9 files changed, 135 insertions, 54 deletions

diff --git a/sys/kern/vfs_bio.c b/sys/kern/vfs_bio.c
index 3e6e6f94783e..a6533956dc5c 100644
--- a/sys/kern/vfs_bio.c
+++ b/sys/kern/vfs_bio.c
@@ -18,7 +18,7 @@
  * 5. Modifications may be freely made to this file if the above conditions
  *    are met.
  *
- * $Id: vfs_bio.c,v 1.105 1996/11/17 02:10:48 dyson Exp $
+ * $Id: vfs_bio.c,v 1.106 1996/11/28 04:26:04 dyson Exp $
  */
 
 /*
@@ -51,6 +51,8 @@
 #include <vm/vm_page.h>
 #include <vm/vm_object.h>
 #include <vm/vm_extern.h>
+#include <vm/lock.h>
+#include <vm/vm_map.h>
 #include <sys/buf.h>
 #include <sys/mount.h>
 #include <sys/malloc.h>
@@ -92,7 +94,6 @@ int vfs_update_wakeup;
 /*
  * buffers base kva
  */
-caddr_t buffers_kva;
 
 /*
  * bogus page -- for I/O to/from partially complete buffers
@@ -134,7 +135,6 @@ bufinit()
 	for (i = 0; i < BUFFER_QUEUES; i++)
 		TAILQ_INIT(&bufqueues[i]);
 
-	buffers_kva = (caddr_t) kmem_alloc_pageable(buffer_map, MAXBSIZE * nbuf);
 	/* finally, initialize each buffer header and stick on empty q */
 	for (i = 0; i < nbuf; i++) {
 		bp = &buf[i];
@@ -145,7 +145,6 @@ bufinit()
 		bp->b_wcred = NOCRED;
 		bp->b_qindex = QUEUE_EMPTY;
 		bp->b_vnbufs.le_next = NOLIST;
-		bp->b_data = buffers_kva + i * MAXBSIZE;
 		TAILQ_INSERT_TAIL(&bufqueues[QUEUE_EMPTY], bp, b_freelist);
 		LIST_INSERT_HEAD(&invalhash, bp, b_hash);
 	}
@@ -177,6 +176,25 @@ bufinit()
 }
 
 /*
+ * Free the kva allocation for a buffer
+ * Must be called only at splbio or higher,
+ *  as this is the only locking for buffer_map.
+ */
+static void
+bfreekva(struct buf * bp)
+{
+	if (bp->b_kvasize == 0)
+		return;
+		
+	vm_map_delete(buffer_map,
+		(vm_offset_t) bp->b_kvabase,
+		(vm_offset_t) bp->b_kvabase + bp->b_kvasize);
+
+	bp->b_kvasize = 0;
+
+}
+
+/*
  * remove the buffer from the appropriate free list
  */
 void
@@ -562,6 +580,10 @@ brelse(struct buf * bp)
 		LIST_REMOVE(bp, b_hash);
 		LIST_INSERT_HEAD(&invalhash, bp, b_hash);
 		bp->b_dev = NODEV;
+		/*
+		 * Get rid of the kva allocation *now*
+		 */
+		bfreekva(bp);
 		if (needsbuffer) {
 			wakeup(&needsbuffer);
 			needsbuffer=0;
@@ -724,7 +746,7 @@ vfs_vmio_release(bp)
 /*
  * Check to see if a block is currently memory resident.
  */
-__inline struct buf *
+struct buf *
 gbincore(struct vnode * vp, daddr_t blkno)
 {
 	struct buf *bp;
@@ -812,10 +834,11 @@ vfs_bio_awrite(struct buf * bp)
  * Find a buffer header which is available for use.
  */
 static struct buf *
-getnewbuf(int slpflag, int slptimeo, int doingvmio)
+getnewbuf(int slpflag, int slptimeo, int size, int maxsize)
 {
 	struct buf *bp;
 	int nbyteswritten = 0;
+	vm_offset_t addr;
 
 start:
 	if (bufspace >= maxbufspace)
@@ -926,15 +949,43 @@ fillbuf:
 	bp->b_resid = 0;
 	bp->b_bcount = 0;
 	bp->b_npages = 0;
-	bp->b_data = buffers_kva + (bp - buf) * MAXBSIZE;
 	bp->b_dirtyoff = bp->b_dirtyend = 0;
 	bp->b_validoff = bp->b_validend = 0;
 	bp->b_usecount = 4;
-	if (bufspace >= maxbufspace + nbyteswritten) {
+
+	maxsize = (maxsize + PAGE_MASK) & ~PAGE_MASK;
+	bfreekva(bp);
+		
+	/*
+	 * See if we have buffer kva space
+	 */
+	if (vm_map_findspace(buffer_map, 0, maxsize, &addr)) {
+		bp->b_flags |= B_INVAL;
+		brelse(bp);
+		goto trytofreespace;
+	}
+
+	/*
+	 * See if we are below are allocated minimum
+	 */
+	if (bufspace >= (maxbufspace + nbyteswritten)) {
 		bp->b_flags |= B_INVAL;
 		brelse(bp);
 		goto trytofreespace;
 	}
+
+	/*
+	 * create a map entry for the buffer -- in essence
+	 * reserving the kva space.
+	 */
+	vm_map_insert(buffer_map, NULL, 0,
+		addr, addr + maxsize,
+		VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
+
+	bp->b_data = (caddr_t) addr;
+	bp->b_kvabase = (caddr_t) addr;
+	bp->b_kvasize = maxsize;
+	
 	return (bp);
 }
 
@@ -1057,6 +1108,18 @@ getblk(struct vnode * vp, daddr_t blkno, int size, int slpflag, int slptimeo)
 	struct buf *bp;
 	int s;
 	struct bufhashhdr *bh;
+	int maxsize;
+
+	if (vp->v_mount) {
+		maxsize = vp->v_mount->mnt_stat.f_iosize;
+		/*
+		 * This happens on mount points.
+		 */
+		if (maxsize < size)
+			maxsize = size;
+	} else {
+		maxsize = size;
+	}
 
 	if (size > MAXBSIZE)
 		panic("getblk: size(%d) > MAXBSIZE(%d)\n", size, MAXBSIZE);
@@ -1086,7 +1149,7 @@ loop:
 		 */
 
 		if (bp->b_bcount != size) {
-			if (bp->b_flags & B_VMIO) {
+			if ((bp->b_flags & B_VMIO) && (size <= bp->b_kvasize)) {
 				allocbuf(bp, size);
 			} else {
 				bp->b_flags |= B_NOCACHE;
@@ -1101,14 +1164,8 @@ loop:
 		return (bp);
 	} else {
 		vm_object_t obj;
-		int doingvmio;
 
-		if ((obj = vp->v_object) && (vp->v_flag & VVMIO)) {
-			doingvmio = 1;
-		} else {
-			doingvmio = 0;
-		}
-		if ((bp = getnewbuf(slpflag, slptimeo, doingvmio)) == 0) {
+		if ((bp = getnewbuf(slpflag, slptimeo, size, maxsize)) == 0) {
 			if (slpflag || slptimeo) {
 				splx(s);
 				return NULL;
@@ -1138,7 +1195,7 @@ loop:
 		bh = BUFHASH(vp, blkno);
 		LIST_INSERT_HEAD(bh, bp, b_hash);
 
-		if (doingvmio) {
+		if ((obj = vp->v_object) && (vp->v_flag & VVMIO)) {
 			bp->b_flags |= (B_VMIO | B_CACHE);
 #if defined(VFS_BIO_DEBUG)
 			if (vp->v_type != VREG && vp->v_type != VBLK)
@@ -1171,7 +1228,7 @@ geteblk(int size)
 	int s;
 
 	s = splbio();
-	while ((bp = getnewbuf(0, 0, 0)) == 0);
+	while ((bp = getnewbuf(0, 0, size, MAXBSIZE)) == 0);
 	splx(s);
 	allocbuf(bp, size);
 	bp->b_flags |= B_INVAL;
@@ -1201,6 +1258,9 @@ allocbuf(struct buf * bp, int size)
 	if (!(bp->b_flags & B_BUSY))
 		panic("allocbuf: buffer not busy");
 
+	if (bp->b_kvasize < size)
+		panic("allocbuf: buffer too small");
+
 	if ((bp->b_flags & B_VMIO) == 0) {
 		caddr_t origbuf;
 		int origbufsize;
@@ -1227,7 +1287,7 @@ allocbuf(struct buf * bp, int size)
 					free(bp->b_data, M_BIOBUF);
 					bufspace -= bp->b_bufsize;
 					bufmallocspace -= bp->b_bufsize;
-					bp->b_data = (caddr_t) buffers_kva + (bp - buf) * MAXBSIZE;
+					bp->b_data = bp->b_kvabase;
 					bp->b_bufsize = 0;
 					bp->b_bcount = 0;
 					bp->b_flags &= ~B_MALLOC;
@@ -1268,7 +1328,7 @@ allocbuf(struct buf * bp, int size)
 			if (bp->b_flags & B_MALLOC) {
 				origbuf = bp->b_data;
 				origbufsize = bp->b_bufsize;
-				bp->b_data = (caddr_t) buffers_kva + (bp - buf) * MAXBSIZE;
+				bp->b_data = bp->b_kvabase;
 				bufspace -= bp->b_bufsize;
 				bufmallocspace -= bp->b_bufsize;
 				bp->b_bufsize = 0;
diff --git a/sys/kern/vfs_cluster.c b/sys/kern/vfs_cluster.c
index 386d9813e5ed..d45663ad6a37 100644
--- a/sys/kern/vfs_cluster.c
+++ b/sys/kern/vfs_cluster.c
@@ -33,7 +33,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)vfs_cluster.c	8.7 (Berkeley) 2/13/94
- * $Id: vfs_cluster.c,v 1.37 1996/07/27 18:49:18 dyson Exp $
+ * $Id: vfs_cluster.c,v 1.38 1996/10/06 07:50:04 dyson Exp $
  */
 
 #include <sys/param.h>
@@ -385,6 +385,10 @@ cluster_rbuild(vp, filesize, lbn, blkno, size, run)
 			VM_PAGE_BITS_ALL)
 			bp->b_pages[j] = bogus_page;
 	}
+	if (bp->b_bufsize > bp->b_kvasize)
+		panic("cluster_rbuild: b_bufsize(%d) > b_kvasize(%d)\n",
+			bp->b_bufsize, bp->b_kvasize);
+	bp->b_kvasize = bp->b_bufsize;
 
 	pmap_qenter(trunc_page((vm_offset_t) bp->b_data),
 		(vm_page_t *)bp->b_pages, bp->b_npages);
@@ -690,6 +694,10 @@ cluster_wbuild(vp, size, start_lbn, len)
 		}
 		pmap_qenter(trunc_page((vm_offset_t) bp->b_data),
 			(vm_page_t *) bp->b_pages, bp->b_npages);
+		if (bp->b_bufsize > bp->b_kvasize)
+			panic("cluster_wbuild: b_bufsize(%d) > b_kvasize(%d)\n",
+				bp->b_bufsize, bp->b_kvasize);
+		bp->b_kvasize = bp->b_bufsize;
 		totalwritten += bp->b_bufsize;
 		bp->b_dirtyoff = 0;
 		bp->b_dirtyend = bp->b_bufsize;
diff --git a/sys/sys/bio.h b/sys/sys/bio.h
index 3244a7e81a1c..ee7d2135e17a 100644
--- a/sys/sys/bio.h
+++ b/sys/sys/bio.h
@@ -36,7 +36,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)buf.h	8.9 (Berkeley) 3/30/95
- * $Id: buf.h,v 1.33 1996/09/06 05:35:00 gibbs Exp $
+ * $Id: buf.h,v 1.34 1996/10/13 14:36:37 phk Exp $
  */
 
 #ifndef _SYS_BUF_H_
@@ -80,6 +80,8 @@ struct buf {
 	struct {
 		caddr_t	b_addr;		/* Memory, superblocks, indirect etc. */
 	} b_un;
+	caddr_t	b_kvabase;		/* base kva for buffer */
+	int	b_kvasize;		/* size of kva for buffer */
 	void	*b_saveaddr;		/* Original b_addr for physio. */
 	daddr_t	b_lblkno;		/* Logical block number. */
 	daddr_t	b_blkno;		/* Underlying physical block number. */
diff --git a/sys/sys/buf.h b/sys/sys/buf.h
index 3244a7e81a1c..ee7d2135e17a 100644
--- a/sys/sys/buf.h
+++ b/sys/sys/buf.h
@@ -36,7 +36,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)buf.h	8.9 (Berkeley) 3/30/95
- * $Id: buf.h,v 1.33 1996/09/06 05:35:00 gibbs Exp $
+ * $Id: buf.h,v 1.34 1996/10/13 14:36:37 phk Exp $
  */
 
 #ifndef _SYS_BUF_H_
@@ -80,6 +80,8 @@ struct buf {
 	struct {
 		caddr_t	b_addr;		/* Memory, superblocks, indirect etc. */
 	} b_un;
+	caddr_t	b_kvabase;		/* base kva for buffer */
+	int	b_kvasize;		/* size of kva for buffer */
 	void	*b_saveaddr;		/* Original b_addr for physio. */
 	daddr_t	b_lblkno;		/* Logical block number. */
 	daddr_t	b_blkno;		/* Underlying physical block number. */
diff --git a/sys/sys/param.h b/sys/sys/param.h
index 7e7a0372c453..60a16ac0c91e 100644
--- a/sys/sys/param.h
+++ b/sys/sys/param.h
@@ -36,7 +36,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)param.h	8.3 (Berkeley) 4/4/95
- * $Id: param.h,v 1.15 1996/05/02 14:21:03 phk Exp $
+ * $Id: param.h,v 1.16 1996/11/28 04:07:44 dyson Exp $
  */
 
 #ifndef _SYS_PARAM_H_
@@ -146,11 +146,8 @@
  * it smaller make make some file systems unmountable.  Also, MAXBSIZE
  * must be less than MAXPHYS!!!
  */
-#if defined(MSDOSFS)
-#define	MAXBSIZE	32768
-#else
-#define	MAXBSIZE	16384
-#endif
+#define	MAXBSIZE	65536
+#define DFLTBSIZE	8192
 #define MAXFRAG 	8
 
 /*
diff --git a/sys/vm/vm_fault.c b/sys/vm/vm_fault.c
index 1119c62d3850..f395d9701d7c 100644
--- a/sys/vm/vm_fault.c
+++ b/sys/vm/vm_fault.c
@@ -66,7 +66,7 @@
  * any improvements or extensions that they make and grant Carnegie the
  * rights to redistribute these changes.
  *
- * $Id: vm_fault.c,v 1.56 1996/07/30 03:08:07 dyson Exp $
+ * $Id: vm_fault.c,v 1.57 1996/09/08 20:44:37 dyson Exp $
  */
 
 /*
@@ -81,6 +81,7 @@
 #include <sys/signalvar.h>
 #include <sys/resourcevar.h>
 #include <sys/vmmeter.h>
+#include <sys/buf.h>
 
 #include <vm/vm.h>
 #include <vm/vm_param.h>
@@ -103,10 +104,6 @@ int vm_fault_additional_pages __P((vm_page_t, int, int, vm_page_t *, int *));
 #define VM_FAULT_READ_BEHIND 3
 #define VM_FAULT_READ (VM_FAULT_READ_AHEAD+VM_FAULT_READ_BEHIND+1)
 
-int vm_fault_free_1;
-int vm_fault_copy_save_1;
-int vm_fault_copy_save_2;
-
 /*
  *	vm_fault:
  *
@@ -200,6 +197,11 @@ RetryFault:;
 		return (result);
 	}
 
+	if (entry->nofault) {
+		panic("vm_fault: fault on nofault entry, addr: %lx",
+			vaddr);
+	}
+
 	vp = vnode_pager_lock(first_object);
 
 	lookup_still_valid = TRUE;
@@ -565,7 +567,6 @@ readrest:
 				first_m = m;
 				m->dirty = VM_PAGE_BITS_ALL;
 				m = NULL;
-				++vm_fault_copy_save_1;
 			} else {
 				/*
 				 * Oh, well, lets copy it.
@@ -639,7 +640,6 @@ readrest:
 								PAGE_WAKEUP(m);
 								vm_page_free(m);
 								m = NULL;
-								++vm_fault_free_1;
 								tm->dirty = VM_PAGE_BITS_ALL;
 								first_m->dirty = VM_PAGE_BITS_ALL;
 							}
@@ -651,7 +651,6 @@ readrest:
 							vm_page_rename(m, other_object, other_pindex);
 							m->dirty = VM_PAGE_BITS_ALL;
 							m->valid = VM_PAGE_BITS_ALL;
-							++vm_fault_copy_save_2;
 						}
 					}
 				}
diff --git a/sys/vm/vm_map.c b/sys/vm/vm_map.c
index 5a04287a7480..48071a0daa57 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 1.56 1996/09/08 23:49:47 dyson Exp $
+ * $Id: vm_map.c,v 1.57 1996/09/14 11:54:55 bde Exp $
  */
 
 /*
@@ -689,6 +689,11 @@ vm_map_insert(map, object, offset, start, end, prot, max, cow)
 	else
 		new_entry->copy_on_write = FALSE;
 
+	if (cow & MAP_NOFAULT)
+		new_entry->nofault = TRUE;
+	else
+		new_entry->nofault = FALSE;
+
 	if (map->is_main_map) {
 		new_entry->inheritance = VM_INHERIT_DEFAULT;
 		new_entry->protection = prot;
diff --git a/sys/vm/vm_map.h b/sys/vm/vm_map.h
index fd04d87de264..40569ef0ed78 100644
--- a/sys/vm/vm_map.h
+++ b/sys/vm/vm_map.h
@@ -61,7 +61,7 @@
  * any improvements or extensions that they make and grant Carnegie the
  * rights to redistribute these changes.
  *
- * $Id: vm_map.h,v 1.14 1996/07/27 03:23:59 dyson Exp $
+ * $Id: vm_map.h,v 1.15 1996/07/30 03:08:11 dyson Exp $
  */
 
 /*
@@ -106,9 +106,9 @@ struct vm_map_entry {
 	vm_ooffset_t offset;		/* offset into object */
 	boolean_t is_a_map:1,		/* Is "object" a map? */
 	 is_sub_map:1,			/* Is "object" a submap? */
-	/* Only in sharing maps: */
 	 copy_on_write:1,		/* is data copy-on-write */
-	 needs_copy:1;			/* does object need to be copied */
+	 needs_copy:1,			/* does object need to be copied */
+	 nofault:1;			/* should never fault */
 	/* Only in task maps: */
 	vm_prot_t protection;		/* protection code */
 	vm_prot_t max_protection;	/* maximum protection */
@@ -208,6 +208,7 @@ typedef struct {
  */
 #define MAP_COPY_NEEDED 0x1
 #define MAP_COPY_ON_WRITE 0x2
+#define MAP_NOFAULT 0x4
 
 #ifdef KERNEL
 extern vm_offset_t kentry_data;
diff --git a/sys/vm/vm_pager.c b/sys/vm/vm_pager.c
index b8db9ac4cb9b..de81090bd6c0 100644
--- a/sys/vm/vm_pager.c
+++ b/sys/vm/vm_pager.c
@@ -61,7 +61,7 @@
  * any improvements or extensions that they make and grant Carnegie the
  * rights to redistribute these changes.
  *
- * $Id: vm_pager.c,v 1.23 1996/05/18 03:38:05 dyson Exp $
+ * $Id: vm_pager.c,v 1.24 1996/09/08 20:44:49 dyson Exp $
  */
 
 /*
@@ -278,6 +278,22 @@ pager_cache(object, should_cache)
 }
 
 /*
+ * initialize a physical buffer
+ */
+
+static void
+initpbuf(struct buf *bp) {
+	bzero(bp, sizeof *bp);
+	bp->b_rcred = NOCRED;
+	bp->b_wcred = NOCRED;
+	bp->b_qindex = QUEUE_NONE;
+	bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
+	bp->b_kvabase = bp->b_data;
+	bp->b_kvasize = MAXPHYS;
+	bp->b_vnbufs.le_next = NOLIST;
+}
+
+/*
  * allocate a physical buffer
  */
 struct buf *
@@ -295,12 +311,7 @@ getpbuf()
 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
 	splx(s);
 
-	bzero(bp, sizeof *bp);
-	bp->b_rcred = NOCRED;
-	bp->b_wcred = NOCRED;
-	bp->b_qindex = QUEUE_NONE;
-	bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
-	bp->b_vnbufs.le_next = NOLIST;
+	initpbuf(bp);
 	return bp;
 }
 
@@ -321,12 +332,8 @@ trypbuf()
 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
 	splx(s);
 
-	bzero(bp, sizeof *bp);
-	bp->b_rcred = NOCRED;
-	bp->b_wcred = NOCRED;
-	bp->b_qindex = QUEUE_NONE;
-	bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
-	bp->b_vnbufs.le_next = NOLIST;
+	initpbuf(bp);
+
 	return bp;
 }