introduce a new km_page allocator that gets pages from kernel_map using

an interrupt safe thread.
use this as the new backend for mbpool and mclpool, eliminating the mb_map.
introduce a sysctl kern.maxclusters which controls the limit of clusters
allocated.
testing by many people, works everywhere but m68k.  ok deraadt@

this essentially deprecates the NMBCLUSTERS option, don't use it.
this should reduce pressure on the kmem_map and the uvm reserve of static
map entries.

7 files changed, 166 insertions, 27 deletions

diff --git a/sys/kern/kern_sysctl.c b/sys/kern/kern_sysctl.c
index 0d785eb7b39..ece5249fa89 100644
--- a/sys/kern/kern_sysctl.c
+++ b/sys/kern/kern_sysctl.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_sysctl.c,v 1.105 2004/04/15 00:22:42 tedu Exp $	*/
+/*	$OpenBSD: kern_sysctl.c,v 1.106 2004/04/19 22:52:33 tedu Exp $	*/
 /*	$NetBSD: kern_sysctl.c,v 1.17 1996/05/20 17:49:05 mrg Exp $	*/
 
 /*-
@@ -90,6 +90,8 @@ extern struct disklist_head disklist;
 extern fixpt_t ccpu;
 extern  long numvnodes;
 
+extern void nmbclust_update(void);
+
 int sysctl_diskinit(int, struct proc *);
 int sysctl_proc_args(int *, u_int, void *, size_t *, struct proc *);
 int sysctl_intrcnt(int *, u_int, void *, size_t *);
@@ -495,6 +497,11 @@ kern_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
 	case KERN_EMUL:
 		return (sysctl_emul(name + 1, namelen - 1, oldp, oldlenp,
 		    newp, newlen));
+	case KERN_MAXCLUSTERS:
+		error = sysctl_int(oldp, oldlenp, newp, newlen, &nmbclust);
+		if (!error)
+			nmbclust_update();
+		return (error);
 	default:
 		return (EOPNOTSUPP);
 	}
diff --git a/sys/kern/uipc_mbuf.c b/sys/kern/uipc_mbuf.c
index 13c00cb32d5..9a8fd13e121 100644
--- a/sys/kern/uipc_mbuf.c
+++ b/sys/kern/uipc_mbuf.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uipc_mbuf.c,v 1.66 2004/04/17 10:18:12 mcbride Exp $	*/
+/*	$OpenBSD: uipc_mbuf.c,v 1.67 2004/04/19 22:52:33 tedu Exp $	*/
 /*	$NetBSD: uipc_mbuf.c,v 1.15.4.1 1996/06/13 17:11:44 cgd Exp $	*/
 
 /*
@@ -102,9 +102,11 @@ int max_datalen;		/* MHLEN - max_hdr */
 void	*mclpool_alloc(struct pool *, int);
 void	mclpool_release(struct pool *, void *);
 struct mbuf *m_copym0(struct mbuf *, int, int, int, int);
+void	nmbclust_update(void);
+
 
 const char *mclpool_warnmsg =
-    "WARNING: mclpool limit reached; increase NMBCLUSTERS";
+    "WARNING: mclpool limit reached; increase kern.maxclusters";
 
 struct pool_allocator mclpool_allocator = {
 	mclpool_alloc, mclpool_release, 0,
@@ -116,24 +118,13 @@ struct pool_allocator mclpool_allocator = {
 void
 mbinit()
 {
-	vaddr_t minaddr, maxaddr;
-
-	minaddr = vm_map_min(kernel_map);
-	mb_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr,
-	    nmbclust*(MCLBYTES), VM_MAP_INTRSAFE, FALSE, NULL);
-
-	pool_init(&mbpool, MSIZE, 0, 0, 0, "mbpl", NULL);
+	pool_init(&mbpool, MSIZE, 0, 0, 0, "mbpl", &mclpool_allocator);
 	pool_init(&mclpool, MCLBYTES, 0, 0, 0, "mclpl", &mclpool_allocator);
 
 	pool_set_drain_hook(&mbpool, m_reclaim, NULL);
 	pool_set_drain_hook(&mclpool, m_reclaim, NULL);
 
-	/*
-	 * Set the hard limit on the mclpool to the number of
-	 * mbuf clusters the kernel is to support.  Log the limit
-	 * reached message max once a minute.
-	 */
-	(void)pool_sethardlimit(&mclpool, nmbclust, mclpool_warnmsg, 60);
+	nmbclust_update();
 
 	/*
 	 * Set a low water mark for both mbufs and clusters.  This should
@@ -145,20 +136,29 @@ mbinit()
 	pool_setlowat(&mclpool, mcllowat);
 }
 
+void
+nmbclust_update(void)
+{
+	/*
+	 * Set the hard limit on the mclpool to the number of
+	 * mbuf clusters the kernel is to support.  Log the limit
+	 * reached message max once a minute.
+	 */
+	(void)pool_sethardlimit(&mclpool, nmbclust, mclpool_warnmsg, 60);
+}
+
+
 
 void *
 mclpool_alloc(struct pool *pp, int flags)
 {
-	boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
-
-	return ((void *)uvm_km_alloc_poolpage1(mb_map, uvmexp.mb_object,
-	    waitok));
+	return uvm_km_getpage();
 }
 
 void
 mclpool_release(struct pool *pp, void *v)
 {
-	uvm_km_free_poolpage1(mb_map, (vaddr_t)v);
+	uvm_km_putpage(v);
 }
 
 void
diff --git a/sys/sys/sysctl.h b/sys/sys/sysctl.h
index 32bb152664b..5bf14091529 100644
--- a/sys/sys/sysctl.h
+++ b/sys/sys/sysctl.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: sysctl.h,v 1.76 2004/02/27 21:46:44 grange Exp $	*/
+/*	$OpenBSD: sysctl.h,v 1.77 2004/04/19 22:52:33 tedu Exp $	*/
 /*	$NetBSD: sysctl.h,v 1.16 1996/04/09 20:55:36 cgd Exp $	*/
 
 /*
@@ -179,7 +179,8 @@ struct ctlname {
 #define	KERN_WATCHDOG		64	/* node: watchdog */
 #define	KERN_EMUL		65	/* node: emuls */
 #define	KERN_PROC2		66	/* struct: process entries */
-#define	KERN_MAXID		67	/* number of valid kern ids */
+#define	KERN_MAXCLUSTERS	67	/* number of mclusters */
+#define	KERN_MAXID		68	/* number of valid kern ids */
 
 #define	CTL_KERN_NAMES { \
 	{ 0, 0 }, \
@@ -249,6 +250,7 @@ struct ctlname {
  	{ "watchdog", CTLTYPE_NODE }, \
  	{ "emul", CTLTYPE_NODE }, \
  	{ "proc2", CTLTYPE_STRUCT }, \
+ 	{ "maxclusters", CTLTYPE_INT }, \
 }
 
 /*
diff --git a/sys/uvm/uvm_extern.h b/sys/uvm/uvm_extern.h
index f1eb2cf02f0..d20e6d25639 100644
--- a/sys/uvm/uvm_extern.h
+++ b/sys/uvm/uvm_extern.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_extern.h,v 1.50 2003/08/10 00:04:50 miod Exp $	*/
+/*	$OpenBSD: uvm_extern.h,v 1.51 2004/04/19 22:52:33 tedu Exp $	*/
 /*	$NetBSD: uvm_extern.h,v 1.57 2001/03/09 01:02:12 chs Exp $	*/
 
 /*
@@ -511,6 +511,8 @@ void			uvm_km_free_poolpage1(vm_map_t, vaddr_t);
 #define	uvm_km_alloc_poolpage(waitok)	uvm_km_alloc_poolpage1(kmem_map, \
 						uvmexp.kmem_object, (waitok))
 #define	uvm_km_free_poolpage(addr)	uvm_km_free_poolpage1(kmem_map, (addr))
+void			*uvm_km_getpage(void);
+void			uvm_km_putpage(void *);
 
 /* uvm_map.c */
 int			uvm_map(vm_map_t, vaddr_t *, vsize_t,
diff --git a/sys/uvm/uvm_init.c b/sys/uvm/uvm_init.c
index a9915e88040..47b9c469bbe 100644
--- a/sys/uvm/uvm_init.c
+++ b/sys/uvm/uvm_init.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_init.c,v 1.10 2001/11/06 01:35:04 art Exp $	*/
+/*	$OpenBSD: uvm_init.c,v 1.11 2004/04/19 22:52:33 tedu Exp $	*/
 /*	$NetBSD: uvm_init.c,v 1.14 2000/06/27 17:29:23 mrg Exp $	*/
 
 /*
@@ -150,6 +150,8 @@ uvm_init()
 	uao_create(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
 	    UAO_FLAG_KERNSWAP);
 
+	uvm_km_page_init();
+
 	/*
 	 * done!
 	 */
diff --git a/sys/uvm/uvm_km.c b/sys/uvm/uvm_km.c
index d343a59e4f9..a07be5b0b8f 100644
--- a/sys/uvm/uvm_km.c
+++ b/sys/uvm/uvm_km.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_km.c,v 1.35 2004/02/23 06:19:32 drahn Exp $	*/
+/*	$OpenBSD: uvm_km.c,v 1.36 2004/04/19 22:52:33 tedu Exp $	*/
 /*	$NetBSD: uvm_km.c,v 1.42 2001/01/14 02:10:01 thorpej Exp $	*/
 
 /* 
@@ -144,6 +144,7 @@
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/proc.h>
+#include <sys/kthread.h>
 
 #include <uvm/uvm.h>
 
@@ -915,3 +916,127 @@ uvm_km_free_poolpage1(map, addr)
 	splx(s);
 #endif /* PMAP_UNMAP_POOLPAGE */
 }
+
+int uvm_km_pages_lowat;
+int uvm_km_pages_free;
+struct km_page {
+	struct km_page *next;
+} *uvm_km_pages_head;
+
+void uvm_km_createthread(void *);
+void uvm_km_thread(void *);
+
+void
+uvm_km_page_init(void)
+{
+	struct km_page *head, *page;
+	int i;
+
+
+	head = NULL;
+	for (i = 0; i < uvm_km_pages_lowat * 4; i++) {
+#if defined(PMAP_MAP_POOLPAGE)
+		struct vm_page *pg;
+		vaddr_t va;
+
+		pg = uvm_pagealloc(NULL, 0, NULL, 0);
+		if (__predict_false(pg == NULL))
+			break;
+
+		va = PMAP_MAP_POOLPAGE(pg);
+		if (__predict_false(va == 0)) {
+			uvm_pagefree(pg);
+			break;
+		}
+		page = (void *)va;
+#else
+		page = (void *)uvm_km_alloc(kernel_map, PAGE_SIZE);
+#endif
+		page->next = head;
+		head = page;
+	}
+	uvm_km_pages_head = head;
+	uvm_km_pages_free = i;
+
+	kthread_create_deferred(uvm_km_createthread, NULL);
+}
+
+void
+uvm_km_createthread(void *arg)
+{
+	kthread_create(uvm_km_thread, NULL, NULL, "kmthread");
+}
+
+void
+uvm_km_thread(void *arg)
+{
+	struct km_page *head, *tail, *page;
+	int i, s, want;
+
+	for (;;) {
+		if (uvm_km_pages_free >= uvm_km_pages_lowat)
+			tsleep(&uvm_km_pages_head, PVM, "kmalloc", 0);
+		want = uvm_km_pages_lowat - uvm_km_pages_free;
+		if (want < 16)
+			want = 16;
+		for (i = 0; i < want; i++) {
+#if defined(PMAP_MAP_POOLPAGE)
+			struct vm_page *pg;
+			vaddr_t va;
+	
+			pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
+			if (__predict_false(pg == NULL))
+				break;
+			va = PMAP_MAP_POOLPAGE(pg);
+			if (__predict_false(va == 0)) {
+				uvm_pagefree(pg);
+				break;
+			}
+			page = (void *)va;
+#else
+			page = (void *)uvm_km_alloc(kernel_map, PAGE_SIZE);
+#endif
+			if (i == 0)
+				head = tail = page;
+			page->next = head;
+			head = page;
+		}
+		s = splvm();
+		tail->next = uvm_km_pages_head;
+		uvm_km_pages_head = head;
+		uvm_km_pages_free += i;
+		splx(s);
+	}
+}
+
+
+void *
+uvm_km_getpage(void)
+{
+	struct km_page *page;
+	int s;
+
+	s = splvm();
+	page = uvm_km_pages_head;
+	if (page) {
+		uvm_km_pages_head = page->next;
+		uvm_km_pages_free--;
+	}
+	splx(s);
+	if (uvm_km_pages_free < uvm_km_pages_lowat)
+		wakeup(&uvm_km_pages_head);
+	return (page);
+}
+
+void
+uvm_km_putpage(void *v)
+{
+	struct km_page *page = v;
+	int s;
+
+	s = splvm();
+	page->next = uvm_km_pages_head;
+	uvm_km_pages_head = page;
+	uvm_km_pages_free++;
+	splx(s);
+}
diff --git a/sys/uvm/uvm_km.h b/sys/uvm/uvm_km.h
index 7c1bc043879..1162b4dc29a 100644
--- a/sys/uvm/uvm_km.h
+++ b/sys/uvm/uvm_km.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_km.h,v 1.7 2002/03/14 01:27:18 millert Exp $	*/
+/*	$OpenBSD: uvm_km.h,v 1.8 2004/04/19 22:52:33 tedu Exp $	*/
 /*	$NetBSD: uvm_km.h,v 1.9 1999/06/21 17:25:11 thorpej Exp $	*/
 
 /*
@@ -49,6 +49,7 @@
  */
 
 void uvm_km_init(vaddr_t, vaddr_t);
+void uvm_km_page_init(void);
 void uvm_km_pgremove(struct uvm_object *, vaddr_t, vaddr_t);
 void uvm_km_pgremove_intrsafe(struct uvm_object *, vaddr_t, vaddr_t);