Fix page prezeroing for SMP, and fix some potential paging-in-progress

hangs.  The paging-in-progress diagnosis was a result of Tor Egge's
excellent detective work.
Submitted by:	Partially from Tor Egge.

11 files changed, 127 insertions, 108 deletions

diff --git a/sys/amd64/amd64/vm_machdep.c b/sys/amd64/amd64/vm_machdep.c
index 17c587b..c28ad33 100644
--- a/sys/amd64/amd64/vm_machdep.c
+++ b/sys/amd64/amd64/vm_machdep.c
@@ -38,7 +38,7 @@
  *
  *	from: @(#)vm_machdep.c	7.3 (Berkeley) 5/13/91
  *	Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
- *	$Id: vm_machdep.c,v 1.99 1998/02/06 12:13:11 eivind Exp $
+ *	$Id: vm_machdep.c,v 1.100 1998/02/13 05:30:18 bde Exp $
  */
 
 #include "npx.h"
@@ -53,6 +53,8 @@
 #include <sys/buf.h>
 #include <sys/vnode.h>
 #include <sys/vmmeter.h>
+#include <sys/kernel.h>
+#include <sys/sysctl.h>
 
 #include <machine/clock.h>
 #include <machine/cpu.h>
@@ -913,6 +915,11 @@ grow(p, sp)
 	return (1);
 }
 
+
+int cnt_prezero;
+
+SYSCTL_INT(_machdep, OID_AUTO, cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
+
 /*
  * Implement the pre-zeroed page mechanism.
  * This routine is called from the idle loop.
@@ -924,10 +931,6 @@ vm_page_zero_idle()
 	vm_page_t m;
 	int s;
 
-#ifdef WRONG
-	if (cnt.v_free_count <= cnt.v_interrupt_free_min) 
-		return (0);
-#endif
 	/*
 	 * XXX
 	 * We stop zeroing pages when there are sufficent prezeroed pages.
@@ -944,36 +947,39 @@ vm_page_zero_idle()
 	if (cnt.v_free_count - vm_page_zero_count <= cnt.v_free_reserved / 2)
 		return (0);
 #ifdef SMP
-	get_mplock();
+	if (try_mplock()) {
+#endif
+		s = splvm();
+		enable_intr();
+		m = vm_page_list_find(PQ_FREE, free_rover);
+		if (m != NULL) {
+			--(*vm_page_queues[m->queue].lcnt);
+			TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq);
+			m->queue = PQ_NONE;
+			splx(s);
+#if 0
+			rel_mplock();
+#endif
+			pmap_zero_page(VM_PAGE_TO_PHYS(m));
+#ifdef 0
+			get_mplock();
 #endif
-	s = splvm();
-	enable_intr();
-	m = vm_page_list_find(PQ_FREE, free_rover);
-	if (m != NULL) {
-		--(*vm_page_queues[m->queue].lcnt);
-		TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq);
-		m->queue = PQ_NONE;
+			(void)splvm();
+			m->queue = PQ_ZERO + m->pc;
+			++(*vm_page_queues[m->queue].lcnt);
+			TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, pageq);
+			free_rover = (free_rover + PQ_PRIME3) & PQ_L2_MASK;
+			++vm_page_zero_count;
+			++cnt_prezero;
+		}
 		splx(s);
+		disable_intr();
 #ifdef SMP
 		rel_mplock();
-#endif
-		pmap_zero_page(VM_PAGE_TO_PHYS(m));
-#ifdef SMP
-		get_mplock();
-#endif
-		(void)splvm();
-		m->queue = PQ_ZERO + m->pc;
-		++(*vm_page_queues[m->queue].lcnt);
-		TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, pageq);
-		free_rover = (free_rover + PQ_PRIME3) & PQ_L2_MASK;
-		++vm_page_zero_count;
+		return (1);
 	}
-	splx(s);
-	disable_intr();
-#ifdef SMP
-	rel_mplock();
 #endif
-	return (1);
+	return (0);
 }
 
 /*
diff --git a/sys/amd64/include/smp.h b/sys/amd64/include/smp.h
index 4d31598..7867fce 100644
--- a/sys/amd64/include/smp.h
+++ b/sys/amd64/include/smp.h
@@ -6,7 +6,7 @@
  * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
  * ----------------------------------------------------------------------------
  *
- * $Id: smp.h,v 1.33 1997/12/08 22:56:44 fsmp Exp $
+ * $Id: smp.h,v 1.34 1998/01/01 08:47:12 bde Exp $
  *
  */
 
@@ -67,11 +67,11 @@ extern u_int			mpintr_lock;
 /* functions in mplock.s */
 void	get_mplock		__P((void));
 void	rel_mplock		__P((void));
-void	try_mplock		__P((void));
+int	try_mplock		__P((void));
 #ifdef RECURSIVE_MPINTRLOCK
 void	get_mpintrlock		__P((void));
 void	rel_mpintrlock		__P((void));
-void	try_mpintrlock		__P((void));
+int	try_mpintrlock		__P((void));
 #endif /*  RECURSIVE_MPINTRLOCK */
 
 /* global data in apic_vector.s */
diff --git a/sys/i386/i386/vm_machdep.c b/sys/i386/i386/vm_machdep.c
index 17c587b..c28ad33 100644
--- a/sys/i386/i386/vm_machdep.c
+++ b/sys/i386/i386/vm_machdep.c
@@ -38,7 +38,7 @@
  *
  *	from: @(#)vm_machdep.c	7.3 (Berkeley) 5/13/91
  *	Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
- *	$Id: vm_machdep.c,v 1.99 1998/02/06 12:13:11 eivind Exp $
+ *	$Id: vm_machdep.c,v 1.100 1998/02/13 05:30:18 bde Exp $
  */
 
 #include "npx.h"
@@ -53,6 +53,8 @@
 #include <sys/buf.h>
 #include <sys/vnode.h>
 #include <sys/vmmeter.h>
+#include <sys/kernel.h>
+#include <sys/sysctl.h>
 
 #include <machine/clock.h>
 #include <machine/cpu.h>
@@ -913,6 +915,11 @@ grow(p, sp)
 	return (1);
 }
 
+
+int cnt_prezero;
+
+SYSCTL_INT(_machdep, OID_AUTO, cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
+
 /*
  * Implement the pre-zeroed page mechanism.
  * This routine is called from the idle loop.
@@ -924,10 +931,6 @@ vm_page_zero_idle()
 	vm_page_t m;
 	int s;
 
-#ifdef WRONG
-	if (cnt.v_free_count <= cnt.v_interrupt_free_min) 
-		return (0);
-#endif
 	/*
 	 * XXX
 	 * We stop zeroing pages when there are sufficent prezeroed pages.
@@ -944,36 +947,39 @@ vm_page_zero_idle()
 	if (cnt.v_free_count - vm_page_zero_count <= cnt.v_free_reserved / 2)
 		return (0);
 #ifdef SMP
-	get_mplock();
+	if (try_mplock()) {
+#endif
+		s = splvm();
+		enable_intr();
+		m = vm_page_list_find(PQ_FREE, free_rover);
+		if (m != NULL) {
+			--(*vm_page_queues[m->queue].lcnt);
+			TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq);
+			m->queue = PQ_NONE;
+			splx(s);
+#if 0
+			rel_mplock();
+#endif
+			pmap_zero_page(VM_PAGE_TO_PHYS(m));
+#ifdef 0
+			get_mplock();
 #endif
-	s = splvm();
-	enable_intr();
-	m = vm_page_list_find(PQ_FREE, free_rover);
-	if (m != NULL) {
-		--(*vm_page_queues[m->queue].lcnt);
-		TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq);
-		m->queue = PQ_NONE;
+			(void)splvm();
+			m->queue = PQ_ZERO + m->pc;
+			++(*vm_page_queues[m->queue].lcnt);
+			TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, pageq);
+			free_rover = (free_rover + PQ_PRIME3) & PQ_L2_MASK;
+			++vm_page_zero_count;
+			++cnt_prezero;
+		}
 		splx(s);
+		disable_intr();
 #ifdef SMP
 		rel_mplock();
-#endif
-		pmap_zero_page(VM_PAGE_TO_PHYS(m));
-#ifdef SMP
-		get_mplock();
-#endif
-		(void)splvm();
-		m->queue = PQ_ZERO + m->pc;
-		++(*vm_page_queues[m->queue].lcnt);
-		TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, pageq);
-		free_rover = (free_rover + PQ_PRIME3) & PQ_L2_MASK;
-		++vm_page_zero_count;
+		return (1);
 	}
-	splx(s);
-	disable_intr();
-#ifdef SMP
-	rel_mplock();
 #endif
-	return (1);
+	return (0);
 }
 
 /*
diff --git a/sys/i386/include/smp.h b/sys/i386/include/smp.h
index 4d31598..7867fce 100644
--- a/sys/i386/include/smp.h
+++ b/sys/i386/include/smp.h
@@ -6,7 +6,7 @@
  * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
  * ----------------------------------------------------------------------------
  *
- * $Id: smp.h,v 1.33 1997/12/08 22:56:44 fsmp Exp $
+ * $Id: smp.h,v 1.34 1998/01/01 08:47:12 bde Exp $
  *
  */
 
@@ -67,11 +67,11 @@ extern u_int			mpintr_lock;
 /* functions in mplock.s */
 void	get_mplock		__P((void));
 void	rel_mplock		__P((void));
-void	try_mplock		__P((void));
+int	try_mplock		__P((void));
 #ifdef RECURSIVE_MPINTRLOCK
 void	get_mpintrlock		__P((void));
 void	rel_mpintrlock		__P((void));
-void	try_mpintrlock		__P((void));
+int	try_mpintrlock		__P((void));
 #endif /*  RECURSIVE_MPINTRLOCK */
 
 /* global data in apic_vector.s */
diff --git a/sys/sys/smp.h b/sys/sys/smp.h
index 4d31598..7867fce 100644
--- a/sys/sys/smp.h
+++ b/sys/sys/smp.h
@@ -6,7 +6,7 @@
  * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
  * ----------------------------------------------------------------------------
  *
- * $Id: smp.h,v 1.33 1997/12/08 22:56:44 fsmp Exp $
+ * $Id: smp.h,v 1.34 1998/01/01 08:47:12 bde Exp $
  *
  */
 
@@ -67,11 +67,11 @@ extern u_int			mpintr_lock;
 /* functions in mplock.s */
 void	get_mplock		__P((void));
 void	rel_mplock		__P((void));
-void	try_mplock		__P((void));
+int	try_mplock		__P((void));
 #ifdef RECURSIVE_MPINTRLOCK
 void	get_mpintrlock		__P((void));
 void	rel_mpintrlock		__P((void));
-void	try_mpintrlock		__P((void));
+int	try_mpintrlock		__P((void));
 #endif /*  RECURSIVE_MPINTRLOCK */
 
 /* global data in apic_vector.s */
diff --git a/sys/vm/swap_pager.c b/sys/vm/swap_pager.c
index a4befb0..05d8a327 100644
--- a/sys/vm/swap_pager.c
+++ b/sys/vm/swap_pager.c
@@ -39,7 +39,7 @@
  * from: Utah $Hdr: swap_pager.c 1.4 91/04/30$
  *
  *	@(#)swap_pager.c	8.9 (Berkeley) 3/21/94
- * $Id: swap_pager.c,v 1.88 1998/02/09 06:11:20 eivind Exp $
+ * $Id: swap_pager.c,v 1.89 1998/02/23 08:22:24 dyson Exp $
  */
 
 /*
@@ -144,7 +144,6 @@ static boolean_t
 					int *before, int *after));
 static int	swap_pager_getpages __P((vm_object_t, vm_page_t *, int, int));
 static void	swap_pager_init __P((void));
-static void	swap_pager_sync __P((void));
 static void spc_free __P((swp_clean_t));
 
 struct pagerops swappagerops = {
@@ -1550,7 +1549,7 @@ swap_pager_putpages(object, m, count, sync, rtvals)
 	return (rv);
 }
 
-static void
+void
 swap_pager_sync()
 {
 	swp_clean_t spc;
diff --git a/sys/vm/swap_pager.h b/sys/vm/swap_pager.h
index 0fa5845..8755547 100644
--- a/sys/vm/swap_pager.h
+++ b/sys/vm/swap_pager.h
@@ -36,7 +36,7 @@
  * SUCH DAMAGE.
  *
  *	from: @(#)swap_pager.h	7.1 (Berkeley) 12/5/90
- *	$Id: swap_pager.h,v 1.18 1997/02/22 09:48:08 peter Exp $
+ *	$Id: swap_pager.h,v 1.19 1998/02/23 08:22:27 dyson Exp $
  */
 
 /*
@@ -78,6 +78,7 @@ void swap_pager_copy __P((vm_object_t, vm_pindex_t, vm_object_t,
 void swap_pager_freespace __P((vm_object_t, vm_pindex_t, vm_size_t));
 void swap_pager_dmzspace __P((vm_object_t, vm_pindex_t, vm_size_t));
 void swap_pager_swap_init __P((void));
+void swap_pager_sync __P((void));
 #endif
 
 #endif				/* _SWAP_PAGER_ */
diff --git a/sys/vm/vm_map.c b/sys/vm/vm_map.c
index cb4b22f..930d687 100644
--- a/sys/vm/vm_map.c
+++ b/sys/vm/vm_map.c
@@ -61,7 +61,7 @@
  * any improvements or extensions that they make and grant Carnegie the
  * rights to redistribute these changes.
  *
- * $Id: vm_map.c,v 1.115 1998/02/20 13:11:54 bde Exp $
+ * $Id: vm_map.c,v 1.116 1998/02/23 08:22:33 dyson Exp $
  */
 
 /*
@@ -2632,12 +2632,7 @@ vm_freeze_copyopts(object, froma, toa)
 
 		vm_object_reference(robject);
 
-		s = splvm();
-		while (robject->paging_in_progress) {
-			robject->flags |= OBJ_PIPWNT;
-			tsleep(robject, PVM, "objfrz", 0);
-		}
-		splx(s);
+		vm_object_pip_wait(robject, "objfrz");
 
 		if (robject->ref_count == 1) {
 			vm_object_deallocate(robject);
diff --git a/sys/vm/vm_object.c b/sys/vm/vm_object.c
index 22b3f7a..ad339d4 100644
--- a/sys/vm/vm_object.c
+++ b/sys/vm/vm_object.c
@@ -61,7 +61,7 @@
  * any improvements or extensions that they make and grant Carnegie the
  * rights to redistribute these changes.
  *
- * $Id: vm_object.c,v 1.112 1998/02/06 12:14:26 eivind Exp $
+ * $Id: vm_object.c,v 1.113 1998/02/09 06:11:30 eivind Exp $
  */
 
 /*
@@ -334,21 +334,23 @@ vm_object_deallocate(object)
 					robject->ref_count++;
 
 			retry:
-					s = splvm();
-					if (robject->paging_in_progress) {
-						robject->flags |= OBJ_PIPWNT;
-						tsleep(robject, PVM, "objde1", 0);
-						splx(s);
-						goto retry;
-					}
-
-					if (object->paging_in_progress) {
-						object->flags |= OBJ_PIPWNT;
-						tsleep(object, PVM, "objde2", 0);
-						splx(s);
+					if (robject->paging_in_progress || object->paging_in_progress) {
+						vm_object_pip_sleep(robject, "objde1");
+						if (robject->paging_in_progress) {
+							if (robject->type == OBJT_SWAP) {
+								swap_pager_sync();
+								goto retry;
+							}
+						}
+
+						vm_object_pip_sleep(object, "objde2");
+						if (object->paging_in_progress) {
+							if (object->type == OBJT_SWAP) {
+								swap_pager_sync();
+							}
+						}
 						goto retry;
 					}
-					splx(s);
 
 					if( robject->ref_count == 1) {
 						robject->ref_count--;
@@ -408,12 +410,7 @@ vm_object_terminate(object)
 	/*
 	 * wait for the pageout daemon to be done with the object
 	 */
-	s = splvm();
-	while (object->paging_in_progress) {
-		object->flags |= OBJ_PIPWNT;
-		tsleep(object, PVM, "objtrm", 0);
-	}
-	splx(s);
+	vm_object_pip_wait(object, "objtrm");
 
 #if defined(DIAGNOSTIC)
 	if (object->paging_in_progress != 0)
diff --git a/sys/vm/vm_object.h b/sys/vm/vm_object.h
index fe41881..76cb4d2 100644
--- a/sys/vm/vm_object.h
+++ b/sys/vm/vm_object.h
@@ -61,7 +61,7 @@
  * any improvements or extensions that they make and grant Carnegie the
  * rights to redistribute these changes.
  *
- * $Id: vm_object.h,v 1.44 1998/01/31 11:56:43 dyson Exp $
+ * $Id: vm_object.h,v 1.45 1998/02/05 03:32:45 dyson Exp $
  */
 
 /*
@@ -164,6 +164,28 @@ vm_object_pip_wakeup(vm_object_t object)
 	}
 }
 
+static __inline void
+vm_object_pip_sleep(vm_object_t object, char *waitid)
+{
+	int s;
+
+	if (object->paging_in_progress) {
+		s = splvm();
+		if (object->paging_in_progress) {
+			object->flags |= OBJ_PIPWNT;
+			tsleep(object, PVM, waitid, 0);
+		}
+		splx(s);
+	}
+}
+
+static __inline void
+vm_object_pip_wait(vm_object_t object, char *waitid)
+{
+	while (object->paging_in_progress)
+		vm_object_pip_sleep(object, waitid);
+}
+
 vm_object_t vm_object_allocate __P((objtype_t, vm_size_t));
 void _vm_object_allocate __P((objtype_t, vm_size_t, vm_object_t));
 boolean_t vm_object_coalesce __P((vm_object_t, vm_pindex_t, vm_size_t, vm_size_t));
diff --git a/sys/vm/vnode_pager.c b/sys/vm/vnode_pager.c
index e7a9efb..98b57b0 100644
--- a/sys/vm/vnode_pager.c
+++ b/sys/vm/vnode_pager.c
@@ -38,7 +38,7 @@
  * SUCH DAMAGE.
  *
  *	from: @(#)vnode_pager.c	7.5 (Berkeley) 4/20/91
- *	$Id: vnode_pager.c,v 1.84 1998/02/06 12:14:30 eivind Exp $
+ *	$Id: vnode_pager.c,v 1.85 1998/02/23 08:22:48 dyson Exp $
  */
 
 /*
@@ -169,14 +169,7 @@ vnode_pager_dealloc(object)
 	if (vp == NULL)
 		panic("vnode_pager_dealloc: pager already dealloced");
 
-	if (object->paging_in_progress) {
-		int s = splvm();
-		while (object->paging_in_progress) {
-			object->flags |= OBJ_PIPWNT;
-			tsleep(object, PVM, "vnpdea", 0);
-		}
-		splx(s);
-	}
+	vm_object_pip_wait(object, "vnpdea");
 
 	object->handle = NULL;
 	object->type = OBJT_DEAD;