Split up fork1():

 - FORK_THREAD handling is a totally separate function, thread_fork(),
   that is only used by sys___tfork() and which loses the flags, func,
   arg, and newprocp parameters and gains tcb parameter to guarantee
   the new thread's TCB is set before the creating thread returns
 - fork1() loses its stack and tidptr parameters
Common bits factor out:
 - struct proc allocation and initialization moves to thread_new()
 - maxthread handling moves to fork_check_maxthread()
 - setting the new thread running moves to fork_thread_start()
The MD cpu_fork() function swaps its unused stacksize parameter for
a tcb parameter.

luna88k testing by aoyama@, alpha testing by dlg@
ok mpi@

11 files changed, 106 insertions, 116 deletions

diff --git a/sys/arch/alpha/alpha/vm_machdep.c b/sys/arch/alpha/alpha/vm_machdep.c
index 7ff834b9ec4..48f6ec11f2a 100644
--- a/sys/arch/alpha/alpha/vm_machdep.c
+++ b/sys/arch/alpha/alpha/vm_machdep.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: vm_machdep.c,v 1.45 2016/10/04 05:55:09 guenther Exp $ */
+/* $OpenBSD: vm_machdep.c,v 1.46 2017/02/12 04:55:08 guenther Exp $ */
 /* $NetBSD: vm_machdep.c,v 1.55 2000/03/29 03:49:48 simonb Exp $ */
 
 /*
@@ -71,24 +71,19 @@ cpu_exit(p)
  * Copy and update the pcb and trap frame, making the child ready to run.
  * 
  * Rig the child's kernel stack so that it will start out in
- * switch_trampoline() and call child_return() with p2 as an
- * argument. This causes the newly-created child process to go
- * directly to user level with an apparent return value of 0 from
- * fork(), while the parent process returns normally.
+ * proc_trampoline() and call 'func' with 'arg' as an argument.
+ * For normal processes this is child_return(), which causes the
+ * child to go directly to user level with an apparent return value
+ * of 0 from fork(), while the parent process returns normally.
+ * For kernel threads this will be a function that never return.
  *
- * p1 is the process being forked;
- *
- * If an alternate user-level stack is requested (with non-NULL stack arg),
- * set up the user stack pointer
- * accordingly.
+ * An alternate user-level stack or TCB can be requested by passing
+ * a non-NULL value; these are poked into the PCB so they're in
+ * effect at the initial return to userspace.
  */
 void
-cpu_fork(p1, p2, stack, stacksize, func, arg)
-	struct proc *p1, *p2;
-	void *stack;
-	size_t stacksize;
-	void (*func)(void *);
-	void *arg;
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
+    void (*func)(void *), void *arg)
 {
 	struct user *up = p2->p_addr;
 
@@ -115,15 +110,17 @@ cpu_fork(p1, p2, stack, stacksize, func, arg)
 
 	/*
 	 * Copy pcb and stack from proc p1 to p2.
-	 * If specified, give the child a different stack.
+	 * If specified, give the child a different stack and/or TCB.
 	 * We do this as cheaply as possible, copying only the active
 	 * part of the stack.  The stack and pcb need to agree;
 	 */
 	up->u_pcb = p1->p_addr->u_pcb;
 	if (stack != NULL)
-		up->u_pcb.pcb_hw.apcb_usp = (u_long)stack + stacksize;
+		up->u_pcb.pcb_hw.apcb_usp = (u_long)stack;
 	else
 		up->u_pcb.pcb_hw.apcb_usp = alpha_pal_rdusp();
+	if (tcb != NULL)
+		up->u_pcb.pcb_hw.apcb_unique = (unsigned long)tcb;
 
 	/*
 	 * Arrange for a non-local goto when the new process
diff --git a/sys/arch/amd64/amd64/vm_machdep.c b/sys/arch/amd64/amd64/vm_machdep.c
index 2b402e5ec74..42dc47f92ea 100644
--- a/sys/arch/amd64/amd64/vm_machdep.c
+++ b/sys/arch/amd64/amd64/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.35 2016/04/03 19:49:35 guenther Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.36 2017/02/12 04:55:08 guenther Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.1 2003/04/26 18:39:33 fvdl Exp $	*/
 
 /*-
@@ -66,10 +66,10 @@ void setredzone(struct proc *);
  * Finish a fork operation, with process p2 nearly set up.
  * Copy and update the kernel stack and pcb, making the child
  * ready to run, and marking it so that it can return differently
- * than the parent.  Returns 1 in the child process, 0 in the parent.
+ * than the parent.
  */
 void
-cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
     void (*func)(void *), void *arg)
 {
 	struct pcb *pcb = &p2->p_addr->u_pcb;
@@ -112,10 +112,12 @@ cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
 	setredzone(p2);
 
 	/*
-	 * If specified, give the child a different stack.
+	 * If specified, give the child a different stack and/or TCB
 	 */
 	if (stack != NULL)
-		tf->tf_rsp = (u_int64_t)stack + stacksize;
+		tf->tf_rsp = (u_int64_t)stack;
+	if (tcb != NULL)
+		pcb->pcb_fsbase = (u_int64_t)tcb;
 
 	sf = (struct switchframe *)tf - 1;
 	sf->sf_r12 = (u_int64_t)func;
diff --git a/sys/arch/arm/arm/vm_machdep.c b/sys/arch/arm/arm/vm_machdep.c
index 38dba6d7d77..9b2a00a92f5 100644
--- a/sys/arch/arm/arm/vm_machdep.c
+++ b/sys/arch/arm/arm/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.18 2016/07/31 09:18:01 jsg Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.19 2017/02/12 04:55:08 guenther Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.31 2004/01/04 11:33:29 jdolecek Exp $	*/
 
 /*
@@ -80,24 +80,21 @@ extern void proc_trampoline	(void);
 /*
  * Finish a fork operation, with process p2 nearly set up.
  * Copy and update the pcb and trap frame, making the child ready to run.
- * 
- * Rig the child's kernel stack so that it will start out in
- * proc_trampoline() and call child_return() with p2 as an
- * argument. This causes the newly-created child process to go
- * directly to user level with an apparent return value of 0 from
- * fork(), while the parent process returns normally.
  *
- * p1 is the process being forked; if p1 == &proc0, we are creating
- * a kernel thread, and the return path and argument are specified with
- * `func' and `arg'.
+ * Rig the child's kernel stack so that it will start out in
+ * proc_trampoline() and call 'func' with 'arg' as an argument.
+ * For normal processes this is child_return(), which causes the
+ * child to go directly to user level with an apparent return value
+ * of 0 from fork(), while the parent process returns normally.
+ * For kernel threads this will be a function that never return.
  *
- * If an alternate user-level stack is requested (with non-zero values
- * in both the stack and stacksize args), set up the user stack pointer
- * accordingly.
+ * An alternate user-level stack or TCB can be requested by passing
+ * a non-NULL value; these are poked into the PCB so they're in
+ * effect at the initial return to userspace.
  */
 void
-cpu_fork(struct proc *p1, struct proc *p2, void *stack,
-    size_t stacksize, void (*func) (void *), void *arg)
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
+    void (*func)(void *), void *arg)
 {
 	struct pcb *pcb = (struct pcb *)&p2->p_addr->u_pcb;
 	struct trapframe *tf;
@@ -130,16 +127,17 @@ cpu_fork(struct proc *p1, struct proc *p2, void *stack,
 
 	pmap_activate(p2);
 
-	p2->p_addr->u_pcb.pcb_tf = tf =
-	    (struct trapframe *)pcb->pcb_un.un_32.pcb32_sp - 1;
+	pcb->pcb_tf = tf = (struct trapframe *)pcb->pcb_un.un_32.pcb32_sp - 1;
 	*tf = *p1->p_addr->u_pcb.pcb_tf;
 
 	/*
-	 * If specified, give the child a different stack (make sure
-	 * it's 8-byte aligned).
+	 * If specified, give the child a different stack and/or TCB.
+	 * Enforce 8-byte alignment on the stack.
 	 */
 	if (stack != NULL)
-		tf->tf_usr_sp = ((vaddr_t)(stack) + stacksize) & -8;
+		tf->tf_usr_sp = (vaddr_t)stack & -8;
+	if (tcb != NULL)
+		p2->p_addr->u_pcb.pcb_tcb = tcb;
 
 	sf = (struct switchframe *)tf - 1;
 	sf->sf_r4 = (u_int)func;
diff --git a/sys/arch/arm64/arm64/vm_machdep.c b/sys/arch/arm64/arm64/vm_machdep.c
index c2e42e53234..3a879b9dc26 100644
--- a/sys/arch/arm64/arm64/vm_machdep.c
+++ b/sys/arch/arm64/arm64/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.1 2016/12/17 23:38:33 patrick Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.2 2017/02/12 04:55:08 guenther Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.1 2003/04/26 18:39:33 fvdl Exp $	*/
 
 /*-
@@ -66,7 +66,7 @@
  * than the parent.  Returns 1 in the child process, 0 in the parent.
  */
 void
-cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
     void (*func)(void *), void *arg)
 {
 	struct pcb *pcb = (struct pcb *)&p2->p_addr->u_pcb;
@@ -86,11 +86,13 @@ cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
 	    - 0x10);
 
 	tf = (struct trapframe *)STACKALIGN(tf);
-	p2->p_addr->u_pcb.pcb_tf = tf;
+	pcb->pcb_tf = tf;
 	*tf = *p1->p_addr->u_pcb.pcb_tf;
 
 	if (stack != NULL)
-		tf->tf_sp = STACKALIGN((u_int)(stack) + stacksize);
+		tf->tf_sp = STACKALIGN((u_int)(stack));
+	if (tcb != NULL)
+		pcb->pcb_tcb = tcb;
 
 	sf = (struct switchframe *)tf - 1;
 	sf->sf_x19 = (uint64_t)func;
diff --git a/sys/arch/hppa/hppa/vm_machdep.c b/sys/arch/hppa/hppa/vm_machdep.c
index 11bceafc94c..eb517d2def7 100644
--- a/sys/arch/hppa/hppa/vm_machdep.c
+++ b/sys/arch/hppa/hppa/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.81 2015/05/05 02:13:46 guenther Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.82 2017/02/12 04:55:08 guenther Exp $	*/
 
 /*
  * Copyright (c) 1999-2004 Michael Shalayeff
@@ -49,7 +49,7 @@
 extern struct pool hppa_fppl;
 
 void
-cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
     void (*func)(void *), void *arg)
 {
 	struct pcb *pcbp;
@@ -102,10 +102,12 @@ cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
 	    (curcpu()->ci_psw & PSL_O);
 
 	/*
-	 * If specified, give the child a different stack.
+	 * If specified, give the child a different stack and/or TCB
 	 */
 	if (stack != NULL)
 		setstack(tf, (u_long)stack, 0);	/* XXX ignore error? */
+	if (tcb != NULL)
+		tf->tf_cr27 = (u_long)tcb;
 
 	/*
 	 * Build stack frames for the cpu_switchto & co.
diff --git a/sys/arch/i386/i386/vm_machdep.c b/sys/arch/i386/i386/vm_machdep.c
index 49901d20f3e..182795c274f 100644
--- a/sys/arch/i386/i386/vm_machdep.c
+++ b/sys/arch/i386/i386/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.64 2016/04/03 17:45:30 guenther Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.65 2017/02/12 04:55:08 guenther Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.61 1996/05/03 19:42:35 christos Exp $	*/
 
 /*-
@@ -67,12 +67,9 @@
  * Copy and update the kernel stack and pcb, making the child
  * ready to run, and marking it so that it can return differently
  * than the parent.  Returns 1 in the child process, 0 in the parent.
- * We currently double-map the user area so that the stack is at the same
- * address in each process; in the future we will probably relocate
- * the frame pointers on the stack after copying.
  */
 void
-cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
     void (*func)(void *), void *arg)
 {
 	struct pcb *pcb = &p2->p_addr->u_pcb;
@@ -104,10 +101,12 @@ cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
 	*tf = *p1->p_md.md_regs;
 
 	/*
-	 * If specified, give the child a different stack.
+	 * If specified, give the child a different stack and/or TCB
 	 */
 	if (stack != NULL)
-		tf->tf_esp = (u_int)stack + stacksize;
+		tf->tf_esp = (u_int)stack;
+	if (tcb != NULL)
+		i386_set_threadbase(p2, (uint32_t)tcb, TSEG_GS);
 
 	sf = (struct switchframe *)tf - 1;
 	sf->sf_esi = (int)func;
diff --git a/sys/arch/m88k/m88k/vm_machdep.c b/sys/arch/m88k/m88k/vm_machdep.c
index d4cb331f49b..c247482c54f 100644
--- a/sys/arch/m88k/m88k/vm_machdep.c
+++ b/sys/arch/m88k/m88k/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.24 2015/05/05 02:13:47 guenther Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.25 2017/02/12 04:55:08 guenther Exp $	*/
 
 /*
  * Copyright (c) 1998 Steve Murphree, Jr.
@@ -67,18 +67,11 @@ extern void switch_exit(struct proc *);
  * Copy and update the kernel stack and pcb, making the child
  * ready to run, and marking it so that it can return differently
  * than the parent.  Returns 1 in the child process, 0 in the parent.
- * We currently double-map the user area so that the stack is at the same
- * address in each process; in the future we will probably relocate
- * the frame pointers on the stack after copying.
  */
 
 void
-cpu_fork(p1, p2, stack, stacksize, func, arg)
-	struct proc *p1, *p2;
-	void *stack;
-	size_t stacksize;
-	void (*func)(void *);
-	void *arg;
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
+    void (*func)(void *), void *arg)
 {
 	struct ksigframe {
 		void (*func)(void *);
@@ -100,10 +93,12 @@ cpu_fork(p1, p2, stack, stacksize, func, arg)
 	p2->p_md.md_tf = (struct trapframe *)USER_REGS(p2);
 
 	/*
-	 * If specified, give the child a different stack.
+	 * If specified, give the child a different stack and/or TCB.
 	 */
 	if (stack != NULL)
-		USER_REGS(p2)->r[31] = (u_int)stack + stacksize;
+		USER_REGS(p2)->r[31] = (u_int)stack;
+	if (tcb != NULL)
+		USER_REGS(p2)->r[27] = (u_int)tcb;
 
 	ksfp = (struct ksigframe *)((char *)p2->p_addr + USPACE) - 1;
 	ksfp->func = func;
diff --git a/sys/arch/mips64/mips64/vm_machdep.c b/sys/arch/mips64/mips64/vm_machdep.c
index f49bed30773..366d7c6969a 100644
--- a/sys/arch/mips64/mips64/vm_machdep.c
+++ b/sys/arch/mips64/mips64/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.33 2016/03/06 19:42:27 mpi Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.34 2017/02/12 04:55:08 guenther Exp $	*/
 /*
  * Copyright (c) 1988 University of Utah.
  * Copyright (c) 1992, 1993
@@ -61,12 +61,8 @@ extern void proc_trampoline(void);
  * Finish a fork operation, with process p2 nearly set up.
  */
 void
-cpu_fork(p1, p2, stack, stacksize, func, arg)
-	struct proc *p1, *p2;
-	void *stack;
-	size_t stacksize;
-	void (*func)(void *);
-	void *arg;
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
+    void (*func)(void *), void *arg)
 {
 	struct cpu_info *ci = curcpu();
 	struct pcb *pcb;
@@ -117,10 +113,12 @@ cpu_fork(p1, p2, stack, stacksize, func, arg)
 	p2->p_md.md_regs = &p2->p_addr->u_pcb.pcb_regs;
 
 	/*
-	 * If specified, give the child a different stack.
+	 * If specified, give the child a different stack and/or TCB.
 	 */
 	if (stack != NULL)
-		p2->p_md.md_regs->sp = (u_int64_t)stack + stacksize;
+		p2->p_md.md_regs->sp = (u_int64_t)stack;
+	if (tcb != NULL)
+		TCB_SET(p2, tcb);
 
 	/*
 	 * Copy the process control block to the new proc and
diff --git a/sys/arch/powerpc/powerpc/vm_machdep.c b/sys/arch/powerpc/powerpc/vm_machdep.c
index 43e18f7401e..fc64451a15c 100644
--- a/sys/arch/powerpc/powerpc/vm_machdep.c
+++ b/sys/arch/powerpc/powerpc/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.48 2015/05/05 02:13:47 guenther Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.49 2017/02/12 04:55:08 guenther Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.1 1996/09/30 16:34:57 ws Exp $	*/
 
 /*
@@ -49,7 +49,7 @@
  * Finish a fork operation, with process p2 nearly set up.
  */
 void
-cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
     void (*func)(void *), void *arg)
 {
 	struct trapframe *tf;
@@ -92,7 +92,11 @@ cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
 	 */
 	if (stack != NULL) {
 		tf = trapframe(p2);
-		tf->fixreg[1] = (register_t)stack + stacksize;
+		tf->fixreg[1] = (register_t)stack;
+	}
+	if (tcb != NULL) {
+		tf = trapframe(p2);
+		tf->fixreg[2] = (register_t)tcb;
 	}
 
 	stktop2 = (caddr_t)((u_long)stktop2 & ~15);  /* Align stack pointer */
diff --git a/sys/arch/sh/sh/vm_machdep.c b/sys/arch/sh/sh/vm_machdep.c
index 77b50487b5b..16645ebfd0d 100644
--- a/sys/arch/sh/sh/vm_machdep.c
+++ b/sys/arch/sh/sh/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.13 2015/05/05 02:13:47 guenther Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.14 2017/02/12 04:55:08 guenther Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.53 2006/08/31 16:49:21 matt Exp $	*/
 
 /*
@@ -120,23 +120,20 @@ extern void proc_trampoline(void);
 /*
  * Finish a fork operation, with process p2 nearly set up.
  * Copy and update the pcb and trap frame, making the child ready to run.
- *
+ * 
  * Rig the child's kernel stack so that it will start out in
- * proc_trampoline() and call child_return() with p2 as an
- * argument. This causes the newly-created child process to go
- * directly to user level with an apparent return value of 0 from
- * fork(), while the parent process returns normally.
- *
- * p1 is the process being forked; if p1 == &proc0, we are creating
- * a kernel thread, and the return path and argument are specified with
- * `func' and `arg'.
+ * proc_trampoline() and call 'func' with 'arg' as an argument.
+ * For normal processes this is child_return(), which causes the
+ * child to go directly to user level with an apparent return value
+ * of 0 from fork(), while the parent process returns normally.
+ * For kernel threads this will be a function that never return.
  *
- * If an alternate user-level stack is requested (with non-zero values
- * in both the stack and stacksize args), set up the user stack pointer
- * accordingly.
+ * An alternate user-level stack or TCB can be requested by passing
+ * a non-NULL value; these are poked into the PCB so they're in
+ * effect at the initial return to userspace.
  */
 void
-cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
     void (*func)(void *), void *arg)
 {
 	struct pcb *pcb;
@@ -213,7 +210,9 @@ cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
 	 * If specified, give the child a different stack.
 	 */
 	if (stack != NULL)
-		tf->tf_r15 = (u_int)stack + stacksize;
+		tf->tf_r15 = (int)stack;
+	if (tcb != NULL)
+		tf->tf_gbr = (int)tcb;
 
 	/* Setup switch frame */
 	sf = &pcb->pcb_sf;
diff --git a/sys/arch/sparc64/sparc64/vm_machdep.c b/sys/arch/sparc64/sparc64/vm_machdep.c
index c30e79a42a7..18e012ed879 100644
--- a/sys/arch/sparc64/sparc64/vm_machdep.c
+++ b/sys/arch/sparc64/sparc64/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.35 2015/11/06 06:33:26 guenther Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.36 2017/02/12 04:55:08 guenther Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.38 2001/06/30 00:02:20 eeh Exp $ */
 
 /*
@@ -153,26 +153,19 @@ char cpu_forkname[] = "cpu_fork()";
  * Copy and update the pcb and trap frame, making the child ready to run.
  * 
  * Rig the child's kernel stack so that it will start out in
- * proc_trampoline() and call child_return() with p2 as an
- * argument. This causes the newly-created child process to go
- * directly to user level with an apparent return value of 0 from
- * fork(), while the parent process returns normally.
+ * proc_trampoline() and call 'func' with 'arg' as an argument.
+ * For normal processes this is child_return(), which causes the
+ * child to go directly to user level with an apparent return value
+ * of 0 from fork(), while the parent process returns normally.
+ * For kernel threads this will be a function that never return.
  *
- * p1 is the process being forked; if p1 == &proc0, we are creating
- * a kernel thread, and the return path and argument are specified with
- * `func' and `arg'.
- *
- * If an alternate user-level stack is requested (with non-zero values
- * in both the stack and stacksize args), set up the user stack pointer
- * accordingly.
+ * An alternate user-level stack or TCB can be requested by passing
+ * a non-NULL value; these are poked into the PCB so they're in
+ * effect at the initial return to userspace.
  */
 void
-cpu_fork(p1, p2, stack, stacksize, func, arg)
-	struct proc *p1, *p2;
-	void *stack;
-	size_t stacksize;
-	void (*func)(void *);
-	void *arg;
+cpu_fork(struct proc *p1, struct proc *p2, void *stack, void *tcb,
+    void (*func)(void *), void *arg)
 {
 	struct pcb *opcb = &p1->p_addr->u_pcb;
 	struct pcb *npcb = &p2->p_addr->u_pcb;
@@ -240,10 +233,11 @@ cpu_fork(p1, p2, stack, stacksize, func, arg)
 	 * with space reserved for the frame, and zero the frame pointer.
 	 */
 	if (stack != NULL) {
-		tf2->tf_out[6] = (u_int64_t)(u_long)stack + stacksize
-		    - (BIAS + CC64FSZ);
+		tf2->tf_out[6] = (u_int64_t)(u_long)stack - (BIAS + CC64FSZ);
 		tf2->tf_in[6] = 0;
 	}
+	if (tcb != NULL)
+		tf2->tf_global[7] = (u_int64_t)tcb;
 
 	/* Construct kernel frame to return to in cpu_switch() */
 	rp = (struct rwindow *)((u_long)npcb + TOPFRAMEOFF);