xref: /netbsd-src/sys/uvm/uvm_map.c (revision 1ad9454efb13a65cd7535ccf867508cb14d9d30e)
1 /*	$NetBSD: uvm_map.c,v 1.229 2006/09/16 07:14:38 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
5  * Copyright (c) 1991, 1993, The Regents of the University of California.
6  *
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * The Mach Operating System project at Carnegie-Mellon University.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by Charles D. Cranor,
23  *      Washington University, the University of California, Berkeley and
24  *      its contributors.
25  * 4. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	@(#)vm_map.c    8.3 (Berkeley) 1/12/94
42  * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp
43  *
44  *
45  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
46  * All rights reserved.
47  *
48  * Permission to use, copy, modify and distribute this software and
49  * its documentation is hereby granted, provided that both the copyright
50  * notice and this permission notice appear in all copies of the
51  * software, derivative works or modified versions, and any portions
52  * thereof, and that both notices appear in supporting documentation.
53  *
54  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
55  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
56  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
57  *
58  * Carnegie Mellon requests users of this software to return to
59  *
60  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
61  *  School of Computer Science
62  *  Carnegie Mellon University
63  *  Pittsburgh PA 15213-3890
64  *
65  * any improvements or extensions that they make and grant Carnegie the
66  * rights to redistribute these changes.
67  */
68 
69 /*
70  * uvm_map.c: uvm map operations
71  */
72 
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: uvm_map.c,v 1.229 2006/09/16 07:14:38 yamt Exp $");
75 
76 #include "opt_ddb.h"
77 #include "opt_uvmhist.h"
78 #include "opt_uvm.h"
79 #include "opt_sysv.h"
80 
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/mman.h>
84 #include <sys/proc.h>
85 #include <sys/malloc.h>
86 #include <sys/pool.h>
87 #include <sys/kernel.h>
88 #include <sys/mount.h>
89 #include <sys/vnode.h>
90 
91 #ifdef SYSVSHM
92 #include <sys/shm.h>
93 #endif
94 
95 #include <uvm/uvm.h>
96 #undef RB_AUGMENT
97 #define	RB_AUGMENT(x)	uvm_rb_augment(x)
98 
99 #ifdef DDB
100 #include <uvm/uvm_ddb.h>
101 #endif
102 
103 #if defined(UVMMAP_NOCOUNTERS)
104 
105 #define	UVMMAP_EVCNT_DEFINE(name)	/* nothing */
106 #define UVMMAP_EVCNT_INCR(ev)		/* nothing */
107 #define UVMMAP_EVCNT_DECR(ev)		/* nothing */
108 
109 #else /* defined(UVMMAP_NOCOUNTERS) */
110 
111 #include <sys/evcnt.h>
112 #define	UVMMAP_EVCNT_DEFINE(name) \
113 struct evcnt uvmmap_evcnt_##name = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, \
114     "uvmmap", #name); \
115 EVCNT_ATTACH_STATIC(uvmmap_evcnt_##name);
116 #define	UVMMAP_EVCNT_INCR(ev)		uvmmap_evcnt_##ev.ev_count++
117 #define	UVMMAP_EVCNT_DECR(ev)		uvmmap_evcnt_##ev.ev_count--
118 
119 #endif /* defined(UVMMAP_NOCOUNTERS) */
120 
121 UVMMAP_EVCNT_DEFINE(ubackmerge)
122 UVMMAP_EVCNT_DEFINE(uforwmerge)
123 UVMMAP_EVCNT_DEFINE(ubimerge)
124 UVMMAP_EVCNT_DEFINE(unomerge)
125 UVMMAP_EVCNT_DEFINE(kbackmerge)
126 UVMMAP_EVCNT_DEFINE(kforwmerge)
127 UVMMAP_EVCNT_DEFINE(kbimerge)
128 UVMMAP_EVCNT_DEFINE(knomerge)
129 UVMMAP_EVCNT_DEFINE(map_call)
130 UVMMAP_EVCNT_DEFINE(mlk_call)
131 UVMMAP_EVCNT_DEFINE(mlk_hint)
132 
133 UVMMAP_EVCNT_DEFINE(uke_alloc)
134 UVMMAP_EVCNT_DEFINE(uke_free)
135 UVMMAP_EVCNT_DEFINE(ukh_alloc)
136 UVMMAP_EVCNT_DEFINE(ukh_free)
137 
138 const char vmmapbsy[] = "vmmapbsy";
139 
140 /*
141  * pool for vmspace structures.
142  */
143 
144 POOL_INIT(uvm_vmspace_pool, sizeof(struct vmspace), 0, 0, 0, "vmsppl",
145     &pool_allocator_nointr);
146 
147 /*
148  * pool for dynamically-allocated map entries.
149  */
150 
151 POOL_INIT(uvm_map_entry_pool, sizeof(struct vm_map_entry), 0, 0, 0, "vmmpepl",
152     &pool_allocator_nointr);
153 
154 MALLOC_DEFINE(M_VMMAP, "VM map", "VM map structures");
155 MALLOC_DEFINE(M_VMPMAP, "VM pmap", "VM pmap");
156 
157 #ifdef PMAP_GROWKERNEL
158 /*
159  * This global represents the end of the kernel virtual address
160  * space.  If we want to exceed this, we must grow the kernel
161  * virtual address space dynamically.
162  *
163  * Note, this variable is locked by kernel_map's lock.
164  */
165 vaddr_t uvm_maxkaddr;
166 #endif
167 
168 /*
169  * macros
170  */
171 
172 /*
173  * VM_MAP_USE_KMAPENT: determine if uvm_kmapent_alloc/free is used
174  * for the vm_map.
175  */
176 extern struct vm_map *pager_map; /* XXX */
177 #define	VM_MAP_USE_KMAPENT_FLAGS(flags) \
178 	(((flags) & VM_MAP_INTRSAFE) != 0)
179 #define	VM_MAP_USE_KMAPENT(map) \
180 	(VM_MAP_USE_KMAPENT_FLAGS((map)->flags) || (map) == kernel_map)
181 
182 /*
183  * UVM_ET_ISCOMPATIBLE: check some requirements for map entry merging
184  */
185 
186 #define	UVM_ET_ISCOMPATIBLE(ent, type, uobj, meflags, \
187     prot, maxprot, inh, adv, wire) \
188 	((ent)->etype == (type) && \
189 	(((ent)->flags ^ (meflags)) & (UVM_MAP_NOMERGE | UVM_MAP_QUANTUM)) \
190 	== 0 && \
191 	(ent)->object.uvm_obj == (uobj) && \
192 	(ent)->protection == (prot) && \
193 	(ent)->max_protection == (maxprot) && \
194 	(ent)->inheritance == (inh) && \
195 	(ent)->advice == (adv) && \
196 	(ent)->wired_count == (wire))
197 
198 /*
199  * uvm_map_entry_link: insert entry into a map
200  *
201  * => map must be locked
202  */
203 #define uvm_map_entry_link(map, after_where, entry) do { \
204 	uvm_mapent_check(entry); \
205 	(map)->nentries++; \
206 	(entry)->prev = (after_where); \
207 	(entry)->next = (after_where)->next; \
208 	(entry)->prev->next = (entry); \
209 	(entry)->next->prev = (entry); \
210 	uvm_rb_insert((map), (entry)); \
211 } while (/*CONSTCOND*/ 0)
212 
213 /*
214  * uvm_map_entry_unlink: remove entry from a map
215  *
216  * => map must be locked
217  */
218 #define uvm_map_entry_unlink(map, entry) do { \
219 	KASSERT((entry) != (map)->first_free); \
220 	KASSERT((entry) != (map)->hint); \
221 	uvm_mapent_check(entry); \
222 	(map)->nentries--; \
223 	(entry)->next->prev = (entry)->prev; \
224 	(entry)->prev->next = (entry)->next; \
225 	uvm_rb_remove((map), (entry)); \
226 } while (/*CONSTCOND*/ 0)
227 
228 /*
229  * SAVE_HINT: saves the specified entry as the hint for future lookups.
230  *
231  * => map need not be locked (protected by hint_lock).
232  */
233 #define SAVE_HINT(map,check,value) do { \
234 	simple_lock(&(map)->hint_lock); \
235 	if ((map)->hint == (check)) \
236 		(map)->hint = (value); \
237 	simple_unlock(&(map)->hint_lock); \
238 } while (/*CONSTCOND*/ 0)
239 
240 /*
241  * clear_hints: ensure that hints don't point to the entry.
242  *
243  * => map must be write-locked.
244  */
245 static void
246 clear_hints(struct vm_map *map, struct vm_map_entry *ent)
247 {
248 
249 	SAVE_HINT(map, ent, ent->prev);
250 	if (map->first_free == ent) {
251 		map->first_free = ent->prev;
252 	}
253 }
254 
255 /*
256  * VM_MAP_RANGE_CHECK: check and correct range
257  *
258  * => map must at least be read locked
259  */
260 
261 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
262 	if (start < vm_map_min(map))		\
263 		start = vm_map_min(map);	\
264 	if (end > vm_map_max(map))		\
265 		end = vm_map_max(map);		\
266 	if (start > end)			\
267 		start = end;			\
268 } while (/*CONSTCOND*/ 0)
269 
270 /*
271  * local prototypes
272  */
273 
274 static struct vm_map_entry *
275 		uvm_mapent_alloc(struct vm_map *, int);
276 static struct vm_map_entry *
277 		uvm_mapent_alloc_split(struct vm_map *,
278 		    const struct vm_map_entry *, int,
279 		    struct uvm_mapent_reservation *);
280 static void	uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *);
281 static void	uvm_mapent_free(struct vm_map_entry *);
282 #if defined(DEBUG)
283 static void	_uvm_mapent_check(const struct vm_map_entry *, const char *,
284 		    int);
285 #define	uvm_mapent_check(map)	_uvm_mapent_check(map, __FILE__, __LINE__)
286 #else /* defined(DEBUG) */
287 #define	uvm_mapent_check(e)	/* nothing */
288 #endif /* defined(DEBUG) */
289 static struct vm_map_entry *
290 		uvm_kmapent_alloc(struct vm_map *, int);
291 static void	uvm_kmapent_free(struct vm_map_entry *);
292 static vsize_t	uvm_kmapent_overhead(vsize_t);
293 
294 static void	uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *);
295 static void	uvm_map_reference_amap(struct vm_map_entry *, int);
296 static int	uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int,
297 		    struct vm_map_entry *);
298 static void	uvm_map_unreference_amap(struct vm_map_entry *, int);
299 
300 int _uvm_map_sanity(struct vm_map *);
301 int _uvm_tree_sanity(struct vm_map *);
302 static vsize_t uvm_rb_subtree_space(const struct vm_map_entry *);
303 
304 static inline int
305 uvm_compare(const struct vm_map_entry *a, const struct vm_map_entry *b)
306 {
307 
308 	if (a->start < b->start)
309 		return (-1);
310 	else if (a->start > b->start)
311 		return (1);
312 
313 	return (0);
314 }
315 
316 static inline void
317 uvm_rb_augment(struct vm_map_entry *entry)
318 {
319 
320 	entry->space = uvm_rb_subtree_space(entry);
321 }
322 
323 RB_PROTOTYPE(uvm_tree, vm_map_entry, rb_entry, uvm_compare);
324 
325 RB_GENERATE(uvm_tree, vm_map_entry, rb_entry, uvm_compare);
326 
327 static inline vsize_t
328 uvm_rb_space(const struct vm_map *map, const struct vm_map_entry *entry)
329 {
330 	/* XXX map is not used */
331 
332 	KASSERT(entry->next != NULL);
333 	return entry->next->start - entry->end;
334 }
335 
336 static vsize_t
337 uvm_rb_subtree_space(const struct vm_map_entry *entry)
338 {
339 	vaddr_t space, tmp;
340 
341 	space = entry->ownspace;
342 	if (RB_LEFT(entry, rb_entry)) {
343 		tmp = RB_LEFT(entry, rb_entry)->space;
344 		if (tmp > space)
345 			space = tmp;
346 	}
347 
348 	if (RB_RIGHT(entry, rb_entry)) {
349 		tmp = RB_RIGHT(entry, rb_entry)->space;
350 		if (tmp > space)
351 			space = tmp;
352 	}
353 
354 	return (space);
355 }
356 
357 static inline void
358 uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry)
359 {
360 	/* We need to traverse to the very top */
361 	do {
362 		entry->ownspace = uvm_rb_space(map, entry);
363 		entry->space = uvm_rb_subtree_space(entry);
364 	} while ((entry = RB_PARENT(entry, rb_entry)) != NULL);
365 }
366 
367 static void
368 uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry)
369 {
370 	vaddr_t space = uvm_rb_space(map, entry);
371 	struct vm_map_entry *tmp;
372 
373 	entry->ownspace = entry->space = space;
374 	tmp = RB_INSERT(uvm_tree, &(map)->rbhead, entry);
375 #ifdef DIAGNOSTIC
376 	if (tmp != NULL)
377 		panic("uvm_rb_insert: duplicate entry?");
378 #endif
379 	uvm_rb_fixup(map, entry);
380 	if (entry->prev != &map->header)
381 		uvm_rb_fixup(map, entry->prev);
382 }
383 
384 static void
385 uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry)
386 {
387 	struct vm_map_entry *parent;
388 
389 	parent = RB_PARENT(entry, rb_entry);
390 	RB_REMOVE(uvm_tree, &(map)->rbhead, entry);
391 	if (entry->prev != &map->header)
392 		uvm_rb_fixup(map, entry->prev);
393 	if (parent)
394 		uvm_rb_fixup(map, parent);
395 }
396 
397 #if defined(DEBUG)
398 int uvm_debug_check_map = 0;
399 int uvm_debug_check_rbtree = 0;
400 #define uvm_map_check(map, name) \
401 	_uvm_map_check((map), (name), __FILE__, __LINE__)
402 static void
403 _uvm_map_check(struct vm_map *map, const char *name,
404     const char *file, int line)
405 {
406 
407 	if ((uvm_debug_check_map && _uvm_map_sanity(map)) ||
408 	    (uvm_debug_check_rbtree && _uvm_tree_sanity(map))) {
409 		panic("uvm_map_check failed: \"%s\" map=%p (%s:%d)",
410 		    name, map, file, line);
411 	}
412 }
413 #else /* defined(DEBUG) */
414 #define uvm_map_check(map, name)	/* nothing */
415 #endif /* defined(DEBUG) */
416 
417 #if defined(DEBUG) || defined(DDB)
418 int
419 _uvm_map_sanity(struct vm_map *map)
420 {
421 	boolean_t first_free_found = FALSE;
422 	boolean_t hint_found = FALSE;
423 	const struct vm_map_entry *e;
424 
425 	e = &map->header;
426 	for (;;) {
427 		if (map->first_free == e) {
428 			first_free_found = TRUE;
429 		} else if (!first_free_found && e->next->start > e->end) {
430 			printf("first_free %p should be %p\n",
431 			    map->first_free, e);
432 			return -1;
433 		}
434 		if (map->hint == e) {
435 			hint_found = TRUE;
436 		}
437 
438 		e = e->next;
439 		if (e == &map->header) {
440 			break;
441 		}
442 	}
443 	if (!first_free_found) {
444 		printf("stale first_free\n");
445 		return -1;
446 	}
447 	if (!hint_found) {
448 		printf("stale hint\n");
449 		return -1;
450 	}
451 	return 0;
452 }
453 
454 int
455 _uvm_tree_sanity(struct vm_map *map)
456 {
457 	struct vm_map_entry *tmp, *trtmp;
458 	int n = 0, i = 1;
459 
460 	RB_FOREACH(tmp, uvm_tree, &map->rbhead) {
461 		if (tmp->ownspace != uvm_rb_space(map, tmp)) {
462 			printf("%d/%d ownspace %lx != %lx %s\n",
463 			    n + 1, map->nentries,
464 			    (ulong)tmp->ownspace, (ulong)uvm_rb_space(map, tmp),
465 			    tmp->next == &map->header ? "(last)" : "");
466 			goto error;
467 		}
468 	}
469 	trtmp = NULL;
470 	RB_FOREACH(tmp, uvm_tree, &map->rbhead) {
471 		if (tmp->space != uvm_rb_subtree_space(tmp)) {
472 			printf("space %lx != %lx\n",
473 			    (ulong)tmp->space,
474 			    (ulong)uvm_rb_subtree_space(tmp));
475 			goto error;
476 		}
477 		if (trtmp != NULL && trtmp->start >= tmp->start) {
478 			printf("corrupt: 0x%lx >= 0x%lx\n",
479 			    trtmp->start, tmp->start);
480 			goto error;
481 		}
482 		n++;
483 
484 		trtmp = tmp;
485 	}
486 
487 	if (n != map->nentries) {
488 		printf("nentries: %d vs %d\n", n, map->nentries);
489 		goto error;
490 	}
491 
492 	for (tmp = map->header.next; tmp && tmp != &map->header;
493 	    tmp = tmp->next, i++) {
494 		trtmp = RB_FIND(uvm_tree, &map->rbhead, tmp);
495 		if (trtmp != tmp) {
496 			printf("lookup: %d: %p - %p: %p\n", i, tmp, trtmp,
497 			    RB_PARENT(tmp, rb_entry));
498 			goto error;
499 		}
500 	}
501 
502 	return (0);
503  error:
504 	return (-1);
505 }
506 #endif /* defined(DEBUG) || defined(DDB) */
507 
508 #ifdef DIAGNOSTIC
509 static struct vm_map *uvm_kmapent_map(struct vm_map_entry *);
510 #endif
511 
512 /*
513  * uvm_mapent_alloc: allocate a map entry
514  */
515 
516 static struct vm_map_entry *
517 uvm_mapent_alloc(struct vm_map *map, int flags)
518 {
519 	struct vm_map_entry *me;
520 	int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK;
521 	UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist);
522 
523 	if (VM_MAP_USE_KMAPENT(map)) {
524 		me = uvm_kmapent_alloc(map, flags);
525 	} else {
526 		me = pool_get(&uvm_map_entry_pool, pflags);
527 		if (__predict_false(me == NULL))
528 			return NULL;
529 		me->flags = 0;
530 	}
531 
532 	UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me,
533 	    ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0);
534 	return (me);
535 }
536 
537 /*
538  * uvm_mapent_alloc_split: allocate a map entry for clipping.
539  */
540 
541 static struct vm_map_entry *
542 uvm_mapent_alloc_split(struct vm_map *map,
543     const struct vm_map_entry *old_entry, int flags,
544     struct uvm_mapent_reservation *umr)
545 {
546 	struct vm_map_entry *me;
547 
548 	KASSERT(!VM_MAP_USE_KMAPENT(map) ||
549 	    (old_entry->flags & UVM_MAP_QUANTUM) || !UMR_EMPTY(umr));
550 
551 	if (old_entry->flags & UVM_MAP_QUANTUM) {
552 		int s;
553 		struct vm_map_kernel *vmk = vm_map_to_kernel(map);
554 
555 		s = splvm();
556 		simple_lock(&uvm.kentry_lock);
557 		me = vmk->vmk_merged_entries;
558 		KASSERT(me);
559 		vmk->vmk_merged_entries = me->next;
560 		simple_unlock(&uvm.kentry_lock);
561 		splx(s);
562 		KASSERT(me->flags & UVM_MAP_QUANTUM);
563 	} else {
564 		me = uvm_mapent_alloc(map, flags);
565 	}
566 
567 	return me;
568 }
569 
570 /*
571  * uvm_mapent_free: free map entry
572  */
573 
574 static void
575 uvm_mapent_free(struct vm_map_entry *me)
576 {
577 	UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist);
578 
579 	UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
580 		me, me->flags, 0, 0);
581 	if (me->flags & UVM_MAP_KERNEL) {
582 		uvm_kmapent_free(me);
583 	} else {
584 		pool_put(&uvm_map_entry_pool, me);
585 	}
586 }
587 
588 /*
589  * uvm_mapent_free_merged: free merged map entry
590  *
591  * => keep the entry if needed.
592  * => caller shouldn't hold map locked if VM_MAP_USE_KMAPENT(map) is true.
593  */
594 
595 static void
596 uvm_mapent_free_merged(struct vm_map *map, struct vm_map_entry *me)
597 {
598 
599 	KASSERT(!(me->flags & UVM_MAP_KERNEL) || uvm_kmapent_map(me) == map);
600 
601 	if (me->flags & UVM_MAP_QUANTUM) {
602 		/*
603 		 * keep this entry for later splitting.
604 		 */
605 		struct vm_map_kernel *vmk;
606 		int s;
607 
608 		KASSERT(VM_MAP_IS_KERNEL(map));
609 		KASSERT(!VM_MAP_USE_KMAPENT(map) ||
610 		    (me->flags & UVM_MAP_KERNEL));
611 
612 		vmk = vm_map_to_kernel(map);
613 		s = splvm();
614 		simple_lock(&uvm.kentry_lock);
615 		me->next = vmk->vmk_merged_entries;
616 		vmk->vmk_merged_entries = me;
617 		simple_unlock(&uvm.kentry_lock);
618 		splx(s);
619 	} else {
620 		uvm_mapent_free(me);
621 	}
622 }
623 
624 /*
625  * uvm_mapent_copy: copy a map entry, preserving flags
626  */
627 
628 static inline void
629 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst)
630 {
631 
632 	memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) -
633 	    ((char *)src));
634 }
635 
636 /*
637  * uvm_mapent_overhead: calculate maximum kva overhead necessary for
638  * map entries.
639  *
640  * => size and flags are the same as uvm_km_suballoc's ones.
641  */
642 
643 vsize_t
644 uvm_mapent_overhead(vsize_t size, int flags)
645 {
646 
647 	if (VM_MAP_USE_KMAPENT_FLAGS(flags)) {
648 		return uvm_kmapent_overhead(size);
649 	}
650 	return 0;
651 }
652 
653 #if defined(DEBUG)
654 static void
655 _uvm_mapent_check(const struct vm_map_entry *entry, const char *file, int line)
656 {
657 
658 	if (entry->start >= entry->end) {
659 		goto bad;
660 	}
661 	if (UVM_ET_ISOBJ(entry)) {
662 		if (entry->object.uvm_obj == NULL) {
663 			goto bad;
664 		}
665 	} else if (UVM_ET_ISSUBMAP(entry)) {
666 		if (entry->object.sub_map == NULL) {
667 			goto bad;
668 		}
669 	} else {
670 		if (entry->object.uvm_obj != NULL ||
671 		    entry->object.sub_map != NULL) {
672 			goto bad;
673 		}
674 	}
675 	if (!UVM_ET_ISOBJ(entry)) {
676 		if (entry->offset != 0) {
677 			goto bad;
678 		}
679 	}
680 
681 	return;
682 
683 bad:
684 	panic("%s: bad entry %p (%s:%d)", __func__, entry, file, line);
685 }
686 #endif /* defined(DEBUG) */
687 
688 /*
689  * uvm_map_entry_unwire: unwire a map entry
690  *
691  * => map should be locked by caller
692  */
693 
694 static inline void
695 uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry)
696 {
697 
698 	entry->wired_count = 0;
699 	uvm_fault_unwire_locked(map, entry->start, entry->end);
700 }
701 
702 
703 /*
704  * wrapper for calling amap_ref()
705  */
706 static inline void
707 uvm_map_reference_amap(struct vm_map_entry *entry, int flags)
708 {
709 
710 	amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff,
711 	    (entry->end - entry->start) >> PAGE_SHIFT, flags);
712 }
713 
714 
715 /*
716  * wrapper for calling amap_unref()
717  */
718 static inline void
719 uvm_map_unreference_amap(struct vm_map_entry *entry, int flags)
720 {
721 
722 	amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff,
723 	    (entry->end - entry->start) >> PAGE_SHIFT, flags);
724 }
725 
726 
727 /*
728  * uvm_map_init: init mapping system at boot time.   note that we allocate
729  * and init the static pool of struct vm_map_entry *'s for the kernel here.
730  */
731 
732 void
733 uvm_map_init(void)
734 {
735 #if defined(UVMHIST)
736 	static struct uvm_history_ent maphistbuf[100];
737 	static struct uvm_history_ent pdhistbuf[100];
738 #endif
739 
740 	/*
741 	 * first, init logging system.
742 	 */
743 
744 	UVMHIST_FUNC("uvm_map_init");
745 	UVMHIST_INIT_STATIC(maphist, maphistbuf);
746 	UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
747 	UVMHIST_CALLED(maphist);
748 	UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
749 
750 	/*
751 	 * initialize the global lock for kernel map entry.
752 	 *
753 	 * XXX is it worth to have per-map lock instead?
754 	 */
755 
756 	simple_lock_init(&uvm.kentry_lock);
757 }
758 
759 /*
760  * clippers
761  */
762 
763 /*
764  * uvm_mapent_splitadj: adjust map entries for splitting, after uvm_mapent_copy.
765  */
766 
767 static void
768 uvm_mapent_splitadj(struct vm_map_entry *entry1, struct vm_map_entry *entry2,
769     vaddr_t splitat)
770 {
771 	vaddr_t adj;
772 
773 	KASSERT(entry1->start < splitat);
774 	KASSERT(splitat < entry1->end);
775 
776 	adj = splitat - entry1->start;
777 	entry1->end = entry2->start = splitat;
778 
779 	if (entry1->aref.ar_amap) {
780 		amap_splitref(&entry1->aref, &entry2->aref, adj);
781 	}
782 	if (UVM_ET_ISSUBMAP(entry1)) {
783 		/* ... unlikely to happen, but play it safe */
784 		 uvm_map_reference(entry1->object.sub_map);
785 	} else if (UVM_ET_ISOBJ(entry1)) {
786 		KASSERT(entry1->object.uvm_obj != NULL); /* suppress coverity */
787 		entry2->offset += adj;
788 		if (entry1->object.uvm_obj->pgops &&
789 		    entry1->object.uvm_obj->pgops->pgo_reference)
790 			entry1->object.uvm_obj->pgops->pgo_reference(
791 			    entry1->object.uvm_obj);
792 	}
793 }
794 
795 /*
796  * uvm_map_clip_start: ensure that the entry begins at or after
797  *	the starting address, if it doesn't we split the entry.
798  *
799  * => caller should use UVM_MAP_CLIP_START macro rather than calling
800  *    this directly
801  * => map must be locked by caller
802  */
803 
804 void
805 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry,
806     vaddr_t start, struct uvm_mapent_reservation *umr)
807 {
808 	struct vm_map_entry *new_entry;
809 
810 	/* uvm_map_simplify_entry(map, entry); */ /* XXX */
811 
812 	uvm_map_check(map, "clip_start entry");
813 	uvm_mapent_check(entry);
814 
815 	/*
816 	 * Split off the front portion.  note that we must insert the new
817 	 * entry BEFORE this one, so that this entry has the specified
818 	 * starting address.
819 	 */
820 	new_entry = uvm_mapent_alloc_split(map, entry, 0, umr);
821 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
822 	uvm_mapent_splitadj(new_entry, entry, start);
823 	uvm_map_entry_link(map, entry->prev, new_entry);
824 
825 	uvm_map_check(map, "clip_start leave");
826 }
827 
828 /*
829  * uvm_map_clip_end: ensure that the entry ends at or before
830  *	the ending address, if it does't we split the reference
831  *
832  * => caller should use UVM_MAP_CLIP_END macro rather than calling
833  *    this directly
834  * => map must be locked by caller
835  */
836 
837 void
838 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end,
839     struct uvm_mapent_reservation *umr)
840 {
841 	struct vm_map_entry *new_entry;
842 
843 	uvm_map_check(map, "clip_end entry");
844 	uvm_mapent_check(entry);
845 
846 	/*
847 	 *	Create a new entry and insert it
848 	 *	AFTER the specified entry
849 	 */
850 	new_entry = uvm_mapent_alloc_split(map, entry, 0, umr);
851 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
852 	uvm_mapent_splitadj(entry, new_entry, end);
853 	uvm_map_entry_link(map, entry, new_entry);
854 
855 	uvm_map_check(map, "clip_end leave");
856 }
857 
858 static void
859 vm_map_drain(struct vm_map *map, uvm_flag_t flags)
860 {
861 
862 	if (!VM_MAP_IS_KERNEL(map)) {
863 		return;
864 	}
865 
866 	uvm_km_va_drain(map, flags);
867 }
868 
869 /*
870  *   M A P   -   m a i n   e n t r y   p o i n t
871  */
872 /*
873  * uvm_map: establish a valid mapping in a map
874  *
875  * => assume startp is page aligned.
876  * => assume size is a multiple of PAGE_SIZE.
877  * => assume sys_mmap provides enough of a "hint" to have us skip
878  *	over text/data/bss area.
879  * => map must be unlocked (we will lock it)
880  * => <uobj,uoffset> value meanings (4 cases):
881  *	 [1] <NULL,uoffset>		== uoffset is a hint for PMAP_PREFER
882  *	 [2] <NULL,UVM_UNKNOWN_OFFSET>	== don't PMAP_PREFER
883  *	 [3] <uobj,uoffset>		== normal mapping
884  *	 [4] <uobj,UVM_UNKNOWN_OFFSET>	== uvm_map finds offset based on VA
885  *
886  *    case [4] is for kernel mappings where we don't know the offset until
887  *    we've found a virtual address.   note that kernel object offsets are
888  *    always relative to vm_map_min(kernel_map).
889  *
890  * => if `align' is non-zero, we align the virtual address to the specified
891  *	alignment.
892  *	this is provided as a mechanism for large pages.
893  *
894  * => XXXCDC: need way to map in external amap?
895  */
896 
897 int
898 uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size,
899     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags)
900 {
901 	struct uvm_map_args args;
902 	struct vm_map_entry *new_entry;
903 	int error;
904 
905 	KASSERT((flags & UVM_FLAG_QUANTUM) == 0 || VM_MAP_IS_KERNEL(map));
906 	KASSERT((size & PAGE_MASK) == 0);
907 
908 	/*
909 	 * for pager_map, allocate the new entry first to avoid sleeping
910 	 * for memory while we have the map locked.
911 	 *
912 	 * besides, because we allocates entries for in-kernel maps
913 	 * a bit differently (cf. uvm_kmapent_alloc/free), we need to
914 	 * allocate them before locking the map.
915 	 */
916 
917 	new_entry = NULL;
918 	if (VM_MAP_USE_KMAPENT(map) || (flags & UVM_FLAG_QUANTUM) ||
919 	    map == pager_map) {
920 		new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT));
921 		if (__predict_false(new_entry == NULL))
922 			return ENOMEM;
923 		if (flags & UVM_FLAG_QUANTUM)
924 			new_entry->flags |= UVM_MAP_QUANTUM;
925 	}
926 	if (map == pager_map)
927 		flags |= UVM_FLAG_NOMERGE;
928 
929 	error = uvm_map_prepare(map, *startp, size, uobj, uoffset, align,
930 	    flags, &args);
931 	if (!error) {
932 		error = uvm_map_enter(map, &args, new_entry);
933 		*startp = args.uma_start;
934 	} else if (new_entry) {
935 		uvm_mapent_free(new_entry);
936 	}
937 
938 #if defined(DEBUG)
939 	if (!error && VM_MAP_IS_KERNEL(map)) {
940 		uvm_km_check_empty(*startp, *startp + size,
941 		    (map->flags & VM_MAP_INTRSAFE) != 0);
942 	}
943 #endif /* defined(DEBUG) */
944 
945 	return error;
946 }
947 
948 int
949 uvm_map_prepare(struct vm_map *map, vaddr_t start, vsize_t size,
950     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags,
951     struct uvm_map_args *args)
952 {
953 	struct vm_map_entry *prev_entry;
954 	vm_prot_t prot = UVM_PROTECTION(flags);
955 	vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
956 
957 	UVMHIST_FUNC("uvm_map_prepare");
958 	UVMHIST_CALLED(maphist);
959 
960 	UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)",
961 	    map, start, size, flags);
962 	UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
963 
964 	/*
965 	 * detect a popular device driver bug.
966 	 */
967 
968 	KASSERT(doing_shutdown || curlwp != NULL ||
969 	    (map->flags & VM_MAP_INTRSAFE));
970 
971 	/*
972 	 * zero-sized mapping doesn't make any sense.
973 	 */
974 	KASSERT(size > 0);
975 
976 	KASSERT((~flags & (UVM_FLAG_NOWAIT | UVM_FLAG_WAITVA)) != 0);
977 
978 	uvm_map_check(map, "map entry");
979 
980 	/*
981 	 * check sanity of protection code
982 	 */
983 
984 	if ((prot & maxprot) != prot) {
985 		UVMHIST_LOG(maphist, "<- prot. failure:  prot=0x%x, max=0x%x",
986 		prot, maxprot,0,0);
987 		return EACCES;
988 	}
989 
990 	/*
991 	 * figure out where to put new VM range
992 	 */
993 
994 retry:
995 	if (vm_map_lock_try(map) == FALSE) {
996 		if (flags & UVM_FLAG_TRYLOCK) {
997 			return EAGAIN;
998 		}
999 		vm_map_lock(map); /* could sleep here */
1000 	}
1001 	prev_entry = uvm_map_findspace(map, start, size, &start,
1002 	    uobj, uoffset, align, flags);
1003 	if (prev_entry == NULL) {
1004 		unsigned int timestamp;
1005 
1006 		timestamp = map->timestamp;
1007 		UVMHIST_LOG(maphist,"waiting va timestamp=0x%x",
1008 			    timestamp,0,0,0);
1009 		simple_lock(&map->flags_lock);
1010 		map->flags |= VM_MAP_WANTVA;
1011 		simple_unlock(&map->flags_lock);
1012 		vm_map_unlock(map);
1013 
1014 		/*
1015 		 * try to reclaim kva and wait until someone does unmap.
1016 		 * XXX fragile locking
1017 		 */
1018 
1019 		vm_map_drain(map, flags);
1020 
1021 		simple_lock(&map->flags_lock);
1022 		while ((map->flags & VM_MAP_WANTVA) != 0 &&
1023 		   map->timestamp == timestamp) {
1024 			if ((flags & UVM_FLAG_WAITVA) == 0) {
1025 				simple_unlock(&map->flags_lock);
1026 				UVMHIST_LOG(maphist,
1027 				    "<- uvm_map_findspace failed!", 0,0,0,0);
1028 				return ENOMEM;
1029 			} else {
1030 				ltsleep(&map->header, PVM, "vmmapva", 0,
1031 				    &map->flags_lock);
1032 			}
1033 		}
1034 		simple_unlock(&map->flags_lock);
1035 		goto retry;
1036 	}
1037 
1038 #ifdef PMAP_GROWKERNEL
1039 	/*
1040 	 * If the kernel pmap can't map the requested space,
1041 	 * then allocate more resources for it.
1042 	 */
1043 	if (map == kernel_map && uvm_maxkaddr < (start + size))
1044 		uvm_maxkaddr = pmap_growkernel(start + size);
1045 #endif
1046 
1047 	UVMMAP_EVCNT_INCR(map_call);
1048 
1049 	/*
1050 	 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
1051 	 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET.   in
1052 	 * either case we want to zero it  before storing it in the map entry
1053 	 * (because it looks strange and confusing when debugging...)
1054 	 *
1055 	 * if uobj is not null
1056 	 *   if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
1057 	 *      and we do not need to change uoffset.
1058 	 *   if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
1059 	 *      now (based on the starting address of the map).   this case is
1060 	 *      for kernel object mappings where we don't know the offset until
1061 	 *      the virtual address is found (with uvm_map_findspace).   the
1062 	 *      offset is the distance we are from the start of the map.
1063 	 */
1064 
1065 	if (uobj == NULL) {
1066 		uoffset = 0;
1067 	} else {
1068 		if (uoffset == UVM_UNKNOWN_OFFSET) {
1069 			KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj));
1070 			uoffset = start - vm_map_min(kernel_map);
1071 		}
1072 	}
1073 
1074 	args->uma_flags = flags;
1075 	args->uma_prev = prev_entry;
1076 	args->uma_start = start;
1077 	args->uma_size = size;
1078 	args->uma_uobj = uobj;
1079 	args->uma_uoffset = uoffset;
1080 
1081 	return 0;
1082 }
1083 
1084 int
1085 uvm_map_enter(struct vm_map *map, const struct uvm_map_args *args,
1086     struct vm_map_entry *new_entry)
1087 {
1088 	struct vm_map_entry *prev_entry = args->uma_prev;
1089 	struct vm_map_entry *dead = NULL;
1090 
1091 	const uvm_flag_t flags = args->uma_flags;
1092 	const vm_prot_t prot = UVM_PROTECTION(flags);
1093 	const vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
1094 	const vm_inherit_t inherit = UVM_INHERIT(flags);
1095 	const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ?
1096 	    AMAP_EXTEND_NOWAIT : 0;
1097 	const int advice = UVM_ADVICE(flags);
1098 	const int meflagval = (flags & UVM_FLAG_QUANTUM) ?
1099 	    UVM_MAP_QUANTUM : 0;
1100 
1101 	vaddr_t start = args->uma_start;
1102 	vsize_t size = args->uma_size;
1103 	struct uvm_object *uobj = args->uma_uobj;
1104 	voff_t uoffset = args->uma_uoffset;
1105 
1106 	const int kmap = (vm_map_pmap(map) == pmap_kernel());
1107 	int merged = 0;
1108 	int error;
1109 	int newetype;
1110 
1111 	UVMHIST_FUNC("uvm_map_enter");
1112 	UVMHIST_CALLED(maphist);
1113 
1114 	UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)",
1115 	    map, start, size, flags);
1116 	UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
1117 
1118 	KASSERT(map->hint == prev_entry); /* bimerge case assumes this */
1119 
1120 	if (flags & UVM_FLAG_QUANTUM) {
1121 		KASSERT(new_entry);
1122 		KASSERT(new_entry->flags & UVM_MAP_QUANTUM);
1123 	}
1124 
1125 	if (uobj)
1126 		newetype = UVM_ET_OBJ;
1127 	else
1128 		newetype = 0;
1129 
1130 	if (flags & UVM_FLAG_COPYONW) {
1131 		newetype |= UVM_ET_COPYONWRITE;
1132 		if ((flags & UVM_FLAG_OVERLAY) == 0)
1133 			newetype |= UVM_ET_NEEDSCOPY;
1134 	}
1135 
1136 	/*
1137 	 * try and insert in map by extending previous entry, if possible.
1138 	 * XXX: we don't try and pull back the next entry.   might be useful
1139 	 * for a stack, but we are currently allocating our stack in advance.
1140 	 */
1141 
1142 	if (flags & UVM_FLAG_NOMERGE)
1143 		goto nomerge;
1144 
1145 	if (prev_entry->end == start &&
1146 	    prev_entry != &map->header &&
1147 	    UVM_ET_ISCOMPATIBLE(prev_entry, newetype, uobj, meflagval,
1148 	    prot, maxprot, inherit, advice, 0)) {
1149 
1150 		if (uobj && prev_entry->offset +
1151 		    (prev_entry->end - prev_entry->start) != uoffset)
1152 			goto forwardmerge;
1153 
1154 		/*
1155 		 * can't extend a shared amap.  note: no need to lock amap to
1156 		 * look at refs since we don't care about its exact value.
1157 		 * if it is one (i.e. we have only reference) it will stay there
1158 		 */
1159 
1160 		if (prev_entry->aref.ar_amap &&
1161 		    amap_refs(prev_entry->aref.ar_amap) != 1) {
1162 			goto forwardmerge;
1163 		}
1164 
1165 		if (prev_entry->aref.ar_amap) {
1166 			error = amap_extend(prev_entry, size,
1167 			    amapwaitflag | AMAP_EXTEND_FORWARDS);
1168 			if (error)
1169 				goto nomerge;
1170 		}
1171 
1172 		if (kmap)
1173 			UVMMAP_EVCNT_INCR(kbackmerge);
1174 		else
1175 			UVMMAP_EVCNT_INCR(ubackmerge);
1176 		UVMHIST_LOG(maphist,"  starting back merge", 0, 0, 0, 0);
1177 
1178 		/*
1179 		 * drop our reference to uobj since we are extending a reference
1180 		 * that we already have (the ref count can not drop to zero).
1181 		 */
1182 
1183 		if (uobj && uobj->pgops->pgo_detach)
1184 			uobj->pgops->pgo_detach(uobj);
1185 
1186 		prev_entry->end += size;
1187 		uvm_rb_fixup(map, prev_entry);
1188 
1189 		uvm_map_check(map, "map backmerged");
1190 
1191 		UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
1192 		merged++;
1193 	}
1194 
1195 forwardmerge:
1196 	if (prev_entry->next->start == (start + size) &&
1197 	    prev_entry->next != &map->header &&
1198 	    UVM_ET_ISCOMPATIBLE(prev_entry->next, newetype, uobj, meflagval,
1199 	    prot, maxprot, inherit, advice, 0)) {
1200 
1201 		if (uobj && prev_entry->next->offset != uoffset + size)
1202 			goto nomerge;
1203 
1204 		/*
1205 		 * can't extend a shared amap.  note: no need to lock amap to
1206 		 * look at refs since we don't care about its exact value.
1207 		 * if it is one (i.e. we have only reference) it will stay there.
1208 		 *
1209 		 * note that we also can't merge two amaps, so if we
1210 		 * merged with the previous entry which has an amap,
1211 		 * and the next entry also has an amap, we give up.
1212 		 *
1213 		 * Interesting cases:
1214 		 * amap, new, amap -> give up second merge (single fwd extend)
1215 		 * amap, new, none -> double forward extend (extend again here)
1216 		 * none, new, amap -> double backward extend (done here)
1217 		 * uobj, new, amap -> single backward extend (done here)
1218 		 *
1219 		 * XXX should we attempt to deal with someone refilling
1220 		 * the deallocated region between two entries that are
1221 		 * backed by the same amap (ie, arefs is 2, "prev" and
1222 		 * "next" refer to it, and adding this allocation will
1223 		 * close the hole, thus restoring arefs to 1 and
1224 		 * deallocating the "next" vm_map_entry)?  -- @@@
1225 		 */
1226 
1227 		if (prev_entry->next->aref.ar_amap &&
1228 		    (amap_refs(prev_entry->next->aref.ar_amap) != 1 ||
1229 		     (merged && prev_entry->aref.ar_amap))) {
1230 			goto nomerge;
1231 		}
1232 
1233 		if (merged) {
1234 			/*
1235 			 * Try to extend the amap of the previous entry to
1236 			 * cover the next entry as well.  If it doesn't work
1237 			 * just skip on, don't actually give up, since we've
1238 			 * already completed the back merge.
1239 			 */
1240 			if (prev_entry->aref.ar_amap) {
1241 				if (amap_extend(prev_entry,
1242 				    prev_entry->next->end -
1243 				    prev_entry->next->start,
1244 				    amapwaitflag | AMAP_EXTEND_FORWARDS))
1245 					goto nomerge;
1246 			}
1247 
1248 			/*
1249 			 * Try to extend the amap of the *next* entry
1250 			 * back to cover the new allocation *and* the
1251 			 * previous entry as well (the previous merge
1252 			 * didn't have an amap already otherwise we
1253 			 * wouldn't be checking here for an amap).  If
1254 			 * it doesn't work just skip on, again, don't
1255 			 * actually give up, since we've already
1256 			 * completed the back merge.
1257 			 */
1258 			else if (prev_entry->next->aref.ar_amap) {
1259 				if (amap_extend(prev_entry->next,
1260 				    prev_entry->end -
1261 				    prev_entry->start,
1262 				    amapwaitflag | AMAP_EXTEND_BACKWARDS))
1263 					goto nomerge;
1264 			}
1265 		} else {
1266 			/*
1267 			 * Pull the next entry's amap backwards to cover this
1268 			 * new allocation.
1269 			 */
1270 			if (prev_entry->next->aref.ar_amap) {
1271 				error = amap_extend(prev_entry->next, size,
1272 				    amapwaitflag | AMAP_EXTEND_BACKWARDS);
1273 				if (error)
1274 					goto nomerge;
1275 			}
1276 		}
1277 
1278 		if (merged) {
1279 			if (kmap) {
1280 				UVMMAP_EVCNT_DECR(kbackmerge);
1281 				UVMMAP_EVCNT_INCR(kbimerge);
1282 			} else {
1283 				UVMMAP_EVCNT_DECR(ubackmerge);
1284 				UVMMAP_EVCNT_INCR(ubimerge);
1285 			}
1286 		} else {
1287 			if (kmap)
1288 				UVMMAP_EVCNT_INCR(kforwmerge);
1289 			else
1290 				UVMMAP_EVCNT_INCR(uforwmerge);
1291 		}
1292 		UVMHIST_LOG(maphist,"  starting forward merge", 0, 0, 0, 0);
1293 
1294 		/*
1295 		 * drop our reference to uobj since we are extending a reference
1296 		 * that we already have (the ref count can not drop to zero).
1297 		 * (if merged, we've already detached)
1298 		 */
1299 		if (uobj && uobj->pgops->pgo_detach && !merged)
1300 			uobj->pgops->pgo_detach(uobj);
1301 
1302 		if (merged) {
1303 			dead = prev_entry->next;
1304 			prev_entry->end = dead->end;
1305 			uvm_map_entry_unlink(map, dead);
1306 			if (dead->aref.ar_amap != NULL) {
1307 				prev_entry->aref = dead->aref;
1308 				dead->aref.ar_amap = NULL;
1309 			}
1310 		} else {
1311 			prev_entry->next->start -= size;
1312 			if (prev_entry != &map->header)
1313 				uvm_rb_fixup(map, prev_entry);
1314 			if (uobj)
1315 				prev_entry->next->offset = uoffset;
1316 		}
1317 
1318 		uvm_map_check(map, "map forwardmerged");
1319 
1320 		UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0);
1321 		merged++;
1322 	}
1323 
1324 nomerge:
1325 	if (!merged) {
1326 		UVMHIST_LOG(maphist,"  allocating new map entry", 0, 0, 0, 0);
1327 		if (kmap)
1328 			UVMMAP_EVCNT_INCR(knomerge);
1329 		else
1330 			UVMMAP_EVCNT_INCR(unomerge);
1331 
1332 		/*
1333 		 * allocate new entry and link it in.
1334 		 */
1335 
1336 		if (new_entry == NULL) {
1337 			new_entry = uvm_mapent_alloc(map,
1338 				(flags & UVM_FLAG_NOWAIT));
1339 			if (__predict_false(new_entry == NULL)) {
1340 				error = ENOMEM;
1341 				goto done;
1342 			}
1343 		}
1344 		new_entry->start = start;
1345 		new_entry->end = new_entry->start + size;
1346 		new_entry->object.uvm_obj = uobj;
1347 		new_entry->offset = uoffset;
1348 
1349 		new_entry->etype = newetype;
1350 
1351 		if (flags & UVM_FLAG_NOMERGE) {
1352 			new_entry->flags |= UVM_MAP_NOMERGE;
1353 		}
1354 
1355 		new_entry->protection = prot;
1356 		new_entry->max_protection = maxprot;
1357 		new_entry->inheritance = inherit;
1358 		new_entry->wired_count = 0;
1359 		new_entry->advice = advice;
1360 		if (flags & UVM_FLAG_OVERLAY) {
1361 
1362 			/*
1363 			 * to_add: for BSS we overallocate a little since we
1364 			 * are likely to extend
1365 			 */
1366 
1367 			vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
1368 				UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
1369 			struct vm_amap *amap = amap_alloc(size, to_add,
1370 			    (flags & UVM_FLAG_NOWAIT));
1371 			if (__predict_false(amap == NULL)) {
1372 				error = ENOMEM;
1373 				goto done;
1374 			}
1375 			new_entry->aref.ar_pageoff = 0;
1376 			new_entry->aref.ar_amap = amap;
1377 		} else {
1378 			new_entry->aref.ar_pageoff = 0;
1379 			new_entry->aref.ar_amap = NULL;
1380 		}
1381 		uvm_map_entry_link(map, prev_entry, new_entry);
1382 
1383 		/*
1384 		 * Update the free space hint
1385 		 */
1386 
1387 		if ((map->first_free == prev_entry) &&
1388 		    (prev_entry->end >= new_entry->start))
1389 			map->first_free = new_entry;
1390 
1391 		new_entry = NULL;
1392 	}
1393 
1394 	map->size += size;
1395 
1396 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1397 
1398 	error = 0;
1399 done:
1400 	vm_map_unlock(map);
1401 	if (new_entry) {
1402 		if (error == 0) {
1403 			KDASSERT(merged);
1404 			uvm_mapent_free_merged(map, new_entry);
1405 		} else {
1406 			uvm_mapent_free(new_entry);
1407 		}
1408 	}
1409 	if (dead) {
1410 		KDASSERT(merged);
1411 		uvm_mapent_free_merged(map, dead);
1412 	}
1413 	return error;
1414 }
1415 
1416 /*
1417  * uvm_map_lookup_entry: find map entry at or before an address
1418  *
1419  * => map must at least be read-locked by caller
1420  * => entry is returned in "entry"
1421  * => return value is true if address is in the returned entry
1422  */
1423 
1424 boolean_t
1425 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address,
1426     struct vm_map_entry **entry	/* OUT */)
1427 {
1428 	struct vm_map_entry *cur;
1429 	boolean_t use_tree = FALSE;
1430 	UVMHIST_FUNC("uvm_map_lookup_entry");
1431 	UVMHIST_CALLED(maphist);
1432 
1433 	UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
1434 	    map, address, entry, 0);
1435 
1436 	/*
1437 	 * start looking either from the head of the
1438 	 * list, or from the hint.
1439 	 */
1440 
1441 	simple_lock(&map->hint_lock);
1442 	cur = map->hint;
1443 	simple_unlock(&map->hint_lock);
1444 
1445 	if (cur == &map->header)
1446 		cur = cur->next;
1447 
1448 	UVMMAP_EVCNT_INCR(mlk_call);
1449 	if (address >= cur->start) {
1450 
1451 		/*
1452 		 * go from hint to end of list.
1453 		 *
1454 		 * but first, make a quick check to see if
1455 		 * we are already looking at the entry we
1456 		 * want (which is usually the case).
1457 		 * note also that we don't need to save the hint
1458 		 * here... it is the same hint (unless we are
1459 		 * at the header, in which case the hint didn't
1460 		 * buy us anything anyway).
1461 		 */
1462 
1463 		if (cur != &map->header && cur->end > address) {
1464 			UVMMAP_EVCNT_INCR(mlk_hint);
1465 			*entry = cur;
1466 			UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
1467 			    cur, 0, 0, 0);
1468 			uvm_mapent_check(*entry);
1469 			return (TRUE);
1470 		}
1471 
1472 		if (map->nentries > 30)
1473 			use_tree = TRUE;
1474 	} else {
1475 
1476 		/*
1477 		 * invalid hint.  use tree.
1478 		 */
1479 		use_tree = TRUE;
1480 	}
1481 
1482 	uvm_map_check(map, __func__);
1483 
1484 	if (use_tree) {
1485 		struct vm_map_entry *prev = &map->header;
1486 		cur = RB_ROOT(&map->rbhead);
1487 
1488 		/*
1489 		 * Simple lookup in the tree.  Happens when the hint is
1490 		 * invalid, or nentries reach a threshold.
1491 		 */
1492 		while (cur) {
1493 			if (address >= cur->start) {
1494 				if (address < cur->end) {
1495 					*entry = cur;
1496 					goto got;
1497 				}
1498 				prev = cur;
1499 				cur = RB_RIGHT(cur, rb_entry);
1500 			} else
1501 				cur = RB_LEFT(cur, rb_entry);
1502 		}
1503 		*entry = prev;
1504 		goto failed;
1505 	}
1506 
1507 	/*
1508 	 * search linearly
1509 	 */
1510 
1511 	while (cur != &map->header) {
1512 		if (cur->end > address) {
1513 			if (address >= cur->start) {
1514 				/*
1515 				 * save this lookup for future
1516 				 * hints, and return
1517 				 */
1518 
1519 				*entry = cur;
1520 got:
1521 				SAVE_HINT(map, map->hint, *entry);
1522 				UVMHIST_LOG(maphist,"<- search got it (0x%x)",
1523 					cur, 0, 0, 0);
1524 				KDASSERT((*entry)->start <= address);
1525 				KDASSERT(address < (*entry)->end);
1526 				uvm_mapent_check(*entry);
1527 				return (TRUE);
1528 			}
1529 			break;
1530 		}
1531 		cur = cur->next;
1532 	}
1533 	*entry = cur->prev;
1534 failed:
1535 	SAVE_HINT(map, map->hint, *entry);
1536 	UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
1537 	KDASSERT((*entry) == &map->header || (*entry)->end <= address);
1538 	KDASSERT((*entry)->next == &map->header ||
1539 	    address < (*entry)->next->start);
1540 	return (FALSE);
1541 }
1542 
1543 /*
1544  * See if the range between start and start + length fits in the gap
1545  * entry->next->start and entry->end.  Returns 1 if fits, 0 if doesn't
1546  * fit, and -1 address wraps around.
1547  */
1548 static int
1549 uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset,
1550     vsize_t align, int topdown, struct vm_map_entry *entry)
1551 {
1552 	vaddr_t end;
1553 
1554 #ifdef PMAP_PREFER
1555 	/*
1556 	 * push start address forward as needed to avoid VAC alias problems.
1557 	 * we only do this if a valid offset is specified.
1558 	 */
1559 
1560 	if (uoffset != UVM_UNKNOWN_OFFSET)
1561 		PMAP_PREFER(uoffset, start, length, topdown);
1562 #endif
1563 	if (align != 0) {
1564 		if ((*start & (align - 1)) != 0) {
1565 			if (topdown)
1566 				*start &= ~(align - 1);
1567 			else
1568 				*start = roundup(*start, align);
1569 		}
1570 		/*
1571 		 * XXX Should we PMAP_PREFER() here again?
1572 		 * eh...i think we're okay
1573 		 */
1574 	}
1575 
1576 	/*
1577 	 * Find the end of the proposed new region.  Be sure we didn't
1578 	 * wrap around the address; if so, we lose.  Otherwise, if the
1579 	 * proposed new region fits before the next entry, we win.
1580 	 */
1581 
1582 	end = *start + length;
1583 	if (end < *start)
1584 		return (-1);
1585 
1586 	if (entry->next->start >= end && *start >= entry->end)
1587 		return (1);
1588 
1589 	return (0);
1590 }
1591 
1592 /*
1593  * uvm_map_findspace: find "length" sized space in "map".
1594  *
1595  * => "hint" is a hint about where we want it, unless UVM_FLAG_FIXED is
1596  *	set in "flags" (in which case we insist on using "hint").
1597  * => "result" is VA returned
1598  * => uobj/uoffset are to be used to handle VAC alignment, if required
1599  * => if "align" is non-zero, we attempt to align to that value.
1600  * => caller must at least have read-locked map
1601  * => returns NULL on failure, or pointer to prev. map entry if success
1602  * => note this is a cross between the old vm_map_findspace and vm_map_find
1603  */
1604 
1605 struct vm_map_entry *
1606 uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length,
1607     vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset,
1608     vsize_t align, int flags)
1609 {
1610 	struct vm_map_entry *entry;
1611 	struct vm_map_entry *child, *prev, *tmp;
1612 	vaddr_t orig_hint;
1613 	const int topdown = map->flags & VM_MAP_TOPDOWN;
1614 	UVMHIST_FUNC("uvm_map_findspace");
1615 	UVMHIST_CALLED(maphist);
1616 
1617 	UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)",
1618 	    map, hint, length, flags);
1619 	KASSERT((align & (align - 1)) == 0);
1620 	KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0);
1621 
1622 	uvm_map_check(map, "map_findspace entry");
1623 
1624 	/*
1625 	 * remember the original hint.  if we are aligning, then we
1626 	 * may have to try again with no alignment constraint if
1627 	 * we fail the first time.
1628 	 */
1629 
1630 	orig_hint = hint;
1631 	if (hint < vm_map_min(map)) {	/* check ranges ... */
1632 		if (flags & UVM_FLAG_FIXED) {
1633 			UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
1634 			return (NULL);
1635 		}
1636 		hint = vm_map_min(map);
1637 	}
1638 	if (hint > vm_map_max(map)) {
1639 		UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
1640 		    hint, vm_map_min(map), vm_map_max(map), 0);
1641 		return (NULL);
1642 	}
1643 
1644 	/*
1645 	 * Look for the first possible address; if there's already
1646 	 * something at this address, we have to start after it.
1647 	 */
1648 
1649 	/*
1650 	 * @@@: there are four, no, eight cases to consider.
1651 	 *
1652 	 * 0: found,     fixed,     bottom up -> fail
1653 	 * 1: found,     fixed,     top down  -> fail
1654 	 * 2: found,     not fixed, bottom up -> start after entry->end,
1655 	 *                                       loop up
1656 	 * 3: found,     not fixed, top down  -> start before entry->start,
1657 	 *                                       loop down
1658 	 * 4: not found, fixed,     bottom up -> check entry->next->start, fail
1659 	 * 5: not found, fixed,     top down  -> check entry->next->start, fail
1660 	 * 6: not found, not fixed, bottom up -> check entry->next->start,
1661 	 *                                       loop up
1662 	 * 7: not found, not fixed, top down  -> check entry->next->start,
1663 	 *                                       loop down
1664 	 *
1665 	 * as you can see, it reduces to roughly five cases, and that
1666 	 * adding top down mapping only adds one unique case (without
1667 	 * it, there would be four cases).
1668 	 */
1669 
1670 	if ((flags & UVM_FLAG_FIXED) == 0 && hint == vm_map_min(map)) {
1671 		entry = map->first_free;
1672 	} else {
1673 		if (uvm_map_lookup_entry(map, hint, &entry)) {
1674 			/* "hint" address already in use ... */
1675 			if (flags & UVM_FLAG_FIXED) {
1676 				UVMHIST_LOG(maphist, "<- fixed & VA in use",
1677 				    0, 0, 0, 0);
1678 				return (NULL);
1679 			}
1680 			if (topdown)
1681 				/* Start from lower gap. */
1682 				entry = entry->prev;
1683 		} else if (flags & UVM_FLAG_FIXED) {
1684 			if (entry->next->start >= hint + length &&
1685 			    hint + length > hint)
1686 				goto found;
1687 
1688 			/* "hint" address is gap but too small */
1689 			UVMHIST_LOG(maphist, "<- fixed mapping failed",
1690 			    0, 0, 0, 0);
1691 			return (NULL); /* only one shot at it ... */
1692 		} else {
1693 			/*
1694 			 * See if given hint fits in this gap.
1695 			 */
1696 			switch (uvm_map_space_avail(&hint, length,
1697 			    uoffset, align, topdown, entry)) {
1698 			case 1:
1699 				goto found;
1700 			case -1:
1701 				goto wraparound;
1702 			}
1703 
1704 			if (topdown) {
1705 				/*
1706 				 * Still there is a chance to fit
1707 				 * if hint > entry->end.
1708 				 */
1709 			} else {
1710 				/* Start from higher gap. */
1711 				entry = entry->next;
1712 				if (entry == &map->header)
1713 					goto notfound;
1714 				goto nextgap;
1715 			}
1716 		}
1717 	}
1718 
1719 	/*
1720 	 * Note that all UVM_FLAGS_FIXED case is already handled.
1721 	 */
1722 	KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1723 
1724 	/* Try to find the space in the red-black tree */
1725 
1726 	/* Check slot before any entry */
1727 	hint = topdown ? entry->next->start - length : entry->end;
1728 	switch (uvm_map_space_avail(&hint, length, uoffset, align,
1729 	    topdown, entry)) {
1730 	case 1:
1731 		goto found;
1732 	case -1:
1733 		goto wraparound;
1734 	}
1735 
1736 nextgap:
1737 	KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1738 	/* If there is not enough space in the whole tree, we fail */
1739 	tmp = RB_ROOT(&map->rbhead);
1740 	if (tmp == NULL || tmp->space < length)
1741 		goto notfound;
1742 
1743 	prev = NULL; /* previous candidate */
1744 
1745 	/* Find an entry close to hint that has enough space */
1746 	for (; tmp;) {
1747 		KASSERT(tmp->next->start == tmp->end + tmp->ownspace);
1748 		if (topdown) {
1749 			if (tmp->next->start < hint + length &&
1750 			    (prev == NULL || tmp->end > prev->end)) {
1751 				if (tmp->ownspace >= length)
1752 					prev = tmp;
1753 				else if ((child = RB_LEFT(tmp, rb_entry))
1754 				    != NULL && child->space >= length)
1755 					prev = tmp;
1756 			}
1757 		} else {
1758 			if (tmp->end >= hint &&
1759 			    (prev == NULL || tmp->end < prev->end)) {
1760 				if (tmp->ownspace >= length)
1761 					prev = tmp;
1762 				else if ((child = RB_RIGHT(tmp, rb_entry))
1763 				    != NULL && child->space >= length)
1764 					prev = tmp;
1765 			}
1766 		}
1767 		if (tmp->next->start < hint + length)
1768 			child = RB_RIGHT(tmp, rb_entry);
1769 		else if (tmp->end > hint)
1770 			child = RB_LEFT(tmp, rb_entry);
1771 		else {
1772 			if (tmp->ownspace >= length)
1773 				break;
1774 			if (topdown)
1775 				child = RB_LEFT(tmp, rb_entry);
1776 			else
1777 				child = RB_RIGHT(tmp, rb_entry);
1778 		}
1779 		if (child == NULL || child->space < length)
1780 			break;
1781 		tmp = child;
1782 	}
1783 
1784 	if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) {
1785 		/*
1786 		 * Check if the entry that we found satifies the
1787 		 * space requirement
1788 		 */
1789 		if (topdown) {
1790 			if (hint > tmp->next->start - length)
1791 				hint = tmp->next->start - length;
1792 		} else {
1793 			if (hint < tmp->end)
1794 				hint = tmp->end;
1795 		}
1796 		switch (uvm_map_space_avail(&hint, length, uoffset, align,
1797 		    topdown, tmp)) {
1798 		case 1:
1799 			entry = tmp;
1800 			goto found;
1801 		case -1:
1802 			goto wraparound;
1803 		}
1804 		if (tmp->ownspace >= length)
1805 			goto listsearch;
1806 	}
1807 	if (prev == NULL)
1808 		goto notfound;
1809 
1810 	if (topdown) {
1811 		KASSERT(orig_hint >= prev->next->start - length ||
1812 		    prev->next->start - length > prev->next->start);
1813 		hint = prev->next->start - length;
1814 	} else {
1815 		KASSERT(orig_hint <= prev->end);
1816 		hint = prev->end;
1817 	}
1818 	switch (uvm_map_space_avail(&hint, length, uoffset, align,
1819 	    topdown, prev)) {
1820 	case 1:
1821 		entry = prev;
1822 		goto found;
1823 	case -1:
1824 		goto wraparound;
1825 	}
1826 	if (prev->ownspace >= length)
1827 		goto listsearch;
1828 
1829 	if (topdown)
1830 		tmp = RB_LEFT(prev, rb_entry);
1831 	else
1832 		tmp = RB_RIGHT(prev, rb_entry);
1833 	for (;;) {
1834 		KASSERT(tmp && tmp->space >= length);
1835 		if (topdown)
1836 			child = RB_RIGHT(tmp, rb_entry);
1837 		else
1838 			child = RB_LEFT(tmp, rb_entry);
1839 		if (child && child->space >= length) {
1840 			tmp = child;
1841 			continue;
1842 		}
1843 		if (tmp->ownspace >= length)
1844 			break;
1845 		if (topdown)
1846 			tmp = RB_LEFT(tmp, rb_entry);
1847 		else
1848 			tmp = RB_RIGHT(tmp, rb_entry);
1849 	}
1850 
1851 	if (topdown) {
1852 		KASSERT(orig_hint >= tmp->next->start - length ||
1853 		    tmp->next->start - length > tmp->next->start);
1854 		hint = tmp->next->start - length;
1855 	} else {
1856 		KASSERT(orig_hint <= tmp->end);
1857 		hint = tmp->end;
1858 	}
1859 	switch (uvm_map_space_avail(&hint, length, uoffset, align,
1860 	    topdown, tmp)) {
1861 	case 1:
1862 		entry = tmp;
1863 		goto found;
1864 	case -1:
1865 		goto wraparound;
1866 	}
1867 
1868 	/*
1869 	 * The tree fails to find an entry because of offset or alignment
1870 	 * restrictions.  Search the list instead.
1871 	 */
1872  listsearch:
1873 	/*
1874 	 * Look through the rest of the map, trying to fit a new region in
1875 	 * the gap between existing regions, or after the very last region.
1876 	 * note: entry->end = base VA of current gap,
1877 	 *	 entry->next->start = VA of end of current gap
1878 	 */
1879 
1880 	for (;;) {
1881 		/* Update hint for current gap. */
1882 		hint = topdown ? entry->next->start - length : entry->end;
1883 
1884 		/* See if it fits. */
1885 		switch (uvm_map_space_avail(&hint, length, uoffset, align,
1886 		    topdown, entry)) {
1887 		case 1:
1888 			goto found;
1889 		case -1:
1890 			goto wraparound;
1891 		}
1892 
1893 		/* Advance to next/previous gap */
1894 		if (topdown) {
1895 			if (entry == &map->header) {
1896 				UVMHIST_LOG(maphist, "<- failed (off start)",
1897 				    0,0,0,0);
1898 				goto notfound;
1899 			}
1900 			entry = entry->prev;
1901 		} else {
1902 			entry = entry->next;
1903 			if (entry == &map->header) {
1904 				UVMHIST_LOG(maphist, "<- failed (off end)",
1905 				    0,0,0,0);
1906 				goto notfound;
1907 			}
1908 		}
1909 	}
1910 
1911  found:
1912 	SAVE_HINT(map, map->hint, entry);
1913 	*result = hint;
1914 	UVMHIST_LOG(maphist,"<- got it!  (result=0x%x)", hint, 0,0,0);
1915 	KASSERT( topdown || hint >= orig_hint);
1916 	KASSERT(!topdown || hint <= orig_hint);
1917 	KASSERT(entry->end <= hint);
1918 	KASSERT(hint + length <= entry->next->start);
1919 	return (entry);
1920 
1921  wraparound:
1922 	UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0);
1923 
1924 	return (NULL);
1925 
1926  notfound:
1927 	UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0);
1928 
1929 	return (NULL);
1930 }
1931 
1932 /*
1933  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
1934  */
1935 
1936 /*
1937  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
1938  *
1939  * => caller must check alignment and size
1940  * => map must be locked by caller
1941  * => we return a list of map entries that we've remove from the map
1942  *    in "entry_list"
1943  */
1944 
1945 void
1946 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end,
1947     struct vm_map_entry **entry_list /* OUT */,
1948     struct uvm_mapent_reservation *umr, int flags)
1949 {
1950 	struct vm_map_entry *entry, *first_entry, *next;
1951 	vaddr_t len;
1952 	UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist);
1953 
1954 	UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
1955 	    map, start, end, 0);
1956 	VM_MAP_RANGE_CHECK(map, start, end);
1957 
1958 	uvm_map_check(map, "unmap_remove entry");
1959 
1960 	/*
1961 	 * find first entry
1962 	 */
1963 
1964 	if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
1965 		/* clip and go... */
1966 		entry = first_entry;
1967 		UVM_MAP_CLIP_START(map, entry, start, umr);
1968 		/* critical!  prevents stale hint */
1969 		SAVE_HINT(map, entry, entry->prev);
1970 	} else {
1971 		entry = first_entry->next;
1972 	}
1973 
1974 	/*
1975 	 * Save the free space hint
1976 	 */
1977 
1978 	if (map->first_free != &map->header && map->first_free->start >= start)
1979 		map->first_free = entry->prev;
1980 
1981 	/*
1982 	 * note: we now re-use first_entry for a different task.  we remove
1983 	 * a number of map entries from the map and save them in a linked
1984 	 * list headed by "first_entry".  once we remove them from the map
1985 	 * the caller should unlock the map and drop the references to the
1986 	 * backing objects [c.f. uvm_unmap_detach].  the object is to
1987 	 * separate unmapping from reference dropping.  why?
1988 	 *   [1] the map has to be locked for unmapping
1989 	 *   [2] the map need not be locked for reference dropping
1990 	 *   [3] dropping references may trigger pager I/O, and if we hit
1991 	 *       a pager that does synchronous I/O we may have to wait for it.
1992 	 *   [4] we would like all waiting for I/O to occur with maps unlocked
1993 	 *       so that we don't block other threads.
1994 	 */
1995 
1996 	first_entry = NULL;
1997 	*entry_list = NULL;
1998 
1999 	/*
2000 	 * break up the area into map entry sized regions and unmap.  note
2001 	 * that all mappings have to be removed before we can even consider
2002 	 * dropping references to amaps or VM objects (otherwise we could end
2003 	 * up with a mapping to a page on the free list which would be very bad)
2004 	 */
2005 
2006 	while ((entry != &map->header) && (entry->start < end)) {
2007 		KASSERT((entry->flags & UVM_MAP_FIRST) == 0);
2008 
2009 		UVM_MAP_CLIP_END(map, entry, end, umr);
2010 		next = entry->next;
2011 		len = entry->end - entry->start;
2012 
2013 		/*
2014 		 * unwire before removing addresses from the pmap; otherwise
2015 		 * unwiring will put the entries back into the pmap (XXX).
2016 		 */
2017 
2018 		if (VM_MAPENT_ISWIRED(entry)) {
2019 			uvm_map_entry_unwire(map, entry);
2020 		}
2021 		if (flags & UVM_FLAG_VAONLY) {
2022 
2023 			/* nothing */
2024 
2025 		} else if ((map->flags & VM_MAP_PAGEABLE) == 0) {
2026 
2027 			/*
2028 			 * if the map is non-pageable, any pages mapped there
2029 			 * must be wired and entered with pmap_kenter_pa(),
2030 			 * and we should free any such pages immediately.
2031 			 * this is mostly used for kmem_map and mb_map.
2032 			 */
2033 
2034 			if ((entry->flags & UVM_MAP_KMAPENT) == 0) {
2035 				uvm_km_pgremove_intrsafe(entry->start,
2036 				    entry->end);
2037 				pmap_kremove(entry->start, len);
2038 			}
2039 		} else if (UVM_ET_ISOBJ(entry) &&
2040 			   UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
2041 			KASSERT(vm_map_pmap(map) == pmap_kernel());
2042 
2043 			/*
2044 			 * note: kernel object mappings are currently used in
2045 			 * two ways:
2046 			 *  [1] "normal" mappings of pages in the kernel object
2047 			 *  [2] uvm_km_valloc'd allocations in which we
2048 			 *      pmap_enter in some non-kernel-object page
2049 			 *      (e.g. vmapbuf).
2050 			 *
2051 			 * for case [1], we need to remove the mapping from
2052 			 * the pmap and then remove the page from the kernel
2053 			 * object (because, once pages in a kernel object are
2054 			 * unmapped they are no longer needed, unlike, say,
2055 			 * a vnode where you might want the data to persist
2056 			 * until flushed out of a queue).
2057 			 *
2058 			 * for case [2], we need to remove the mapping from
2059 			 * the pmap.  there shouldn't be any pages at the
2060 			 * specified offset in the kernel object [but it
2061 			 * doesn't hurt to call uvm_km_pgremove just to be
2062 			 * safe?]
2063 			 *
2064 			 * uvm_km_pgremove currently does the following:
2065 			 *   for pages in the kernel object in range:
2066 			 *     - drops the swap slot
2067 			 *     - uvm_pagefree the page
2068 			 */
2069 
2070 			/*
2071 			 * remove mappings from pmap and drop the pages
2072 			 * from the object.  offsets are always relative
2073 			 * to vm_map_min(kernel_map).
2074 			 */
2075 
2076 			pmap_remove(pmap_kernel(), entry->start,
2077 			    entry->start + len);
2078 			uvm_km_pgremove(entry->start, entry->end);
2079 
2080 			/*
2081 			 * null out kernel_object reference, we've just
2082 			 * dropped it
2083 			 */
2084 
2085 			entry->etype &= ~UVM_ET_OBJ;
2086 			entry->object.uvm_obj = NULL;
2087 		} else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) {
2088 
2089 			/*
2090 			 * remove mappings the standard way.
2091 			 */
2092 
2093 			pmap_remove(map->pmap, entry->start, entry->end);
2094 		}
2095 
2096 #if defined(DEBUG)
2097 		if ((entry->flags & UVM_MAP_KMAPENT) == 0) {
2098 
2099 			/*
2100 			 * check if there's remaining mapping,
2101 			 * which is a bug in caller.
2102 			 */
2103 
2104 			vaddr_t va;
2105 			for (va = entry->start; va < entry->end;
2106 			    va += PAGE_SIZE) {
2107 				if (pmap_extract(vm_map_pmap(map), va, NULL)) {
2108 					panic("uvm_unmap_remove: has mapping");
2109 				}
2110 			}
2111 
2112 			if (VM_MAP_IS_KERNEL(map)) {
2113 				uvm_km_check_empty(entry->start, entry->end,
2114 				    (map->flags & VM_MAP_INTRSAFE) != 0);
2115 			}
2116 		}
2117 #endif /* defined(DEBUG) */
2118 
2119 		/*
2120 		 * remove entry from map and put it on our list of entries
2121 		 * that we've nuked.  then go to next entry.
2122 		 */
2123 
2124 		UVMHIST_LOG(maphist, "  removed map entry 0x%x", entry, 0, 0,0);
2125 
2126 		/* critical!  prevents stale hint */
2127 		SAVE_HINT(map, entry, entry->prev);
2128 
2129 		uvm_map_entry_unlink(map, entry);
2130 		KASSERT(map->size >= len);
2131 		map->size -= len;
2132 		entry->prev = NULL;
2133 		entry->next = first_entry;
2134 		first_entry = entry;
2135 		entry = next;
2136 	}
2137 	if ((map->flags & VM_MAP_DYING) == 0) {
2138 		pmap_update(vm_map_pmap(map));
2139 	}
2140 
2141 	uvm_map_check(map, "unmap_remove leave");
2142 
2143 	/*
2144 	 * now we've cleaned up the map and are ready for the caller to drop
2145 	 * references to the mapped objects.
2146 	 */
2147 
2148 	*entry_list = first_entry;
2149 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
2150 
2151 	simple_lock(&map->flags_lock);
2152 	if (map->flags & VM_MAP_WANTVA) {
2153 		map->flags &= ~VM_MAP_WANTVA;
2154 		wakeup(&map->header);
2155 	}
2156 	simple_unlock(&map->flags_lock);
2157 }
2158 
2159 /*
2160  * uvm_unmap_detach: drop references in a chain of map entries
2161  *
2162  * => we will free the map entries as we traverse the list.
2163  */
2164 
2165 void
2166 uvm_unmap_detach(struct vm_map_entry *first_entry, int flags)
2167 {
2168 	struct vm_map_entry *next_entry;
2169 	UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
2170 
2171 	while (first_entry) {
2172 		KASSERT(!VM_MAPENT_ISWIRED(first_entry));
2173 		UVMHIST_LOG(maphist,
2174 		    "  detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
2175 		    first_entry, first_entry->aref.ar_amap,
2176 		    first_entry->object.uvm_obj,
2177 		    UVM_ET_ISSUBMAP(first_entry));
2178 
2179 		/*
2180 		 * drop reference to amap, if we've got one
2181 		 */
2182 
2183 		if (first_entry->aref.ar_amap)
2184 			uvm_map_unreference_amap(first_entry, flags);
2185 
2186 		/*
2187 		 * drop reference to our backing object, if we've got one
2188 		 */
2189 
2190 		KASSERT(!UVM_ET_ISSUBMAP(first_entry));
2191 		if (UVM_ET_ISOBJ(first_entry) &&
2192 		    first_entry->object.uvm_obj->pgops->pgo_detach) {
2193 			(*first_entry->object.uvm_obj->pgops->pgo_detach)
2194 				(first_entry->object.uvm_obj);
2195 		}
2196 		next_entry = first_entry->next;
2197 		uvm_mapent_free(first_entry);
2198 		first_entry = next_entry;
2199 	}
2200 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
2201 }
2202 
2203 /*
2204  *   E X T R A C T I O N   F U N C T I O N S
2205  */
2206 
2207 /*
2208  * uvm_map_reserve: reserve space in a vm_map for future use.
2209  *
2210  * => we reserve space in a map by putting a dummy map entry in the
2211  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
2212  * => map should be unlocked (we will write lock it)
2213  * => we return true if we were able to reserve space
2214  * => XXXCDC: should be inline?
2215  */
2216 
2217 int
2218 uvm_map_reserve(struct vm_map *map, vsize_t size,
2219     vaddr_t offset	/* hint for pmap_prefer */,
2220     vsize_t align	/* alignment hint */,
2221     vaddr_t *raddr	/* IN:hint, OUT: reserved VA */,
2222     uvm_flag_t flags	/* UVM_FLAG_FIXED or 0 */)
2223 {
2224 	UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
2225 
2226 	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
2227 	    map,size,offset,raddr);
2228 
2229 	size = round_page(size);
2230 
2231 	/*
2232 	 * reserve some virtual space.
2233 	 */
2234 
2235 	if (uvm_map(map, raddr, size, NULL, offset, 0,
2236 	    UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
2237 	    UVM_ADV_RANDOM, UVM_FLAG_NOMERGE|flags)) != 0) {
2238 	    UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
2239 		return (FALSE);
2240 	}
2241 
2242 	UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
2243 	return (TRUE);
2244 }
2245 
2246 /*
2247  * uvm_map_replace: replace a reserved (blank) area of memory with
2248  * real mappings.
2249  *
2250  * => caller must WRITE-LOCK the map
2251  * => we return TRUE if replacement was a success
2252  * => we expect the newents chain to have nnewents entrys on it and
2253  *    we expect newents->prev to point to the last entry on the list
2254  * => note newents is allowed to be NULL
2255  */
2256 
2257 int
2258 uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end,
2259     struct vm_map_entry *newents, int nnewents)
2260 {
2261 	struct vm_map_entry *oldent, *last;
2262 
2263 	uvm_map_check(map, "map_replace entry");
2264 
2265 	/*
2266 	 * first find the blank map entry at the specified address
2267 	 */
2268 
2269 	if (!uvm_map_lookup_entry(map, start, &oldent)) {
2270 		return (FALSE);
2271 	}
2272 
2273 	/*
2274 	 * check to make sure we have a proper blank entry
2275 	 */
2276 
2277 	if (end < oldent->end && !VM_MAP_USE_KMAPENT(map)) {
2278 		UVM_MAP_CLIP_END(map, oldent, end, NULL);
2279 	}
2280 	if (oldent->start != start || oldent->end != end ||
2281 	    oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
2282 		return (FALSE);
2283 	}
2284 
2285 #ifdef DIAGNOSTIC
2286 
2287 	/*
2288 	 * sanity check the newents chain
2289 	 */
2290 
2291 	{
2292 		struct vm_map_entry *tmpent = newents;
2293 		int nent = 0;
2294 		vaddr_t cur = start;
2295 
2296 		while (tmpent) {
2297 			nent++;
2298 			if (tmpent->start < cur)
2299 				panic("uvm_map_replace1");
2300 			if (tmpent->start > tmpent->end || tmpent->end > end) {
2301 		printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
2302 			    tmpent->start, tmpent->end, end);
2303 				panic("uvm_map_replace2");
2304 			}
2305 			cur = tmpent->end;
2306 			if (tmpent->next) {
2307 				if (tmpent->next->prev != tmpent)
2308 					panic("uvm_map_replace3");
2309 			} else {
2310 				if (newents->prev != tmpent)
2311 					panic("uvm_map_replace4");
2312 			}
2313 			tmpent = tmpent->next;
2314 		}
2315 		if (nent != nnewents)
2316 			panic("uvm_map_replace5");
2317 	}
2318 #endif
2319 
2320 	/*
2321 	 * map entry is a valid blank!   replace it.   (this does all the
2322 	 * work of map entry link/unlink...).
2323 	 */
2324 
2325 	if (newents) {
2326 		last = newents->prev;
2327 
2328 		/* critical: flush stale hints out of map */
2329 		SAVE_HINT(map, map->hint, newents);
2330 		if (map->first_free == oldent)
2331 			map->first_free = last;
2332 
2333 		last->next = oldent->next;
2334 		last->next->prev = last;
2335 
2336 		/* Fix RB tree */
2337 		uvm_rb_remove(map, oldent);
2338 
2339 		newents->prev = oldent->prev;
2340 		newents->prev->next = newents;
2341 		map->nentries = map->nentries + (nnewents - 1);
2342 
2343 		/* Fixup the RB tree */
2344 		{
2345 			int i;
2346 			struct vm_map_entry *tmp;
2347 
2348 			tmp = newents;
2349 			for (i = 0; i < nnewents && tmp; i++) {
2350 				uvm_rb_insert(map, tmp);
2351 				tmp = tmp->next;
2352 			}
2353 		}
2354 	} else {
2355 		/* NULL list of new entries: just remove the old one */
2356 		clear_hints(map, oldent);
2357 		uvm_map_entry_unlink(map, oldent);
2358 	}
2359 
2360 	uvm_map_check(map, "map_replace leave");
2361 
2362 	/*
2363 	 * now we can free the old blank entry and return.
2364 	 */
2365 
2366 	uvm_mapent_free(oldent);
2367 	return (TRUE);
2368 }
2369 
2370 /*
2371  * uvm_map_extract: extract a mapping from a map and put it somewhere
2372  *	(maybe removing the old mapping)
2373  *
2374  * => maps should be unlocked (we will write lock them)
2375  * => returns 0 on success, error code otherwise
2376  * => start must be page aligned
2377  * => len must be page sized
2378  * => flags:
2379  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
2380  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
2381  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
2382  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
2383  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
2384  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
2385  *             be used from within the kernel in a kernel level map <<<
2386  */
2387 
2388 int
2389 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len,
2390     struct vm_map *dstmap, vaddr_t *dstaddrp, int flags)
2391 {
2392 	vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge;
2393 	struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry,
2394 	    *deadentry, *oldentry;
2395 	vsize_t elen;
2396 	int nchain, error, copy_ok;
2397 	UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
2398 
2399 	UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
2400 	    len,0);
2401 	UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
2402 
2403 	uvm_map_check(srcmap, "map_extract src enter");
2404 	uvm_map_check(dstmap, "map_extract dst enter");
2405 
2406 	/*
2407 	 * step 0: sanity check: start must be on a page boundary, length
2408 	 * must be page sized.  can't ask for CONTIG/QREF if you asked for
2409 	 * REMOVE.
2410 	 */
2411 
2412 	KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0);
2413 	KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 ||
2414 		(flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0);
2415 
2416 	/*
2417 	 * step 1: reserve space in the target map for the extracted area
2418 	 */
2419 
2420 	if ((flags & UVM_EXTRACT_RESERVED) == 0) {
2421 		dstaddr = vm_map_min(dstmap);
2422 		if (!uvm_map_reserve(dstmap, len, start, 0, &dstaddr, 0))
2423 			return (ENOMEM);
2424 		*dstaddrp = dstaddr;	/* pass address back to caller */
2425 		UVMHIST_LOG(maphist, "  dstaddr=0x%x", dstaddr,0,0,0);
2426 	} else {
2427 		dstaddr = *dstaddrp;
2428 	}
2429 
2430 	/*
2431 	 * step 2: setup for the extraction process loop by init'ing the
2432 	 * map entry chain, locking src map, and looking up the first useful
2433 	 * entry in the map.
2434 	 */
2435 
2436 	end = start + len;
2437 	newend = dstaddr + len;
2438 	chain = endchain = NULL;
2439 	nchain = 0;
2440 	vm_map_lock(srcmap);
2441 
2442 	if (uvm_map_lookup_entry(srcmap, start, &entry)) {
2443 
2444 		/* "start" is within an entry */
2445 		if (flags & UVM_EXTRACT_QREF) {
2446 
2447 			/*
2448 			 * for quick references we don't clip the entry, so
2449 			 * the entry may map space "before" the starting
2450 			 * virtual address... this is the "fudge" factor
2451 			 * (which can be non-zero only the first time
2452 			 * through the "while" loop in step 3).
2453 			 */
2454 
2455 			fudge = start - entry->start;
2456 		} else {
2457 
2458 			/*
2459 			 * normal reference: we clip the map to fit (thus
2460 			 * fudge is zero)
2461 			 */
2462 
2463 			UVM_MAP_CLIP_START(srcmap, entry, start, NULL);
2464 			SAVE_HINT(srcmap, srcmap->hint, entry->prev);
2465 			fudge = 0;
2466 		}
2467 	} else {
2468 
2469 		/* "start" is not within an entry ... skip to next entry */
2470 		if (flags & UVM_EXTRACT_CONTIG) {
2471 			error = EINVAL;
2472 			goto bad;    /* definite hole here ... */
2473 		}
2474 
2475 		entry = entry->next;
2476 		fudge = 0;
2477 	}
2478 
2479 	/* save values from srcmap for step 6 */
2480 	orig_entry = entry;
2481 	orig_fudge = fudge;
2482 
2483 	/*
2484 	 * step 3: now start looping through the map entries, extracting
2485 	 * as we go.
2486 	 */
2487 
2488 	while (entry->start < end && entry != &srcmap->header) {
2489 
2490 		/* if we are not doing a quick reference, clip it */
2491 		if ((flags & UVM_EXTRACT_QREF) == 0)
2492 			UVM_MAP_CLIP_END(srcmap, entry, end, NULL);
2493 
2494 		/* clear needs_copy (allow chunking) */
2495 		if (UVM_ET_ISNEEDSCOPY(entry)) {
2496 			amap_copy(srcmap, entry,
2497 			    AMAP_COPY_NOWAIT|AMAP_COPY_NOMERGE, start, end);
2498 			if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
2499 				error = ENOMEM;
2500 				goto bad;
2501 			}
2502 
2503 			/* amap_copy could clip (during chunk)!  update fudge */
2504 			if (fudge) {
2505 				fudge = start - entry->start;
2506 				orig_fudge = fudge;
2507 			}
2508 		}
2509 
2510 		/* calculate the offset of this from "start" */
2511 		oldoffset = (entry->start + fudge) - start;
2512 
2513 		/* allocate a new map entry */
2514 		newentry = uvm_mapent_alloc(dstmap, 0);
2515 		if (newentry == NULL) {
2516 			error = ENOMEM;
2517 			goto bad;
2518 		}
2519 
2520 		/* set up new map entry */
2521 		newentry->next = NULL;
2522 		newentry->prev = endchain;
2523 		newentry->start = dstaddr + oldoffset;
2524 		newentry->end =
2525 		    newentry->start + (entry->end - (entry->start + fudge));
2526 		if (newentry->end > newend || newentry->end < newentry->start)
2527 			newentry->end = newend;
2528 		newentry->object.uvm_obj = entry->object.uvm_obj;
2529 		if (newentry->object.uvm_obj) {
2530 			if (newentry->object.uvm_obj->pgops->pgo_reference)
2531 				newentry->object.uvm_obj->pgops->
2532 				    pgo_reference(newentry->object.uvm_obj);
2533 				newentry->offset = entry->offset + fudge;
2534 		} else {
2535 			newentry->offset = 0;
2536 		}
2537 		newentry->etype = entry->etype;
2538 		newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
2539 			entry->max_protection : entry->protection;
2540 		newentry->max_protection = entry->max_protection;
2541 		newentry->inheritance = entry->inheritance;
2542 		newentry->wired_count = 0;
2543 		newentry->aref.ar_amap = entry->aref.ar_amap;
2544 		if (newentry->aref.ar_amap) {
2545 			newentry->aref.ar_pageoff =
2546 			    entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
2547 			uvm_map_reference_amap(newentry, AMAP_SHARED |
2548 			    ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
2549 		} else {
2550 			newentry->aref.ar_pageoff = 0;
2551 		}
2552 		newentry->advice = entry->advice;
2553 
2554 		/* now link it on the chain */
2555 		nchain++;
2556 		if (endchain == NULL) {
2557 			chain = endchain = newentry;
2558 		} else {
2559 			endchain->next = newentry;
2560 			endchain = newentry;
2561 		}
2562 
2563 		/* end of 'while' loop! */
2564 		if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
2565 		    (entry->next == &srcmap->header ||
2566 		    entry->next->start != entry->end)) {
2567 			error = EINVAL;
2568 			goto bad;
2569 		}
2570 		entry = entry->next;
2571 		fudge = 0;
2572 	}
2573 
2574 	/*
2575 	 * step 4: close off chain (in format expected by uvm_map_replace)
2576 	 */
2577 
2578 	if (chain)
2579 		chain->prev = endchain;
2580 
2581 	/*
2582 	 * step 5: attempt to lock the dest map so we can pmap_copy.
2583 	 * note usage of copy_ok:
2584 	 *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
2585 	 *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
2586 	 */
2587 
2588 	if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
2589 		copy_ok = 1;
2590 		if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
2591 		    nchain)) {
2592 			if (srcmap != dstmap)
2593 				vm_map_unlock(dstmap);
2594 			error = EIO;
2595 			goto bad;
2596 		}
2597 	} else {
2598 		copy_ok = 0;
2599 		/* replace defered until step 7 */
2600 	}
2601 
2602 	/*
2603 	 * step 6: traverse the srcmap a second time to do the following:
2604 	 *  - if we got a lock on the dstmap do pmap_copy
2605 	 *  - if UVM_EXTRACT_REMOVE remove the entries
2606 	 * we make use of orig_entry and orig_fudge (saved in step 2)
2607 	 */
2608 
2609 	if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
2610 
2611 		/* purge possible stale hints from srcmap */
2612 		if (flags & UVM_EXTRACT_REMOVE) {
2613 			SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev);
2614 			if (srcmap->first_free != &srcmap->header &&
2615 			    srcmap->first_free->start >= start)
2616 				srcmap->first_free = orig_entry->prev;
2617 		}
2618 
2619 		entry = orig_entry;
2620 		fudge = orig_fudge;
2621 		deadentry = NULL;	/* for UVM_EXTRACT_REMOVE */
2622 
2623 		while (entry->start < end && entry != &srcmap->header) {
2624 			if (copy_ok) {
2625 				oldoffset = (entry->start + fudge) - start;
2626 				elen = MIN(end, entry->end) -
2627 				    (entry->start + fudge);
2628 				pmap_copy(dstmap->pmap, srcmap->pmap,
2629 				    dstaddr + oldoffset, elen,
2630 				    entry->start + fudge);
2631 			}
2632 
2633 			/* we advance "entry" in the following if statement */
2634 			if (flags & UVM_EXTRACT_REMOVE) {
2635 				pmap_remove(srcmap->pmap, entry->start,
2636 						entry->end);
2637 				oldentry = entry;	/* save entry */
2638 				entry = entry->next;	/* advance */
2639 				uvm_map_entry_unlink(srcmap, oldentry);
2640 							/* add to dead list */
2641 				oldentry->next = deadentry;
2642 				deadentry = oldentry;
2643 			} else {
2644 				entry = entry->next;		/* advance */
2645 			}
2646 
2647 			/* end of 'while' loop */
2648 			fudge = 0;
2649 		}
2650 		pmap_update(srcmap->pmap);
2651 
2652 		/*
2653 		 * unlock dstmap.  we will dispose of deadentry in
2654 		 * step 7 if needed
2655 		 */
2656 
2657 		if (copy_ok && srcmap != dstmap)
2658 			vm_map_unlock(dstmap);
2659 
2660 	} else {
2661 		deadentry = NULL;
2662 	}
2663 
2664 	/*
2665 	 * step 7: we are done with the source map, unlock.   if copy_ok
2666 	 * is 0 then we have not replaced the dummy mapping in dstmap yet
2667 	 * and we need to do so now.
2668 	 */
2669 
2670 	vm_map_unlock(srcmap);
2671 	if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
2672 		uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
2673 
2674 	/* now do the replacement if we didn't do it in step 5 */
2675 	if (copy_ok == 0) {
2676 		vm_map_lock(dstmap);
2677 		error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
2678 		    nchain);
2679 		vm_map_unlock(dstmap);
2680 
2681 		if (error == FALSE) {
2682 			error = EIO;
2683 			goto bad2;
2684 		}
2685 	}
2686 
2687 	uvm_map_check(srcmap, "map_extract src leave");
2688 	uvm_map_check(dstmap, "map_extract dst leave");
2689 
2690 	return (0);
2691 
2692 	/*
2693 	 * bad: failure recovery
2694 	 */
2695 bad:
2696 	vm_map_unlock(srcmap);
2697 bad2:			/* src already unlocked */
2698 	if (chain)
2699 		uvm_unmap_detach(chain,
2700 		    (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
2701 
2702 	uvm_map_check(srcmap, "map_extract src err leave");
2703 	uvm_map_check(dstmap, "map_extract dst err leave");
2704 
2705 	if ((flags & UVM_EXTRACT_RESERVED) == 0) {
2706 		uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
2707 	}
2708 	return (error);
2709 }
2710 
2711 /* end of extraction functions */
2712 
2713 /*
2714  * uvm_map_submap: punch down part of a map into a submap
2715  *
2716  * => only the kernel_map is allowed to be submapped
2717  * => the purpose of submapping is to break up the locking granularity
2718  *	of a larger map
2719  * => the range specified must have been mapped previously with a uvm_map()
2720  *	call [with uobj==NULL] to create a blank map entry in the main map.
2721  *	[And it had better still be blank!]
2722  * => maps which contain submaps should never be copied or forked.
2723  * => to remove a submap, use uvm_unmap() on the main map
2724  *	and then uvm_map_deallocate() the submap.
2725  * => main map must be unlocked.
2726  * => submap must have been init'd and have a zero reference count.
2727  *	[need not be locked as we don't actually reference it]
2728  */
2729 
2730 int
2731 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end,
2732     struct vm_map *submap)
2733 {
2734 	struct vm_map_entry *entry;
2735 	struct uvm_mapent_reservation umr;
2736 	int error;
2737 
2738 	uvm_mapent_reserve(map, &umr, 2, 0);
2739 
2740 	vm_map_lock(map);
2741 	VM_MAP_RANGE_CHECK(map, start, end);
2742 
2743 	if (uvm_map_lookup_entry(map, start, &entry)) {
2744 		UVM_MAP_CLIP_START(map, entry, start, &umr);
2745 		UVM_MAP_CLIP_END(map, entry, end, &umr);	/* to be safe */
2746 	} else {
2747 		entry = NULL;
2748 	}
2749 
2750 	if (entry != NULL &&
2751 	    entry->start == start && entry->end == end &&
2752 	    entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
2753 	    !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
2754 		entry->etype |= UVM_ET_SUBMAP;
2755 		entry->object.sub_map = submap;
2756 		entry->offset = 0;
2757 		uvm_map_reference(submap);
2758 		error = 0;
2759 	} else {
2760 		error = EINVAL;
2761 	}
2762 	vm_map_unlock(map);
2763 
2764 	uvm_mapent_unreserve(map, &umr);
2765 
2766 	return error;
2767 }
2768 
2769 /*
2770  * uvm_map_setup_kernel: init in-kernel map
2771  *
2772  * => map must not be in service yet.
2773  */
2774 
2775 void
2776 uvm_map_setup_kernel(struct vm_map_kernel *map,
2777     vaddr_t vmin, vaddr_t vmax, int flags)
2778 {
2779 
2780 	uvm_map_setup(&map->vmk_map, vmin, vmax, flags);
2781 
2782 	callback_head_init(&map->vmk_reclaim_callback);
2783 	LIST_INIT(&map->vmk_kentry_free);
2784 	map->vmk_merged_entries = NULL;
2785 }
2786 
2787 
2788 /*
2789  * uvm_map_protect: change map protection
2790  *
2791  * => set_max means set max_protection.
2792  * => map must be unlocked.
2793  */
2794 
2795 #define MASK(entry)	(UVM_ET_ISCOPYONWRITE(entry) ? \
2796 			 ~VM_PROT_WRITE : VM_PROT_ALL)
2797 
2798 int
2799 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end,
2800     vm_prot_t new_prot, boolean_t set_max)
2801 {
2802 	struct vm_map_entry *current, *entry;
2803 	int error = 0;
2804 	UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
2805 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
2806 		    map, start, end, new_prot);
2807 
2808 	vm_map_lock(map);
2809 	VM_MAP_RANGE_CHECK(map, start, end);
2810 	if (uvm_map_lookup_entry(map, start, &entry)) {
2811 		UVM_MAP_CLIP_START(map, entry, start, NULL);
2812 	} else {
2813 		entry = entry->next;
2814 	}
2815 
2816 	/*
2817 	 * make a first pass to check for protection violations.
2818 	 */
2819 
2820 	current = entry;
2821 	while ((current != &map->header) && (current->start < end)) {
2822 		if (UVM_ET_ISSUBMAP(current)) {
2823 			error = EINVAL;
2824 			goto out;
2825 		}
2826 		if ((new_prot & current->max_protection) != new_prot) {
2827 			error = EACCES;
2828 			goto out;
2829 		}
2830 		/*
2831 		 * Don't allow VM_PROT_EXECUTE to be set on entries that
2832 		 * point to vnodes that are associated with a NOEXEC file
2833 		 * system.
2834 		 */
2835 		if (UVM_ET_ISOBJ(current) &&
2836 		    UVM_OBJ_IS_VNODE(current->object.uvm_obj)) {
2837 			struct vnode *vp =
2838 			    (struct vnode *) current->object.uvm_obj;
2839 
2840 			if ((new_prot & VM_PROT_EXECUTE) != 0 &&
2841 			    (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
2842 				error = EACCES;
2843 				goto out;
2844 			}
2845 		}
2846 
2847 		current = current->next;
2848 	}
2849 
2850 	/* go back and fix up protections (no need to clip this time). */
2851 
2852 	current = entry;
2853 	while ((current != &map->header) && (current->start < end)) {
2854 		vm_prot_t old_prot;
2855 
2856 		UVM_MAP_CLIP_END(map, current, end, NULL);
2857 		old_prot = current->protection;
2858 		if (set_max)
2859 			current->protection =
2860 			    (current->max_protection = new_prot) & old_prot;
2861 		else
2862 			current->protection = new_prot;
2863 
2864 		/*
2865 		 * update physical map if necessary.  worry about copy-on-write
2866 		 * here -- CHECK THIS XXX
2867 		 */
2868 
2869 		if (current->protection != old_prot) {
2870 			/* update pmap! */
2871 			pmap_protect(map->pmap, current->start, current->end,
2872 			    current->protection & MASK(entry));
2873 
2874 			/*
2875 			 * If this entry points at a vnode, and the
2876 			 * protection includes VM_PROT_EXECUTE, mark
2877 			 * the vnode as VEXECMAP.
2878 			 */
2879 			if (UVM_ET_ISOBJ(current)) {
2880 				struct uvm_object *uobj =
2881 				    current->object.uvm_obj;
2882 
2883 				if (UVM_OBJ_IS_VNODE(uobj) &&
2884 				    (current->protection & VM_PROT_EXECUTE))
2885 					vn_markexec((struct vnode *) uobj);
2886 			}
2887 		}
2888 
2889 		/*
2890 		 * If the map is configured to lock any future mappings,
2891 		 * wire this entry now if the old protection was VM_PROT_NONE
2892 		 * and the new protection is not VM_PROT_NONE.
2893 		 */
2894 
2895 		if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
2896 		    VM_MAPENT_ISWIRED(entry) == 0 &&
2897 		    old_prot == VM_PROT_NONE &&
2898 		    new_prot != VM_PROT_NONE) {
2899 			if (uvm_map_pageable(map, entry->start,
2900 			    entry->end, FALSE,
2901 			    UVM_LK_ENTER|UVM_LK_EXIT) != 0) {
2902 
2903 				/*
2904 				 * If locking the entry fails, remember the
2905 				 * error if it's the first one.  Note we
2906 				 * still continue setting the protection in
2907 				 * the map, but will return the error
2908 				 * condition regardless.
2909 				 *
2910 				 * XXX Ignore what the actual error is,
2911 				 * XXX just call it a resource shortage
2912 				 * XXX so that it doesn't get confused
2913 				 * XXX what uvm_map_protect() itself would
2914 				 * XXX normally return.
2915 				 */
2916 
2917 				error = ENOMEM;
2918 			}
2919 		}
2920 		current = current->next;
2921 	}
2922 	pmap_update(map->pmap);
2923 
2924  out:
2925 	vm_map_unlock(map);
2926 
2927 	UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0);
2928 	return error;
2929 }
2930 
2931 #undef  MASK
2932 
2933 /*
2934  * uvm_map_inherit: set inheritance code for range of addrs in map.
2935  *
2936  * => map must be unlocked
2937  * => note that the inherit code is used during a "fork".  see fork
2938  *	code for details.
2939  */
2940 
2941 int
2942 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end,
2943     vm_inherit_t new_inheritance)
2944 {
2945 	struct vm_map_entry *entry, *temp_entry;
2946 	UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
2947 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
2948 	    map, start, end, new_inheritance);
2949 
2950 	switch (new_inheritance) {
2951 	case MAP_INHERIT_NONE:
2952 	case MAP_INHERIT_COPY:
2953 	case MAP_INHERIT_SHARE:
2954 		break;
2955 	default:
2956 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2957 		return EINVAL;
2958 	}
2959 
2960 	vm_map_lock(map);
2961 	VM_MAP_RANGE_CHECK(map, start, end);
2962 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
2963 		entry = temp_entry;
2964 		UVM_MAP_CLIP_START(map, entry, start, NULL);
2965 	}  else {
2966 		entry = temp_entry->next;
2967 	}
2968 	while ((entry != &map->header) && (entry->start < end)) {
2969 		UVM_MAP_CLIP_END(map, entry, end, NULL);
2970 		entry->inheritance = new_inheritance;
2971 		entry = entry->next;
2972 	}
2973 	vm_map_unlock(map);
2974 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
2975 	return 0;
2976 }
2977 
2978 /*
2979  * uvm_map_advice: set advice code for range of addrs in map.
2980  *
2981  * => map must be unlocked
2982  */
2983 
2984 int
2985 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice)
2986 {
2987 	struct vm_map_entry *entry, *temp_entry;
2988 	UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
2989 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
2990 	    map, start, end, new_advice);
2991 
2992 	vm_map_lock(map);
2993 	VM_MAP_RANGE_CHECK(map, start, end);
2994 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
2995 		entry = temp_entry;
2996 		UVM_MAP_CLIP_START(map, entry, start, NULL);
2997 	} else {
2998 		entry = temp_entry->next;
2999 	}
3000 
3001 	/*
3002 	 * XXXJRT: disallow holes?
3003 	 */
3004 
3005 	while ((entry != &map->header) && (entry->start < end)) {
3006 		UVM_MAP_CLIP_END(map, entry, end, NULL);
3007 
3008 		switch (new_advice) {
3009 		case MADV_NORMAL:
3010 		case MADV_RANDOM:
3011 		case MADV_SEQUENTIAL:
3012 			/* nothing special here */
3013 			break;
3014 
3015 		default:
3016 			vm_map_unlock(map);
3017 			UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
3018 			return EINVAL;
3019 		}
3020 		entry->advice = new_advice;
3021 		entry = entry->next;
3022 	}
3023 
3024 	vm_map_unlock(map);
3025 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
3026 	return 0;
3027 }
3028 
3029 /*
3030  * uvm_map_pageable: sets the pageability of a range in a map.
3031  *
3032  * => wires map entries.  should not be used for transient page locking.
3033  *	for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
3034  * => regions specified as not pageable require lock-down (wired) memory
3035  *	and page tables.
3036  * => map must never be read-locked
3037  * => if islocked is TRUE, map is already write-locked
3038  * => we always unlock the map, since we must downgrade to a read-lock
3039  *	to call uvm_fault_wire()
3040  * => XXXCDC: check this and try and clean it up.
3041  */
3042 
3043 int
3044 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end,
3045     boolean_t new_pageable, int lockflags)
3046 {
3047 	struct vm_map_entry *entry, *start_entry, *failed_entry;
3048 	int rv;
3049 #ifdef DIAGNOSTIC
3050 	u_int timestamp_save;
3051 #endif
3052 	UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
3053 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
3054 		    map, start, end, new_pageable);
3055 	KASSERT(map->flags & VM_MAP_PAGEABLE);
3056 
3057 	if ((lockflags & UVM_LK_ENTER) == 0)
3058 		vm_map_lock(map);
3059 	VM_MAP_RANGE_CHECK(map, start, end);
3060 
3061 	/*
3062 	 * only one pageability change may take place at one time, since
3063 	 * uvm_fault_wire assumes it will be called only once for each
3064 	 * wiring/unwiring.  therefore, we have to make sure we're actually
3065 	 * changing the pageability for the entire region.  we do so before
3066 	 * making any changes.
3067 	 */
3068 
3069 	if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
3070 		if ((lockflags & UVM_LK_EXIT) == 0)
3071 			vm_map_unlock(map);
3072 
3073 		UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0);
3074 		return EFAULT;
3075 	}
3076 	entry = start_entry;
3077 
3078 	/*
3079 	 * handle wiring and unwiring separately.
3080 	 */
3081 
3082 	if (new_pageable) {		/* unwire */
3083 		UVM_MAP_CLIP_START(map, entry, start, NULL);
3084 
3085 		/*
3086 		 * unwiring.  first ensure that the range to be unwired is
3087 		 * really wired down and that there are no holes.
3088 		 */
3089 
3090 		while ((entry != &map->header) && (entry->start < end)) {
3091 			if (entry->wired_count == 0 ||
3092 			    (entry->end < end &&
3093 			     (entry->next == &map->header ||
3094 			      entry->next->start > entry->end))) {
3095 				if ((lockflags & UVM_LK_EXIT) == 0)
3096 					vm_map_unlock(map);
3097 				UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0);
3098 				return EINVAL;
3099 			}
3100 			entry = entry->next;
3101 		}
3102 
3103 		/*
3104 		 * POSIX 1003.1b - a single munlock call unlocks a region,
3105 		 * regardless of the number of mlock calls made on that
3106 		 * region.
3107 		 */
3108 
3109 		entry = start_entry;
3110 		while ((entry != &map->header) && (entry->start < end)) {
3111 			UVM_MAP_CLIP_END(map, entry, end, NULL);
3112 			if (VM_MAPENT_ISWIRED(entry))
3113 				uvm_map_entry_unwire(map, entry);
3114 			entry = entry->next;
3115 		}
3116 		if ((lockflags & UVM_LK_EXIT) == 0)
3117 			vm_map_unlock(map);
3118 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
3119 		return 0;
3120 	}
3121 
3122 	/*
3123 	 * wire case: in two passes [XXXCDC: ugly block of code here]
3124 	 *
3125 	 * 1: holding the write lock, we create any anonymous maps that need
3126 	 *    to be created.  then we clip each map entry to the region to
3127 	 *    be wired and increment its wiring count.
3128 	 *
3129 	 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
3130 	 *    in the pages for any newly wired area (wired_count == 1).
3131 	 *
3132 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
3133 	 *    deadlock with another thread that may have faulted on one of
3134 	 *    the pages to be wired (it would mark the page busy, blocking
3135 	 *    us, then in turn block on the map lock that we hold).  because
3136 	 *    of problems in the recursive lock package, we cannot upgrade
3137 	 *    to a write lock in vm_map_lookup.  thus, any actions that
3138 	 *    require the write lock must be done beforehand.  because we
3139 	 *    keep the read lock on the map, the copy-on-write status of the
3140 	 *    entries we modify here cannot change.
3141 	 */
3142 
3143 	while ((entry != &map->header) && (entry->start < end)) {
3144 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3145 
3146 			/*
3147 			 * perform actions of vm_map_lookup that need the
3148 			 * write lock on the map: create an anonymous map
3149 			 * for a copy-on-write region, or an anonymous map
3150 			 * for a zero-fill region.  (XXXCDC: submap case
3151 			 * ok?)
3152 			 */
3153 
3154 			if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
3155 				if (UVM_ET_ISNEEDSCOPY(entry) &&
3156 				    ((entry->max_protection & VM_PROT_WRITE) ||
3157 				     (entry->object.uvm_obj == NULL))) {
3158 					amap_copy(map, entry, 0, start, end);
3159 					/* XXXCDC: wait OK? */
3160 				}
3161 			}
3162 		}
3163 		UVM_MAP_CLIP_START(map, entry, start, NULL);
3164 		UVM_MAP_CLIP_END(map, entry, end, NULL);
3165 		entry->wired_count++;
3166 
3167 		/*
3168 		 * Check for holes
3169 		 */
3170 
3171 		if (entry->protection == VM_PROT_NONE ||
3172 		    (entry->end < end &&
3173 		     (entry->next == &map->header ||
3174 		      entry->next->start > entry->end))) {
3175 
3176 			/*
3177 			 * found one.  amap creation actions do not need to
3178 			 * be undone, but the wired counts need to be restored.
3179 			 */
3180 
3181 			while (entry != &map->header && entry->end > start) {
3182 				entry->wired_count--;
3183 				entry = entry->prev;
3184 			}
3185 			if ((lockflags & UVM_LK_EXIT) == 0)
3186 				vm_map_unlock(map);
3187 			UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
3188 			return EINVAL;
3189 		}
3190 		entry = entry->next;
3191 	}
3192 
3193 	/*
3194 	 * Pass 2.
3195 	 */
3196 
3197 #ifdef DIAGNOSTIC
3198 	timestamp_save = map->timestamp;
3199 #endif
3200 	vm_map_busy(map);
3201 	vm_map_downgrade(map);
3202 
3203 	rv = 0;
3204 	entry = start_entry;
3205 	while (entry != &map->header && entry->start < end) {
3206 		if (entry->wired_count == 1) {
3207 			rv = uvm_fault_wire(map, entry->start, entry->end,
3208 			    entry->max_protection, 1);
3209 			if (rv) {
3210 
3211 				/*
3212 				 * wiring failed.  break out of the loop.
3213 				 * we'll clean up the map below, once we
3214 				 * have a write lock again.
3215 				 */
3216 
3217 				break;
3218 			}
3219 		}
3220 		entry = entry->next;
3221 	}
3222 
3223 	if (rv) {	/* failed? */
3224 
3225 		/*
3226 		 * Get back to an exclusive (write) lock.
3227 		 */
3228 
3229 		vm_map_upgrade(map);
3230 		vm_map_unbusy(map);
3231 
3232 #ifdef DIAGNOSTIC
3233 		if (timestamp_save != map->timestamp)
3234 			panic("uvm_map_pageable: stale map");
3235 #endif
3236 
3237 		/*
3238 		 * first drop the wiring count on all the entries
3239 		 * which haven't actually been wired yet.
3240 		 */
3241 
3242 		failed_entry = entry;
3243 		while (entry != &map->header && entry->start < end) {
3244 			entry->wired_count--;
3245 			entry = entry->next;
3246 		}
3247 
3248 		/*
3249 		 * now, unwire all the entries that were successfully
3250 		 * wired above.
3251 		 */
3252 
3253 		entry = start_entry;
3254 		while (entry != failed_entry) {
3255 			entry->wired_count--;
3256 			if (VM_MAPENT_ISWIRED(entry) == 0)
3257 				uvm_map_entry_unwire(map, entry);
3258 			entry = entry->next;
3259 		}
3260 		if ((lockflags & UVM_LK_EXIT) == 0)
3261 			vm_map_unlock(map);
3262 		UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
3263 		return (rv);
3264 	}
3265 
3266 	/* We are holding a read lock here. */
3267 	if ((lockflags & UVM_LK_EXIT) == 0) {
3268 		vm_map_unbusy(map);
3269 		vm_map_unlock_read(map);
3270 	} else {
3271 
3272 		/*
3273 		 * Get back to an exclusive (write) lock.
3274 		 */
3275 
3276 		vm_map_upgrade(map);
3277 		vm_map_unbusy(map);
3278 	}
3279 
3280 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
3281 	return 0;
3282 }
3283 
3284 /*
3285  * uvm_map_pageable_all: special case of uvm_map_pageable - affects
3286  * all mapped regions.
3287  *
3288  * => map must not be locked.
3289  * => if no flags are specified, all regions are unwired.
3290  * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
3291  */
3292 
3293 int
3294 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit)
3295 {
3296 	struct vm_map_entry *entry, *failed_entry;
3297 	vsize_t size;
3298 	int rv;
3299 #ifdef DIAGNOSTIC
3300 	u_int timestamp_save;
3301 #endif
3302 	UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
3303 	UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
3304 
3305 	KASSERT(map->flags & VM_MAP_PAGEABLE);
3306 
3307 	vm_map_lock(map);
3308 
3309 	/*
3310 	 * handle wiring and unwiring separately.
3311 	 */
3312 
3313 	if (flags == 0) {			/* unwire */
3314 
3315 		/*
3316 		 * POSIX 1003.1b -- munlockall unlocks all regions,
3317 		 * regardless of how many times mlockall has been called.
3318 		 */
3319 
3320 		for (entry = map->header.next; entry != &map->header;
3321 		     entry = entry->next) {
3322 			if (VM_MAPENT_ISWIRED(entry))
3323 				uvm_map_entry_unwire(map, entry);
3324 		}
3325 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
3326 		vm_map_unlock(map);
3327 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
3328 		return 0;
3329 	}
3330 
3331 	if (flags & MCL_FUTURE) {
3332 
3333 		/*
3334 		 * must wire all future mappings; remember this.
3335 		 */
3336 
3337 		vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
3338 	}
3339 
3340 	if ((flags & MCL_CURRENT) == 0) {
3341 
3342 		/*
3343 		 * no more work to do!
3344 		 */
3345 
3346 		UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
3347 		vm_map_unlock(map);
3348 		return 0;
3349 	}
3350 
3351 	/*
3352 	 * wire case: in three passes [XXXCDC: ugly block of code here]
3353 	 *
3354 	 * 1: holding the write lock, count all pages mapped by non-wired
3355 	 *    entries.  if this would cause us to go over our limit, we fail.
3356 	 *
3357 	 * 2: still holding the write lock, we create any anonymous maps that
3358 	 *    need to be created.  then we increment its wiring count.
3359 	 *
3360 	 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
3361 	 *    in the pages for any newly wired area (wired_count == 1).
3362 	 *
3363 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
3364 	 *    deadlock with another thread that may have faulted on one of
3365 	 *    the pages to be wired (it would mark the page busy, blocking
3366 	 *    us, then in turn block on the map lock that we hold).  because
3367 	 *    of problems in the recursive lock package, we cannot upgrade
3368 	 *    to a write lock in vm_map_lookup.  thus, any actions that
3369 	 *    require the write lock must be done beforehand.  because we
3370 	 *    keep the read lock on the map, the copy-on-write status of the
3371 	 *    entries we modify here cannot change.
3372 	 */
3373 
3374 	for (size = 0, entry = map->header.next; entry != &map->header;
3375 	     entry = entry->next) {
3376 		if (entry->protection != VM_PROT_NONE &&
3377 		    VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3378 			size += entry->end - entry->start;
3379 		}
3380 	}
3381 
3382 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
3383 		vm_map_unlock(map);
3384 		return ENOMEM;
3385 	}
3386 
3387 	if (limit != 0 &&
3388 	    (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
3389 		vm_map_unlock(map);
3390 		return ENOMEM;
3391 	}
3392 
3393 	/*
3394 	 * Pass 2.
3395 	 */
3396 
3397 	for (entry = map->header.next; entry != &map->header;
3398 	     entry = entry->next) {
3399 		if (entry->protection == VM_PROT_NONE)
3400 			continue;
3401 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3402 
3403 			/*
3404 			 * perform actions of vm_map_lookup that need the
3405 			 * write lock on the map: create an anonymous map
3406 			 * for a copy-on-write region, or an anonymous map
3407 			 * for a zero-fill region.  (XXXCDC: submap case
3408 			 * ok?)
3409 			 */
3410 
3411 			if (!UVM_ET_ISSUBMAP(entry)) {	/* not submap */
3412 				if (UVM_ET_ISNEEDSCOPY(entry) &&
3413 				    ((entry->max_protection & VM_PROT_WRITE) ||
3414 				     (entry->object.uvm_obj == NULL))) {
3415 					amap_copy(map, entry, 0, entry->start,
3416 					    entry->end);
3417 					/* XXXCDC: wait OK? */
3418 				}
3419 			}
3420 		}
3421 		entry->wired_count++;
3422 	}
3423 
3424 	/*
3425 	 * Pass 3.
3426 	 */
3427 
3428 #ifdef DIAGNOSTIC
3429 	timestamp_save = map->timestamp;
3430 #endif
3431 	vm_map_busy(map);
3432 	vm_map_downgrade(map);
3433 
3434 	rv = 0;
3435 	for (entry = map->header.next; entry != &map->header;
3436 	     entry = entry->next) {
3437 		if (entry->wired_count == 1) {
3438 			rv = uvm_fault_wire(map, entry->start, entry->end,
3439 			    entry->max_protection, 1);
3440 			if (rv) {
3441 
3442 				/*
3443 				 * wiring failed.  break out of the loop.
3444 				 * we'll clean up the map below, once we
3445 				 * have a write lock again.
3446 				 */
3447 
3448 				break;
3449 			}
3450 		}
3451 	}
3452 
3453 	if (rv) {
3454 
3455 		/*
3456 		 * Get back an exclusive (write) lock.
3457 		 */
3458 
3459 		vm_map_upgrade(map);
3460 		vm_map_unbusy(map);
3461 
3462 #ifdef DIAGNOSTIC
3463 		if (timestamp_save != map->timestamp)
3464 			panic("uvm_map_pageable_all: stale map");
3465 #endif
3466 
3467 		/*
3468 		 * first drop the wiring count on all the entries
3469 		 * which haven't actually been wired yet.
3470 		 *
3471 		 * Skip VM_PROT_NONE entries like we did above.
3472 		 */
3473 
3474 		failed_entry = entry;
3475 		for (/* nothing */; entry != &map->header;
3476 		     entry = entry->next) {
3477 			if (entry->protection == VM_PROT_NONE)
3478 				continue;
3479 			entry->wired_count--;
3480 		}
3481 
3482 		/*
3483 		 * now, unwire all the entries that were successfully
3484 		 * wired above.
3485 		 *
3486 		 * Skip VM_PROT_NONE entries like we did above.
3487 		 */
3488 
3489 		for (entry = map->header.next; entry != failed_entry;
3490 		     entry = entry->next) {
3491 			if (entry->protection == VM_PROT_NONE)
3492 				continue;
3493 			entry->wired_count--;
3494 			if (VM_MAPENT_ISWIRED(entry))
3495 				uvm_map_entry_unwire(map, entry);
3496 		}
3497 		vm_map_unlock(map);
3498 		UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
3499 		return (rv);
3500 	}
3501 
3502 	/* We are holding a read lock here. */
3503 	vm_map_unbusy(map);
3504 	vm_map_unlock_read(map);
3505 
3506 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
3507 	return 0;
3508 }
3509 
3510 /*
3511  * uvm_map_clean: clean out a map range
3512  *
3513  * => valid flags:
3514  *   if (flags & PGO_CLEANIT): dirty pages are cleaned first
3515  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
3516  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
3517  *   if (flags & PGO_FREE): any cached pages are freed after clean
3518  * => returns an error if any part of the specified range isn't mapped
3519  * => never a need to flush amap layer since the anonymous memory has
3520  *	no permanent home, but may deactivate pages there
3521  * => called from sys_msync() and sys_madvise()
3522  * => caller must not write-lock map (read OK).
3523  * => we may sleep while cleaning if SYNCIO [with map read-locked]
3524  */
3525 
3526 int
3527 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
3528 {
3529 	struct vm_map_entry *current, *entry;
3530 	struct uvm_object *uobj;
3531 	struct vm_amap *amap;
3532 	struct vm_anon *anon;
3533 	struct vm_page *pg;
3534 	vaddr_t offset;
3535 	vsize_t size;
3536 	voff_t uoff;
3537 	int error, refs;
3538 	UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
3539 
3540 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
3541 		    map, start, end, flags);
3542 	KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) !=
3543 		(PGO_FREE|PGO_DEACTIVATE));
3544 
3545 	vm_map_lock_read(map);
3546 	VM_MAP_RANGE_CHECK(map, start, end);
3547 	if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
3548 		vm_map_unlock_read(map);
3549 		return EFAULT;
3550 	}
3551 
3552 	/*
3553 	 * Make a first pass to check for holes and wiring problems.
3554 	 */
3555 
3556 	for (current = entry; current->start < end; current = current->next) {
3557 		if (UVM_ET_ISSUBMAP(current)) {
3558 			vm_map_unlock_read(map);
3559 			return EINVAL;
3560 		}
3561 		if ((flags & PGO_FREE) != 0 && VM_MAPENT_ISWIRED(entry)) {
3562 			vm_map_unlock_read(map);
3563 			return EBUSY;
3564 		}
3565 		if (end <= current->end) {
3566 			break;
3567 		}
3568 		if (current->end != current->next->start) {
3569 			vm_map_unlock_read(map);
3570 			return EFAULT;
3571 		}
3572 	}
3573 
3574 	error = 0;
3575 	for (current = entry; start < end; current = current->next) {
3576 		amap = current->aref.ar_amap;	/* top layer */
3577 		uobj = current->object.uvm_obj;	/* bottom layer */
3578 		KASSERT(start >= current->start);
3579 
3580 		/*
3581 		 * No amap cleaning necessary if:
3582 		 *
3583 		 *	(1) There's no amap.
3584 		 *
3585 		 *	(2) We're not deactivating or freeing pages.
3586 		 */
3587 
3588 		if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
3589 			goto flush_object;
3590 
3591 		amap_lock(amap);
3592 		offset = start - current->start;
3593 		size = MIN(end, current->end) - start;
3594 		for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) {
3595 			anon = amap_lookup(&current->aref, offset);
3596 			if (anon == NULL)
3597 				continue;
3598 
3599 			simple_lock(&anon->an_lock);
3600 			pg = anon->an_page;
3601 			if (pg == NULL) {
3602 				simple_unlock(&anon->an_lock);
3603 				continue;
3604 			}
3605 
3606 			switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
3607 
3608 			/*
3609 			 * In these first 3 cases, we just deactivate the page.
3610 			 */
3611 
3612 			case PGO_CLEANIT|PGO_FREE:
3613 			case PGO_CLEANIT|PGO_DEACTIVATE:
3614 			case PGO_DEACTIVATE:
3615  deactivate_it:
3616 				/*
3617 				 * skip the page if it's loaned or wired,
3618 				 * since it shouldn't be on a paging queue
3619 				 * at all in these cases.
3620 				 */
3621 
3622 				uvm_lock_pageq();
3623 				if (pg->loan_count != 0 ||
3624 				    pg->wire_count != 0) {
3625 					uvm_unlock_pageq();
3626 					simple_unlock(&anon->an_lock);
3627 					continue;
3628 				}
3629 				KASSERT(pg->uanon == anon);
3630 				pmap_clear_reference(pg);
3631 				uvm_pagedeactivate(pg);
3632 				uvm_unlock_pageq();
3633 				simple_unlock(&anon->an_lock);
3634 				continue;
3635 
3636 			case PGO_FREE:
3637 
3638 				/*
3639 				 * If there are multiple references to
3640 				 * the amap, just deactivate the page.
3641 				 */
3642 
3643 				if (amap_refs(amap) > 1)
3644 					goto deactivate_it;
3645 
3646 				/* skip the page if it's wired */
3647 				if (pg->wire_count != 0) {
3648 					simple_unlock(&anon->an_lock);
3649 					continue;
3650 				}
3651 				amap_unadd(&current->aref, offset);
3652 				refs = --anon->an_ref;
3653 				simple_unlock(&anon->an_lock);
3654 				if (refs == 0)
3655 					uvm_anfree(anon);
3656 				continue;
3657 			}
3658 		}
3659 		amap_unlock(amap);
3660 
3661  flush_object:
3662 		/*
3663 		 * flush pages if we've got a valid backing object.
3664 		 * note that we must always clean object pages before
3665 		 * freeing them since otherwise we could reveal stale
3666 		 * data from files.
3667 		 */
3668 
3669 		uoff = current->offset + (start - current->start);
3670 		size = MIN(end, current->end) - start;
3671 		if (uobj != NULL) {
3672 			simple_lock(&uobj->vmobjlock);
3673 			if (uobj->pgops->pgo_put != NULL)
3674 				error = (uobj->pgops->pgo_put)(uobj, uoff,
3675 				    uoff + size, flags | PGO_CLEANIT);
3676 			else
3677 				error = 0;
3678 		}
3679 		start += size;
3680 	}
3681 	vm_map_unlock_read(map);
3682 	return (error);
3683 }
3684 
3685 
3686 /*
3687  * uvm_map_checkprot: check protection in map
3688  *
3689  * => must allow specified protection in a fully allocated region.
3690  * => map must be read or write locked by caller.
3691  */
3692 
3693 boolean_t
3694 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end,
3695     vm_prot_t protection)
3696 {
3697 	struct vm_map_entry *entry;
3698 	struct vm_map_entry *tmp_entry;
3699 
3700 	if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
3701 		return (FALSE);
3702 	}
3703 	entry = tmp_entry;
3704 	while (start < end) {
3705 		if (entry == &map->header) {
3706 			return (FALSE);
3707 		}
3708 
3709 		/*
3710 		 * no holes allowed
3711 		 */
3712 
3713 		if (start < entry->start) {
3714 			return (FALSE);
3715 		}
3716 
3717 		/*
3718 		 * check protection associated with entry
3719 		 */
3720 
3721 		if ((entry->protection & protection) != protection) {
3722 			return (FALSE);
3723 		}
3724 		start = entry->end;
3725 		entry = entry->next;
3726 	}
3727 	return (TRUE);
3728 }
3729 
3730 /*
3731  * uvmspace_alloc: allocate a vmspace structure.
3732  *
3733  * - structure includes vm_map and pmap
3734  * - XXX: no locking on this structure
3735  * - refcnt set to 1, rest must be init'd by caller
3736  */
3737 struct vmspace *
3738 uvmspace_alloc(vaddr_t vmin, vaddr_t vmax)
3739 {
3740 	struct vmspace *vm;
3741 	UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
3742 
3743 	vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
3744 	uvmspace_init(vm, NULL, vmin, vmax);
3745 	UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
3746 	return (vm);
3747 }
3748 
3749 /*
3750  * uvmspace_init: initialize a vmspace structure.
3751  *
3752  * - XXX: no locking on this structure
3753  * - refcnt set to 1, rest must be init'd by caller
3754  */
3755 void
3756 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t vmin, vaddr_t vmax)
3757 {
3758 	UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
3759 
3760 	memset(vm, 0, sizeof(*vm));
3761 	uvm_map_setup(&vm->vm_map, vmin, vmax, VM_MAP_PAGEABLE
3762 #ifdef __USING_TOPDOWN_VM
3763 	    | VM_MAP_TOPDOWN
3764 #endif
3765 	    );
3766 	if (pmap)
3767 		pmap_reference(pmap);
3768 	else
3769 		pmap = pmap_create();
3770 	vm->vm_map.pmap = pmap;
3771 	vm->vm_refcnt = 1;
3772 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3773 }
3774 
3775 /*
3776  * uvmspace_share: share a vmspace between two processes
3777  *
3778  * - used for vfork, threads(?)
3779  */
3780 
3781 void
3782 uvmspace_share(struct proc *p1, struct proc *p2)
3783 {
3784 
3785 	uvmspace_addref(p1->p_vmspace);
3786 	p2->p_vmspace = p1->p_vmspace;
3787 }
3788 
3789 /*
3790  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
3791  *
3792  * - XXX: no locking on vmspace
3793  */
3794 
3795 void
3796 uvmspace_unshare(struct lwp *l)
3797 {
3798 	struct proc *p = l->l_proc;
3799 	struct vmspace *nvm, *ovm = p->p_vmspace;
3800 
3801 	if (ovm->vm_refcnt == 1)
3802 		/* nothing to do: vmspace isn't shared in the first place */
3803 		return;
3804 
3805 	/* make a new vmspace, still holding old one */
3806 	nvm = uvmspace_fork(ovm);
3807 
3808 	pmap_deactivate(l);		/* unbind old vmspace */
3809 	p->p_vmspace = nvm;
3810 	pmap_activate(l);		/* switch to new vmspace */
3811 
3812 	uvmspace_free(ovm);		/* drop reference to old vmspace */
3813 }
3814 
3815 /*
3816  * uvmspace_exec: the process wants to exec a new program
3817  */
3818 
3819 void
3820 uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end)
3821 {
3822 	struct proc *p = l->l_proc;
3823 	struct vmspace *nvm, *ovm = p->p_vmspace;
3824 	struct vm_map *map = &ovm->vm_map;
3825 
3826 #ifdef __sparc__
3827 	/* XXX cgd 960926: the sparc #ifdef should be a MD hook */
3828 	kill_user_windows(l);   /* before stack addresses go away */
3829 #endif
3830 
3831 	/*
3832 	 * see if more than one process is using this vmspace...
3833 	 */
3834 
3835 	if (ovm->vm_refcnt == 1) {
3836 
3837 		/*
3838 		 * if p is the only process using its vmspace then we can safely
3839 		 * recycle that vmspace for the program that is being exec'd.
3840 		 */
3841 
3842 #ifdef SYSVSHM
3843 		/*
3844 		 * SYSV SHM semantics require us to kill all segments on an exec
3845 		 */
3846 
3847 		if (ovm->vm_shm)
3848 			shmexit(ovm);
3849 #endif
3850 
3851 		/*
3852 		 * POSIX 1003.1b -- "lock future mappings" is revoked
3853 		 * when a process execs another program image.
3854 		 */
3855 
3856 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
3857 
3858 		/*
3859 		 * now unmap the old program
3860 		 */
3861 
3862 		pmap_remove_all(map->pmap);
3863 		uvm_unmap(map, vm_map_min(map), vm_map_max(map));
3864 		KASSERT(map->header.prev == &map->header);
3865 		KASSERT(map->nentries == 0);
3866 
3867 		/*
3868 		 * resize the map
3869 		 */
3870 
3871 		vm_map_setmin(map, start);
3872 		vm_map_setmax(map, end);
3873 	} else {
3874 
3875 		/*
3876 		 * p's vmspace is being shared, so we can't reuse it for p since
3877 		 * it is still being used for others.   allocate a new vmspace
3878 		 * for p
3879 		 */
3880 
3881 		nvm = uvmspace_alloc(start, end);
3882 
3883 		/*
3884 		 * install new vmspace and drop our ref to the old one.
3885 		 */
3886 
3887 		pmap_deactivate(l);
3888 		p->p_vmspace = nvm;
3889 		pmap_activate(l);
3890 
3891 		uvmspace_free(ovm);
3892 	}
3893 }
3894 
3895 /*
3896  * uvmspace_addref: add a referece to a vmspace.
3897  */
3898 
3899 void
3900 uvmspace_addref(struct vmspace *vm)
3901 {
3902 	struct vm_map *map = &vm->vm_map;
3903 
3904 	KASSERT((map->flags & VM_MAP_DYING) == 0);
3905 
3906 	simple_lock(&map->ref_lock);
3907 	KASSERT(vm->vm_refcnt > 0);
3908 	vm->vm_refcnt++;
3909 	simple_unlock(&map->ref_lock);
3910 }
3911 
3912 /*
3913  * uvmspace_free: free a vmspace data structure
3914  */
3915 
3916 void
3917 uvmspace_free(struct vmspace *vm)
3918 {
3919 	struct vm_map_entry *dead_entries;
3920 	struct vm_map *map = &vm->vm_map;
3921 	int n;
3922 
3923 	UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
3924 
3925 	UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
3926 	simple_lock(&map->ref_lock);
3927 	n = --vm->vm_refcnt;
3928 	simple_unlock(&map->ref_lock);
3929 	if (n > 0)
3930 		return;
3931 
3932 	/*
3933 	 * at this point, there should be no other references to the map.
3934 	 * delete all of the mappings, then destroy the pmap.
3935 	 */
3936 
3937 	map->flags |= VM_MAP_DYING;
3938 	pmap_remove_all(map->pmap);
3939 #ifdef SYSVSHM
3940 	/* Get rid of any SYSV shared memory segments. */
3941 	if (vm->vm_shm != NULL)
3942 		shmexit(vm);
3943 #endif
3944 	if (map->nentries) {
3945 		uvm_unmap_remove(map, vm_map_min(map), vm_map_max(map),
3946 		    &dead_entries, NULL, 0);
3947 		if (dead_entries != NULL)
3948 			uvm_unmap_detach(dead_entries, 0);
3949 	}
3950 	KASSERT(map->nentries == 0);
3951 	KASSERT(map->size == 0);
3952 	pmap_destroy(map->pmap);
3953 	pool_put(&uvm_vmspace_pool, vm);
3954 }
3955 
3956 /*
3957  *   F O R K   -   m a i n   e n t r y   p o i n t
3958  */
3959 /*
3960  * uvmspace_fork: fork a process' main map
3961  *
3962  * => create a new vmspace for child process from parent.
3963  * => parent's map must not be locked.
3964  */
3965 
3966 struct vmspace *
3967 uvmspace_fork(struct vmspace *vm1)
3968 {
3969 	struct vmspace *vm2;
3970 	struct vm_map *old_map = &vm1->vm_map;
3971 	struct vm_map *new_map;
3972 	struct vm_map_entry *old_entry;
3973 	struct vm_map_entry *new_entry;
3974 	UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
3975 
3976 	vm_map_lock(old_map);
3977 
3978 	vm2 = uvmspace_alloc(vm_map_min(old_map), vm_map_max(old_map));
3979 	memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
3980 	    (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
3981 	new_map = &vm2->vm_map;		  /* XXX */
3982 
3983 	old_entry = old_map->header.next;
3984 	new_map->size = old_map->size;
3985 
3986 	/*
3987 	 * go entry-by-entry
3988 	 */
3989 
3990 	while (old_entry != &old_map->header) {
3991 
3992 		/*
3993 		 * first, some sanity checks on the old entry
3994 		 */
3995 
3996 		KASSERT(!UVM_ET_ISSUBMAP(old_entry));
3997 		KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) ||
3998 			!UVM_ET_ISNEEDSCOPY(old_entry));
3999 
4000 		switch (old_entry->inheritance) {
4001 		case MAP_INHERIT_NONE:
4002 
4003 			/*
4004 			 * drop the mapping, modify size
4005 			 */
4006 			new_map->size -= old_entry->end - old_entry->start;
4007 			break;
4008 
4009 		case MAP_INHERIT_SHARE:
4010 
4011 			/*
4012 			 * share the mapping: this means we want the old and
4013 			 * new entries to share amaps and backing objects.
4014 			 */
4015 			/*
4016 			 * if the old_entry needs a new amap (due to prev fork)
4017 			 * then we need to allocate it now so that we have
4018 			 * something we own to share with the new_entry.   [in
4019 			 * other words, we need to clear needs_copy]
4020 			 */
4021 
4022 			if (UVM_ET_ISNEEDSCOPY(old_entry)) {
4023 				/* get our own amap, clears needs_copy */
4024 				amap_copy(old_map, old_entry, AMAP_COPY_NOCHUNK,
4025 				    0, 0);
4026 				/* XXXCDC: WAITOK??? */
4027 			}
4028 
4029 			new_entry = uvm_mapent_alloc(new_map, 0);
4030 			/* old_entry -> new_entry */
4031 			uvm_mapent_copy(old_entry, new_entry);
4032 
4033 			/* new pmap has nothing wired in it */
4034 			new_entry->wired_count = 0;
4035 
4036 			/*
4037 			 * gain reference to object backing the map (can't
4038 			 * be a submap, already checked this case).
4039 			 */
4040 
4041 			if (new_entry->aref.ar_amap)
4042 				uvm_map_reference_amap(new_entry, AMAP_SHARED);
4043 
4044 			if (new_entry->object.uvm_obj &&
4045 			    new_entry->object.uvm_obj->pgops->pgo_reference)
4046 				new_entry->object.uvm_obj->
4047 				    pgops->pgo_reference(
4048 				        new_entry->object.uvm_obj);
4049 
4050 			/* insert entry at end of new_map's entry list */
4051 			uvm_map_entry_link(new_map, new_map->header.prev,
4052 			    new_entry);
4053 
4054 			break;
4055 
4056 		case MAP_INHERIT_COPY:
4057 
4058 			/*
4059 			 * copy-on-write the mapping (using mmap's
4060 			 * MAP_PRIVATE semantics)
4061 			 *
4062 			 * allocate new_entry, adjust reference counts.
4063 			 * (note that new references are read-only).
4064 			 */
4065 
4066 			new_entry = uvm_mapent_alloc(new_map, 0);
4067 			/* old_entry -> new_entry */
4068 			uvm_mapent_copy(old_entry, new_entry);
4069 
4070 			if (new_entry->aref.ar_amap)
4071 				uvm_map_reference_amap(new_entry, 0);
4072 
4073 			if (new_entry->object.uvm_obj &&
4074 			    new_entry->object.uvm_obj->pgops->pgo_reference)
4075 				new_entry->object.uvm_obj->pgops->pgo_reference
4076 				    (new_entry->object.uvm_obj);
4077 
4078 			/* new pmap has nothing wired in it */
4079 			new_entry->wired_count = 0;
4080 
4081 			new_entry->etype |=
4082 			    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
4083 			uvm_map_entry_link(new_map, new_map->header.prev,
4084 			    new_entry);
4085 
4086 			/*
4087 			 * the new entry will need an amap.  it will either
4088 			 * need to be copied from the old entry or created
4089 			 * from scratch (if the old entry does not have an
4090 			 * amap).  can we defer this process until later
4091 			 * (by setting "needs_copy") or do we need to copy
4092 			 * the amap now?
4093 			 *
4094 			 * we must copy the amap now if any of the following
4095 			 * conditions hold:
4096 			 * 1. the old entry has an amap and that amap is
4097 			 *    being shared.  this means that the old (parent)
4098 			 *    process is sharing the amap with another
4099 			 *    process.  if we do not clear needs_copy here
4100 			 *    we will end up in a situation where both the
4101 			 *    parent and child process are refering to the
4102 			 *    same amap with "needs_copy" set.  if the
4103 			 *    parent write-faults, the fault routine will
4104 			 *    clear "needs_copy" in the parent by allocating
4105 			 *    a new amap.   this is wrong because the
4106 			 *    parent is supposed to be sharing the old amap
4107 			 *    and the new amap will break that.
4108 			 *
4109 			 * 2. if the old entry has an amap and a non-zero
4110 			 *    wire count then we are going to have to call
4111 			 *    amap_cow_now to avoid page faults in the
4112 			 *    parent process.   since amap_cow_now requires
4113 			 *    "needs_copy" to be clear we might as well
4114 			 *    clear it here as well.
4115 			 *
4116 			 */
4117 
4118 			if (old_entry->aref.ar_amap != NULL) {
4119 				if ((amap_flags(old_entry->aref.ar_amap) &
4120 				     AMAP_SHARED) != 0 ||
4121 				    VM_MAPENT_ISWIRED(old_entry)) {
4122 
4123 					amap_copy(new_map, new_entry,
4124 					    AMAP_COPY_NOCHUNK, 0, 0);
4125 					/* XXXCDC: M_WAITOK ... ok? */
4126 				}
4127 			}
4128 
4129 			/*
4130 			 * if the parent's entry is wired down, then the
4131 			 * parent process does not want page faults on
4132 			 * access to that memory.  this means that we
4133 			 * cannot do copy-on-write because we can't write
4134 			 * protect the old entry.   in this case we
4135 			 * resolve all copy-on-write faults now, using
4136 			 * amap_cow_now.   note that we have already
4137 			 * allocated any needed amap (above).
4138 			 */
4139 
4140 			if (VM_MAPENT_ISWIRED(old_entry)) {
4141 
4142 			  /*
4143 			   * resolve all copy-on-write faults now
4144 			   * (note that there is nothing to do if
4145 			   * the old mapping does not have an amap).
4146 			   */
4147 			  if (old_entry->aref.ar_amap)
4148 			    amap_cow_now(new_map, new_entry);
4149 
4150 			} else {
4151 
4152 			  /*
4153 			   * setup mappings to trigger copy-on-write faults
4154 			   * we must write-protect the parent if it has
4155 			   * an amap and it is not already "needs_copy"...
4156 			   * if it is already "needs_copy" then the parent
4157 			   * has already been write-protected by a previous
4158 			   * fork operation.
4159 			   */
4160 
4161 			  if (old_entry->aref.ar_amap &&
4162 			      !UVM_ET_ISNEEDSCOPY(old_entry)) {
4163 			      if (old_entry->max_protection & VM_PROT_WRITE) {
4164 				pmap_protect(old_map->pmap,
4165 					     old_entry->start,
4166 					     old_entry->end,
4167 					     old_entry->protection &
4168 					     ~VM_PROT_WRITE);
4169 				pmap_update(old_map->pmap);
4170 			      }
4171 			      old_entry->etype |= UVM_ET_NEEDSCOPY;
4172 			  }
4173 			}
4174 			break;
4175 		}  /* end of switch statement */
4176 		old_entry = old_entry->next;
4177 	}
4178 
4179 	vm_map_unlock(old_map);
4180 
4181 #ifdef SYSVSHM
4182 	if (vm1->vm_shm)
4183 		shmfork(vm1, vm2);
4184 #endif
4185 
4186 #ifdef PMAP_FORK
4187 	pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
4188 #endif
4189 
4190 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
4191 	return (vm2);
4192 }
4193 
4194 
4195 /*
4196  * in-kernel map entry allocation.
4197  */
4198 
4199 struct uvm_kmapent_hdr {
4200 	LIST_ENTRY(uvm_kmapent_hdr) ukh_listq;
4201 	int ukh_nused;
4202 	struct vm_map_entry *ukh_freelist;
4203 	struct vm_map *ukh_map;
4204 	struct vm_map_entry ukh_entries[0];
4205 };
4206 
4207 #define	UVM_KMAPENT_CHUNK				\
4208 	((PAGE_SIZE - sizeof(struct uvm_kmapent_hdr))	\
4209 	/ sizeof(struct vm_map_entry))
4210 
4211 #define	UVM_KHDR_FIND(entry)	\
4212 	((struct uvm_kmapent_hdr *)(((vaddr_t)entry) & ~PAGE_MASK))
4213 
4214 
4215 #ifdef DIAGNOSTIC
4216 static struct vm_map *
4217 uvm_kmapent_map(struct vm_map_entry *entry)
4218 {
4219 	const struct uvm_kmapent_hdr *ukh;
4220 
4221 	ukh = UVM_KHDR_FIND(entry);
4222 	return ukh->ukh_map;
4223 }
4224 #endif
4225 
4226 static inline struct vm_map_entry *
4227 uvm_kmapent_get(struct uvm_kmapent_hdr *ukh)
4228 {
4229 	struct vm_map_entry *entry;
4230 
4231 	KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
4232 	KASSERT(ukh->ukh_nused >= 0);
4233 
4234 	entry = ukh->ukh_freelist;
4235 	if (entry) {
4236 		KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT))
4237 		    == UVM_MAP_KERNEL);
4238 		ukh->ukh_freelist = entry->next;
4239 		ukh->ukh_nused++;
4240 		KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
4241 	} else {
4242 		KASSERT(ukh->ukh_nused == UVM_KMAPENT_CHUNK);
4243 	}
4244 
4245 	return entry;
4246 }
4247 
4248 static inline void
4249 uvm_kmapent_put(struct uvm_kmapent_hdr *ukh, struct vm_map_entry *entry)
4250 {
4251 
4252 	KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT))
4253 	    == UVM_MAP_KERNEL);
4254 	KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
4255 	KASSERT(ukh->ukh_nused > 0);
4256 	KASSERT(ukh->ukh_freelist != NULL ||
4257 	    ukh->ukh_nused == UVM_KMAPENT_CHUNK);
4258 	KASSERT(ukh->ukh_freelist == NULL ||
4259 	    ukh->ukh_nused < UVM_KMAPENT_CHUNK);
4260 
4261 	ukh->ukh_nused--;
4262 	entry->next = ukh->ukh_freelist;
4263 	ukh->ukh_freelist = entry;
4264 }
4265 
4266 /*
4267  * uvm_kmapent_alloc: allocate a map entry for in-kernel map
4268  */
4269 
4270 static struct vm_map_entry *
4271 uvm_kmapent_alloc(struct vm_map *map, int flags)
4272 {
4273 	struct vm_page *pg;
4274 	struct uvm_map_args args;
4275 	struct uvm_kmapent_hdr *ukh;
4276 	struct vm_map_entry *entry;
4277 	uvm_flag_t mapflags = UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL,
4278 	    UVM_INH_NONE, UVM_ADV_RANDOM, flags | UVM_FLAG_NOMERGE);
4279 	vaddr_t va;
4280 	int error;
4281 	int i;
4282 	int s;
4283 
4284 	KDASSERT(UVM_KMAPENT_CHUNK > 2);
4285 	KDASSERT(kernel_map != NULL);
4286 	KASSERT(vm_map_pmap(map) == pmap_kernel());
4287 
4288 	UVMMAP_EVCNT_INCR(uke_alloc);
4289 	entry = NULL;
4290 again:
4291 	/*
4292 	 * try to grab an entry from freelist.
4293 	 */
4294 	s = splvm();
4295 	simple_lock(&uvm.kentry_lock);
4296 	ukh = LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free);
4297 	if (ukh) {
4298 		entry = uvm_kmapent_get(ukh);
4299 		if (ukh->ukh_nused == UVM_KMAPENT_CHUNK)
4300 			LIST_REMOVE(ukh, ukh_listq);
4301 	}
4302 	simple_unlock(&uvm.kentry_lock);
4303 	splx(s);
4304 
4305 	if (entry)
4306 		return entry;
4307 
4308 	/*
4309 	 * there's no free entry for this vm_map.
4310 	 * now we need to allocate some vm_map_entry.
4311 	 * for simplicity, always allocate one page chunk of them at once.
4312 	 */
4313 
4314 	pg = uvm_pagealloc(NULL, 0, NULL, 0);
4315 	if (__predict_false(pg == NULL)) {
4316 		if (flags & UVM_FLAG_NOWAIT)
4317 			return NULL;
4318 		uvm_wait("kme_alloc");
4319 		goto again;
4320 	}
4321 
4322 	error = uvm_map_prepare(map, 0, PAGE_SIZE, NULL, 0, 0, mapflags, &args);
4323 	if (error) {
4324 		uvm_pagefree(pg);
4325 		return NULL;
4326 	}
4327 
4328 	va = args.uma_start;
4329 
4330 	pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), VM_PROT_READ|VM_PROT_WRITE);
4331 	pmap_update(vm_map_pmap(map));
4332 
4333 	ukh = (void *)va;
4334 
4335 	/*
4336 	 * use the first entry for ukh itsself.
4337 	 */
4338 
4339 	entry = &ukh->ukh_entries[0];
4340 	entry->flags = UVM_MAP_KERNEL | UVM_MAP_KMAPENT;
4341 	error = uvm_map_enter(map, &args, entry);
4342 	KASSERT(error == 0);
4343 
4344 	ukh->ukh_nused = UVM_KMAPENT_CHUNK;
4345 	ukh->ukh_map = map;
4346 	ukh->ukh_freelist = NULL;
4347 	for (i = UVM_KMAPENT_CHUNK - 1; i >= 2; i--) {
4348 		struct vm_map_entry *xentry = &ukh->ukh_entries[i];
4349 
4350 		xentry->flags = UVM_MAP_KERNEL;
4351 		uvm_kmapent_put(ukh, xentry);
4352 	}
4353 	KASSERT(ukh->ukh_nused == 2);
4354 
4355 	s = splvm();
4356 	simple_lock(&uvm.kentry_lock);
4357 	LIST_INSERT_HEAD(&vm_map_to_kernel(map)->vmk_kentry_free,
4358 	    ukh, ukh_listq);
4359 	simple_unlock(&uvm.kentry_lock);
4360 	splx(s);
4361 
4362 	/*
4363 	 * return second entry.
4364 	 */
4365 
4366 	entry = &ukh->ukh_entries[1];
4367 	entry->flags = UVM_MAP_KERNEL;
4368 	UVMMAP_EVCNT_INCR(ukh_alloc);
4369 	return entry;
4370 }
4371 
4372 /*
4373  * uvm_mapent_free: free map entry for in-kernel map
4374  */
4375 
4376 static void
4377 uvm_kmapent_free(struct vm_map_entry *entry)
4378 {
4379 	struct uvm_kmapent_hdr *ukh;
4380 	struct vm_page *pg;
4381 	struct vm_map *map;
4382 	struct pmap *pmap;
4383 	vaddr_t va;
4384 	paddr_t pa;
4385 	struct vm_map_entry *deadentry;
4386 	int s;
4387 
4388 	UVMMAP_EVCNT_INCR(uke_free);
4389 	ukh = UVM_KHDR_FIND(entry);
4390 	map = ukh->ukh_map;
4391 
4392 	s = splvm();
4393 	simple_lock(&uvm.kentry_lock);
4394 	uvm_kmapent_put(ukh, entry);
4395 	if (ukh->ukh_nused > 1) {
4396 		if (ukh->ukh_nused == UVM_KMAPENT_CHUNK - 1)
4397 			LIST_INSERT_HEAD(
4398 			    &vm_map_to_kernel(map)->vmk_kentry_free,
4399 			    ukh, ukh_listq);
4400 		simple_unlock(&uvm.kentry_lock);
4401 		splx(s);
4402 		return;
4403 	}
4404 
4405 	/*
4406 	 * now we can free this ukh.
4407 	 *
4408 	 * however, keep an empty ukh to avoid ping-pong.
4409 	 */
4410 
4411 	if (LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free) == ukh &&
4412 	    LIST_NEXT(ukh, ukh_listq) == NULL) {
4413 		simple_unlock(&uvm.kentry_lock);
4414 		splx(s);
4415 		return;
4416 	}
4417 	LIST_REMOVE(ukh, ukh_listq);
4418 	simple_unlock(&uvm.kentry_lock);
4419 	splx(s);
4420 
4421 	KASSERT(ukh->ukh_nused == 1);
4422 
4423 	/*
4424 	 * remove map entry for ukh itsself.
4425 	 */
4426 
4427 	va = (vaddr_t)ukh;
4428 	KASSERT((va & PAGE_MASK) == 0);
4429 	vm_map_lock(map);
4430 	uvm_unmap_remove(map, va, va + PAGE_SIZE, &deadentry, NULL, 0);
4431 	KASSERT(deadentry->flags & UVM_MAP_KERNEL);
4432 	KASSERT(deadentry->flags & UVM_MAP_KMAPENT);
4433 	KASSERT(deadentry->next == NULL);
4434 	KASSERT(deadentry == &ukh->ukh_entries[0]);
4435 
4436 	/*
4437 	 * unmap the page from pmap and free it.
4438 	 */
4439 
4440 	pmap = vm_map_pmap(map);
4441 	KASSERT(pmap == pmap_kernel());
4442 	if (!pmap_extract(pmap, va, &pa))
4443 		panic("%s: no mapping", __func__);
4444 	pmap_kremove(va, PAGE_SIZE);
4445 	vm_map_unlock(map);
4446 	pg = PHYS_TO_VM_PAGE(pa);
4447 	uvm_pagefree(pg);
4448 	UVMMAP_EVCNT_INCR(ukh_free);
4449 }
4450 
4451 static vsize_t
4452 uvm_kmapent_overhead(vsize_t size)
4453 {
4454 
4455 	/*
4456 	 * - the max number of unmerged entries is howmany(size, PAGE_SIZE)
4457 	 *   as the min allocation unit is PAGE_SIZE.
4458 	 * - UVM_KMAPENT_CHUNK "kmapent"s are allocated from a page.
4459 	 *   one of them are used to map the page itself.
4460 	 */
4461 
4462 	return howmany(howmany(size, PAGE_SIZE), (UVM_KMAPENT_CHUNK - 1)) *
4463 	    PAGE_SIZE;
4464 }
4465 
4466 /*
4467  * map entry reservation
4468  */
4469 
4470 /*
4471  * uvm_mapent_reserve: reserve map entries for clipping before locking map.
4472  *
4473  * => needed when unmapping entries allocated without UVM_FLAG_QUANTUM.
4474  * => caller shouldn't hold map locked.
4475  */
4476 int
4477 uvm_mapent_reserve(struct vm_map *map, struct uvm_mapent_reservation *umr,
4478     int nentries, int flags)
4479 {
4480 
4481 	umr->umr_nentries = 0;
4482 
4483 	if ((flags & UVM_FLAG_QUANTUM) != 0)
4484 		return 0;
4485 
4486 	if (!VM_MAP_USE_KMAPENT(map))
4487 		return 0;
4488 
4489 	while (nentries--) {
4490 		struct vm_map_entry *ent;
4491 		ent = uvm_kmapent_alloc(map, flags);
4492 		if (!ent) {
4493 			uvm_mapent_unreserve(map, umr);
4494 			return ENOMEM;
4495 		}
4496 		UMR_PUTENTRY(umr, ent);
4497 	}
4498 
4499 	return 0;
4500 }
4501 
4502 /*
4503  * uvm_mapent_unreserve:
4504  *
4505  * => caller shouldn't hold map locked.
4506  * => never fail or sleep.
4507  */
4508 void
4509 uvm_mapent_unreserve(struct vm_map *map, struct uvm_mapent_reservation *umr)
4510 {
4511 
4512 	while (!UMR_EMPTY(umr))
4513 		uvm_kmapent_free(UMR_GETENTRY(umr));
4514 }
4515 
4516 /*
4517  * uvm_mapent_trymerge: try to merge an entry with its neighbors.
4518  *
4519  * => called with map locked.
4520  * => return non zero if successfully merged.
4521  */
4522 
4523 int
4524 uvm_mapent_trymerge(struct vm_map *map, struct vm_map_entry *entry, int flags)
4525 {
4526 	struct uvm_object *uobj;
4527 	struct vm_map_entry *next;
4528 	struct vm_map_entry *prev;
4529 	vsize_t size;
4530 	int merged = 0;
4531 	boolean_t copying;
4532 	int newetype;
4533 
4534 	if (VM_MAP_USE_KMAPENT(map)) {
4535 		return 0;
4536 	}
4537 	if (entry->aref.ar_amap != NULL) {
4538 		return 0;
4539 	}
4540 	if ((entry->flags & UVM_MAP_NOMERGE) != 0) {
4541 		return 0;
4542 	}
4543 
4544 	uobj = entry->object.uvm_obj;
4545 	size = entry->end - entry->start;
4546 	copying = (flags & UVM_MERGE_COPYING) != 0;
4547 	newetype = copying ? (entry->etype & ~UVM_ET_NEEDSCOPY) : entry->etype;
4548 
4549 	next = entry->next;
4550 	if (next != &map->header &&
4551 	    next->start == entry->end &&
4552 	    ((copying && next->aref.ar_amap != NULL &&
4553 	    amap_refs(next->aref.ar_amap) == 1) ||
4554 	    (!copying && next->aref.ar_amap == NULL)) &&
4555 	    UVM_ET_ISCOMPATIBLE(next, newetype,
4556 	    uobj, entry->flags, entry->protection,
4557 	    entry->max_protection, entry->inheritance, entry->advice,
4558 	    entry->wired_count) &&
4559 	    (uobj == NULL || entry->offset + size == next->offset)) {
4560 		int error;
4561 
4562 		if (copying) {
4563 			error = amap_extend(next, size,
4564 			    AMAP_EXTEND_NOWAIT|AMAP_EXTEND_BACKWARDS);
4565 		} else {
4566 			error = 0;
4567 		}
4568 		if (error == 0) {
4569 			if (uobj) {
4570 				if (uobj->pgops->pgo_detach) {
4571 					uobj->pgops->pgo_detach(uobj);
4572 				}
4573 			}
4574 
4575 			entry->end = next->end;
4576 			clear_hints(map, next);
4577 			uvm_map_entry_unlink(map, next);
4578 			if (copying) {
4579 				entry->aref = next->aref;
4580 				entry->etype &= ~UVM_ET_NEEDSCOPY;
4581 			}
4582 			uvm_map_check(map, "trymerge forwardmerge");
4583 			uvm_mapent_free_merged(map, next);
4584 			merged++;
4585 		}
4586 	}
4587 
4588 	prev = entry->prev;
4589 	if (prev != &map->header &&
4590 	    prev->end == entry->start &&
4591 	    ((copying && !merged && prev->aref.ar_amap != NULL &&
4592 	    amap_refs(prev->aref.ar_amap) == 1) ||
4593 	    (!copying && prev->aref.ar_amap == NULL)) &&
4594 	    UVM_ET_ISCOMPATIBLE(prev, newetype,
4595 	    uobj, entry->flags, entry->protection,
4596 	    entry->max_protection, entry->inheritance, entry->advice,
4597 	    entry->wired_count) &&
4598 	    (uobj == NULL ||
4599 	    prev->offset + prev->end - prev->start == entry->offset)) {
4600 		int error;
4601 
4602 		if (copying) {
4603 			error = amap_extend(prev, size,
4604 			    AMAP_EXTEND_NOWAIT|AMAP_EXTEND_FORWARDS);
4605 		} else {
4606 			error = 0;
4607 		}
4608 		if (error == 0) {
4609 			if (uobj) {
4610 				if (uobj->pgops->pgo_detach) {
4611 					uobj->pgops->pgo_detach(uobj);
4612 				}
4613 				entry->offset = prev->offset;
4614 			}
4615 
4616 			entry->start = prev->start;
4617 			clear_hints(map, prev);
4618 			uvm_map_entry_unlink(map, prev);
4619 			if (copying) {
4620 				entry->aref = prev->aref;
4621 				entry->etype &= ~UVM_ET_NEEDSCOPY;
4622 			}
4623 			uvm_map_check(map, "trymerge backmerge");
4624 			uvm_mapent_free_merged(map, prev);
4625 			merged++;
4626 		}
4627 	}
4628 
4629 	return merged;
4630 }
4631 
4632 #if defined(DDB)
4633 
4634 /*
4635  * DDB hooks
4636  */
4637 
4638 /*
4639  * uvm_map_printit: actually prints the map
4640  */
4641 
4642 void
4643 uvm_map_printit(struct vm_map *map, boolean_t full,
4644     void (*pr)(const char *, ...))
4645 {
4646 	struct vm_map_entry *entry;
4647 
4648 	(*pr)("MAP %p: [0x%lx->0x%lx]\n", map, vm_map_min(map),
4649 	    vm_map_max(map));
4650 	(*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
4651 	    map->nentries, map->size, map->ref_count, map->timestamp,
4652 	    map->flags);
4653 	(*pr)("\tpmap=%p(resident=%ld, wired=%ld)\n", map->pmap,
4654 	    pmap_resident_count(map->pmap), pmap_wired_count(map->pmap));
4655 	if (!full)
4656 		return;
4657 	for (entry = map->header.next; entry != &map->header;
4658 	    entry = entry->next) {
4659 		(*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
4660 		    entry, entry->start, entry->end, entry->object.uvm_obj,
4661 		    (long long)entry->offset, entry->aref.ar_amap,
4662 		    entry->aref.ar_pageoff);
4663 		(*pr)(
4664 		    "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, "
4665 		    "wc=%d, adv=%d\n",
4666 		    (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
4667 		    (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
4668 		    (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
4669 		    entry->protection, entry->max_protection,
4670 		    entry->inheritance, entry->wired_count, entry->advice);
4671 	}
4672 }
4673 
4674 /*
4675  * uvm_object_printit: actually prints the object
4676  */
4677 
4678 void
4679 uvm_object_printit(struct uvm_object *uobj, boolean_t full,
4680     void (*pr)(const char *, ...))
4681 {
4682 	struct vm_page *pg;
4683 	int cnt = 0;
4684 
4685 	(*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
4686 	    uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
4687 	if (UVM_OBJ_IS_KERN_OBJECT(uobj))
4688 		(*pr)("refs=<SYSTEM>\n");
4689 	else
4690 		(*pr)("refs=%d\n", uobj->uo_refs);
4691 
4692 	if (!full) {
4693 		return;
4694 	}
4695 	(*pr)("  PAGES <pg,offset>:\n  ");
4696 	TAILQ_FOREACH(pg, &uobj->memq, listq) {
4697 		cnt++;
4698 		(*pr)("<%p,0x%llx> ", pg, (long long)pg->offset);
4699 		if ((cnt % 3) == 0) {
4700 			(*pr)("\n  ");
4701 		}
4702 	}
4703 	if ((cnt % 3) != 0) {
4704 		(*pr)("\n");
4705 	}
4706 }
4707 
4708 /*
4709  * uvm_page_printit: actually print the page
4710  */
4711 
4712 static const char page_flagbits[] = UVM_PGFLAGBITS;
4713 static const char page_pqflagbits[] = UVM_PQFLAGBITS;
4714 
4715 void
4716 uvm_page_printit(struct vm_page *pg, boolean_t full,
4717     void (*pr)(const char *, ...))
4718 {
4719 	struct vm_page *tpg;
4720 	struct uvm_object *uobj;
4721 	struct pglist *pgl;
4722 	char pgbuf[128];
4723 	char pqbuf[128];
4724 
4725 	(*pr)("PAGE %p:\n", pg);
4726 	bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf));
4727 	bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
4728 	(*pr)("  flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n",
4729 	    pgbuf, pqbuf, pg->wire_count, (long)VM_PAGE_TO_PHYS(pg));
4730 	(*pr)("  uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
4731 	    pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
4732 #if defined(UVM_PAGE_TRKOWN)
4733 	if (pg->flags & PG_BUSY)
4734 		(*pr)("  owning process = %d, tag=%s\n",
4735 		    pg->owner, pg->owner_tag);
4736 	else
4737 		(*pr)("  page not busy, no owner\n");
4738 #else
4739 	(*pr)("  [page ownership tracking disabled]\n");
4740 #endif
4741 
4742 	if (!full)
4743 		return;
4744 
4745 	/* cross-verify object/anon */
4746 	if ((pg->pqflags & PQ_FREE) == 0) {
4747 		if (pg->pqflags & PQ_ANON) {
4748 			if (pg->uanon == NULL || pg->uanon->an_page != pg)
4749 			    (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
4750 				(pg->uanon) ? pg->uanon->an_page : NULL);
4751 			else
4752 				(*pr)("  anon backpointer is OK\n");
4753 		} else {
4754 			uobj = pg->uobject;
4755 			if (uobj) {
4756 				(*pr)("  checking object list\n");
4757 				TAILQ_FOREACH(tpg, &uobj->memq, listq) {
4758 					if (tpg == pg) {
4759 						break;
4760 					}
4761 				}
4762 				if (tpg)
4763 					(*pr)("  page found on object list\n");
4764 				else
4765 			(*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
4766 			}
4767 		}
4768 	}
4769 
4770 	/* cross-verify page queue */
4771 	if (pg->pqflags & PQ_FREE) {
4772 		int fl = uvm_page_lookup_freelist(pg);
4773 		int color = VM_PGCOLOR_BUCKET(pg);
4774 		pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[
4775 		    ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN];
4776 	} else {
4777 		pgl = NULL;
4778 	}
4779 
4780 	if (pgl) {
4781 		(*pr)("  checking pageq list\n");
4782 		TAILQ_FOREACH(tpg, pgl, pageq) {
4783 			if (tpg == pg) {
4784 				break;
4785 			}
4786 		}
4787 		if (tpg)
4788 			(*pr)("  page found on pageq list\n");
4789 		else
4790 			(*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
4791 	}
4792 }
4793 
4794 /*
4795  * uvm_pages_printthem - print a summary of all managed pages
4796  */
4797 
4798 void
4799 uvm_page_printall(void (*pr)(const char *, ...))
4800 {
4801 	unsigned i;
4802 	struct vm_page *pg;
4803 
4804 	(*pr)("%18s %4s %2s %18s %18s"
4805 #ifdef UVM_PAGE_TRKOWN
4806 	    " OWNER"
4807 #endif
4808 	    "\n", "PAGE", "FLAG", "PQ", "UOBJECT", "UANON");
4809 	for (i = 0; i < vm_nphysseg; i++) {
4810 		for (pg = vm_physmem[i].pgs; pg <= vm_physmem[i].lastpg; pg++) {
4811 			(*pr)("%18p %04x %02x %18p %18p",
4812 			    pg, pg->flags, pg->pqflags, pg->uobject,
4813 			    pg->uanon);
4814 #ifdef UVM_PAGE_TRKOWN
4815 			if (pg->flags & PG_BUSY)
4816 				(*pr)(" %d [%s]", pg->owner, pg->owner_tag);
4817 #endif
4818 			(*pr)("\n");
4819 		}
4820 	}
4821 }
4822 
4823 #endif
4824 
4825 /*
4826  * uvm_map_create: create map
4827  */
4828 
4829 struct vm_map *
4830 uvm_map_create(pmap_t pmap, vaddr_t vmin, vaddr_t vmax, int flags)
4831 {
4832 	struct vm_map *result;
4833 
4834 	MALLOC(result, struct vm_map *, sizeof(struct vm_map),
4835 	    M_VMMAP, M_WAITOK);
4836 	uvm_map_setup(result, vmin, vmax, flags);
4837 	result->pmap = pmap;
4838 	return(result);
4839 }
4840 
4841 /*
4842  * uvm_map_setup: init map
4843  *
4844  * => map must not be in service yet.
4845  */
4846 
4847 void
4848 uvm_map_setup(struct vm_map *map, vaddr_t vmin, vaddr_t vmax, int flags)
4849 {
4850 
4851 	RB_INIT(&map->rbhead);
4852 	map->header.next = map->header.prev = &map->header;
4853 	map->nentries = 0;
4854 	map->size = 0;
4855 	map->ref_count = 1;
4856 	vm_map_setmin(map, vmin);
4857 	vm_map_setmax(map, vmax);
4858 	map->flags = flags;
4859 	map->first_free = &map->header;
4860 	map->hint = &map->header;
4861 	map->timestamp = 0;
4862 	lockinit(&map->lock, PVM, "vmmaplk", 0, 0);
4863 	simple_lock_init(&map->ref_lock);
4864 	simple_lock_init(&map->hint_lock);
4865 	simple_lock_init(&map->flags_lock);
4866 }
4867 
4868 
4869 /*
4870  *   U N M A P   -   m a i n   e n t r y   p o i n t
4871  */
4872 
4873 /*
4874  * uvm_unmap1: remove mappings from a vm_map (from "start" up to "stop")
4875  *
4876  * => caller must check alignment and size
4877  * => map must be unlocked (we will lock it)
4878  * => flags is UVM_FLAG_QUANTUM or 0.
4879  */
4880 
4881 void
4882 uvm_unmap1(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
4883 {
4884 	struct vm_map_entry *dead_entries;
4885 	struct uvm_mapent_reservation umr;
4886 	UVMHIST_FUNC("uvm_unmap"); UVMHIST_CALLED(maphist);
4887 
4888 	UVMHIST_LOG(maphist, "  (map=0x%x, start=0x%x, end=0x%x)",
4889 	    map, start, end, 0);
4890 	/*
4891 	 * work now done by helper functions.   wipe the pmap's and then
4892 	 * detach from the dead entries...
4893 	 */
4894 	uvm_mapent_reserve(map, &umr, 2, flags);
4895 	vm_map_lock(map);
4896 	uvm_unmap_remove(map, start, end, &dead_entries, &umr, flags);
4897 	vm_map_unlock(map);
4898 	uvm_mapent_unreserve(map, &umr);
4899 
4900 	if (dead_entries != NULL)
4901 		uvm_unmap_detach(dead_entries, 0);
4902 
4903 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
4904 }
4905 
4906 
4907 /*
4908  * uvm_map_reference: add reference to a map
4909  *
4910  * => map need not be locked (we use ref_lock).
4911  */
4912 
4913 void
4914 uvm_map_reference(struct vm_map *map)
4915 {
4916 	simple_lock(&map->ref_lock);
4917 	map->ref_count++;
4918 	simple_unlock(&map->ref_lock);
4919 }
4920 
4921 struct vm_map_kernel *
4922 vm_map_to_kernel(struct vm_map *map)
4923 {
4924 
4925 	KASSERT(VM_MAP_IS_KERNEL(map));
4926 
4927 	return (struct vm_map_kernel *)map;
4928 }
4929 
4930 boolean_t
4931 vm_map_starved_p(struct vm_map *map)
4932 {
4933 
4934 	if ((map->flags & VM_MAP_WANTVA) != 0) {
4935 		return TRUE;
4936 	}
4937 	/* XXX */
4938 	if ((vm_map_max(map) - vm_map_min(map)) / 16 * 15 < map->size) {
4939 		return TRUE;
4940 	}
4941 	return FALSE;
4942 }
4943