NULL) {
 		struct timespec now, ats;
 
-		if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0 ||
-		    (error = clock_gettime(p, SCARG(uap, clock_id), &now)) != 0)
-			return (error);
+		if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) ||
+		    (error = clock_gettime(p, SCARG(uap, clock_id), &now))) {
+			*retval = error;
+			return (0);
+		}
 
 		if (timespeccmp(&ats, &now, <)) {
 			/* already passed: still do the unlock */
 			if (lock) {
 				if ((error = copyout(&unlocked, lock,
-				    sizeof(unlocked))) != 0)
-					return (error);
+				    sizeof(unlocked))) != 0) {
+					*retval = error;
+					return (0);
+				}
 			}
-			return (EWOULDBLOCK);
+			*retval = EWOULDBLOCK;
+			return (0);
 		}
 
 		timespecsub(&ats, &now, &ats);
@@ -450,23 +462,41 @@ sys_thrsleep(struct proc *p, void *v, register_t *revtal)
 
 	if (lock) {
 		if ((error = copyout(&unlocked, lock, sizeof(unlocked))) != 0)
-			return (error);
+			goto out;
 	}
-	error = tsleep(&p->p_thrslpid, PUSER | PCATCH, "thrsleep",
-	    (int)to_ticks);
+
+	if (SCARG(uap, abort) != NULL) {
+		if ((error = copyin(SCARG(uap, abort), &abort,
+		    sizeof(abort))) != 0)
+			goto out;
+		if (abort) {
+			error = EINTR;
+			goto out;
+		}
+	}
+
+	if (p->p_thrslpid == 0)
+		error = 0;
+	else
+		error = tsleep(&p->p_thrslpid, PUSER | PCATCH, "thrsleep",
+		    (int)to_ticks);
+
+out:
+	p->p_thrslpid = 0;
 
 	if (error == ERESTART)
 		error = EINTR;
 
-	return (error);
+	*retval = error;
+	return (0);
 
 }
 
 int
-sys_thrwakeup(struct proc *p, void *v, register_t *retval)
+sys___thrwakeup(struct proc *p, void *v, register_t *retval)
 {
-	struct sys_thrwakeup_args /* {
-		syscallarg(void *) ident;
+	struct sys___thrwakeup_args /* {
+		syscallarg(const volatile void *) ident;
 		syscallarg(int) n;
 	} */ *uap = v;
 	long ident = (long)SCARG(uap, ident);
@@ -475,17 +505,20 @@ sys_thrwakeup(struct proc *p, void *v, register_t *retval)
 	int found = 0;
 
 	if (!rthreads_enabled)
-		return (ENOTSUP);
-	TAILQ_FOREACH(q, &p->p_p->ps_threads, p_thr_link) {
-		if (q->p_thrslpid == ident) {
-			wakeup_one(&q->p_thrslpid);
-			q->p_thrslpid = 0;
-			if (++found == n)
-				return (0);
+		*retval = ENOTSUP;
+	else if (ident == 0)
+		*retval = EINVAL;
+	else {
+		TAILQ_FOREACH(q, &p->p_p->ps_threads, p_thr_link) {
+			if (q->p_thrslpid == ident) {
+				wakeup_one(&q->p_thrslpid);
+				q->p_thrslpid = 0;
+				if (++found == n)
+					break;
+			}
 		}
+		*retval = found ? 0 : ESRCH;
 	}
-	if (!found)
-		return (ESRCH);
 
 	return (0);
 }
diff --git a/sys/kern/syscalls.c b/sys/kern/syscalls.c
index 29bad4c5909..a7880b9ef64 100644
--- a/sys/kern/syscalls.c
+++ b/sys/kern/syscalls.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: syscalls.c,v 1.132 2011/10/15 23:36:31 guenther Exp $	*/
+/*	$OpenBSD: syscalls.c,v 1.133 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * System call names.
@@ -382,10 +382,10 @@ char *syscallnames[] = {
 #endif
 	"sched_yield",			/* 298 = sched_yield */
 	"getthrid",			/* 299 = getthrid */
-	"thrsleep",			/* 300 = thrsleep */
-	"thrwakeup",			/* 301 = thrwakeup */
-	"threxit",			/* 302 = threxit */
-	"thrsigdivert",			/* 303 = thrsigdivert */
+	"__thrsleep",			/* 300 = __thrsleep */
+	"__thrwakeup",			/* 301 = __thrwakeup */
+	"__threxit",			/* 302 = __threxit */
+	"__thrsigdivert",			/* 303 = __thrsigdivert */
 	"__getcwd",			/* 304 = __getcwd */
 	"adjfreq",			/* 305 = adjfreq */
 	"getfsstat",			/* 306 = getfsstat */
diff --git a/sys/kern/syscalls.master b/sys/kern/syscalls.master
index c8011d97d42..af289aab619 100644
--- a/sys/kern/syscalls.master
+++ b/sys/kern/syscalls.master
@@ -1,4 +1,4 @@
-;	$OpenBSD: syscalls.master,v 1.119 2011/10/15 23:35:29 guenther Exp $
+;	$OpenBSD: syscalls.master,v 1.120 2012/01/17 02:34:18 guenther Exp $
 ;	$NetBSD: syscalls.master,v 1.32 1996/04/23 10:24:21 mycroft Exp $
 
 ;	@(#)syscalls.master	8.2 (Berkeley) 1/13/94
@@ -524,11 +524,13 @@
 #endif
 298	STD NOLOCK	{ int sys_sched_yield(void); }
 299	STD NOLOCK	{ pid_t sys_getthrid(void); }
-300	STD		{ int sys_thrsleep(void *ident, clockid_t clock_id, \
-			    const struct timespec *tp, void *lock); }
-301	STD		{ int sys_thrwakeup(void *ident, int n); }
-302	STD		{ void sys_threxit(pid_t *notdead); }
-303	STD		{ int sys_thrsigdivert(sigset_t sigmask, \
+300	STD		{ int sys___thrsleep(const volatile void *ident, \
+			    clockid_t clock_id, const struct timespec *tp, \
+			    void *lock, const int *abort); }
+301	STD		{ int sys___thrwakeup(const volatile void *ident, \
+			    int n); }
+302	STD		{ void sys___threxit(pid_t *notdead); }
+303	STD		{ int sys___thrsigdivert(sigset_t sigmask, \
 			    siginfo_t *info, const struct timespec *timeout); }
 304	STD		{ int sys___getcwd(char *buf, size_t len); }
 305	STD		{ int sys_adjfreq(const int64_t *freq, \
diff --git a/sys/sys/syscall.h b/sys/sys/syscall.h
index 0145f63cc6b..ccd89b97435 100644
--- a/sys/sys/syscall.h
+++ b/sys/sys/syscall.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: syscall.h,v 1.131 2011/10/15 23:36:31 guenther Exp $	*/
+/*	$OpenBSD: syscall.h,v 1.132 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * System call numbers.
@@ -589,17 +589,17 @@
 /* syscall: "getthrid" ret: "pid_t" args: */
 #define	SYS_getthrid	299
 
-/* syscall: "thrsleep" ret: "int" args: "void *" "clockid_t" "const struct timespec *" "void *" */
-#define	SYS_thrsleep	300
+/* syscall: "__thrsleep" ret: "int" args: "const volatile void *" "clockid_t" "const struct timespec *" "void *" "const int *" */
+#define	SYS___thrsleep	300
 
-/* syscall: "thrwakeup" ret: "int" args: "void *" "int" */
-#define	SYS_thrwakeup	301
+/* syscall: "__thrwakeup" ret: "int" args: "const volatile void *" "int" */
+#define	SYS___thrwakeup	301
 
-/* syscall: "threxit" ret: "void" args: "pid_t *" */
-#define	SYS_threxit	302
+/* syscall: "__threxit" ret: "void" args: "pid_t *" */
+#define	SYS___threxit	302
 
-/* syscall: "thrsigdivert" ret: "int" args: "sigset_t" "siginfo_t *" "const struct timespec *" */
-#define	SYS_thrsigdivert	303
+/* syscall: "__thrsigdivert" ret: "int" args: "sigset_t" "siginfo_t *" "const struct timespec *" */
+#define	SYS___thrsigdivert	303
 
 /* syscall: "__getcwd" ret: "int" args: "char *" "size_t" */
 #define	SYS___getcwd	304
diff --git a/sys/sys/syscallargs.h b/sys/sys/syscallargs.h
index 5e3077f901a..f43f66d46f3 100644
--- a/sys/sys/syscallargs.h
+++ b/sys/sys/syscallargs.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: syscallargs.h,v 1.133 2011/10/15 23:36:31 guenther Exp $	*/
+/*	$OpenBSD: syscallargs.h,v 1.134 2012/01/17 02:34:18 guenther Exp $	*/
 
 /*
  * System call argument lists.
@@ -880,23 +880,24 @@ struct sys_msgctl_args {
 	syscallarg(struct msqid_ds *) buf;
 };
 
-struct sys_thrsleep_args {
-	syscallarg(void *) ident;
+struct sys___thrsleep_args {
+	syscallarg(const volatile void *) ident;
 	syscallarg(clockid_t) clock_id;
 	syscallarg(const struct timespec *) tp;
 	syscallarg(void *) lock;
+	syscallarg(const int *) abort;
 };
 
-struct sys_thrwakeup_args {
-	syscallarg(void *) ident;
+struct sys___thrwakeup_args {
+	syscallarg(const volatile void *) ident;
 	syscallarg(int) n;
 };
 
-struct sys_threxit_args {
+struct sys___threxit_args {
 	syscallarg(pid_t *) notdead;
 };
 
-struct sys_thrsigdivert_args {
+struct sys___thrsigdivert_args {
 	syscallarg(sigset_t) sigmask;
 	syscallarg(siginfo_t *) info;
 	syscallarg(const struct timespec *) timeout;
@@ -1279,10 +1280,10 @@ int	sys_msgctl(struct proc *, void *, register_t *);
 #endif
 int	sys_sched_yield(struct proc *, void *, register_t *);
 int	sys_getthrid(struct proc *, void *, register_t *);
-int	sys_thrsleep(struct proc *, void *, register_t *);
-int	sys_thrwakeup(struct proc *, void *, register_t *);
-int	sys_threxit(struct proc *, void *, register_t *);
-int	sys_thrsigdivert(struct proc *, void *, register_t *);
+int	sys___thrsleep(struct proc *, void *, register_t *);
+int	sys___thrwakeup(struct proc *, void *, register_t *);
+int	sys___threxit(struct proc *, void *, register_t *);
+int	sys___thrsigdivert(struct proc *, void *, register_t *);
 int	sys___getcwd(struct proc *, void *, register_t *);
 int	sys_adjfreq(struct proc *, void *, register_t *);
 int	sys_getfsstat(struct proc *, void *, register_t *);