1 files changed, 184 insertions, 25 deletions

diff --git a/sys/uvm/uvm_anon.c b/sys/uvm/uvm_anon.c
index b841a4f1e65..f24114aee9e 100644
--- a/sys/uvm/uvm_anon.c
+++ b/sys/uvm/uvm_anon.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_anon.c,v 1.22 2006/06/21 16:20:05 mickey Exp $	*/
+/*	$OpenBSD: uvm_anon.c,v 1.23 2006/07/13 22:51:26 deraadt Exp $	*/
 /*	$NetBSD: uvm_anon.c,v 1.10 2000/11/25 06:27:59 chs Exp $	*/
 
 /*
@@ -47,7 +47,22 @@
 #include <uvm/uvm.h>
 #include <uvm/uvm_swap.h>
 
-struct pool uvm_anon_pool;
+/*
+ * anonblock_list: global list of anon blocks,
+ * locked by swap_syscall_lock (since we never remove
+ * anything from this list and we only add to it via swapctl(2)).
+ */
+
+struct uvm_anonblock {
+	LIST_ENTRY(uvm_anonblock) list;
+	int count;
+	struct vm_anon *anons;
+};
+static LIST_HEAD(anonlist, uvm_anonblock) anonblock_list;
+
+
+static boolean_t anon_pagein(struct vm_anon *);
+
 
 /*
  * allocate anons
@@ -55,28 +70,104 @@ struct pool uvm_anon_pool;
 void
 uvm_anon_init()
 {
-	pool_init(&uvm_anon_pool, sizeof(struct vm_anon), 0, 0, 0, "anonpl",
-	    &pool_allocator_nointr);
-	pool_sethiwat(&uvm_anon_pool, uvmexp.free / 16);
+	int nanon = uvmexp.free - (uvmexp.free / 16); /* XXXCDC ??? */
+
+	simple_lock_init(&uvm.afreelock);
+	LIST_INIT(&anonblock_list);
+
+	/*
+	 * Allocate the initial anons.
+	 */
+	uvm_anon_add(nanon);
 }
 
 /*
- * allocate an anon
+ * add some more anons to the free pool.  called when we add
+ * more swap space.
+ *
+ * => swap_syscall_lock should be held (protects anonblock_list).
  */
-struct vm_anon *
-uvm_analloc()
+int
+uvm_anon_add(count)
+	int	count;
 {
+	struct uvm_anonblock *anonblock;
 	struct vm_anon *anon;
+	int lcv, needed;
+
+	simple_lock(&uvm.afreelock);
+	uvmexp.nanonneeded += count;
+	needed = uvmexp.nanonneeded - uvmexp.nanon;
+	simple_unlock(&uvm.afreelock);
 
-	anon = pool_get(&uvm_anon_pool, PR_NOWAIT);
-	if (anon) {
+	if (needed <= 0) {
+		return 0;
+	}
+ 
+	anon = (void *)uvm_km_alloc(kernel_map, sizeof(*anon) * needed);
+
+	/* XXX Should wait for VM to free up. */
+	if (anon == NULL) {
+		printf("uvm_anon_add: can not allocate %d anons\n", needed);
+		panic("uvm_anon_add");
+	}
+
+	MALLOC(anonblock, void *, sizeof(*anonblock), M_UVMAMAP, M_WAITOK);
+
+	anonblock->count = needed;
+	anonblock->anons = anon;
+	LIST_INSERT_HEAD(&anonblock_list, anonblock, list);
+	memset(anon, 0, sizeof(*anon) * needed);
+ 
+	simple_lock(&uvm.afreelock);
+	uvmexp.nanon += needed;
+	uvmexp.nfreeanon += needed;
+	for (lcv = 0; lcv < needed; lcv++) {
 		simple_lock_init(&anon->an_lock);
-		anon->an_ref = 1;
-		anon->an_page = NULL;
-		anon->an_swslot = 0;
-		simple_lock(&anon->an_lock);
+		anon[lcv].u.an_nxt = uvm.afree;
+		uvm.afree = &anon[lcv];
+		simple_lock_init(&uvm.afree->an_lock);
 	}
-	return(anon);
+	simple_unlock(&uvm.afreelock);
+	return 0;
+}
+
+/*
+ * remove anons from the free pool.
+ */
+void
+uvm_anon_remove(count)
+	int count;
+{
+	/*
+	 * we never actually free any anons, to avoid allocation overhead.
+	 * XXX someday we might want to try to free anons.
+	 */
+
+	simple_lock(&uvm.afreelock);
+	uvmexp.nanonneeded -= count;
+	simple_unlock(&uvm.afreelock);
+}
+
+/*
+ * allocate an anon
+ */
+struct vm_anon *
+uvm_analloc()
+{
+	struct vm_anon *a;
+
+	simple_lock(&uvm.afreelock);
+	a = uvm.afree;
+	if (a) {
+		uvm.afree = a->u.an_nxt;
+		uvmexp.nfreeanon--;
+		a->an_ref = 1;
+		a->an_swslot = 0;
+		a->u.an_page = NULL;		/* so we can free quickly */
+	}
+	simple_unlock(&uvm.afreelock);
+	return(a);
 }
 
 /*
@@ -99,7 +190,7 @@ uvm_anfree(anon)
 	 * get page
 	 */
 
-	pg = anon->an_page;
+	pg = anon->u.an_page;
 
 	/*
 	 * if there is a resident page and it is loaned, then anon may not
@@ -166,10 +257,11 @@ uvm_anfree(anon)
 	 * now that we've stripped the data areas from the anon, free the anon
 	 * itself!
 	 */
-	KASSERT(anon->an_page == NULL);
-	KASSERT(anon->an_swslot == 0);
-
-	pool_put(&uvm_anon_pool, anon);
+	simple_lock(&uvm.afreelock);
+	anon->u.an_nxt = uvm.afree;
+	uvm.afree = anon;
+	uvmexp.nfreeanon++;
+	simple_unlock(&uvm.afreelock);
 	UVMHIST_LOG(maphist,"<- done!",0,0,0,0);
 }
 
@@ -192,7 +284,7 @@ uvm_anon_dropswap(anon)
 	uvm_swap_free(anon->an_swslot, 1);
 	anon->an_swslot = 0;
 
-	if (anon->an_page == NULL) {
+	if (anon->u.an_page == NULL) {
 		/* this page is no longer only in swap. */
 		simple_lock(&uvm.swap_data_lock);
 		uvmexp.swpgonly--;
@@ -233,7 +325,7 @@ uvm_anon_lockloanpg(anon)
 	 * not produce an incorrect result.
 	 */
 
-	while (((pg = anon->an_page) != NULL) && pg->loan_count != 0) {
+	while (((pg = anon->u.an_page) != NULL) && pg->loan_count != 0) {
 
 		/*
 		 * quickly check to see if the page has an object before
@@ -295,6 +387,73 @@ uvm_anon_lockloanpg(anon)
 	return(pg);
 }
 
+
+
+/*
+ * page in every anon that is paged out to a range of swslots.
+ * 
+ * swap_syscall_lock should be held (protects anonblock_list).
+ */
+
+boolean_t
+anon_swap_off(startslot, endslot)
+	int startslot, endslot;
+{
+	struct uvm_anonblock *anonblock;
+
+	for (anonblock = LIST_FIRST(&anonblock_list);
+	     anonblock != NULL;
+	     anonblock = LIST_NEXT(anonblock, list)) {
+		int i;
+
+		/*
+		 * loop thru all the anons in the anonblock,
+		 * paging in where needed.
+		 */
+
+		for (i = 0; i < anonblock->count; i++) {
+			struct vm_anon *anon = &anonblock->anons[i];
+			int slot;
+
+			/*
+			 * lock anon to work on it.
+			 */
+
+			simple_lock(&anon->an_lock);
+
+			/*
+			 * is this anon's swap slot in range?
+			 */
+
+			slot = anon->an_swslot;
+			if (slot >= startslot && slot < endslot) {
+				boolean_t rv;
+
+				/*
+				 * yup, page it in.
+				 */
+
+				/* locked: anon */
+				rv = anon_pagein(anon);
+				/* unlocked: anon */
+
+				if (rv) {
+					return rv;
+				}
+			} else {
+
+				/*
+				 * nope, unlock and proceed.
+				 */
+
+				simple_unlock(&anon->an_lock);
+			}
+		}
+	}
+	return FALSE;
+}
+
+
 /*
  * fetch an anon's page.
  *
@@ -302,8 +461,8 @@ uvm_anon_lockloanpg(anon)
  * => returns TRUE if pagein was aborted due to lack of memory.
  */
 
-boolean_t
-uvm_anon_pagein(anon)
+static boolean_t
+anon_pagein(anon)
 	struct vm_anon *anon;
 {
 	struct vm_page *pg;
@@ -345,7 +504,7 @@ uvm_anon_pagein(anon)
 	 * mark it as dirty, clear its swslot and un-busy it.
 	 */
 
-	pg = anon->an_page;
+	pg = anon->u.an_page;
 	uobj = pg->uobject;
 	uvm_swap_free(anon->an_swslot, 1);
 	anon->an_swslot = 0;