Move part of the mp lock logic into per-cpu callbacks; on MVME197DP we need

to disable NMI sources in addition to interrupt sources, and we can not
use a quick sequence with shadowing frozen as done for atomic ops.

This lets GENERIC.MP boot multiuser on MVME197DP boards, and is so far stable
enough to be able to recompile a kernel from scratch (with make -j2).

9 files changed, 144 insertions, 38 deletions

diff --git a/sys/arch/m88k/include/cpu.h b/sys/arch/m88k/include/cpu.h
index b85a7e93dad..90dec8e0e56 100644
--- a/sys/arch/m88k/include/cpu.h
+++ b/sys/arch/m88k/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.43 2009/02/21 18:35:20 miod Exp $ */
+/*	$OpenBSD: cpu.h,v 1.44 2009/02/21 18:37:47 miod Exp $ */
 /*
  * Copyright (c) 1996 Nivas Madhur
  * Copyright (c) 1992, 1993
@@ -79,6 +79,8 @@
 
 extern u_int max_cpus;
 
+#include <machine/lock.h>
+
 /*
  * Per-CPU data structure
  */
@@ -94,6 +96,15 @@ struct cpu_info {
 	u_int		 ci_cpuid;		/* cpu number */
 
 	/*
+	 * Function pointers used within mplock to ensure
+	 * non-interruptability.
+	 */
+	uint32_t	(*ci_mp_atomic_begin)
+			    (__cpu_simple_lock_t *lock, uint *csr);
+	void		(*ci_mp_atomic_end)
+			    (uint32_t psr, __cpu_simple_lock_t *lock, uint csr);
+
+	/*
 	 * The following fields are used differently depending on
 	 * the processor type.  Think of them as an anonymous union
 	 * of two anonymous structs.
@@ -104,8 +115,6 @@ struct cpu_info {
 	u_int		 ci_cpudep3;
 	u_int		 ci_cpudep4;
 	u_int		 ci_cpudep5;
-	u_int		 ci_cpudep6;
-	u_int		 ci_cpudep7;
 
 	/* 88100 fields */
 #define	ci_pfsr_i0	 ci_cpudep0		/* instruction... */
diff --git a/sys/arch/m88k/include/m88100.h b/sys/arch/m88k/include/m88100.h
index ded8b635150..245e2af244f 100644
--- a/sys/arch/m88k/include/m88100.h
+++ b/sys/arch/m88k/include/m88100.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: m88100.h,v 1.4 2007/11/17 05:36:23 miod Exp $ */
+/*	$OpenBSD: m88100.h,v 1.5 2009/02/21 18:37:47 miod Exp $ */
 /*
  * Mach Operating System
  * Copyright (c) 1993-1992 Carnegie Mellon University
@@ -71,6 +71,7 @@ u_int32_t do_xmem_word(vaddr_t, u_int32_t, int);
 u_int8_t  do_xmem_byte(vaddr_t, u_int8_t, int);
 
 void	m88100_apply_patches(void);
+void	m88100_smp_setup(struct cpu_info *);
 
 #endif
 
diff --git a/sys/arch/m88k/m88k/m88100_machdep.c b/sys/arch/m88k/m88k/m88100_machdep.c
index 359019d215d..e97306ca658 100644
--- a/sys/arch/m88k/m88k/m88100_machdep.c
+++ b/sys/arch/m88k/m88k/m88100_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: m88100_machdep.c,v 1.6 2007/12/02 21:17:17 miod Exp $	*/
+/*	$OpenBSD: m88100_machdep.c,v 1.7 2009/02/21 18:37:48 miod Exp $	*/
 /*
  * Mach Operating System
  * Copyright (c) 1993-1991 Carnegie Mellon University
@@ -32,6 +32,11 @@
 #include <machine/asm_macro.h>
 #include <m88k/m88100.h>
 
+#ifdef MULTIPROCESSOR
+uint32_t m88100_mp_atomic_begin(__cpu_simple_lock_t *, uint *);
+void	m88100_mp_atomic_end(uint32_t, __cpu_simple_lock_t *, uint);
+#endif
+
 /*
  *  Data Access Emulation for M88100 exceptions
  */
@@ -305,3 +310,35 @@ m88100_apply_patches()
 	}
 #endif
 }
+
+#ifdef MULTIPROCESSOR
+void
+m88100_smp_setup(struct cpu_info *ci)
+{
+	/*
+	 * Setup function pointers for mplock operation.
+	 */
+
+	ci->ci_mp_atomic_begin = m88100_mp_atomic_begin;
+	ci->ci_mp_atomic_end = m88100_mp_atomic_end;
+}
+
+uint32_t
+m88100_mp_atomic_begin(__cpu_simple_lock_t *lock, uint *csr)
+{
+	uint32_t psr;
+
+	psr = get_psr();
+	set_psr(psr | PSR_IND);
+	__cpu_simple_lock(lock);
+
+	return psr;
+}
+
+void
+m88100_mp_atomic_end(uint32_t psr, __cpu_simple_lock_t *lock, uint csr)
+{
+	__cpu_simple_unlock(lock);
+	set_psr(psr);
+}
+#endif
diff --git a/sys/arch/m88k/m88k/mplock.c b/sys/arch/m88k/m88k/mplock.c
index 8e8903be465..7bb0fddcbcd 100644
--- a/sys/arch/m88k/m88k/mplock.c
+++ b/sys/arch/m88k/m88k/mplock.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mplock.c,v 1.1 2007/12/02 21:20:17 miod Exp $	*/
+/*	$OpenBSD: mplock.c,v 1.2 2009/02/21 18:37:48 miod Exp $	*/
 
 /*
  * Copyright (c) 2004 Niklas Hallqvist.  All rights reserved.
@@ -67,6 +67,8 @@ void
 __mp_lock(struct __mp_lock *mpl)
 {
 	struct cpu_info *ci = curcpu();
+	uint32_t psr;
+	uint gcsr;
 
 	/*
 	 * Please notice that mpl_count gets incremented twice for the
@@ -81,22 +83,18 @@ __mp_lock(struct __mp_lock *mpl)
 	 */
 
 	for (;;) {
-		u_int32_t psr = get_psr();
+		psr = (*ci->ci_mp_atomic_begin)(&mpl->mpl_lock, &gcsr);
 
-		set_psr(psr | PSR_IND);
-		__cpu_simple_lock(&mpl->mpl_lock);
 		if (mpl->mpl_count == 0) {
 			mpl->mpl_count = 1;
 			mpl->mpl_cpu = ci;
 		}
 		if (mpl->mpl_cpu == ci) {
 			mpl->mpl_count++;
-			__cpu_simple_unlock(&mpl->mpl_lock);
-			set_psr(psr);
+			(*ci->ci_mp_atomic_end)(psr, &mpl->mpl_lock, gcsr);
 			break;
 		}
-		__cpu_simple_unlock(&mpl->mpl_lock);
-		set_psr(psr);
+		(*ci->ci_mp_atomic_end)(psr, &mpl->mpl_lock, gcsr);
 
 		__mp_lock_spin(mpl);
 	}
@@ -105,45 +103,45 @@ __mp_lock(struct __mp_lock *mpl)
 void
 __mp_unlock(struct __mp_lock *mpl)
 {
-	u_int32_t psr = get_psr();
+	struct cpu_info *ci = curcpu();
+	u_int32_t psr;
+	uint gcsr;
 
 #ifdef MP_LOCKDEBUG
-	if (mpl->mpl_cpu != curcpu()) {
+	if (mpl->mpl_cpu != ci) {
 		db_printf("__mp_unlock(%p): not held lock\n", mpl);
 		Debugger();
 	}
 #endif
 
-	set_psr(psr | PSR_IND);
-	__cpu_simple_lock(&mpl->mpl_lock);
+	psr = (*ci->ci_mp_atomic_begin)(&mpl->mpl_lock, &gcsr);
 	if (--mpl->mpl_count == 1) {
 		mpl->mpl_cpu = NULL;
 		mpl->mpl_count = 0;
 	}
-	__cpu_simple_unlock(&mpl->mpl_lock);
-	set_psr(psr);
+	(*ci->ci_mp_atomic_end)(psr, &mpl->mpl_lock, gcsr);
 }
 
 int
 __mp_release_all(struct __mp_lock *mpl)
 {
-	u_int32_t psr = get_psr();
+	struct cpu_info *ci = curcpu();
+	u_int32_t psr;
+	uint gcsr;
 	int rv;
 
 #ifdef MP_LOCKDEBUG
-	if (mpl->mpl_cpu != curcpu()) {
+	if (mpl->mpl_cpu != ci) {
 		db_printf("__mp_release_all(%p): not held lock\n", mpl);
 		Debugger();
 	}
 #endif
 
-	set_psr(psr | PSR_IND);
-	__cpu_simple_lock(&mpl->mpl_lock);
+	psr = (*ci->ci_mp_atomic_begin)(&mpl->mpl_lock, &gcsr);
 	rv = mpl->mpl_count - 1;
 	mpl->mpl_cpu = NULL;
 	mpl->mpl_count = 0;
-	__cpu_simple_unlock(&mpl->mpl_lock);
-	set_psr(psr);
+	(*ci->ci_mp_atomic_end)(psr, &mpl->mpl_lock, gcsr);
 
 	return (rv);
 }
@@ -151,22 +149,22 @@ __mp_release_all(struct __mp_lock *mpl)
 int
 __mp_release_all_but_one(struct __mp_lock *mpl)
 {
-	u_int32_t psr = get_psr();
+	struct cpu_info *ci = curcpu();
+	u_int32_t psr;
+	uint gcsr;
 	int rv;
 
 #ifdef MP_LOCKDEBUG
-	if (mpl->mpl_cpu != curcpu()) {
+	if (mpl->mpl_cpu != ci) {
 		db_printf("__mp_release_all_but_one(%p): not held lock\n", mpl);
 		Debugger();
 	}
 #endif
 
-	set_psr(psr | PSR_IND);
-	__cpu_simple_lock(&mpl->mpl_lock);
+	psr = (*ci->ci_mp_atomic_begin)(&mpl->mpl_lock, &gcsr);
 	rv = mpl->mpl_count - 2;
 	mpl->mpl_count = 2;
-	__cpu_simple_unlock(&mpl->mpl_lock);
-	set_psr(psr);
+	(*ci->ci_mp_atomic_end)(psr, &mpl->mpl_lock, gcsr);
 
 	return (rv);
 }
diff --git a/sys/arch/mvme88k/include/cpu.h b/sys/arch/mvme88k/include/cpu.h
index c386d65e2de..a4e5de1a1f3 100644
--- a/sys/arch/mvme88k/include/cpu.h
+++ b/sys/arch/mvme88k/include/cpu.h
@@ -1,4 +1,4 @@
-/* $OpenBSD: cpu.h,v 1.41 2009/02/16 23:03:33 miod Exp $ */
+/* $OpenBSD: cpu.h,v 1.42 2009/02/21 18:37:48 miod Exp $ */
 /*
  * Copyright (c) 1996 Nivas Madhur
  * Copyright (c) 1992, 1993
@@ -57,6 +57,7 @@ extern void	(*md_init_clocks)(void);
 extern void	(*md_send_ipi)(int, cpuid_t);
 extern void	(*md_delay)(int);
 extern void	(*md_soft_ipi)(void);
+extern void	(*md_smp_setup)(struct cpu_info *);
 
 struct intrhand {
 	SLIST_ENTRY(intrhand) ih_link;
diff --git a/sys/arch/mvme88k/mvme88k/locore.S b/sys/arch/mvme88k/mvme88k/locore.S
index 479fc43280f..49691210319 100644
--- a/sys/arch/mvme88k/mvme88k/locore.S
+++ b/sys/arch/mvme88k/mvme88k/locore.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.S,v 1.64 2009/02/21 18:35:22 miod Exp $	*/
+/*	$OpenBSD: locore.S,v 1.65 2009/02/21 18:37:48 miod Exp $	*/
 /*
  * Copyright (c) 2005, Miodrag Vallat.
  * Copyright (c) 1998 Steve Murphree, Jr.
@@ -230,6 +230,14 @@ GLOBAL(secondary_start)
 	bsr.n	_C_LABEL(secondary_main)
 	 addu	r31, r2, USPACE			/* switch to startup stack */
 
+	/*
+	 * Dummy mp_atomic_begin() and mp_atomic_end() routine, so that
+	 * we can interact with ddb if things go wrong very early during
+	 * bootstrap. Of course this should never happen (-:
+	 */
+ASLOCAL(dummy_mplock)
+	jmp	r1
+
 #endif	/* MULTIPROCESSOR */
 
 /*
@@ -350,7 +358,10 @@ ASLOCAL(dummy_cpu)
 	word	0				/* ci_curproc */
 	word	0				/* ci_curpcb */
 	word	0				/* ci_cpuid */
-	space	CPU_INFO_SIZEOF - 4 * 4
+	word	_ASM_LABEL(dummy_mplock)	/* ci_mp_atomic_begin */
+	word	_ASM_LABEL(dummy_mplock)	/* ci_mp_atomic_end */
+
+	space	CPU_INFO_SIZEOF - 6 * 4
 #endif	/* MULTIPROCESSOR */
 
 #if defined(DDB) || NKSYMS > 0
diff --git a/sys/arch/mvme88k/mvme88k/m188_machdep.c b/sys/arch/mvme88k/mvme88k/m188_machdep.c
index 31806f1f708..9ed61c605d1 100644
--- a/sys/arch/mvme88k/mvme88k/m188_machdep.c
+++ b/sys/arch/mvme88k/mvme88k/m188_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: m188_machdep.c,v 1.49 2009/02/20 23:35:11 miod Exp $	*/
+/*	$OpenBSD: m188_machdep.c,v 1.50 2009/02/21 18:37:48 miod Exp $	*/
 /*
  * Copyright (c) 1998, 1999, 2000, 2001 Steve Murphree, Jr.
  * Copyright (c) 1996 Nivas Madhur
@@ -224,6 +224,9 @@ m188_bootstrap()
 	md_soft_ipi = m188_soft_ipi;
 #endif
 	md_delay = m188_delay;
+#ifdef MULTIPROCESSOR
+	md_smp_setup = m88100_smp_setup;
+#endif
 
 	/* clear and disable all interrupts */
 	*(volatile u_int32_t *)MVME188_IENALL = 0;
diff --git a/sys/arch/mvme88k/mvme88k/m197_machdep.c b/sys/arch/mvme88k/mvme88k/m197_machdep.c
index 1614020694a..027fad9e43b 100644
--- a/sys/arch/mvme88k/mvme88k/m197_machdep.c
+++ b/sys/arch/mvme88k/mvme88k/m197_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: m197_machdep.c,v 1.35 2009/02/21 18:35:22 miod Exp $	*/
+/*	$OpenBSD: m197_machdep.c,v 1.36 2009/02/21 18:37:49 miod Exp $	*/
 
 /*
  * Copyright (c) 2009 Miodrag Vallat.
@@ -89,9 +89,12 @@ void	m197_ext_int(struct trapframe *);
 u_int	m197_getipl(void);
 void	m197_ipi_handler(struct trapframe *);
 vaddr_t	m197_memsize(void);
+uint32_t m197_mp_atomic_begin(__cpu_simple_lock_t *, uint *);
+void	m197_mp_atomic_end(uint32_t, __cpu_simple_lock_t *, uint);
 void	m197_nmi(struct trapframe *);
 u_int	m197_raiseipl(u_int);
 u_int	m197_setipl(u_int);
+void	m197_smp_setup(struct cpu_info *);
 void	m197_soft_ipi(void);
 void	m197_startup(void);
 
@@ -406,6 +409,7 @@ m197_bootstrap()
 	md_send_ipi = m197_send_ipi;
 	md_soft_ipi = m197_soft_ipi;
 	md_delay = m197_delay;
+	md_smp_setup = m197_smp_setup;
 #else
 	md_delay = m1x7_delay;
 #endif
@@ -647,4 +651,40 @@ m197_delay(int us)
 	}
 }
 
+void
+m197_smp_setup(struct cpu_info *ci)
+{
+	/*
+	 * Setup function pointers for mplock operation.
+	 */
+
+	ci->ci_mp_atomic_begin = m197_mp_atomic_begin;
+	ci->ci_mp_atomic_end = m197_mp_atomic_end;
+}
+
+uint32_t
+m197_mp_atomic_begin(__cpu_simple_lock_t *lock, uint *csr)
+{
+	uint32_t psr;
+
+	psr = get_psr();
+	set_psr(psr | PSR_IND);
+
+	*csr = *(volatile uint8_t *)(BS_BASE + BS_CPINT);
+	*(volatile uint8_t *)(BS_BASE + BS_CPINT) = 0;
+
+	__cpu_simple_lock(lock);
+
+	return psr;
+}
+
+void
+m197_mp_atomic_end(uint32_t psr, __cpu_simple_lock_t *lock, uint csr)
+{
+	__cpu_simple_unlock(lock);
+
+	*(volatile uint8_t *)(BS_BASE + BS_CPINT) = csr & BS_CPI_IEN;
+
+	set_psr(psr);
+}
 #endif	/* MULTIPROCESSOR */
diff --git a/sys/arch/mvme88k/mvme88k/machdep.c b/sys/arch/mvme88k/mvme88k/machdep.c
index 17228943858..e175e5c5b13 100644
--- a/sys/arch/mvme88k/mvme88k/machdep.c
+++ b/sys/arch/mvme88k/mvme88k/machdep.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: machdep.c,v 1.224 2009/02/21 18:35:22 miod Exp $	*/
+/* $OpenBSD: machdep.c,v 1.225 2009/02/21 18:37:49 miod Exp $	*/
 /*
  * Copyright (c) 1998, 1999, 2000, 2001 Steve Murphree, Jr.
  * Copyright (c) 1996 Nivas Madhur
@@ -136,6 +136,9 @@ void (*md_send_ipi)(int, cpuid_t);
 void (*md_soft_ipi)(void);
 #endif
 void (*md_delay)(int) = dumb_delay;
+#ifdef MULTIPROCESSOR
+void (*md_smp_setup)(struct cpu_info *);
+#endif
 
 int physmem;	  /* available physical memory, in pages */
 
@@ -714,6 +717,7 @@ secondary_pre_main()
 	set_cpu_number(cmmu_cpu_number()); /* Determine cpu number by CMMU */
 	ci = curcpu();
 	ci->ci_curproc = &proc0;
+	(*md_smp_setup)(ci);
 
 	splhigh();
 
@@ -1011,9 +1015,11 @@ mvme_bootstrap()
 	setup_board_config();
 	master_cpu = cmmu_init();
 	set_cpu_number(master_cpu);
+#ifdef MULTIPROCESSOR
+	(*md_smp_setup)(curcpu());
+#endif
 	SET(curcpu()->ci_flags, CIF_ALIVE | CIF_PRIMARY);
 
-
 #ifdef M88100
 	if (CPU_IS88100) {
 		m88100_apply_patches();