Back out all the buffer cache changes I committed during c2k9. This reverts three

commits:

1) The sysctl allowing bufcachepercent to be changed at boot time.
2) The change moving the buffer cache hash chains to a red-black tree
3) The dynamic buffer cache (Which depended on the earlier too).

ok on the backout from marco and todd

27 files changed, 176 insertions, 254 deletions

diff --git a/sys/arch/amd64/amd64/machdep.c b/sys/arch/amd64/amd64/machdep.c
index 4631a8f8c3e..bb3c8e97c59 100644
--- a/sys/arch/amd64/amd64/machdep.c
+++ b/sys/arch/amd64/amd64/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.95 2009/06/14 03:04:07 deraadt Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.96 2009/06/15 17:01:25 beck Exp $	*/
 /*	$NetBSD: machdep.c,v 1.3 2003/05/07 22:58:18 fvdl Exp $	*/
 
 /*-
@@ -399,6 +399,12 @@ setup_buffers()
 	 */
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
+
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
 }
 
 /*
diff --git a/sys/arch/aviion/aviion/machdep.c b/sys/arch/aviion/aviion/machdep.c
index 3b8237656c5..ebff8ba9465 100644
--- a/sys/arch/aviion/aviion/machdep.c
+++ b/sys/arch/aviion/aviion/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.28 2009/06/03 21:30:19 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.29 2009/06/15 17:01:25 beck Exp $	*/
 /*
  * Copyright (c) 2007 Miodrag Vallat.
  *
@@ -293,6 +293,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/hp300/hp300/machdep.c b/sys/arch/hp300/hp300/machdep.c
index d560a1b7c58..c4b391071c6 100644
--- a/sys/arch/hp300/hp300/machdep.c
+++ b/sys/arch/hp300/hp300/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.121 2009/06/03 21:30:19 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.122 2009/06/15 17:01:25 beck Exp $	*/
 /*	$NetBSD: machdep.c,v 1.121 1999/03/26 23:41:29 mycroft Exp $	*/
 
 /*
@@ -290,6 +290,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/hppa/hppa/machdep.c b/sys/arch/hppa/hppa/machdep.c
index f7c0af3f823..843dac586e3 100644
--- a/sys/arch/hppa/hppa/machdep.c
+++ b/sys/arch/hppa/hppa/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.172 2009/06/03 21:30:19 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.173 2009/06/15 17:01:25 beck Exp $	*/
 
 /*
  * Copyright (c) 1999-2003 Michael Shalayeff
@@ -653,6 +653,12 @@ cpu_startup(void)
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/hppa64/hppa64/machdep.c b/sys/arch/hppa64/hppa64/machdep.c
index e3385e60162..33a408a7a02 100644
--- a/sys/arch/hppa64/hppa64/machdep.c
+++ b/sys/arch/hppa64/hppa64/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.14 2009/06/03 21:30:19 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.15 2009/06/15 17:01:25 beck Exp $	*/
 
 /*
  * Copyright (c) 2005 Michael Shalayeff
@@ -406,6 +406,12 @@ cpu_startup(void)
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 printf("here3\n");
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
diff --git a/sys/arch/i386/i386/machdep.c b/sys/arch/i386/i386/machdep.c
index ad3a266aa0f..2969cc3413c 100644
--- a/sys/arch/i386/i386/machdep.c
+++ b/sys/arch/i386/i386/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.452 2009/06/03 21:30:19 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.453 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: machdep.c,v 1.214 1996/11/10 03:16:17 thorpej Exp $	*/
 
 /*-
@@ -547,6 +547,11 @@ setup_buffers()
 	if (bufpages == 0)
 		bufpages = atop(avail_end) * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4)
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
 }
 
 /*
diff --git a/sys/arch/luna88k/luna88k/machdep.c b/sys/arch/luna88k/luna88k/machdep.c
index 27a6f7ae066..03281cec0cb 100644
--- a/sys/arch/luna88k/luna88k/machdep.c
+++ b/sys/arch/luna88k/luna88k/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.63 2009/06/03 21:30:19 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.64 2009/06/15 17:01:26 beck Exp $	*/
 /*
  * Copyright (c) 1998, 1999, 2000, 2001 Steve Murphree, Jr.
  * Copyright (c) 1996 Nivas Madhur
@@ -470,6 +470,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/mac68k/mac68k/machdep.c b/sys/arch/mac68k/mac68k/machdep.c
index a86c163e416..9f992582dca 100644
--- a/sys/arch/mac68k/mac68k/machdep.c
+++ b/sys/arch/mac68k/mac68k/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.148 2009/06/03 21:30:20 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.149 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: machdep.c,v 1.207 1998/07/08 04:39:34 thorpej Exp $	*/
 
 /*
@@ -419,6 +419,12 @@ cpu_startup(void)
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/macppc/macppc/machdep.c b/sys/arch/macppc/macppc/machdep.c
index 0a3a0b876e4..ab9e2862361 100644
--- a/sys/arch/macppc/macppc/machdep.c
+++ b/sys/arch/macppc/macppc/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.107 2009/06/09 01:12:38 deraadt Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.108 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: machdep.c,v 1.4 1996/10/16 19:33:11 ws Exp $	*/
 
 /*
@@ -509,6 +509,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/mvme68k/mvme68k/machdep.c b/sys/arch/mvme68k/mvme68k/machdep.c
index 836bec8ec95..db6c36898db 100644
--- a/sys/arch/mvme68k/mvme68k/machdep.c
+++ b/sys/arch/mvme68k/mvme68k/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.115 2009/06/03 21:30:20 beck Exp $ */
+/*	$OpenBSD: machdep.c,v 1.116 2009/06/15 17:01:26 beck Exp $ */
 
 /*
  * Copyright (c) 1995 Theo de Raadt
@@ -265,6 +265,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/mvme88k/mvme88k/machdep.c b/sys/arch/mvme88k/mvme88k/machdep.c
index 32c28247423..063acd72a66 100644
--- a/sys/arch/mvme88k/mvme88k/machdep.c
+++ b/sys/arch/mvme88k/mvme88k/machdep.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: machdep.c,v 1.230 2009/06/03 21:30:20 beck Exp $	*/
+/* $OpenBSD: machdep.c,v 1.231 2009/06/15 17:01:26 beck Exp $	*/
 /*
  * Copyright (c) 1998, 1999, 2000, 2001 Steve Murphree, Jr.
  * Copyright (c) 1996 Nivas Madhur
@@ -404,6 +404,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/mvmeppc/mvmeppc/machdep.c b/sys/arch/mvmeppc/mvmeppc/machdep.c
index 42da01b655a..a2432cce471 100644
--- a/sys/arch/mvmeppc/mvmeppc/machdep.c
+++ b/sys/arch/mvmeppc/mvmeppc/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.57 2009/06/03 21:30:20 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.58 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: machdep.c,v 1.4 1996/10/16 19:33:11 ws Exp $	*/
 
 /*
@@ -415,6 +415,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/sgi/sgi/machdep.c b/sys/arch/sgi/sgi/machdep.c
index c44bd038eec..f296ef4b2b8 100644
--- a/sys/arch/sgi/sgi/machdep.c
+++ b/sys/arch/sgi/sgi/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.73 2009/06/13 21:48:03 miod Exp $ */
+/*	$OpenBSD: machdep.c,v 1.74 2009/06/15 17:01:26 beck Exp $ */
 
 /*
  * Copyright (c) 2003-2004 Opsycon AB  (www.opsycon.se / www.opsycon.com)
@@ -768,6 +768,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm. */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments. This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/socppc/socppc/machdep.c b/sys/arch/socppc/socppc/machdep.c
index f1da92fde61..fc2a738d51b 100644
--- a/sys/arch/socppc/socppc/machdep.c
+++ b/sys/arch/socppc/socppc/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.12 2009/06/09 01:12:38 deraadt Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.13 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: machdep.c,v 1.4 1996/10/16 19:33:11 ws Exp $	*/
 
 /*
@@ -772,6 +772,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/solbourne/solbourne/machdep.c b/sys/arch/solbourne/solbourne/machdep.c
index cc3e64cf9a7..0c90df97b4d 100644
--- a/sys/arch/solbourne/solbourne/machdep.c
+++ b/sys/arch/solbourne/solbourne/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.9 2009/06/03 21:30:20 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.10 2009/06/15 17:01:26 beck Exp $	*/
 /*	OpenBSD: machdep.c,v 1.105 2005/04/11 15:13:01 deraadt Exp 	*/
 
 /*
@@ -190,6 +190,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/sparc/sparc/machdep.c b/sys/arch/sparc/sparc/machdep.c
index ea63610774f..e1a7178feea 100644
--- a/sys/arch/sparc/sparc/machdep.c
+++ b/sys/arch/sparc/sparc/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.117 2009/06/03 21:30:20 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.118 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: machdep.c,v 1.85 1997/09/12 08:55:02 pk Exp $ */
 
 /*
@@ -203,6 +203,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively
 	 * limits the number of processes exec'ing at any time.
diff --git a/sys/arch/vax/vax/machdep.c b/sys/arch/vax/vax/machdep.c
index fd9433bcbba..21aba9fe1d3 100644
--- a/sys/arch/vax/vax/machdep.c
+++ b/sys/arch/vax/vax/machdep.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: machdep.c,v 1.100 2009/06/03 21:30:20 beck Exp $ */
+/* $OpenBSD: machdep.c,v 1.101 2009/06/15 17:01:26 beck Exp $ */
 /* $NetBSD: machdep.c,v 1.108 2000/09/13 15:00:23 thorpej Exp $	 */
 
 /*
@@ -212,6 +212,12 @@ cpu_startup()
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
+	/* Restrict to at most 25% filled kvm */
+	if (bufpages >
+	    (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4) 
+		bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
+		    PAGE_SIZE / 4;
+
 	/*
 	 * Allocate a submap for exec arguments.  This map effectively limits
 	 * the number of processes exec'ing at any time.
diff --git a/sys/kern/kern_sysctl.c b/sys/kern/kern_sysctl.c
index d7a6e82b2dc..3844f123316 100644
--- a/sys/kern/kern_sysctl.c
+++ b/sys/kern/kern_sysctl.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_sysctl.c,v 1.173 2009/06/08 00:52:23 deraadt Exp $	*/
+/*	$OpenBSD: kern_sysctl.c,v 1.174 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: kern_sysctl.c,v 1.17 1996/05/20 17:49:05 mrg Exp $	*/
 
 /*-
@@ -554,24 +554,6 @@ kern_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
 	case KERN_CPTIME2:
 		return (sysctl_cptime2(name + 1, namelen -1, oldp, oldlenp,
 		    newp, newlen));
-	case KERN_CACHEPCT: {
-		int opct, pgs;
-		opct = bufcachepercent;
-		error = sysctl_int(oldp, oldlenp, newp, newlen,
-		    &bufcachepercent);
-		if (error)
-			return(error);
-		if (bufcachepercent > 90 || bufcachepercent < 5) {
-			bufcachepercent = opct;
-			return (EINVAL);
-		}
-		if (bufcachepercent != opct) {
-			pgs = bufcachepercent * physmem / 100;
-			bufadjust(pgs); /* adjust bufpages */
-			bufhighpages = bufpages; /* set high water mark */
-		}
-		return(0);
-	}
 	default:
 		return (EOPNOTSUPP);
 	}
diff --git a/sys/kern/vfs_bio.c b/sys/kern/vfs_bio.c
index 42435d9e05f..0e7c21786ec 100644
--- a/sys/kern/vfs_bio.c
+++ b/sys/kern/vfs_bio.c
@@ -1,7 +1,7 @@
-/*	$OpenBSD: vfs_bio.c,v 1.116 2009/06/06 18:06:22 art Exp $	*/
+/*	$OpenBSD: vfs_bio.c,v 1.117 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: vfs_bio.c,v 1.44 1996/06/11 11:15:36 pk Exp $	*/
 
-/*
+/*-
  * Copyright (c) 1994 Christopher G. Demetriou
  * Copyright (c) 1982, 1986, 1989, 1993
  *	The Regents of the University of California.  All rights reserved.
@@ -62,6 +62,20 @@
 #include <miscfs/specfs/specdev.h>
 
 /*
+ * Definitions for the buffer hash lists.
+ */
+#define	BUFHASH(dvp, lbn)	\
+	(&bufhashtbl[((long)(dvp) / sizeof(*(dvp)) + (int)(lbn)) & bufhash])
+LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash;
+u_long	bufhash;
+
+/*
+ * Insq/Remq for the buffer hash lists.
+ */
+#define	binshash(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_hash)
+#define	bremhash(bp)		LIST_REMOVE(bp, b_hash)
+
+/*
  * Definitions for the buffer free lists.
  */
 #define	BQUEUES		2		/* number of free buffer queues */
@@ -109,9 +123,6 @@ long hidirtypages;
 long locleanpages;
 long hicleanpages;
 long maxcleanpages;
-long backoffpages;	/* backoff counter for page allocations */
-long buflowpages;	/* bufpages low water mark */
-long bufhighpages; 	/* bufpages high water mark */
 
 /* XXX - should be defined here. */
 extern int bufcachepercent;
@@ -171,13 +182,9 @@ buf_put(struct buf *bp)
 		panic("buf_put: b_dep is not empty");
 #endif
 
+	bremhash(bp);
 	LIST_REMOVE(bp, b_list);
 	bcstats.numbufs--;
-	if (backoffpages) {
-		backoffpages -= atop(bp->b_bufsize);
-		if (backoffpages < 0)
-			backoffpages = 0;
-	}
 
 	if (buf_dealloc_mem(bp) != 0)
 		return;
@@ -193,7 +200,7 @@ bufinit(void)
 	struct bqueues *dp;
 
 	/* XXX - for now */
-	bufhighpages = buflowpages = bufpages = bufcachepercent = bufkvm = 0;
+	bufpages = bufcachepercent = bufkvm = 0;
 
 	/*
 	 * If MD code doesn't say otherwise, use 10% of kvm for mappings and
@@ -204,16 +211,6 @@ bufinit(void)
 	if (bufpages == 0)
 		bufpages = physmem * bufcachepercent / 100;
 
-	bufhighpages = bufpages;
-
-	/*
-	 * set the base backoff level for the buffer cache to bufpages.
-	 * we will not allow uvm to steal back more than this number of
-	 * pages
-	 */
-	buflowpages = physmem * 10 / 100;
-
-
 	if (bufkvm == 0)
 		bufkvm = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 10;
 
@@ -240,6 +237,7 @@ bufinit(void)
  	 */
 	buf_mem_init(bufkvm);
 
+	bufhashtbl = hashinit(bufpages / 4, M_CACHE, M_WAITOK, &bufhash);
 	hidirtypages = (bufpages / 4) * 3;
 	lodirtypages = bufpages / 2;
 
@@ -253,104 +251,6 @@ bufinit(void)
 	maxcleanpages = locleanpages;
 }
 
-/*
- * Change cachepct
- */
-void
-bufadjust(int newbufpages)
-{
-	/*
-	 * XXX - note, bufkvm was allocated once, based on 10% of physmem
-	 * see above.
-	 */
-	struct buf *bp;
-	int s;
-
-	s = splbio();
-	bufpages = newbufpages;
-
-	hidirtypages = (bufpages / 4) * 3;
-	lodirtypages = bufpages / 2;
-
-	/*
-	 * When we hit 95% of pages being clean, we bring them down to
-	 * 90% to have some slack.
-	 */
-	hicleanpages = bufpages - (bufpages / 20);
-	locleanpages = bufpages - (bufpages / 10);
-
-	maxcleanpages = locleanpages;
-
-	/*
-	 * If we we have more buffers allocated than bufpages,
-	 * free them up to get back down. this may possibly consume
-	 * all our clean pages...
-	 */
-	while ((bp = TAILQ_FIRST(&bufqueues[BQ_CLEAN])) &&
-	    (bcstats.numbufpages > bufpages)) {
-		bremfree(bp);
-		if (bp->b_vp) {
-			RB_REMOVE(buf_rb_bufs,
-			    &bp->b_vp->v_bufs_tree, bp);
-			brelvp(bp);
-		}
-		buf_put(bp);
-	}
-
-	/*
-	 * Wake up cleaner if we're getting low on pages. We might
-	 * now have too much dirty, or have fallen below our low
-	 * water mark on clean pages so we need to free more stuff
-	 * up.
-	 */
-	if (bcstats.numdirtypages >= hidirtypages ||
-	    bcstats.numcleanpages <= locleanpages)
-		wakeup(&bd_req);
-
-	/*
-	 * if immediate action has not freed up enough goo for us
-	 * to proceed - we tsleep and wait for the cleaner above
-	 * to do it's work and get us reduced down to sanity.
-	 */
-	while (bcstats.numbufpages > bufpages) {
-		tsleep(&needbuffer, PRIBIO, "needbuffer", 0);
-	}
-	splx(s);
-}
-
-/*
- * Make the buffer cache back off from cachepct.
- */
-int
-bufbackoff()
-{
-	/*
-	 * Back off the amount of buffer cache pages. Called by the page
-	 * daemon to consume buffer cache pages rather than swapping.
-	 *
-	 * On success, it frees N pages from the buffer cache, and sets
-	 * a flag so that the next N allocations from buf_get will recycle
-	 * a buffer rather than allocate a new one. It then returns 0 to the
-	 * caller. 
-	 *
-	 * on failure, it could free no pages from the buffer cache, does
-	 * nothing and returns -1 to the caller. 
-	 */
-	long d;
-
-	if (bufpages <= buflowpages) 
-		return(-1);
-
-	if (bufpages - BACKPAGES >= buflowpages)
-		d = BACKPAGES;
-	else
-		d = bufpages - buflowpages;
-	backoffpages = BACKPAGES;
-	bufadjust(bufpages - d);
-	backoffpages = BACKPAGES;
-	return(0);
-}
-
 struct buf *
 bio_doread(struct vnode *vp, daddr64_t blkno, int size, int async)
 {
@@ -776,12 +676,10 @@ brelse(struct buf *bp)
 			CLR(bp->b_flags, B_DELWRI);
 		}
 
-		if (bp->b_vp) {
-			RB_REMOVE(buf_rb_bufs, &bp->b_vp->v_bufs_tree,
-			    bp);
+		if (bp->b_vp)
 			brelvp(bp);
-		}
-		bp->b_vp = NULL;
+		bremhash(bp);
+		binshash(bp, &invalhash);
 
 		/*
 		 * If the buffer has no associated data, place it back in the
@@ -799,9 +697,6 @@ brelse(struct buf *bp)
 				CLR(bp->b_flags, B_WANTED);
 				wakeup(bp);
 			}
-			if (bp->b_vp != NULL)
-				RB_REMOVE(buf_rb_bufs,
-				    &bp->b_vp->v_bufs_tree, bp);
 			buf_put(bp);
 			splx(s);
 			return;
@@ -863,14 +758,15 @@ struct buf *
 incore(struct vnode *vp, daddr64_t blkno)
 {
 	struct buf *bp;
-	struct buf b;
-
-	/* Search buf lookup tree */
-	b.b_lblkno = blkno;
-	bp = RB_FIND(buf_rb_bufs, &vp->v_bufs_tree, &b);
-	if (bp && !ISSET(bp->b_flags, B_INVAL))
-		return(bp);
-	return(NULL);
+
+	/* Search hash chain */
+	LIST_FOREACH(bp, BUFHASH(vp, blkno), b_hash) {
+		if (bp->b_lblkno == blkno && bp->b_vp == vp &&
+		    !ISSET(bp->b_flags, B_INVAL))
+			return (bp);
+	}
+
+	return (NULL);
 }
 
 /*
@@ -885,7 +781,6 @@ struct buf *
 getblk(struct vnode *vp, daddr64_t blkno, int size, int slpflag, int slptimeo)
 {
 	struct buf *bp;
-	struct buf b;
 	int s, error;
 
 	/*
@@ -899,9 +794,9 @@ getblk(struct vnode *vp, daddr64_t blkno, int size, int slpflag, int slptimeo)
 	 * the block until the write is finished.
 	 */
 start:
-	b.b_lblkno = blkno;
-	bp = RB_FIND(buf_rb_bufs, &vp->v_bufs_tree, &b);
-	if (bp != NULL) {
+	LIST_FOREACH(bp, BUFHASH(vp, blkno), b_hash) {
+		if (bp->b_lblkno != blkno || bp->b_vp != vp)
+			continue;
 
 		s = splbio();
 		if (ISSET(bp->b_flags, B_BUSY)) {
@@ -950,24 +845,11 @@ geteblk(int size)
 struct buf *
 buf_get(struct vnode *vp, daddr64_t blkno, size_t size)
 {
-	static int gcount = 0;
 	struct buf *bp;
 	int poolwait = size == 0 ? PR_NOWAIT : PR_WAITOK;
 	int npages;
 	int s;
 
-	/*
-	 * if we were previously backed off, slowly climb back up
-	 * to the high water mark again.
-	 */
-	if ((backoffpages == 0) && (bufpages < bufhighpages)) {
-		if ( gcount == 0 )  {
-			bufadjust(bufpages + BACKPAGES);
-			gcount += BACKPAGES;
-		} else
-			gcount--;
-	}
-
 	s = splbio();
 	if (size) {
 		/*
@@ -985,11 +867,8 @@ buf_get(struct vnode *vp, daddr64_t blkno, size_t size)
 			while (bcstats.numcleanpages > locleanpages) {
 				bp = TAILQ_FIRST(&bufqueues[BQ_CLEAN]);
 				bremfree(bp);
-				if (bp->b_vp) {
-					RB_REMOVE(buf_rb_bufs,
-					    &bp->b_vp->v_bufs_tree, bp);
+				if (bp->b_vp)
 					brelvp(bp);
-				}
 				buf_put(bp);
 			}
 		}
@@ -999,21 +878,16 @@ buf_get(struct vnode *vp, daddr64_t blkno, size_t size)
 		/*
 		 * Free some buffers until we have enough space.
 		 */
-		while ((bcstats.numbufpages + npages > bufpages)
-		    || backoffpages) {
+		while (bcstats.numbufpages + npages > bufpages) {
 			int freemax = 5;
 			int i = freemax;
 			while ((bp = TAILQ_FIRST(&bufqueues[BQ_CLEAN])) && i--) {
 				bremfree(bp);
-				if (bp->b_vp) {
-					RB_REMOVE(buf_rb_bufs,
-					    &bp->b_vp->v_bufs_tree, bp);
+				if (bp->b_vp)
 					brelvp(bp);
-				}
 				buf_put(bp);
 			}
-			if (freemax == i &&
-			    (bcstats.numbufpages + npages > bufpages)) {
+			if (freemax == i) {
 				needbuffer++;
 				tsleep(&needbuffer, PRIBIO, "needbuffer", 0);
 				splx(s);
@@ -1054,12 +928,11 @@ buf_get(struct vnode *vp, daddr64_t blkno, size_t size)
 
 		bp->b_blkno = bp->b_lblkno = blkno;
 		bgetvp(vp, bp);
-		if (RB_INSERT(buf_rb_bufs, &vp->v_bufs_tree, bp))
-			panic("buf_get: dup lblk vp %p bp %p", vp, bp);
+		binshash(bp, BUFHASH(vp, blkno));
 	} else {
 		bp->b_vnbufs.le_next = NOLIST;
 		SET(bp->b_flags, B_INVAL);
-		bp->b_vp = NULL;
+		binshash(bp, &invalhash);
 	}
 
 	LIST_INSERT_HEAD(&bufhead, bp, b_list);
diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c
index ced2bdc291e..4d09d4f20a0 100644
--- a/sys/kern/vfs_subr.c
+++ b/sys/kern/vfs_subr.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vfs_subr.c,v 1.177 2009/06/06 18:06:22 art Exp $	*/
+/*	$OpenBSD: vfs_subr.c,v 1.178 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: vfs_subr.c,v 1.53 1996/04/22 01:39:13 christos Exp $	*/
 
 /*
@@ -59,7 +59,6 @@
 #include <sys/mbuf.h>
 #include <sys/syscallargs.h>
 #include <sys/pool.h>
-#include <sys/tree.h>
 
 #include <uvm/uvm_extern.h>
 #include <sys/sysctl.h>
@@ -116,19 +115,6 @@ void printlockedvnodes(void);
 
 struct pool vnode_pool;
 
-static int rb_buf_compare(struct buf *b1, struct buf *b2);
-RB_GENERATE(buf_rb_bufs, buf, b_rbbufs, rb_buf_compare);
-
-static int
-rb_buf_compare(struct buf *b1, struct buf *b2)
-{
-	if (b1->b_lblkno < b2->b_lblkno)
-		return(-1);
-	if (b1->b_lblkno > b2->b_lblkno)
-		return(1);
-	return(0);
-}
-
 /*
  * Initialize the vnode management data structures.
  */
@@ -359,7 +345,6 @@ getnewvnode(enum vtagtype tag, struct mount *mp, int (**vops)(void *),
 	    ((TAILQ_FIRST(listhd = &vnode_hold_list) == NULL) || toggle))) {
 		splx(s);
 		vp = pool_get(&vnode_pool, PR_WAITOK | PR_ZERO);
-		RB_INIT(&vp->v_bufs_tree);
 		numvnodes++;
 	} else {
 		for (vp = TAILQ_FIRST(listhd); vp != NULLVP;
diff --git a/sys/sys/buf.h b/sys/sys/buf.h
index d8b021a5f81..522a320711f 100644
--- a/sys/sys/buf.h
+++ b/sys/sys/buf.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: buf.h,v 1.64 2009/06/03 22:09:30 thib Exp $	*/
+/*	$OpenBSD: buf.h,v 1.65 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: buf.h,v 1.25 1997/04/09 21:12:17 mycroft Exp $	*/
 
 /*
@@ -40,16 +40,12 @@
 #ifndef _SYS_BUF_H_
 #define	_SYS_BUF_H_
 #include <sys/queue.h>
-#include <sys/tree.h>
 
 #define NOLIST ((struct buf *)0x87654321)
 
 struct buf;
 struct vnode;
 
-struct buf_rb_bufs;
-RB_PROTOTYPE(buf_rb_bufs, buf, b_rbbufs, rb_buf_compare);
-
 LIST_HEAD(bufhead, buf);
 
 /*
@@ -76,8 +72,8 @@ extern struct bio_ops {
  * The buffer header describes an I/O operation in the kernel.
  */
 struct buf {
-	RB_ENTRY(buf) b_rbbufs;		/* vnode "hash" tree */
 	LIST_ENTRY(buf) b_list;		/* All allocated buffers. */
+	LIST_ENTRY(buf) b_hash;		/* Hash chain. */
 	LIST_ENTRY(buf) b_vnbufs;	/* Buffer's associated vnode. */
 	TAILQ_ENTRY(buf) b_freelist;	/* Free list position if not active. */
 	time_t	b_synctime;		/* Time this buffer should be flushed */
diff --git a/sys/sys/mount.h b/sys/sys/mount.h
index 402a926e398..b856a781827 100644
--- a/sys/sys/mount.h
+++ b/sys/sys/mount.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mount.h,v 1.94 2009/06/05 04:29:14 beck Exp $	*/
+/*	$OpenBSD: mount.h,v 1.95 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: mount.h,v 1.48 1996/02/18 11:55:47 fvdl Exp $	*/
 
 /*
@@ -504,14 +504,8 @@ struct bcachestats {
 	int64_t cachehits;		/* total reads found in cache */
 };
 #ifdef _KERNEL
-#define BACKPAGES 100
 extern struct bcachestats bcstats;
-extern long buflowpages, bufhighpages;
-#define BUFPAGES_DEFICIT (((buflowpages - bcstats.numbufpages) < 0) ? 0 \
-    : buflowpages - bcstats.numbufpages)
-extern int bufcachepercent;
-extern void bufadjust(int);
-extern int bufbackoff(void);
+#define BUFPAGES_DEFICIT (bufpages - bcstats.numbufpages)
 #endif
 
 /*
diff --git a/sys/sys/syscall.h b/sys/sys/syscall.h
index 408f1c047ac..69c415c8a51 100644
--- a/sys/sys/syscall.h
+++ b/sys/sys/syscall.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: syscall.h,v 1.104 2009/06/03 15:47:13 jj Exp $	*/
+/*	$OpenBSD: syscall.h,v 1.105 2009/06/15 17:01:26 beck Exp $	*/
 
 /*
  * System call numbers.
diff --git a/sys/sys/syscallargs.h b/sys/sys/syscallargs.h
index 38df51dff5d..8aa98f64558 100644
--- a/sys/sys/syscallargs.h
+++ b/sys/sys/syscallargs.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: syscallargs.h,v 1.106 2009/06/03 15:47:13 jj Exp $	*/
+/*	$OpenBSD: syscallargs.h,v 1.107 2009/06/15 17:01:26 beck Exp $	*/
 
 /*
  * System call argument lists.
diff --git a/sys/sys/sysctl.h b/sys/sys/sysctl.h
index 93fc0798768..b1cf095d9cf 100644
--- a/sys/sys/sysctl.h
+++ b/sys/sys/sysctl.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: sysctl.h,v 1.98 2009/06/07 03:07:19 millert Exp $	*/
+/*	$OpenBSD: sysctl.h,v 1.99 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: sysctl.h,v 1.16 1996/04/09 20:55:36 cgd Exp $	*/
 
 /*
@@ -184,9 +184,8 @@ struct ctlname {
 #define	KERN_TIMECOUNTER	69	/* node: timecounter */
 #define	KERN_MAXLOCKSPERUID	70	/* int: locks per uid */
 #define	KERN_CPTIME2		71	/* array: cp_time2 */
-#define	KERN_CACHEPCT		72	/* buffer cache % of physmem */
-#define	KERN_FILE2		73	/* struct: file entries */
-#define	KERN_MAXID		74	/* number of valid kern ids */
+#define	KERN_FILE2		72	/* struct: file entries */
+#define	KERN_MAXID		73	/* number of valid kern ids */
 
 #define	CTL_KERN_NAMES { \
 	{ 0, 0 }, \
@@ -261,7 +260,6 @@ struct ctlname {
  	{ "timecounter", CTLTYPE_NODE }, \
  	{ "maxlocksperuid", CTLTYPE_INT }, \
  	{ "cp_time2", CTLTYPE_STRUCT }, \
- 	{ "bufcachepercent", CTLTYPE_INT }, \
 	{ "file2", CTLTYPE_STRUCT }, \
 }
 
diff --git a/sys/sys/vnode.h b/sys/sys/vnode.h
index 76a5fe81c41..e640dc36027 100644
--- a/sys/sys/vnode.h
+++ b/sys/sys/vnode.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vnode.h,v 1.99 2009/06/03 14:45:55 jj Exp $	*/
+/*	$OpenBSD: vnode.h,v 1.100 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: vnode.h,v 1.38 1996/02/29 20:59:05 cgd Exp $	*/
 
 /*
@@ -32,12 +32,10 @@
  *	@(#)vnode.h	8.11 (Berkeley) 11/21/94
  */
 
-#include <sys/buf.h>
 #include <sys/types.h>
 #include <sys/queue.h>
 #include <sys/lock.h>
 #include <sys/selinfo.h>
-#include <sys/tree.h>
 
 #include <uvm/uvm.h>
 #include <uvm/uvm_vnode.h>
@@ -81,8 +79,6 @@ enum vtagtype	{
  */
 LIST_HEAD(buflists, buf);
 
-RB_HEAD(buf_rb_bufs, buf);
-
 struct vnode {
 	struct uvm_vnode v_uvm;			/* uvm data */
 	int	(**v_op)(void *);		/* vnode operations vector */
@@ -98,7 +94,6 @@ struct vnode {
 	struct	mount *v_mount;			/* ptr to vfs we are in */
 	TAILQ_ENTRY(vnode) v_freelist;		/* vnode freelist */
 	LIST_ENTRY(vnode) v_mntvnodes;		/* vnodes for mount point */
-	struct	buf_rb_bufs v_bufs_tree;	/* lookup of all bufs */
 	struct	buflists v_cleanblkhd;		/* clean blocklist head */
 	struct	buflists v_dirtyblkhd;		/* dirty blocklist head */
 	u_int   v_numoutput;			/* num of writes in progress */
diff --git a/sys/uvm/uvm_pdaemon.c b/sys/uvm/uvm_pdaemon.c
index 3e88675842b..b30cf1e0a98 100644
--- a/sys/uvm/uvm_pdaemon.c
+++ b/sys/uvm/uvm_pdaemon.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_pdaemon.c,v 1.47 2009/06/06 23:35:08 art Exp $	*/
+/*	$OpenBSD: uvm_pdaemon.c,v 1.48 2009/06/15 17:01:26 beck Exp $	*/
 /*	$NetBSD: uvm_pdaemon.c,v 1.23 2000/08/20 10:24:14 bjh21 Exp $	*/
 
 /* 
@@ -214,8 +214,8 @@ uvm_pageout(void *arg)
 	for (;;) {
 		uvm_lock_fpageq();
 		UVMHIST_LOG(pdhist,"  <<SLEEPING>>",0,0,0,0);
-		msleep(&uvm.pagedaemon_proc, &uvm.fpageqlock,
-		    PVM | PNORELOCK, "pgdaemon", 0);
+		msleep(&uvm.pagedaemon_proc, &uvm.fpageqlock, PVM | PNORELOCK,
+		    "pgdaemon", 0);
 		uvmexp.pdwoke++;
 		UVMHIST_LOG(pdhist,"  <<WOKE UP>>",0,0,0,0);
 
@@ -239,13 +239,11 @@ uvm_pageout(void *arg)
 		    uvmexp.inactarg);
 
 		/*
-		 * get pages from the buffer cache, or scan if needed
+		 * scan if needed
 		 */
-		if (uvmexp.inactive < uvmexp.inactarg)
+		if ((uvmexp.free - BUFPAGES_DEFICIT) < uvmexp.freetarg ||
+		    uvmexp.inactive < uvmexp.inactarg) {
 			uvmpd_scan();
-		else if ((uvmexp.free - BUFPAGES_DEFICIT) < uvmexp.freetarg) {
-			if (bufbackoff() == -1)
-				uvmpd_scan();
 		}
 
 		/*