Add the ability for the allocflag argument of the vm_page_grab() to

specify the increment of vm_pageout_deficit when sleeping due to page
shortage. Then, in allocbuf(), the code to allocate pages when extending
vmio buffer can be replaced by a call to vm_page_grab().

Suggested and reviewed by:	alc
MFC after:	2 weeks

3 files changed, 24 insertions, 52 deletions

diff --git a/sys/kern/vfs_bio.c b/sys/kern/vfs_bio.c
index 95f8929..762b253 100644
--- a/sys/kern/vfs_bio.c
+++ b/sys/kern/vfs_bio.c
@@ -3013,63 +3013,24 @@ allocbuf(struct buf *bp, int size)
 			VM_OBJECT_LOCK(obj);
 			while (bp->b_npages < desiredpages) {
 				vm_page_t m;
-				vm_pindex_t pi;
-
-				pi = OFF_TO_IDX(bp->b_offset) + bp->b_npages;
-				if ((m = vm_page_lookup(obj, pi)) == NULL) {
-					/*
-					 * note: must allocate system pages
-					 * since blocking here could intefere
-					 * with paging I/O, no matter which
-					 * process we are.
-					 */
-					m = vm_page_alloc(obj, pi,
-					    VM_ALLOC_NOBUSY | VM_ALLOC_SYSTEM |
-					    VM_ALLOC_WIRED);
-					if (m == NULL) {
-						atomic_add_int(&vm_pageout_deficit,
-						    desiredpages - bp->b_npages);
-						VM_OBJECT_UNLOCK(obj);
-						VM_WAIT;
-						VM_OBJECT_LOCK(obj);
-					} else {
-						if (m->valid == 0)
-							bp->b_flags &= ~B_CACHE;
-						bp->b_pages[bp->b_npages] = m;
-						++bp->b_npages;
-					}
-					continue;
-				}
 
 				/*
-				 * We found a page.  If we have to sleep on it,
-				 * retry because it might have gotten freed out
-				 * from under us.
+				 * We must allocate system pages since blocking
+				 * here could intefere with paging I/O, no
+				 * matter which process we are.
 				 *
 				 * We can only test VPO_BUSY here.  Blocking on
 				 * m->busy might lead to a deadlock:
-				 *
 				 *  vm_fault->getpages->cluster_read->allocbuf
-				 *
-				 */
-				if ((m->oflags & VPO_BUSY) != 0) {
-					/*
-					 * Reference the page before unlocking
-					 * and sleeping so that the page daemon
-					 * is less likely to reclaim it.  
-					 */
-					vm_page_lock_queues();
-					vm_page_flag_set(m, PG_REFERENCED);
-					vm_page_sleep(m, "pgtblk");
-					continue;
-				}
-
-				/*
-				 * We have a good page.
+				 * Thus, we specify VM_ALLOC_IGN_SBUSY.
 				 */
-				vm_page_lock(m);
-				vm_page_wire(m);
-				vm_page_unlock(m);
+				m = vm_page_grab(obj, OFF_TO_IDX(bp->b_offset) +
+				    bp->b_npages, VM_ALLOC_NOBUSY |
+				    VM_ALLOC_SYSTEM | VM_ALLOC_WIRED |
+				    VM_ALLOC_RETRY | VM_ALLOC_IGN_SBUSY |
+				    VM_ALLOC_COUNT(desiredpages - bp->b_npages));
+				if (m->valid == 0)
+					bp->b_flags &= ~B_CACHE;
 				bp->b_pages[bp->b_npages] = m;
 				++bp->b_npages;
 			}
diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c
index 6659e76..810338d 100644
--- a/sys/vm/vm_page.c
+++ b/sys/vm/vm_page.c
@@ -2038,11 +2038,13 @@ vm_page_t
 vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
 {
 	vm_page_t m;
+	u_int count;
 
 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
 retrylookup:
 	if ((m = vm_page_lookup(object, pindex)) != NULL) {
-		if ((m->oflags & VPO_BUSY) != 0 || m->busy != 0) {
+		if ((m->oflags & VPO_BUSY) != 0 ||
+		    ((allocflags & VM_ALLOC_IGN_SBUSY) == 0 && m->busy != 0)) {
 			if ((allocflags & VM_ALLOC_RETRY) != 0) {
 				/*
 				 * Reference the page before unlocking and
@@ -2067,9 +2069,13 @@ retrylookup:
 			return (m);
 		}
 	}
-	m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY);
+	m = vm_page_alloc(object, pindex, allocflags & ~(VM_ALLOC_RETRY |
+	    VM_ALLOC_IGN_SBUSY | VM_ALLOC_COUNT_MASK));
 	if (m == NULL) {
 		VM_OBJECT_UNLOCK(object);
+		count = (u_int)allocflags >> VM_ALLOC_COUNT_SHIFT;
+		if (count > 0)
+			atomic_add_int(&vm_pageout_deficit, count);
 		VM_WAIT;
 		VM_OBJECT_LOCK(object);
 		if ((allocflags & VM_ALLOC_RETRY) == 0)
diff --git a/sys/vm/vm_page.h b/sys/vm/vm_page.h
index 1b2f39a..0829025 100644
--- a/sys/vm/vm_page.h
+++ b/sys/vm/vm_page.h
@@ -317,6 +317,11 @@ extern struct vpglocks vm_page_queue_lock;
 #define	VM_ALLOC_NOBUSY		0x0200	/* Do not busy the page */
 #define	VM_ALLOC_IFCACHED	0x0400	/* Fail if the page is not cached */
 #define	VM_ALLOC_IFNOTCACHED	0x0800	/* Fail if the page is cached */
+#define	VM_ALLOC_IGN_SBUSY	0x1000	/* vm_page_grab() only */
+
+#define	VM_ALLOC_COUNT_SHIFT	16
+#define	VM_ALLOC_COUNT(count)	((count) << VM_ALLOC_COUNT_SHIFT)
+#define	VM_ALLOC_COUNT_MASK	VM_ALLOC_COUNT(0xffff)
 
 void vm_page_flag_set(vm_page_t m, unsigned short bits);
 void vm_page_flag_clear(vm_page_t m, unsigned short bits);