always create a stack frame in cpu_switch() and put a stub

for doing page idle zeroing.
store the kernele stack pointer into pcb instead of trapframe
for debuggers doing stack traces; from kettenis@

4 files changed, 96 insertions, 78 deletions

diff --git a/sys/arch/hppa/hppa/genassym.cf b/sys/arch/hppa/hppa/genassym.cf
index 35626288f50..ee718b10e25 100644
--- a/sys/arch/hppa/hppa/genassym.cf
+++ b/sys/arch/hppa/hppa/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.24 2003/07/15 18:15:41 mickey Exp $
+#	$OpenBSD: genassym.cf,v 1.25 2004/06/08 22:00:25 mickey Exp $
 
 #
 # Copyright (c) 1982, 1990, 1993
@@ -148,10 +148,14 @@ member	pcb_fpregs
 member	pcb_onfault
 member	pcb_space
 member	pcb_uva
+member	pcb_ksp
 
 struct	user
 member	u_pcb
 
+struct	uvm
+member	page_idle_zero
+
 # system calls
 export	SYSCALLGATE
 export	SYS_exit
diff --git a/sys/arch/hppa/hppa/locore.S b/sys/arch/hppa/hppa/locore.S
index 3bd089e3632..883c4d3eab8 100644
--- a/sys/arch/hppa/hppa/locore.S
+++ b/sys/arch/hppa/hppa/locore.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.S,v 1.132 2004/05/25 22:14:02 mickey Exp $	*/
+/*	$OpenBSD: locore.S,v 1.133 2004/06/08 22:00:25 mickey Exp $	*/
 
 /*
  * Copyright (c) 1998-2004 Michael Shalayeff
@@ -2708,40 +2708,55 @@ EXIT(remrunqueue)
 ENTRY(cpu_switch,128)
 	ldil	L%cpl, t1
 	ldw	R%cpl(t1), ret0
-	stw	ret0, HPPA_FRAME_SL(sp)
+	copy	r3, r1
 	stw	rp, HPPA_FRAME_CRP(sp)
+	copy	sp, r3
+	stw	ret0, HPPA_FRAME_SL(sp)
+	stwm	r1, HPPA_FRAME_SIZE+20*4(sp)
 
 	/*
 	 * Clear curproc so that we don't accumulate system time while idle.
 	 */
 	ldil	L%curproc, t1
 	ldw	R%curproc(t1), arg2
+	b	switch_search
 	stw	r0, R%curproc(t1)
 	/* remain on the old (curproc)'s stack until we have a better choice */
 
+cpu_idle
+	copy	r0, arg0
+	break	HPPA_BREAK_KERNEL, HPPA_BREAK_SPLLOWER
+	.import uvm, data
+	ldil	L%(uvm + PAGE_IDLE_ZERO), t3
+	ldw	R%(uvm + PAGE_IDLE_ZERO)(t3), t4
+	sub,<>	r0, t4, r0
+	b	cpu_loop
+
+	stw	arg2, 4(r3)
+	ldil	L%uvm_pageidlezero, t1
+	ldo	R%uvm_pageidlezero(t1), t2
+	.call
+	ble	0(sr0, t2)
+	copy	r31, rp
+
+	ldw	HPPA_FRAME_SL(r3), ret0
+	ldw	4(r3), arg2
+
+cpu_loop
+	ldil	L%cpl, arg0
+	stw	ret0, R%cpl(arg0)
+
+switch_search
 	/*
 	 * t1:   &whichqs
 	 * arg2: old curproc
 	 *
 	 */
-
-switch_search
 	ldil	L%whichqs, t1
 	ldw	R%whichqs(t1), t3
-	comb,<>	r0, t3, gotprocs
-	nop
-
+	comb,=,n r0, t3, cpu_idle
 	copy	r0, arg0
-	break	HPPA_BREAK_KERNEL, HPPA_BREAK_SPLLOWER
-
-	/* XXX do idle work here */
-	nop ! nop ! nop ! nop ! nop ! nop ! nop ! nop
-
-	ldil	L%cpl, arg0
-	b	switch_search
-	stw	ret0, R%cpl(arg0)
 
-gotprocs
 	ldi	0, t4
 getbit
 	mtsar	t4
@@ -2797,9 +2812,8 @@ link_ok
 	 * the new process' kernel stack must be reasonable.
 	 */
 	comb,=,n arg1, arg2, kstack_ok
-	ldw     P_MD_REGS(arg1), t1
 	ldw     P_ADDR(arg1), arg0
-	ldw     TF_R30(t1), t1
+	ldw	U_PCB+PCB_KSP(arg0), t1
 	ldo     NBPG(arg0), arg0
 	comb,>>,n arg0, t1, switch_error
 	copy    arg1, t2
@@ -2821,30 +2835,27 @@ kstack_ok
 	 *
 	 * arg2: old proc
 	 */
-	ldw	P_MD_REGS(arg2), t1
-	ldw	TF_R30(t1), t3
-	copy	sp, t2
-	stw,ma	r3, HPPA_FRAME_SIZE+20*4(sp)
-	stw	sp, TF_R30(t1)
-	stw	t2, HPPA_FRAME_PSP(sp)
-	stw	t3, HPPA_FRAME_ARG(0)(sp)	/* real user's stack */
-	fdc	r0(t1)
+	ldw	P_ADDR(arg2), t1
 	/* save callee-save registers */
-	stw	r4,   1*4(t2)
-	stw	r5,   2*4(t2)
-	stw	r6,   3*4(t2)
-	stw	r7,   4*4(t2)
-	stw	r8,   5*4(t2)
-	stw	r9,   6*4(t2)
-	stw	r10,  7*4(t2)
-	stw	r11,  8*4(t2)
-	stw	r12,  9*4(t2)
-	stw	r13, 10*4(t2)
-	stw	r14, 11*4(t2)
-	stw	r15, 12*4(t2)
-	stw	r16, 13*4(t2)
-	stw	r17, 14*4(t2)
-	stw	r18, 15*4(t2)
+	stw	r4,   1*4(r3)
+	stw	sp, U_PCB+PCB_KSP(t1)
+	stw	r5,   2*4(r3)
+	stw	r6,   3*4(r3)
+	stw	r7,   4*4(r3)
+	stw	r8,   5*4(r3)
+	stw	r9,   6*4(r3)
+	stw	r10,  7*4(r3)
+	stw	r11,  8*4(r3)
+	stw	r12,  9*4(r3)
+	stw	r13, 10*4(r3)
+	stw	r14, 11*4(r3)
+	stw	r15, 12*4(r3)
+	stw	r16, 13*4(r3)
+	stw	r17, 14*4(r3)
+	stw	r18, 15*4(r3)
+	fdc	r0(t1)
+	stw	r0, HPPA_FRAME_ARG(1)(sp)	/* say no trampoline */
+	sync
 
 	/* don't need old curproc (arg2) starting from here */
 switch_exited
@@ -2853,44 +2864,46 @@ switch_exited
 	 *
 	 * arg1: new proc
 	 */
-	mtctl	r0, ccr			/* disable FPU */
+	ldw	P_ADDR(arg1), t2
 	ldw	P_MD_REGS(arg1), t1
+	ldw	U_PCB+PCB_KSP(t2), sp
+	mtctl	r0, ccr			/* disable FPU */
 	ldw	TF_CR30(t1), t2
-	ldw	TF_R30(t1), sp
 	ldw	TF_CR9(t1), t3
 	mtctl	t2, cr30
 	mtctl	t3, pidr2
-	ldw	HPPA_FRAME_PSP(sp), t2
-	ldw	HPPA_FRAME_ARG(0)(sp), t3
+	ldo	-(HPPA_FRAME_SIZE+20*4)(sp), r3
+	ldw	HPPA_FRAME_ARG(0)(sp), arg0
 	ldw	HPPA_FRAME_ARG(1)(sp), t4 /* in case we're on trampoline */
-	ldw	HPPA_FRAME_ARG(2)(sp), arg0
-	ldw	HPPA_FRAME_SL(t2), ret0
-	ldw	HPPA_FRAME_CRP(t2), rp
-	ldw	 0*4(t2), r3
-	ldw	 1*4(t2), r4
-	ldw	 2*4(t2), r5
-	ldw	 3*4(t2), r6
-	ldw	 4*4(t2), r7
-	ldw	 5*4(t2), r8
-	ldw	 6*4(t2), r9
-	ldw	 7*4(t2), r10
-	ldw	 8*4(t2), r11
-	ldw	 9*4(t2), r12
-	ldw	10*4(t2), r13
-	ldw	11*4(t2), r14
-	ldw	12*4(t2), r15
-	ldw	13*4(t2), r16
-	ldw	14*4(t2), r17
-	ldw	15*4(t2), r18
-	stw	t3, TF_R30(t1)
-	copy	t2, sp
+	sub,=	r0, t4, r0
+	b	switch_gonnajump
+	ldw	 1*4(r3), r4
+	ldw	 2*4(r3), r5
+	ldw	 3*4(r3), r6
+	ldw	 4*4(r3), r7
+	ldw	 5*4(r3), r8
+	ldw	 6*4(r3), r9
+	ldw	 7*4(r3), r10
+	ldw	 8*4(r3), r11
+	ldw	 9*4(r3), r12
+	ldw	10*4(r3), r13
+	ldw	11*4(r3), r14
+	ldw	12*4(r3), r15
+	ldw	13*4(r3), r16
+	ldw	14*4(r3), r17
+	ldw	15*4(r3), r18
+switch_gonnajump
+	ldw	HPPA_FRAME_SL(r3), ret0
 	ldil	L%cpl, t1
 	stw	ret0, R%cpl(t1)
+	sync
 
 switch_return
 	ldil	L%curproc, t1
-	bv	0(rp)
 	stw	arg1, R%curproc(t1)
+	ldw	HPPA_FRAME_CRP(r3), rp
+	bv	0(rp)
+	ldwm	-(HPPA_FRAME_SIZE+20*4)(sp), r3
 EXIT(cpu_switch)
 
 ENTRY(switch_trampoline,0)
diff --git a/sys/arch/hppa/hppa/vm_machdep.c b/sys/arch/hppa/hppa/vm_machdep.c
index 90259aacc68..c16b394c608 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 1.53 2004/04/21 23:09:30 mickey Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.54 2004/06/08 22:00:25 mickey Exp $	*/
 
 /*
  * Copyright (c) 1999-2004 Michael Shalayeff
@@ -206,18 +206,18 @@ cpu_fork(p1, p2, stack, stacksize, func, arg)
 		tf->tf_sp = (register_t)stack;
 
 	/*
-	 * Build a stack frame for the cpu_switch & co.
+	 * Build stack frames for the cpu_switch & co.
 	 */
 	osp = sp + HPPA_FRAME_SIZE;
-	sp += 2*HPPA_FRAME_SIZE + 20*4; /* std frame + calee-save registers */
-	*HPPA_FRAME_CARG(0, sp) = tf->tf_sp;
-	*HPPA_FRAME_CARG(1, sp) = KERNMODE(func);
-	*HPPA_FRAME_CARG(2, sp) = (register_t)arg;
-	*(register_t*)(osp) = (sp - HPPA_FRAME_SIZE);
-	*(register_t*)(sp + HPPA_FRAME_PSP) = osp;
+	*(register_t*)(osp - HPPA_FRAME_SIZE) = 0;
 	*(register_t*)(osp + HPPA_FRAME_CRP) = (register_t)&switch_trampoline;
 	*(register_t*)(osp + HPPA_FRAME_SL) = 0;	/* cpl */
-	tf->tf_sp = sp;
+	*(register_t*)(osp) = (osp - HPPA_FRAME_SIZE);
+
+	sp = osp + HPPA_FRAME_SIZE + 20*4; /* frame + calee-save registers */
+	*HPPA_FRAME_CARG(0, sp) = (register_t)arg;
+	*HPPA_FRAME_CARG(1, sp) = KERNMODE(func);
+	pcbp->pcb_ksp = sp;
 	fdcache(HPPA_SID_KERNEL, (vaddr_t)p2->p_addr, sp - (vaddr_t)p2->p_addr);
 }
 
diff --git a/sys/arch/hppa/include/pcb.h b/sys/arch/hppa/include/pcb.h
index 7d2c62bb708..16da7dd0053 100644
--- a/sys/arch/hppa/include/pcb.h
+++ b/sys/arch/hppa/include/pcb.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pcb.h,v 1.8 2004/04/07 18:24:19 mickey Exp $	*/
+/*	$OpenBSD: pcb.h,v 1.9 2004/06/08 22:00:25 mickey Exp $	*/
 
 /*
  * Copyright (c) 1999-2004 Michael Shalayeff
@@ -37,6 +37,7 @@ struct pcb {
 	u_int pcb_onfault;		/* SW copy fault handler */
 	pa_space_t pcb_space;		/* copy pmap_space, for asm's sake */
 	vaddr_t pcb_uva;		/* KVA for U-area */
+	u_int pcb_ksp;			/* kernel sp for ctxsw */
 };
 
 struct md_coredump {