xref: /netbsd-src/sys/uvm/uvm_map.c (revision e5548b402ae4c44fb816de42c7bba9581ce23ef5)
1 /*	$NetBSD: uvm_map.c,v 1.205 2005/12/11 12:25:29 christos 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 1.205 2005/12/11 12:25:29 christos 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 #define UVM_MAP_C
96 #include <uvm/uvm.h>
97 #undef RB_AUGMENT
98 #define	RB_AUGMENT(x)	uvm_rb_augment(x)
99 
100 #ifdef DDB
101 #include <uvm/uvm_ddb.h>
102 #endif
103 
104 #ifndef UVMMAP_NOCOUNTERS
105 #include <sys/device.h>
106 struct evcnt map_ubackmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
107     "uvmmap", "ubackmerge");
108 struct evcnt map_uforwmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
109     "uvmmap", "uforwmerge");
110 struct evcnt map_ubimerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
111     "uvmmap", "ubimerge");
112 struct evcnt map_unomerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
113     "uvmmap", "unomerge");
114 struct evcnt map_kbackmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
115     "uvmmap", "kbackmerge");
116 struct evcnt map_kforwmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
117     "uvmmap", "kforwmerge");
118 struct evcnt map_kbimerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
119     "uvmmap", "kbimerge");
120 struct evcnt map_knomerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
121     "uvmmap", "knomerge");
122 struct evcnt uvm_map_call = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
123     "uvmmap", "map_call");
124 struct evcnt uvm_mlk_call = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
125     "uvmmap", "mlk_call");
126 struct evcnt uvm_mlk_hint = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
127     "uvmmap", "mlk_hint");
128 
129 EVCNT_ATTACH_STATIC(map_ubackmerge);
130 EVCNT_ATTACH_STATIC(map_uforwmerge);
131 EVCNT_ATTACH_STATIC(map_ubimerge);
132 EVCNT_ATTACH_STATIC(map_unomerge);
133 EVCNT_ATTACH_STATIC(map_kbackmerge);
134 EVCNT_ATTACH_STATIC(map_kforwmerge);
135 EVCNT_ATTACH_STATIC(map_kbimerge);
136 EVCNT_ATTACH_STATIC(map_knomerge);
137 EVCNT_ATTACH_STATIC(uvm_map_call);
138 EVCNT_ATTACH_STATIC(uvm_mlk_call);
139 EVCNT_ATTACH_STATIC(uvm_mlk_hint);
140 
141 #define UVMCNT_INCR(ev)		ev.ev_count++
142 #define UVMCNT_DECR(ev)		ev.ev_count--
143 #else
144 #define UVMCNT_INCR(ev)
145 #define UVMCNT_DECR(ev)
146 #endif
147 
148 const char vmmapbsy[] = "vmmapbsy";
149 
150 /*
151  * pool for vmspace structures.
152  */
153 
154 POOL_INIT(uvm_vmspace_pool, sizeof(struct vmspace), 0, 0, 0, "vmsppl",
155     &pool_allocator_nointr);
156 
157 /*
158  * pool for dynamically-allocated map entries.
159  */
160 
161 POOL_INIT(uvm_map_entry_pool, sizeof(struct vm_map_entry), 0, 0, 0, "vmmpepl",
162     &pool_allocator_nointr);
163 
164 MALLOC_DEFINE(M_VMMAP, "VM map", "VM map structures");
165 MALLOC_DEFINE(M_VMPMAP, "VM pmap", "VM pmap");
166 
167 #ifdef PMAP_GROWKERNEL
168 /*
169  * This global represents the end of the kernel virtual address
170  * space.  If we want to exceed this, we must grow the kernel
171  * virtual address space dynamically.
172  *
173  * Note, this variable is locked by kernel_map's lock.
174  */
175 vaddr_t uvm_maxkaddr;
176 #endif
177 
178 /*
179  * macros
180  */
181 
182 /*
183  * VM_MAP_USE_KMAPENT: determine if uvm_kmapent_alloc/free is used
184  * for the vm_map.
185  */
186 extern struct vm_map *pager_map; /* XXX */
187 #define	VM_MAP_USE_KMAPENT(map) \
188 	(((map)->flags & VM_MAP_INTRSAFE) || (map) == kernel_map)
189 
190 /*
191  * UVM_ET_ISCOMPATIBLE: check some requirements for map entry merging
192  */
193 
194 #define	UVM_ET_ISCOMPATIBLE(ent, type, uobj, meflags, \
195     prot, maxprot, inh, adv, wire) \
196 	((ent)->etype == (type) && \
197 	(((ent)->flags ^ (meflags)) & (UVM_MAP_NOMERGE | UVM_MAP_QUANTUM)) \
198 	== 0 && \
199 	(ent)->object.uvm_obj == (uobj) && \
200 	(ent)->protection == (prot) && \
201 	(ent)->max_protection == (maxprot) && \
202 	(ent)->inheritance == (inh) && \
203 	(ent)->advice == (adv) && \
204 	(ent)->wired_count == (wire))
205 
206 /*
207  * uvm_map_entry_link: insert entry into a map
208  *
209  * => map must be locked
210  */
211 #define uvm_map_entry_link(map, after_where, entry) do { \
212 	KASSERT(entry->start < entry->end); \
213 	(map)->nentries++; \
214 	(entry)->prev = (after_where); \
215 	(entry)->next = (after_where)->next; \
216 	(entry)->prev->next = (entry); \
217 	(entry)->next->prev = (entry); \
218 	uvm_rb_insert((map), (entry)); \
219 } while (/*CONSTCOND*/ 0)
220 
221 /*
222  * uvm_map_entry_unlink: remove entry from a map
223  *
224  * => map must be locked
225  */
226 #define uvm_map_entry_unlink(map, entry) do { \
227 	(map)->nentries--; \
228 	(entry)->next->prev = (entry)->prev; \
229 	(entry)->prev->next = (entry)->next; \
230 	uvm_rb_remove((map), (entry)); \
231 } while (/*CONSTCOND*/ 0)
232 
233 /*
234  * SAVE_HINT: saves the specified entry as the hint for future lookups.
235  *
236  * => map need not be locked (protected by hint_lock).
237  */
238 #define SAVE_HINT(map,check,value) do { \
239 	simple_lock(&(map)->hint_lock); \
240 	if ((map)->hint == (check)) \
241 		(map)->hint = (value); \
242 	simple_unlock(&(map)->hint_lock); \
243 } while (/*CONSTCOND*/ 0)
244 
245 /*
246  * VM_MAP_RANGE_CHECK: check and correct range
247  *
248  * => map must at least be read locked
249  */
250 
251 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
252 	if (start < vm_map_min(map))		\
253 		start = vm_map_min(map);	\
254 	if (end > vm_map_max(map))		\
255 		end = vm_map_max(map);		\
256 	if (start > end)			\
257 		start = end;			\
258 } while (/*CONSTCOND*/ 0)
259 
260 /*
261  * local prototypes
262  */
263 
264 static struct vm_map_entry *
265 		uvm_mapent_alloc(struct vm_map *, int);
266 static struct vm_map_entry *
267 		uvm_mapent_alloc_split(struct vm_map *,
268 		    const struct vm_map_entry *, int,
269 		    struct uvm_mapent_reservation *);
270 static void	uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *);
271 static void	uvm_mapent_free(struct vm_map_entry *);
272 static struct vm_map_entry *
273 		uvm_kmapent_alloc(struct vm_map *, int);
274 static void	uvm_kmapent_free(struct vm_map_entry *);
275 static void	uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *);
276 static void	uvm_map_reference_amap(struct vm_map_entry *, int);
277 static int	uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int,
278 		    struct vm_map_entry *);
279 static void	uvm_map_unreference_amap(struct vm_map_entry *, int);
280 
281 int _uvm_tree_sanity(struct vm_map *, const char *);
282 static vsize_t uvm_rb_subtree_space(const struct vm_map_entry *);
283 
284 static __inline int
285 uvm_compare(const struct vm_map_entry *a, const struct vm_map_entry *b)
286 {
287 
288 	if (a->start < b->start)
289 		return (-1);
290 	else if (a->start > b->start)
291 		return (1);
292 
293 	return (0);
294 }
295 
296 static __inline void
297 uvm_rb_augment(struct vm_map_entry *entry)
298 {
299 
300 	entry->space = uvm_rb_subtree_space(entry);
301 }
302 
303 RB_PROTOTYPE(uvm_tree, vm_map_entry, rb_entry, uvm_compare);
304 
305 RB_GENERATE(uvm_tree, vm_map_entry, rb_entry, uvm_compare);
306 
307 static __inline vsize_t
308 uvm_rb_space(const struct vm_map *map, const struct vm_map_entry *entry)
309 {
310 	/* XXX map is not used */
311 
312 	KASSERT(entry->next != NULL);
313 	return entry->next->start - entry->end;
314 }
315 
316 static vsize_t
317 uvm_rb_subtree_space(const struct vm_map_entry *entry)
318 {
319 	vaddr_t space, tmp;
320 
321 	space = entry->ownspace;
322 	if (RB_LEFT(entry, rb_entry)) {
323 		tmp = RB_LEFT(entry, rb_entry)->space;
324 		if (tmp > space)
325 			space = tmp;
326 	}
327 
328 	if (RB_RIGHT(entry, rb_entry)) {
329 		tmp = RB_RIGHT(entry, rb_entry)->space;
330 		if (tmp > space)
331 			space = tmp;
332 	}
333 
334 	return (space);
335 }
336 
337 static __inline void
338 uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry)
339 {
340 	/* We need to traverse to the very top */
341 	do {
342 		entry->ownspace = uvm_rb_space(map, entry);
343 		entry->space = uvm_rb_subtree_space(entry);
344 	} while ((entry = RB_PARENT(entry, rb_entry)) != NULL);
345 }
346 
347 static void
348 uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry)
349 {
350 	vaddr_t space = uvm_rb_space(map, entry);
351 	struct vm_map_entry *tmp;
352 
353 	entry->ownspace = entry->space = space;
354 	tmp = RB_INSERT(uvm_tree, &(map)->rbhead, entry);
355 #ifdef DIAGNOSTIC
356 	if (tmp != NULL)
357 		panic("uvm_rb_insert: duplicate entry?");
358 #endif
359 	uvm_rb_fixup(map, entry);
360 	if (entry->prev != &map->header)
361 		uvm_rb_fixup(map, entry->prev);
362 }
363 
364 static void
365 uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry)
366 {
367 	struct vm_map_entry *parent;
368 
369 	parent = RB_PARENT(entry, rb_entry);
370 	RB_REMOVE(uvm_tree, &(map)->rbhead, entry);
371 	if (entry->prev != &map->header)
372 		uvm_rb_fixup(map, entry->prev);
373 	if (parent)
374 		uvm_rb_fixup(map, parent);
375 }
376 
377 #ifdef DEBUG
378 int uvm_debug_check_rbtree = 0;
379 #define uvm_tree_sanity(x,y)		\
380 	if (uvm_debug_check_rbtree)	\
381 		_uvm_tree_sanity(x,y)
382 #else
383 #define uvm_tree_sanity(x,y)
384 #endif
385 
386 int
387 _uvm_tree_sanity(struct vm_map *map, const char *name)
388 {
389 	struct vm_map_entry *tmp, *trtmp;
390 	int n = 0, i = 1;
391 
392 	RB_FOREACH(tmp, uvm_tree, &map->rbhead) {
393 		if (tmp->ownspace != uvm_rb_space(map, tmp)) {
394 			printf("%s: %d/%d ownspace %lx != %lx %s\n",
395 			    name, n + 1, map->nentries,
396 			    (ulong)tmp->ownspace, (ulong)uvm_rb_space(map, tmp),
397 			    tmp->next == &map->header ? "(last)" : "");
398 			goto error;
399 		}
400 	}
401 	trtmp = NULL;
402 	RB_FOREACH(tmp, uvm_tree, &map->rbhead) {
403 		if (tmp->space != uvm_rb_subtree_space(tmp)) {
404 			printf("%s: space %lx != %lx\n",
405 			    name, (ulong)tmp->space,
406 			    (ulong)uvm_rb_subtree_space(tmp));
407 			goto error;
408 		}
409 		if (trtmp != NULL && trtmp->start >= tmp->start) {
410 			printf("%s: corrupt: 0x%lx >= 0x%lx\n",
411 			    name, trtmp->start, tmp->start);
412 			goto error;
413 		}
414 		n++;
415 
416 		trtmp = tmp;
417 	}
418 
419 	if (n != map->nentries) {
420 		printf("%s: nentries: %d vs %d\n",
421 		    name, n, map->nentries);
422 		goto error;
423 	}
424 
425 	for (tmp = map->header.next; tmp && tmp != &map->header;
426 	    tmp = tmp->next, i++) {
427 		trtmp = RB_FIND(uvm_tree, &map->rbhead, tmp);
428 		if (trtmp != tmp) {
429 			printf("%s: lookup: %d: %p - %p: %p\n",
430 			    name, i, tmp, trtmp,
431 			    RB_PARENT(tmp, rb_entry));
432 			goto error;
433 		}
434 	}
435 
436 	return (0);
437  error:
438 #if defined(DDB) && __GNUC__ < 4
439 	/* handy breakpoint location for error case */
440 	__asm(".globl treesanity_label\ntreesanity_label:");
441 #endif
442 	return (-1);
443 }
444 
445 #ifdef DIAGNOSTIC
446 static struct vm_map *uvm_kmapent_map(struct vm_map_entry *);
447 #endif
448 
449 /*
450  * uvm_mapent_alloc: allocate a map entry
451  */
452 
453 static struct vm_map_entry *
454 uvm_mapent_alloc(struct vm_map *map, int flags)
455 {
456 	struct vm_map_entry *me;
457 	int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK;
458 	UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist);
459 
460 	if (VM_MAP_USE_KMAPENT(map)) {
461 		me = uvm_kmapent_alloc(map, flags);
462 	} else {
463 		me = pool_get(&uvm_map_entry_pool, pflags);
464 		if (__predict_false(me == NULL))
465 			return NULL;
466 		me->flags = 0;
467 	}
468 
469 	UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me,
470 	    ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0);
471 	return (me);
472 }
473 
474 /*
475  * uvm_mapent_alloc_split: allocate a map entry for clipping.
476  */
477 
478 static struct vm_map_entry *
479 uvm_mapent_alloc_split(struct vm_map *map,
480     const struct vm_map_entry *old_entry, int flags,
481     struct uvm_mapent_reservation *umr)
482 {
483 	struct vm_map_entry *me;
484 
485 	KASSERT(!VM_MAP_USE_KMAPENT(map) ||
486 	    (old_entry->flags & UVM_MAP_QUANTUM) || !UMR_EMPTY(umr));
487 
488 	if (old_entry->flags & UVM_MAP_QUANTUM) {
489 		int s;
490 		struct vm_map_kernel *vmk = vm_map_to_kernel(map);
491 
492 		s = splvm();
493 		simple_lock(&uvm.kentry_lock);
494 		me = vmk->vmk_merged_entries;
495 		KASSERT(me);
496 		vmk->vmk_merged_entries = me->next;
497 		simple_unlock(&uvm.kentry_lock);
498 		splx(s);
499 		KASSERT(me->flags & UVM_MAP_QUANTUM);
500 	} else {
501 		me = uvm_mapent_alloc(map, flags);
502 	}
503 
504 	return me;
505 }
506 
507 /*
508  * uvm_mapent_free: free map entry
509  */
510 
511 static void
512 uvm_mapent_free(struct vm_map_entry *me)
513 {
514 	UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist);
515 
516 	UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
517 		me, me->flags, 0, 0);
518 	if (me->flags & UVM_MAP_KERNEL) {
519 		uvm_kmapent_free(me);
520 	} else {
521 		pool_put(&uvm_map_entry_pool, me);
522 	}
523 }
524 
525 /*
526  * uvm_mapent_free_merged: free merged map entry
527  *
528  * => keep the entry if needed.
529  * => caller shouldn't hold map locked if VM_MAP_USE_KMAPENT(map) is true.
530  */
531 
532 static void
533 uvm_mapent_free_merged(struct vm_map *map, struct vm_map_entry *me)
534 {
535 
536 	KASSERT(!(me->flags & UVM_MAP_KERNEL) || uvm_kmapent_map(me) == map);
537 
538 	if (me->flags & UVM_MAP_QUANTUM) {
539 		/*
540 		 * keep this entry for later splitting.
541 		 */
542 		struct vm_map_kernel *vmk;
543 		int s;
544 
545 		KASSERT(VM_MAP_IS_KERNEL(map));
546 		KASSERT(!VM_MAP_USE_KMAPENT(map) ||
547 		    (me->flags & UVM_MAP_KERNEL));
548 
549 		vmk = vm_map_to_kernel(map);
550 		s = splvm();
551 		simple_lock(&uvm.kentry_lock);
552 		me->next = vmk->vmk_merged_entries;
553 		vmk->vmk_merged_entries = me;
554 		simple_unlock(&uvm.kentry_lock);
555 		splx(s);
556 	} else {
557 		uvm_mapent_free(me);
558 	}
559 }
560 
561 /*
562  * uvm_mapent_copy: copy a map entry, preserving flags
563  */
564 
565 static __inline void
566 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst)
567 {
568 
569 	memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) -
570 	    ((char *)src));
571 }
572 
573 /*
574  * uvm_map_entry_unwire: unwire a map entry
575  *
576  * => map should be locked by caller
577  */
578 
579 static __inline void
580 uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry)
581 {
582 
583 	entry->wired_count = 0;
584 	uvm_fault_unwire_locked(map, entry->start, entry->end);
585 }
586 
587 
588 /*
589  * wrapper for calling amap_ref()
590  */
591 static __inline void
592 uvm_map_reference_amap(struct vm_map_entry *entry, int flags)
593 {
594 
595 	amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff,
596 	    (entry->end - entry->start) >> PAGE_SHIFT, flags);
597 }
598 
599 
600 /*
601  * wrapper for calling amap_unref()
602  */
603 static __inline void
604 uvm_map_unreference_amap(struct vm_map_entry *entry, int flags)
605 {
606 
607 	amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff,
608 	    (entry->end - entry->start) >> PAGE_SHIFT, flags);
609 }
610 
611 
612 /*
613  * uvm_map_init: init mapping system at boot time.   note that we allocate
614  * and init the static pool of struct vm_map_entry *'s for the kernel here.
615  */
616 
617 void
618 uvm_map_init(void)
619 {
620 #if defined(UVMHIST)
621 	static struct uvm_history_ent maphistbuf[100];
622 	static struct uvm_history_ent pdhistbuf[100];
623 #endif
624 
625 	/*
626 	 * first, init logging system.
627 	 */
628 
629 	UVMHIST_FUNC("uvm_map_init");
630 	UVMHIST_INIT_STATIC(maphist, maphistbuf);
631 	UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
632 	UVMHIST_CALLED(maphist);
633 	UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
634 
635 	/*
636 	 * initialize the global lock for kernel map entry.
637 	 *
638 	 * XXX is it worth to have per-map lock instead?
639 	 */
640 
641 	simple_lock_init(&uvm.kentry_lock);
642 }
643 
644 /*
645  * clippers
646  */
647 
648 /*
649  * uvm_map_clip_start: ensure that the entry begins at or after
650  *	the starting address, if it doesn't we split the entry.
651  *
652  * => caller should use UVM_MAP_CLIP_START macro rather than calling
653  *    this directly
654  * => map must be locked by caller
655  */
656 
657 void
658 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry,
659     vaddr_t start, struct uvm_mapent_reservation *umr)
660 {
661 	struct vm_map_entry *new_entry;
662 	vaddr_t new_adj;
663 
664 	/* uvm_map_simplify_entry(map, entry); */ /* XXX */
665 
666 	uvm_tree_sanity(map, "clip_start entry");
667 
668 	/*
669 	 * Split off the front portion.  note that we must insert the new
670 	 * entry BEFORE this one, so that this entry has the specified
671 	 * starting address.
672 	 */
673 	new_entry = uvm_mapent_alloc_split(map, entry, 0, umr);
674 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
675 
676 	new_entry->end = start;
677 	new_adj = start - new_entry->start;
678 	if (entry->object.uvm_obj)
679 		entry->offset += new_adj;	/* shift start over */
680 
681 	/* Does not change order for the RB tree */
682 	entry->start = start;
683 
684 	if (new_entry->aref.ar_amap) {
685 		amap_splitref(&new_entry->aref, &entry->aref, new_adj);
686 	}
687 
688 	uvm_map_entry_link(map, entry->prev, new_entry);
689 
690 	if (UVM_ET_ISSUBMAP(entry)) {
691 		/* ... unlikely to happen, but play it safe */
692 		 uvm_map_reference(new_entry->object.sub_map);
693 	} else {
694 		if (UVM_ET_ISOBJ(entry) &&
695 		    entry->object.uvm_obj->pgops &&
696 		    entry->object.uvm_obj->pgops->pgo_reference)
697 			entry->object.uvm_obj->pgops->pgo_reference(
698 			    entry->object.uvm_obj);
699 	}
700 
701 	uvm_tree_sanity(map, "clip_start leave");
702 }
703 
704 /*
705  * uvm_map_clip_end: ensure that the entry ends at or before
706  *	the ending address, if it does't we split the reference
707  *
708  * => caller should use UVM_MAP_CLIP_END macro rather than calling
709  *    this directly
710  * => map must be locked by caller
711  */
712 
713 void
714 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end,
715     struct uvm_mapent_reservation *umr)
716 {
717 	struct vm_map_entry *	new_entry;
718 	vaddr_t new_adj; /* #bytes we move start forward */
719 
720 	uvm_tree_sanity(map, "clip_end entry");
721 
722 	/*
723 	 *	Create a new entry and insert it
724 	 *	AFTER the specified entry
725 	 */
726 	new_entry = uvm_mapent_alloc_split(map, entry, 0, umr);
727 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
728 
729 	new_entry->start = entry->end = end;
730 	new_adj = end - entry->start;
731 	if (new_entry->object.uvm_obj)
732 		new_entry->offset += new_adj;
733 
734 	if (entry->aref.ar_amap)
735 		amap_splitref(&entry->aref, &new_entry->aref, new_adj);
736 
737 	uvm_rb_fixup(map, entry);
738 
739 	uvm_map_entry_link(map, entry, new_entry);
740 
741 	if (UVM_ET_ISSUBMAP(entry)) {
742 		/* ... unlikely to happen, but play it safe */
743 		uvm_map_reference(new_entry->object.sub_map);
744 	} else {
745 		if (UVM_ET_ISOBJ(entry) &&
746 		    entry->object.uvm_obj->pgops &&
747 		    entry->object.uvm_obj->pgops->pgo_reference)
748 			entry->object.uvm_obj->pgops->pgo_reference(
749 			    entry->object.uvm_obj);
750 	}
751 
752 	uvm_tree_sanity(map, "clip_end leave");
753 }
754 
755 
756 /*
757  *   M A P   -   m a i n   e n t r y   p o i n t
758  */
759 /*
760  * uvm_map: establish a valid mapping in a map
761  *
762  * => assume startp is page aligned.
763  * => assume size is a multiple of PAGE_SIZE.
764  * => assume sys_mmap provides enough of a "hint" to have us skip
765  *	over text/data/bss area.
766  * => map must be unlocked (we will lock it)
767  * => <uobj,uoffset> value meanings (4 cases):
768  *	 [1] <NULL,uoffset>		== uoffset is a hint for PMAP_PREFER
769  *	 [2] <NULL,UVM_UNKNOWN_OFFSET>	== don't PMAP_PREFER
770  *	 [3] <uobj,uoffset>		== normal mapping
771  *	 [4] <uobj,UVM_UNKNOWN_OFFSET>	== uvm_map finds offset based on VA
772  *
773  *    case [4] is for kernel mappings where we don't know the offset until
774  *    we've found a virtual address.   note that kernel object offsets are
775  *    always relative to vm_map_min(kernel_map).
776  *
777  * => if `align' is non-zero, we align the virtual address to the specified
778  *	alignment.
779  *	this is provided as a mechanism for large pages.
780  *
781  * => XXXCDC: need way to map in external amap?
782  */
783 
784 int
785 uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size,
786     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags)
787 {
788 	struct uvm_map_args args;
789 	struct vm_map_entry *new_entry;
790 	int error;
791 
792 	KASSERT((flags & UVM_FLAG_QUANTUM) == 0 || VM_MAP_IS_KERNEL(map));
793 	KASSERT((size & PAGE_MASK) == 0);
794 
795 	/*
796 	 * for pager_map, allocate the new entry first to avoid sleeping
797 	 * for memory while we have the map locked.
798 	 *
799 	 * besides, because we allocates entries for in-kernel maps
800 	 * a bit differently (cf. uvm_kmapent_alloc/free), we need to
801 	 * allocate them before locking the map.
802 	 */
803 
804 	new_entry = NULL;
805 	if (VM_MAP_USE_KMAPENT(map) || (flags & UVM_FLAG_QUANTUM) ||
806 	    map == pager_map) {
807 		new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT));
808 		if (__predict_false(new_entry == NULL))
809 			return ENOMEM;
810 		if (flags & UVM_FLAG_QUANTUM)
811 			new_entry->flags |= UVM_MAP_QUANTUM;
812 	}
813 	if (map == pager_map)
814 		flags |= UVM_FLAG_NOMERGE;
815 
816 	error = uvm_map_prepare(map, *startp, size, uobj, uoffset, align,
817 	    flags, &args);
818 	if (!error) {
819 		error = uvm_map_enter(map, &args, new_entry);
820 		*startp = args.uma_start;
821 	} else if (new_entry) {
822 		uvm_mapent_free(new_entry);
823 	}
824 
825 #if defined(DEBUG)
826 	if (!error && VM_MAP_IS_KERNEL(map)) {
827 		uvm_km_check_empty(*startp, *startp + size,
828 		    (map->flags & VM_MAP_INTRSAFE) != 0);
829 	}
830 #endif /* defined(DEBUG) */
831 
832 	return error;
833 }
834 
835 int
836 uvm_map_prepare(struct vm_map *map, vaddr_t start, vsize_t size,
837     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags,
838     struct uvm_map_args *args)
839 {
840 	struct vm_map_entry *prev_entry;
841 	vm_prot_t prot = UVM_PROTECTION(flags);
842 	vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
843 
844 	UVMHIST_FUNC("uvm_map_prepare");
845 	UVMHIST_CALLED(maphist);
846 
847 	UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)",
848 	    map, start, size, flags);
849 	UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
850 
851 	/*
852 	 * detect a popular device driver bug.
853 	 */
854 
855 	KASSERT(doing_shutdown || curlwp != NULL ||
856 	    (map->flags & VM_MAP_INTRSAFE));
857 
858 	/*
859 	 * zero-sized mapping doesn't make any sense.
860 	 */
861 	KASSERT(size > 0);
862 
863 	KASSERT((~flags & (UVM_FLAG_NOWAIT | UVM_FLAG_WAITVA)) != 0);
864 
865 	uvm_tree_sanity(map, "map entry");
866 
867 	/*
868 	 * check sanity of protection code
869 	 */
870 
871 	if ((prot & maxprot) != prot) {
872 		UVMHIST_LOG(maphist, "<- prot. failure:  prot=0x%x, max=0x%x",
873 		prot, maxprot,0,0);
874 		return EACCES;
875 	}
876 
877 	/*
878 	 * figure out where to put new VM range
879 	 */
880 
881 retry:
882 	if (vm_map_lock_try(map) == FALSE) {
883 		if (flags & UVM_FLAG_TRYLOCK) {
884 			return EAGAIN;
885 		}
886 		vm_map_lock(map); /* could sleep here */
887 	}
888 	if ((prev_entry = uvm_map_findspace(map, start, size, &start,
889 	    uobj, uoffset, align, flags)) == NULL) {
890 		unsigned int timestamp;
891 
892 		if ((flags & UVM_FLAG_WAITVA) == 0) {
893 			UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",
894 			    0,0,0,0);
895 			vm_map_unlock(map);
896 			return ENOMEM;
897 		}
898 		timestamp = map->timestamp;
899 		UVMHIST_LOG(maphist,"waiting va timestamp=0x%x",
900 			    timestamp,0,0,0);
901 		simple_lock(&map->flags_lock);
902 		map->flags |= VM_MAP_WANTVA;
903 		simple_unlock(&map->flags_lock);
904 		vm_map_unlock(map);
905 
906 		/*
907 		 * wait until someone does unmap.
908 		 * XXX fragile locking
909 		 */
910 
911 		simple_lock(&map->flags_lock);
912 		while ((map->flags & VM_MAP_WANTVA) != 0 &&
913 		   map->timestamp == timestamp) {
914 			ltsleep(&map->header, PVM, "vmmapva", 0,
915 			    &map->flags_lock);
916 		}
917 		simple_unlock(&map->flags_lock);
918 		goto retry;
919 	}
920 
921 #ifdef PMAP_GROWKERNEL
922 	/*
923 	 * If the kernel pmap can't map the requested space,
924 	 * then allocate more resources for it.
925 	 */
926 	if (map == kernel_map && uvm_maxkaddr < (start + size))
927 		uvm_maxkaddr = pmap_growkernel(start + size);
928 #endif
929 
930 	UVMCNT_INCR(uvm_map_call);
931 
932 	/*
933 	 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
934 	 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET.   in
935 	 * either case we want to zero it  before storing it in the map entry
936 	 * (because it looks strange and confusing when debugging...)
937 	 *
938 	 * if uobj is not null
939 	 *   if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
940 	 *      and we do not need to change uoffset.
941 	 *   if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
942 	 *      now (based on the starting address of the map).   this case is
943 	 *      for kernel object mappings where we don't know the offset until
944 	 *      the virtual address is found (with uvm_map_findspace).   the
945 	 *      offset is the distance we are from the start of the map.
946 	 */
947 
948 	if (uobj == NULL) {
949 		uoffset = 0;
950 	} else {
951 		if (uoffset == UVM_UNKNOWN_OFFSET) {
952 			KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj));
953 			uoffset = start - vm_map_min(kernel_map);
954 		}
955 	}
956 
957 	args->uma_flags = flags;
958 	args->uma_prev = prev_entry;
959 	args->uma_start = start;
960 	args->uma_size = size;
961 	args->uma_uobj = uobj;
962 	args->uma_uoffset = uoffset;
963 
964 	return 0;
965 }
966 
967 int
968 uvm_map_enter(struct vm_map *map, const struct uvm_map_args *args,
969     struct vm_map_entry *new_entry)
970 {
971 	struct vm_map_entry *prev_entry = args->uma_prev;
972 	struct vm_map_entry *dead = NULL;
973 
974 	const uvm_flag_t flags = args->uma_flags;
975 	const vm_prot_t prot = UVM_PROTECTION(flags);
976 	const vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
977 	const vm_inherit_t inherit = UVM_INHERIT(flags);
978 	const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ?
979 	    AMAP_EXTEND_NOWAIT : 0;
980 	const int advice = UVM_ADVICE(flags);
981 	const int meflagval = (flags & UVM_FLAG_QUANTUM) ?
982 	    UVM_MAP_QUANTUM : 0;
983 
984 	vaddr_t start = args->uma_start;
985 	vsize_t size = args->uma_size;
986 	struct uvm_object *uobj = args->uma_uobj;
987 	voff_t uoffset = args->uma_uoffset;
988 
989 	const int kmap = (vm_map_pmap(map) == pmap_kernel());
990 	int merged = 0;
991 	int error;
992 	int newetype;
993 
994 	UVMHIST_FUNC("uvm_map_enter");
995 	UVMHIST_CALLED(maphist);
996 
997 	UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)",
998 	    map, start, size, flags);
999 	UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
1000 
1001 	if (flags & UVM_FLAG_QUANTUM) {
1002 		KASSERT(new_entry);
1003 		KASSERT(new_entry->flags & UVM_MAP_QUANTUM);
1004 	}
1005 
1006 	if (uobj)
1007 		newetype = UVM_ET_OBJ;
1008 	else
1009 		newetype = 0;
1010 
1011 	if (flags & UVM_FLAG_COPYONW) {
1012 		newetype |= UVM_ET_COPYONWRITE;
1013 		if ((flags & UVM_FLAG_OVERLAY) == 0)
1014 			newetype |= UVM_ET_NEEDSCOPY;
1015 	}
1016 
1017 	/*
1018 	 * try and insert in map by extending previous entry, if possible.
1019 	 * XXX: we don't try and pull back the next entry.   might be useful
1020 	 * for a stack, but we are currently allocating our stack in advance.
1021 	 */
1022 
1023 	if (flags & UVM_FLAG_NOMERGE)
1024 		goto nomerge;
1025 
1026 	if (prev_entry->end == start &&
1027 	    prev_entry != &map->header &&
1028 	    UVM_ET_ISCOMPATIBLE(prev_entry, newetype, uobj, meflagval,
1029 	    prot, maxprot, inherit, advice, 0)) {
1030 
1031 		if (uobj && prev_entry->offset +
1032 		    (prev_entry->end - prev_entry->start) != uoffset)
1033 			goto forwardmerge;
1034 
1035 		/*
1036 		 * can't extend a shared amap.  note: no need to lock amap to
1037 		 * look at refs since we don't care about its exact value.
1038 		 * if it is one (i.e. we have only reference) it will stay there
1039 		 */
1040 
1041 		if (prev_entry->aref.ar_amap &&
1042 		    amap_refs(prev_entry->aref.ar_amap) != 1) {
1043 			goto forwardmerge;
1044 		}
1045 
1046 		if (prev_entry->aref.ar_amap) {
1047 			error = amap_extend(prev_entry, size,
1048 			    amapwaitflag | AMAP_EXTEND_FORWARDS);
1049 			if (error)
1050 				goto nomerge;
1051 		}
1052 
1053 		if (kmap)
1054 			UVMCNT_INCR(map_kbackmerge);
1055 		else
1056 			UVMCNT_INCR(map_ubackmerge);
1057 		UVMHIST_LOG(maphist,"  starting back merge", 0, 0, 0, 0);
1058 
1059 		/*
1060 		 * drop our reference to uobj since we are extending a reference
1061 		 * that we already have (the ref count can not drop to zero).
1062 		 */
1063 
1064 		if (uobj && uobj->pgops->pgo_detach)
1065 			uobj->pgops->pgo_detach(uobj);
1066 
1067 		prev_entry->end += size;
1068 		uvm_rb_fixup(map, prev_entry);
1069 
1070 		uvm_tree_sanity(map, "map backmerged");
1071 
1072 		UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
1073 		merged++;
1074 	}
1075 
1076 forwardmerge:
1077 	if (prev_entry->next->start == (start + size) &&
1078 	    prev_entry->next != &map->header &&
1079 	    UVM_ET_ISCOMPATIBLE(prev_entry->next, newetype, uobj, meflagval,
1080 	    prot, maxprot, inherit, advice, 0)) {
1081 
1082 		if (uobj && prev_entry->next->offset != uoffset + size)
1083 			goto nomerge;
1084 
1085 		/*
1086 		 * can't extend a shared amap.  note: no need to lock amap to
1087 		 * look at refs since we don't care about its exact value.
1088 		 * if it is one (i.e. we have only reference) it will stay there.
1089 		 *
1090 		 * note that we also can't merge two amaps, so if we
1091 		 * merged with the previous entry which has an amap,
1092 		 * and the next entry also has an amap, we give up.
1093 		 *
1094 		 * Interesting cases:
1095 		 * amap, new, amap -> give up second merge (single fwd extend)
1096 		 * amap, new, none -> double forward extend (extend again here)
1097 		 * none, new, amap -> double backward extend (done here)
1098 		 * uobj, new, amap -> single backward extend (done here)
1099 		 *
1100 		 * XXX should we attempt to deal with someone refilling
1101 		 * the deallocated region between two entries that are
1102 		 * backed by the same amap (ie, arefs is 2, "prev" and
1103 		 * "next" refer to it, and adding this allocation will
1104 		 * close the hole, thus restoring arefs to 1 and
1105 		 * deallocating the "next" vm_map_entry)?  -- @@@
1106 		 */
1107 
1108 		if (prev_entry->next->aref.ar_amap &&
1109 		    (amap_refs(prev_entry->next->aref.ar_amap) != 1 ||
1110 		     (merged && prev_entry->aref.ar_amap))) {
1111 			goto nomerge;
1112 		}
1113 
1114 		if (merged) {
1115 			/*
1116 			 * Try to extend the amap of the previous entry to
1117 			 * cover the next entry as well.  If it doesn't work
1118 			 * just skip on, don't actually give up, since we've
1119 			 * already completed the back merge.
1120 			 */
1121 			if (prev_entry->aref.ar_amap) {
1122 				if (amap_extend(prev_entry,
1123 				    prev_entry->next->end -
1124 				    prev_entry->next->start,
1125 				    amapwaitflag | AMAP_EXTEND_FORWARDS))
1126 					goto nomerge;
1127 			}
1128 
1129 			/*
1130 			 * Try to extend the amap of the *next* entry
1131 			 * back to cover the new allocation *and* the
1132 			 * previous entry as well (the previous merge
1133 			 * didn't have an amap already otherwise we
1134 			 * wouldn't be checking here for an amap).  If
1135 			 * it doesn't work just skip on, again, don't
1136 			 * actually give up, since we've already
1137 			 * completed the back merge.
1138 			 */
1139 			else if (prev_entry->next->aref.ar_amap) {
1140 				if (amap_extend(prev_entry->next,
1141 				    prev_entry->end -
1142 				    prev_entry->start,
1143 				    amapwaitflag | AMAP_EXTEND_BACKWARDS))
1144 					goto nomerge;
1145 			}
1146 		} else {
1147 			/*
1148 			 * Pull the next entry's amap backwards to cover this
1149 			 * new allocation.
1150 			 */
1151 			if (prev_entry->next->aref.ar_amap) {
1152 				error = amap_extend(prev_entry->next, size,
1153 				    amapwaitflag | AMAP_EXTEND_BACKWARDS);
1154 				if (error)
1155 					goto nomerge;
1156 			}
1157 		}
1158 
1159 		if (merged) {
1160 			if (kmap) {
1161 				UVMCNT_DECR(map_kbackmerge);
1162 				UVMCNT_INCR(map_kbimerge);
1163 			} else {
1164 				UVMCNT_DECR(map_ubackmerge);
1165 				UVMCNT_INCR(map_ubimerge);
1166 			}
1167 		} else {
1168 			if (kmap)
1169 				UVMCNT_INCR(map_kforwmerge);
1170 			else
1171 				UVMCNT_INCR(map_uforwmerge);
1172 		}
1173 		UVMHIST_LOG(maphist,"  starting forward merge", 0, 0, 0, 0);
1174 
1175 		/*
1176 		 * drop our reference to uobj since we are extending a reference
1177 		 * that we already have (the ref count can not drop to zero).
1178 		 * (if merged, we've already detached)
1179 		 */
1180 		if (uobj && uobj->pgops->pgo_detach && !merged)
1181 			uobj->pgops->pgo_detach(uobj);
1182 
1183 		if (merged) {
1184 			dead = prev_entry->next;
1185 			prev_entry->end = dead->end;
1186 			uvm_map_entry_unlink(map, dead);
1187 			if (dead->aref.ar_amap != NULL) {
1188 				prev_entry->aref = dead->aref;
1189 				dead->aref.ar_amap = NULL;
1190 			}
1191 		} else {
1192 			prev_entry->next->start -= size;
1193 			if (prev_entry != &map->header)
1194 				uvm_rb_fixup(map, prev_entry);
1195 			if (uobj)
1196 				prev_entry->next->offset = uoffset;
1197 		}
1198 
1199 		uvm_tree_sanity(map, "map forwardmerged");
1200 
1201 		UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0);
1202 		merged++;
1203 	}
1204 
1205 nomerge:
1206 	if (!merged) {
1207 		UVMHIST_LOG(maphist,"  allocating new map entry", 0, 0, 0, 0);
1208 		if (kmap)
1209 			UVMCNT_INCR(map_knomerge);
1210 		else
1211 			UVMCNT_INCR(map_unomerge);
1212 
1213 		/*
1214 		 * allocate new entry and link it in.
1215 		 */
1216 
1217 		if (new_entry == NULL) {
1218 			new_entry = uvm_mapent_alloc(map,
1219 				(flags & UVM_FLAG_NOWAIT));
1220 			if (__predict_false(new_entry == NULL)) {
1221 				error = ENOMEM;
1222 				goto done;
1223 			}
1224 		}
1225 		new_entry->start = start;
1226 		new_entry->end = new_entry->start + size;
1227 		new_entry->object.uvm_obj = uobj;
1228 		new_entry->offset = uoffset;
1229 
1230 		new_entry->etype = newetype;
1231 
1232 		if (flags & UVM_FLAG_NOMERGE) {
1233 			new_entry->flags |= UVM_MAP_NOMERGE;
1234 		}
1235 
1236 		new_entry->protection = prot;
1237 		new_entry->max_protection = maxprot;
1238 		new_entry->inheritance = inherit;
1239 		new_entry->wired_count = 0;
1240 		new_entry->advice = advice;
1241 		if (flags & UVM_FLAG_OVERLAY) {
1242 
1243 			/*
1244 			 * to_add: for BSS we overallocate a little since we
1245 			 * are likely to extend
1246 			 */
1247 
1248 			vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
1249 				UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
1250 			struct vm_amap *amap = amap_alloc(size, to_add,
1251 			    (flags & UVM_FLAG_NOWAIT) ? M_NOWAIT : M_WAITOK);
1252 			if (__predict_false(amap == NULL)) {
1253 				error = ENOMEM;
1254 				goto done;
1255 			}
1256 			new_entry->aref.ar_pageoff = 0;
1257 			new_entry->aref.ar_amap = amap;
1258 		} else {
1259 			new_entry->aref.ar_pageoff = 0;
1260 			new_entry->aref.ar_amap = NULL;
1261 		}
1262 		uvm_map_entry_link(map, prev_entry, new_entry);
1263 
1264 		/*
1265 		 * Update the free space hint
1266 		 */
1267 
1268 		if ((map->first_free == prev_entry) &&
1269 		    (prev_entry->end >= new_entry->start))
1270 			map->first_free = new_entry;
1271 
1272 		new_entry = NULL;
1273 	}
1274 
1275 	map->size += size;
1276 
1277 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1278 
1279 	error = 0;
1280 done:
1281 	vm_map_unlock(map);
1282 	if (new_entry) {
1283 		if (error == 0) {
1284 			KDASSERT(merged);
1285 			uvm_mapent_free_merged(map, new_entry);
1286 		} else {
1287 			uvm_mapent_free(new_entry);
1288 		}
1289 	}
1290 	if (dead) {
1291 		KDASSERT(merged);
1292 		uvm_mapent_free_merged(map, dead);
1293 	}
1294 	return error;
1295 }
1296 
1297 /*
1298  * uvm_map_lookup_entry: find map entry at or before an address
1299  *
1300  * => map must at least be read-locked by caller
1301  * => entry is returned in "entry"
1302  * => return value is true if address is in the returned entry
1303  */
1304 
1305 boolean_t
1306 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address,
1307     struct vm_map_entry **entry	/* OUT */)
1308 {
1309 	struct vm_map_entry *cur;
1310 	boolean_t use_tree = FALSE;
1311 	UVMHIST_FUNC("uvm_map_lookup_entry");
1312 	UVMHIST_CALLED(maphist);
1313 
1314 	UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
1315 	    map, address, entry, 0);
1316 
1317 	/*
1318 	 * start looking either from the head of the
1319 	 * list, or from the hint.
1320 	 */
1321 
1322 	simple_lock(&map->hint_lock);
1323 	cur = map->hint;
1324 	simple_unlock(&map->hint_lock);
1325 
1326 	if (cur == &map->header)
1327 		cur = cur->next;
1328 
1329 	UVMCNT_INCR(uvm_mlk_call);
1330 	if (address >= cur->start) {
1331 
1332 		/*
1333 		 * go from hint to end of list.
1334 		 *
1335 		 * but first, make a quick check to see if
1336 		 * we are already looking at the entry we
1337 		 * want (which is usually the case).
1338 		 * note also that we don't need to save the hint
1339 		 * here... it is the same hint (unless we are
1340 		 * at the header, in which case the hint didn't
1341 		 * buy us anything anyway).
1342 		 */
1343 
1344 		if (cur != &map->header && cur->end > address) {
1345 			UVMCNT_INCR(uvm_mlk_hint);
1346 			*entry = cur;
1347 			UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
1348 			    cur, 0, 0, 0);
1349 			return (TRUE);
1350 		}
1351 
1352 		if (map->nentries > 30)
1353 			use_tree = TRUE;
1354 	} else {
1355 
1356 		/*
1357 		 * invalid hint.  use tree.
1358 		 */
1359 		use_tree = TRUE;
1360 	}
1361 
1362 	uvm_tree_sanity(map, __func__);
1363 
1364 	if (use_tree) {
1365 		struct vm_map_entry *prev = &map->header;
1366 		cur = RB_ROOT(&map->rbhead);
1367 
1368 		/*
1369 		 * Simple lookup in the tree.  Happens when the hint is
1370 		 * invalid, or nentries reach a threshold.
1371 		 */
1372 		while (cur) {
1373 			if (address >= cur->start) {
1374 				if (address < cur->end) {
1375 					*entry = cur;
1376 					goto got;
1377 				}
1378 				prev = cur;
1379 				cur = RB_RIGHT(cur, rb_entry);
1380 			} else
1381 				cur = RB_LEFT(cur, rb_entry);
1382 		}
1383 		*entry = prev;
1384 		goto failed;
1385 	}
1386 
1387 	/*
1388 	 * search linearly
1389 	 */
1390 
1391 	while (cur != &map->header) {
1392 		if (cur->end > address) {
1393 			if (address >= cur->start) {
1394 				/*
1395 				 * save this lookup for future
1396 				 * hints, and return
1397 				 */
1398 
1399 				*entry = cur;
1400 got:
1401 				SAVE_HINT(map, map->hint, *entry);
1402 				UVMHIST_LOG(maphist,"<- search got it (0x%x)",
1403 					cur, 0, 0, 0);
1404 				KDASSERT((*entry)->start <= address);
1405 				KDASSERT(address < (*entry)->end);
1406 				return (TRUE);
1407 			}
1408 			break;
1409 		}
1410 		cur = cur->next;
1411 	}
1412 	*entry = cur->prev;
1413 failed:
1414 	SAVE_HINT(map, map->hint, *entry);
1415 	UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
1416 	KDASSERT((*entry) == &map->header || (*entry)->end <= address);
1417 	KDASSERT((*entry)->next == &map->header ||
1418 	    address < (*entry)->next->start);
1419 	return (FALSE);
1420 }
1421 
1422 /*
1423  * See if the range between start and start + length fits in the gap
1424  * entry->next->start and entry->end.  Returns 1 if fits, 0 if doesn't
1425  * fit, and -1 address wraps around.
1426  */
1427 static int
1428 uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset,
1429     vsize_t align, int topdown, struct vm_map_entry *entry)
1430 {
1431 	vaddr_t end;
1432 
1433 #ifdef PMAP_PREFER
1434 	/*
1435 	 * push start address forward as needed to avoid VAC alias problems.
1436 	 * we only do this if a valid offset is specified.
1437 	 */
1438 
1439 	if (uoffset != UVM_UNKNOWN_OFFSET)
1440 		PMAP_PREFER(uoffset, start, length, topdown);
1441 #endif
1442 	if (align != 0) {
1443 		if ((*start & (align - 1)) != 0) {
1444 			if (topdown)
1445 				*start &= ~(align - 1);
1446 			else
1447 				*start = roundup(*start, align);
1448 		}
1449 		/*
1450 		 * XXX Should we PMAP_PREFER() here again?
1451 		 * eh...i think we're okay
1452 		 */
1453 	}
1454 
1455 	/*
1456 	 * Find the end of the proposed new region.  Be sure we didn't
1457 	 * wrap around the address; if so, we lose.  Otherwise, if the
1458 	 * proposed new region fits before the next entry, we win.
1459 	 */
1460 
1461 	end = *start + length;
1462 	if (end < *start)
1463 		return (-1);
1464 
1465 	if (entry->next->start >= end && *start >= entry->end)
1466 		return (1);
1467 
1468 	return (0);
1469 }
1470 
1471 /*
1472  * uvm_map_findspace: find "length" sized space in "map".
1473  *
1474  * => "hint" is a hint about where we want it, unless UVM_FLAG_FIXED is
1475  *	set in "flags" (in which case we insist on using "hint").
1476  * => "result" is VA returned
1477  * => uobj/uoffset are to be used to handle VAC alignment, if required
1478  * => if "align" is non-zero, we attempt to align to that value.
1479  * => caller must at least have read-locked map
1480  * => returns NULL on failure, or pointer to prev. map entry if success
1481  * => note this is a cross between the old vm_map_findspace and vm_map_find
1482  */
1483 
1484 struct vm_map_entry *
1485 uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length,
1486     vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset,
1487     vsize_t align, int flags)
1488 {
1489 	struct vm_map_entry *entry;
1490 	struct vm_map_entry *child, *prev, *tmp;
1491 	vaddr_t orig_hint;
1492 	const int topdown = map->flags & VM_MAP_TOPDOWN;
1493 	UVMHIST_FUNC("uvm_map_findspace");
1494 	UVMHIST_CALLED(maphist);
1495 
1496 	UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)",
1497 	    map, hint, length, flags);
1498 	KASSERT((align & (align - 1)) == 0);
1499 	KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0);
1500 
1501 	uvm_tree_sanity(map, "map_findspace entry");
1502 
1503 	/*
1504 	 * remember the original hint.  if we are aligning, then we
1505 	 * may have to try again with no alignment constraint if
1506 	 * we fail the first time.
1507 	 */
1508 
1509 	orig_hint = hint;
1510 	if (hint < vm_map_min(map)) {	/* check ranges ... */
1511 		if (flags & UVM_FLAG_FIXED) {
1512 			UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
1513 			return (NULL);
1514 		}
1515 		hint = vm_map_min(map);
1516 	}
1517 	if (hint > vm_map_max(map)) {
1518 		UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
1519 		    hint, vm_map_min(map), vm_map_max(map), 0);
1520 		return (NULL);
1521 	}
1522 
1523 	/*
1524 	 * Look for the first possible address; if there's already
1525 	 * something at this address, we have to start after it.
1526 	 */
1527 
1528 	/*
1529 	 * @@@: there are four, no, eight cases to consider.
1530 	 *
1531 	 * 0: found,     fixed,     bottom up -> fail
1532 	 * 1: found,     fixed,     top down  -> fail
1533 	 * 2: found,     not fixed, bottom up -> start after entry->end,
1534 	 *                                       loop up
1535 	 * 3: found,     not fixed, top down  -> start before entry->start,
1536 	 *                                       loop down
1537 	 * 4: not found, fixed,     bottom up -> check entry->next->start, fail
1538 	 * 5: not found, fixed,     top down  -> check entry->next->start, fail
1539 	 * 6: not found, not fixed, bottom up -> check entry->next->start,
1540 	 *                                       loop up
1541 	 * 7: not found, not fixed, top down  -> check entry->next->start,
1542 	 *                                       loop down
1543 	 *
1544 	 * as you can see, it reduces to roughly five cases, and that
1545 	 * adding top down mapping only adds one unique case (without
1546 	 * it, there would be four cases).
1547 	 */
1548 
1549 	if ((flags & UVM_FLAG_FIXED) == 0 && hint == vm_map_min(map)) {
1550 		entry = map->first_free;
1551 	} else {
1552 		if (uvm_map_lookup_entry(map, hint, &entry)) {
1553 			/* "hint" address already in use ... */
1554 			if (flags & UVM_FLAG_FIXED) {
1555 				UVMHIST_LOG(maphist, "<- fixed & VA in use",
1556 				    0, 0, 0, 0);
1557 				return (NULL);
1558 			}
1559 			if (topdown)
1560 				/* Start from lower gap. */
1561 				entry = entry->prev;
1562 		} else if (flags & UVM_FLAG_FIXED) {
1563 			if (entry->next->start >= hint + length &&
1564 			    hint + length > hint)
1565 				goto found;
1566 
1567 			/* "hint" address is gap but too small */
1568 			UVMHIST_LOG(maphist, "<- fixed mapping failed",
1569 			    0, 0, 0, 0);
1570 			return (NULL); /* only one shot at it ... */
1571 		} else {
1572 			/*
1573 			 * See if given hint fits in this gap.
1574 			 */
1575 			switch (uvm_map_space_avail(&hint, length,
1576 			    uoffset, align, topdown, entry)) {
1577 			case 1:
1578 				goto found;
1579 			case -1:
1580 				goto wraparound;
1581 			}
1582 
1583 			if (topdown) {
1584 				/*
1585 				 * Still there is a chance to fit
1586 				 * if hint > entry->end.
1587 				 */
1588 			} else {
1589 				/* Start from higher gap. */
1590 				entry = entry->next;
1591 				if (entry == &map->header)
1592 					goto notfound;
1593 				goto nextgap;
1594 			}
1595 		}
1596 	}
1597 
1598 	/*
1599 	 * Note that all UVM_FLAGS_FIXED case is already handled.
1600 	 */
1601 	KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1602 
1603 	/* Try to find the space in the red-black tree */
1604 
1605 	/* Check slot before any entry */
1606 	hint = topdown ? entry->next->start - length : entry->end;
1607 	switch (uvm_map_space_avail(&hint, length, uoffset, align,
1608 	    topdown, entry)) {
1609 	case 1:
1610 		goto found;
1611 	case -1:
1612 		goto wraparound;
1613 	}
1614 
1615 nextgap:
1616 	KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1617 	/* If there is not enough space in the whole tree, we fail */
1618 	tmp = RB_ROOT(&map->rbhead);
1619 	if (tmp == NULL || tmp->space < length)
1620 		goto notfound;
1621 
1622 	prev = NULL; /* previous candidate */
1623 
1624 	/* Find an entry close to hint that has enough space */
1625 	for (; tmp;) {
1626 		KASSERT(tmp->next->start == tmp->end + tmp->ownspace);
1627 		if (topdown) {
1628 			if (tmp->next->start < hint + length &&
1629 			    (prev == NULL || tmp->end > prev->end)) {
1630 				if (tmp->ownspace >= length)
1631 					prev = tmp;
1632 				else if ((child = RB_LEFT(tmp, rb_entry))
1633 				    != NULL && child->space >= length)
1634 					prev = tmp;
1635 			}
1636 		} else {
1637 			if (tmp->end >= hint &&
1638 			    (prev == NULL || tmp->end < prev->end)) {
1639 				if (tmp->ownspace >= length)
1640 					prev = tmp;
1641 				else if ((child = RB_RIGHT(tmp, rb_entry))
1642 				    != NULL && child->space >= length)
1643 					prev = tmp;
1644 			}
1645 		}
1646 		if (tmp->next->start < hint + length)
1647 			child = RB_RIGHT(tmp, rb_entry);
1648 		else if (tmp->end > hint)
1649 			child = RB_LEFT(tmp, rb_entry);
1650 		else {
1651 			if (tmp->ownspace >= length)
1652 				break;
1653 			if (topdown)
1654 				child = RB_LEFT(tmp, rb_entry);
1655 			else
1656 				child = RB_RIGHT(tmp, rb_entry);
1657 		}
1658 		if (child == NULL || child->space < length)
1659 			break;
1660 		tmp = child;
1661 	}
1662 
1663 	if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) {
1664 		/*
1665 		 * Check if the entry that we found satifies the
1666 		 * space requirement
1667 		 */
1668 		if (topdown) {
1669 			if (hint > tmp->next->start - length)
1670 				hint = tmp->next->start - length;
1671 		} else {
1672 			if (hint < tmp->end)
1673 				hint = tmp->end;
1674 		}
1675 		switch (uvm_map_space_avail(&hint, length, uoffset, align,
1676 		    topdown, tmp)) {
1677 		case 1:
1678 			entry = tmp;
1679 			goto found;
1680 		case -1:
1681 			goto wraparound;
1682 		}
1683 		if (tmp->ownspace >= length)
1684 			goto listsearch;
1685 	}
1686 	if (prev == NULL)
1687 		goto notfound;
1688 
1689 	if (topdown) {
1690 		KASSERT(orig_hint >= prev->next->start - length ||
1691 		    prev->next->start - length > prev->next->start);
1692 		hint = prev->next->start - length;
1693 	} else {
1694 		KASSERT(orig_hint <= prev->end);
1695 		hint = prev->end;
1696 	}
1697 	switch (uvm_map_space_avail(&hint, length, uoffset, align,
1698 	    topdown, prev)) {
1699 	case 1:
1700 		entry = prev;
1701 		goto found;
1702 	case -1:
1703 		goto wraparound;
1704 	}
1705 	if (prev->ownspace >= length)
1706 		goto listsearch;
1707 
1708 	if (topdown)
1709 		tmp = RB_LEFT(prev, rb_entry);
1710 	else
1711 		tmp = RB_RIGHT(prev, rb_entry);
1712 	for (;;) {
1713 		KASSERT(tmp && tmp->space >= length);
1714 		if (topdown)
1715 			child = RB_RIGHT(tmp, rb_entry);
1716 		else
1717 			child = RB_LEFT(tmp, rb_entry);
1718 		if (child && child->space >= length) {
1719 			tmp = child;
1720 			continue;
1721 		}
1722 		if (tmp->ownspace >= length)
1723 			break;
1724 		if (topdown)
1725 			tmp = RB_LEFT(tmp, rb_entry);
1726 		else
1727 			tmp = RB_RIGHT(tmp, rb_entry);
1728 	}
1729 
1730 	if (topdown) {
1731 		KASSERT(orig_hint >= tmp->next->start - length ||
1732 		    tmp->next->start - length > tmp->next->start);
1733 		hint = tmp->next->start - length;
1734 	} else {
1735 		KASSERT(orig_hint <= tmp->end);
1736 		hint = tmp->end;
1737 	}
1738 	switch (uvm_map_space_avail(&hint, length, uoffset, align,
1739 	    topdown, tmp)) {
1740 	case 1:
1741 		entry = tmp;
1742 		goto found;
1743 	case -1:
1744 		goto wraparound;
1745 	}
1746 
1747 	/*
1748 	 * The tree fails to find an entry because of offset or alignment
1749 	 * restrictions.  Search the list instead.
1750 	 */
1751  listsearch:
1752 	/*
1753 	 * Look through the rest of the map, trying to fit a new region in
1754 	 * the gap between existing regions, or after the very last region.
1755 	 * note: entry->end = base VA of current gap,
1756 	 *	 entry->next->start = VA of end of current gap
1757 	 */
1758 
1759 	for (;;) {
1760 		/* Update hint for current gap. */
1761 		hint = topdown ? entry->next->start - length : entry->end;
1762 
1763 		/* See if it fits. */
1764 		switch (uvm_map_space_avail(&hint, length, uoffset, align,
1765 		    topdown, entry)) {
1766 		case 1:
1767 			goto found;
1768 		case -1:
1769 			goto wraparound;
1770 		}
1771 
1772 		/* Advance to next/previous gap */
1773 		if (topdown) {
1774 			if (entry == &map->header) {
1775 				UVMHIST_LOG(maphist, "<- failed (off start)",
1776 				    0,0,0,0);
1777 				goto notfound;
1778 			}
1779 			entry = entry->prev;
1780 		} else {
1781 			entry = entry->next;
1782 			if (entry == &map->header) {
1783 				UVMHIST_LOG(maphist, "<- failed (off end)",
1784 				    0,0,0,0);
1785 				goto notfound;
1786 			}
1787 		}
1788 	}
1789 
1790  found:
1791 	SAVE_HINT(map, map->hint, entry);
1792 	*result = hint;
1793 	UVMHIST_LOG(maphist,"<- got it!  (result=0x%x)", hint, 0,0,0);
1794 	KASSERT( topdown || hint >= orig_hint);
1795 	KASSERT(!topdown || hint <= orig_hint);
1796 	KASSERT(entry->end <= hint);
1797 	KASSERT(hint + length <= entry->next->start);
1798 	return (entry);
1799 
1800  wraparound:
1801 	UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0);
1802 
1803 	return (NULL);
1804 
1805  notfound:
1806 	UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0);
1807 
1808 	return (NULL);
1809 }
1810 
1811 /*
1812  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
1813  */
1814 
1815 /*
1816  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
1817  *
1818  * => caller must check alignment and size
1819  * => map must be locked by caller
1820  * => we return a list of map entries that we've remove from the map
1821  *    in "entry_list"
1822  */
1823 
1824 void
1825 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end,
1826     struct vm_map_entry **entry_list /* OUT */,
1827     struct uvm_mapent_reservation *umr, int flags)
1828 {
1829 	struct vm_map_entry *entry, *first_entry, *next;
1830 	vaddr_t len;
1831 	UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist);
1832 
1833 	UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
1834 	    map, start, end, 0);
1835 	VM_MAP_RANGE_CHECK(map, start, end);
1836 
1837 	uvm_tree_sanity(map, "unmap_remove entry");
1838 
1839 	/*
1840 	 * find first entry
1841 	 */
1842 
1843 	if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
1844 		/* clip and go... */
1845 		entry = first_entry;
1846 		UVM_MAP_CLIP_START(map, entry, start, umr);
1847 		/* critical!  prevents stale hint */
1848 		SAVE_HINT(map, entry, entry->prev);
1849 	} else {
1850 		entry = first_entry->next;
1851 	}
1852 
1853 	/*
1854 	 * Save the free space hint
1855 	 */
1856 
1857 	if (map->first_free->start >= start)
1858 		map->first_free = entry->prev;
1859 
1860 	/*
1861 	 * note: we now re-use first_entry for a different task.  we remove
1862 	 * a number of map entries from the map and save them in a linked
1863 	 * list headed by "first_entry".  once we remove them from the map
1864 	 * the caller should unlock the map and drop the references to the
1865 	 * backing objects [c.f. uvm_unmap_detach].  the object is to
1866 	 * separate unmapping from reference dropping.  why?
1867 	 *   [1] the map has to be locked for unmapping
1868 	 *   [2] the map need not be locked for reference dropping
1869 	 *   [3] dropping references may trigger pager I/O, and if we hit
1870 	 *       a pager that does synchronous I/O we may have to wait for it.
1871 	 *   [4] we would like all waiting for I/O to occur with maps unlocked
1872 	 *       so that we don't block other threads.
1873 	 */
1874 
1875 	first_entry = NULL;
1876 	*entry_list = NULL;
1877 
1878 	/*
1879 	 * break up the area into map entry sized regions and unmap.  note
1880 	 * that all mappings have to be removed before we can even consider
1881 	 * dropping references to amaps or VM objects (otherwise we could end
1882 	 * up with a mapping to a page on the free list which would be very bad)
1883 	 */
1884 
1885 	while ((entry != &map->header) && (entry->start < end)) {
1886 		KASSERT((entry->flags & UVM_MAP_FIRST) == 0);
1887 
1888 		UVM_MAP_CLIP_END(map, entry, end, umr);
1889 		next = entry->next;
1890 		len = entry->end - entry->start;
1891 
1892 		/*
1893 		 * unwire before removing addresses from the pmap; otherwise
1894 		 * unwiring will put the entries back into the pmap (XXX).
1895 		 */
1896 
1897 		if (VM_MAPENT_ISWIRED(entry)) {
1898 			uvm_map_entry_unwire(map, entry);
1899 		}
1900 		if (flags & UVM_FLAG_VAONLY) {
1901 
1902 			/* nothing */
1903 
1904 		} else if ((map->flags & VM_MAP_PAGEABLE) == 0) {
1905 
1906 			/*
1907 			 * if the map is non-pageable, any pages mapped there
1908 			 * must be wired and entered with pmap_kenter_pa(),
1909 			 * and we should free any such pages immediately.
1910 			 * this is mostly used for kmem_map and mb_map.
1911 			 */
1912 
1913 			if ((entry->flags & UVM_MAP_KMAPENT) == 0) {
1914 				uvm_km_pgremove_intrsafe(entry->start,
1915 				    entry->end);
1916 				pmap_kremove(entry->start, len);
1917 			}
1918 		} else if (UVM_ET_ISOBJ(entry) &&
1919 			   UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
1920 			KASSERT(vm_map_pmap(map) == pmap_kernel());
1921 
1922 			/*
1923 			 * note: kernel object mappings are currently used in
1924 			 * two ways:
1925 			 *  [1] "normal" mappings of pages in the kernel object
1926 			 *  [2] uvm_km_valloc'd allocations in which we
1927 			 *      pmap_enter in some non-kernel-object page
1928 			 *      (e.g. vmapbuf).
1929 			 *
1930 			 * for case [1], we need to remove the mapping from
1931 			 * the pmap and then remove the page from the kernel
1932 			 * object (because, once pages in a kernel object are
1933 			 * unmapped they are no longer needed, unlike, say,
1934 			 * a vnode where you might want the data to persist
1935 			 * until flushed out of a queue).
1936 			 *
1937 			 * for case [2], we need to remove the mapping from
1938 			 * the pmap.  there shouldn't be any pages at the
1939 			 * specified offset in the kernel object [but it
1940 			 * doesn't hurt to call uvm_km_pgremove just to be
1941 			 * safe?]
1942 			 *
1943 			 * uvm_km_pgremove currently does the following:
1944 			 *   for pages in the kernel object in range:
1945 			 *     - drops the swap slot
1946 			 *     - uvm_pagefree the page
1947 			 */
1948 
1949 			/*
1950 			 * remove mappings from pmap and drop the pages
1951 			 * from the object.  offsets are always relative
1952 			 * to vm_map_min(kernel_map).
1953 			 */
1954 
1955 			pmap_remove(pmap_kernel(), entry->start,
1956 			    entry->start + len);
1957 			uvm_km_pgremove(entry->start, entry->end);
1958 
1959 			/*
1960 			 * null out kernel_object reference, we've just
1961 			 * dropped it
1962 			 */
1963 
1964 			entry->etype &= ~UVM_ET_OBJ;
1965 			entry->object.uvm_obj = NULL;
1966 		} else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) {
1967 
1968 			/*
1969 			 * remove mappings the standard way.
1970 			 */
1971 
1972 			pmap_remove(map->pmap, entry->start, entry->end);
1973 		}
1974 
1975 #if defined(DEBUG)
1976 		if ((entry->flags & UVM_MAP_KMAPENT) == 0) {
1977 
1978 			/*
1979 			 * check if there's remaining mapping,
1980 			 * which is a bug in caller.
1981 			 */
1982 
1983 			vaddr_t va;
1984 			for (va = entry->start; va < entry->end;
1985 			    va += PAGE_SIZE) {
1986 				if (pmap_extract(vm_map_pmap(map), va, NULL)) {
1987 					panic("uvm_unmap_remove: has mapping");
1988 				}
1989 			}
1990 
1991 			if (VM_MAP_IS_KERNEL(map)) {
1992 				uvm_km_check_empty(entry->start, entry->end,
1993 				    (map->flags & VM_MAP_INTRSAFE) != 0);
1994 			}
1995 		}
1996 #endif /* defined(DEBUG) */
1997 
1998 		/*
1999 		 * remove entry from map and put it on our list of entries
2000 		 * that we've nuked.  then go to next entry.
2001 		 */
2002 
2003 		UVMHIST_LOG(maphist, "  removed map entry 0x%x", entry, 0, 0,0);
2004 
2005 		/* critical!  prevents stale hint */
2006 		SAVE_HINT(map, entry, entry->prev);
2007 
2008 		uvm_map_entry_unlink(map, entry);
2009 		KASSERT(map->size >= len);
2010 		map->size -= len;
2011 		entry->prev = NULL;
2012 		entry->next = first_entry;
2013 		first_entry = entry;
2014 		entry = next;
2015 	}
2016 	if ((map->flags & VM_MAP_DYING) == 0) {
2017 		pmap_update(vm_map_pmap(map));
2018 	}
2019 
2020 	uvm_tree_sanity(map, "unmap_remove leave");
2021 
2022 	/*
2023 	 * now we've cleaned up the map and are ready for the caller to drop
2024 	 * references to the mapped objects.
2025 	 */
2026 
2027 	*entry_list = first_entry;
2028 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
2029 
2030 	simple_lock(&map->flags_lock);
2031 	if (map->flags & VM_MAP_WANTVA) {
2032 		map->flags &= ~VM_MAP_WANTVA;
2033 		wakeup(&map->header);
2034 	}
2035 	simple_unlock(&map->flags_lock);
2036 }
2037 
2038 /*
2039  * uvm_unmap_detach: drop references in a chain of map entries
2040  *
2041  * => we will free the map entries as we traverse the list.
2042  */
2043 
2044 void
2045 uvm_unmap_detach(struct vm_map_entry *first_entry, int flags)
2046 {
2047 	struct vm_map_entry *next_entry;
2048 	UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
2049 
2050 	while (first_entry) {
2051 		KASSERT(!VM_MAPENT_ISWIRED(first_entry));
2052 		UVMHIST_LOG(maphist,
2053 		    "  detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
2054 		    first_entry, first_entry->aref.ar_amap,
2055 		    first_entry->object.uvm_obj,
2056 		    UVM_ET_ISSUBMAP(first_entry));
2057 
2058 		/*
2059 		 * drop reference to amap, if we've got one
2060 		 */
2061 
2062 		if (first_entry->aref.ar_amap)
2063 			uvm_map_unreference_amap(first_entry, flags);
2064 
2065 		/*
2066 		 * drop reference to our backing object, if we've got one
2067 		 */
2068 
2069 		KASSERT(!UVM_ET_ISSUBMAP(first_entry));
2070 		if (UVM_ET_ISOBJ(first_entry) &&
2071 		    first_entry->object.uvm_obj->pgops->pgo_detach) {
2072 			(*first_entry->object.uvm_obj->pgops->pgo_detach)
2073 				(first_entry->object.uvm_obj);
2074 		}
2075 		next_entry = first_entry->next;
2076 		uvm_mapent_free(first_entry);
2077 		first_entry = next_entry;
2078 	}
2079 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
2080 }
2081 
2082 /*
2083  *   E X T R A C T I O N   F U N C T I O N S
2084  */
2085 
2086 /*
2087  * uvm_map_reserve: reserve space in a vm_map for future use.
2088  *
2089  * => we reserve space in a map by putting a dummy map entry in the
2090  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
2091  * => map should be unlocked (we will write lock it)
2092  * => we return true if we were able to reserve space
2093  * => XXXCDC: should be inline?
2094  */
2095 
2096 int
2097 uvm_map_reserve(struct vm_map *map, vsize_t size,
2098     vaddr_t offset	/* hint for pmap_prefer */,
2099     vsize_t align	/* alignment hint */,
2100     vaddr_t *raddr	/* IN:hint, OUT: reserved VA */)
2101 {
2102 	UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
2103 
2104 	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
2105 	    map,size,offset,raddr);
2106 
2107 	size = round_page(size);
2108 	if (*raddr < vm_map_min(map))
2109 		*raddr = vm_map_min(map);		/* hint */
2110 
2111 	/*
2112 	 * reserve some virtual space.
2113 	 */
2114 
2115 	if (uvm_map(map, raddr, size, NULL, offset, 0,
2116 	    UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
2117 	    UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) {
2118 	    UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
2119 		return (FALSE);
2120 	}
2121 
2122 	UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
2123 	return (TRUE);
2124 }
2125 
2126 /*
2127  * uvm_map_replace: replace a reserved (blank) area of memory with
2128  * real mappings.
2129  *
2130  * => caller must WRITE-LOCK the map
2131  * => we return TRUE if replacement was a success
2132  * => we expect the newents chain to have nnewents entrys on it and
2133  *    we expect newents->prev to point to the last entry on the list
2134  * => note newents is allowed to be NULL
2135  */
2136 
2137 int
2138 uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end,
2139     struct vm_map_entry *newents, int nnewents)
2140 {
2141 	struct vm_map_entry *oldent, *last;
2142 
2143 	uvm_tree_sanity(map, "map_replace entry");
2144 
2145 	/*
2146 	 * first find the blank map entry at the specified address
2147 	 */
2148 
2149 	if (!uvm_map_lookup_entry(map, start, &oldent)) {
2150 		return (FALSE);
2151 	}
2152 
2153 	/*
2154 	 * check to make sure we have a proper blank entry
2155 	 */
2156 
2157 	if (oldent->start != start || oldent->end != end ||
2158 	    oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
2159 		return (FALSE);
2160 	}
2161 
2162 #ifdef DIAGNOSTIC
2163 
2164 	/*
2165 	 * sanity check the newents chain
2166 	 */
2167 
2168 	{
2169 		struct vm_map_entry *tmpent = newents;
2170 		int nent = 0;
2171 		vaddr_t cur = start;
2172 
2173 		while (tmpent) {
2174 			nent++;
2175 			if (tmpent->start < cur)
2176 				panic("uvm_map_replace1");
2177 			if (tmpent->start > tmpent->end || tmpent->end > end) {
2178 		printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
2179 			    tmpent->start, tmpent->end, end);
2180 				panic("uvm_map_replace2");
2181 			}
2182 			cur = tmpent->end;
2183 			if (tmpent->next) {
2184 				if (tmpent->next->prev != tmpent)
2185 					panic("uvm_map_replace3");
2186 			} else {
2187 				if (newents->prev != tmpent)
2188 					panic("uvm_map_replace4");
2189 			}
2190 			tmpent = tmpent->next;
2191 		}
2192 		if (nent != nnewents)
2193 			panic("uvm_map_replace5");
2194 	}
2195 #endif
2196 
2197 	/*
2198 	 * map entry is a valid blank!   replace it.   (this does all the
2199 	 * work of map entry link/unlink...).
2200 	 */
2201 
2202 	if (newents) {
2203 		last = newents->prev;
2204 
2205 		/* critical: flush stale hints out of map */
2206 		SAVE_HINT(map, map->hint, newents);
2207 		if (map->first_free == oldent)
2208 			map->first_free = last;
2209 
2210 		last->next = oldent->next;
2211 		last->next->prev = last;
2212 
2213 		/* Fix RB tree */
2214 		uvm_rb_remove(map, oldent);
2215 
2216 		newents->prev = oldent->prev;
2217 		newents->prev->next = newents;
2218 		map->nentries = map->nentries + (nnewents - 1);
2219 
2220 		/* Fixup the RB tree */
2221 		{
2222 			int i;
2223 			struct vm_map_entry *tmp;
2224 
2225 			tmp = newents;
2226 			for (i = 0; i < nnewents && tmp; i++) {
2227 				uvm_rb_insert(map, tmp);
2228 				tmp = tmp->next;
2229 			}
2230 		}
2231 	} else {
2232 
2233 		/* critical: flush stale hints out of map */
2234 		SAVE_HINT(map, map->hint, oldent->prev);
2235 		if (map->first_free == oldent)
2236 			map->first_free = oldent->prev;
2237 
2238 		/* NULL list of new entries: just remove the old one */
2239 		uvm_map_entry_unlink(map, oldent);
2240 	}
2241 
2242 	uvm_tree_sanity(map, "map_replace leave");
2243 
2244 	/*
2245 	 * now we can free the old blank entry, unlock the map and return.
2246 	 */
2247 
2248 	uvm_mapent_free(oldent);
2249 	return (TRUE);
2250 }
2251 
2252 /*
2253  * uvm_map_extract: extract a mapping from a map and put it somewhere
2254  *	(maybe removing the old mapping)
2255  *
2256  * => maps should be unlocked (we will write lock them)
2257  * => returns 0 on success, error code otherwise
2258  * => start must be page aligned
2259  * => len must be page sized
2260  * => flags:
2261  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
2262  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
2263  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
2264  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
2265  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
2266  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
2267  *             be used from within the kernel in a kernel level map <<<
2268  */
2269 
2270 int
2271 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len,
2272     struct vm_map *dstmap, vaddr_t *dstaddrp, int flags)
2273 {
2274 	vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge;
2275 	struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry,
2276 	    *deadentry, *oldentry;
2277 	vsize_t elen;
2278 	int nchain, error, copy_ok;
2279 	UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
2280 
2281 	UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
2282 	    len,0);
2283 	UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
2284 
2285 	uvm_tree_sanity(srcmap, "map_extract src enter");
2286 	uvm_tree_sanity(dstmap, "map_extract dst enter");
2287 
2288 	/*
2289 	 * step 0: sanity check: start must be on a page boundary, length
2290 	 * must be page sized.  can't ask for CONTIG/QREF if you asked for
2291 	 * REMOVE.
2292 	 */
2293 
2294 	KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0);
2295 	KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 ||
2296 		(flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0);
2297 
2298 	/*
2299 	 * step 1: reserve space in the target map for the extracted area
2300 	 */
2301 
2302 	dstaddr = vm_map_min(dstmap);
2303 	if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE)
2304 		return (ENOMEM);
2305 	*dstaddrp = dstaddr;	/* pass address back to caller */
2306 	UVMHIST_LOG(maphist, "  dstaddr=0x%x", dstaddr,0,0,0);
2307 
2308 	/*
2309 	 * step 2: setup for the extraction process loop by init'ing the
2310 	 * map entry chain, locking src map, and looking up the first useful
2311 	 * entry in the map.
2312 	 */
2313 
2314 	end = start + len;
2315 	newend = dstaddr + len;
2316 	chain = endchain = NULL;
2317 	nchain = 0;
2318 	vm_map_lock(srcmap);
2319 
2320 	if (uvm_map_lookup_entry(srcmap, start, &entry)) {
2321 
2322 		/* "start" is within an entry */
2323 		if (flags & UVM_EXTRACT_QREF) {
2324 
2325 			/*
2326 			 * for quick references we don't clip the entry, so
2327 			 * the entry may map space "before" the starting
2328 			 * virtual address... this is the "fudge" factor
2329 			 * (which can be non-zero only the first time
2330 			 * through the "while" loop in step 3).
2331 			 */
2332 
2333 			fudge = start - entry->start;
2334 		} else {
2335 
2336 			/*
2337 			 * normal reference: we clip the map to fit (thus
2338 			 * fudge is zero)
2339 			 */
2340 
2341 			UVM_MAP_CLIP_START(srcmap, entry, start, NULL);
2342 			SAVE_HINT(srcmap, srcmap->hint, entry->prev);
2343 			fudge = 0;
2344 		}
2345 	} else {
2346 
2347 		/* "start" is not within an entry ... skip to next entry */
2348 		if (flags & UVM_EXTRACT_CONTIG) {
2349 			error = EINVAL;
2350 			goto bad;    /* definite hole here ... */
2351 		}
2352 
2353 		entry = entry->next;
2354 		fudge = 0;
2355 	}
2356 
2357 	/* save values from srcmap for step 6 */
2358 	orig_entry = entry;
2359 	orig_fudge = fudge;
2360 
2361 	/*
2362 	 * step 3: now start looping through the map entries, extracting
2363 	 * as we go.
2364 	 */
2365 
2366 	while (entry->start < end && entry != &srcmap->header) {
2367 
2368 		/* if we are not doing a quick reference, clip it */
2369 		if ((flags & UVM_EXTRACT_QREF) == 0)
2370 			UVM_MAP_CLIP_END(srcmap, entry, end, NULL);
2371 
2372 		/* clear needs_copy (allow chunking) */
2373 		if (UVM_ET_ISNEEDSCOPY(entry)) {
2374 			amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
2375 			if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
2376 				error = ENOMEM;
2377 				goto bad;
2378 			}
2379 
2380 			/* amap_copy could clip (during chunk)!  update fudge */
2381 			if (fudge) {
2382 				fudge = start - entry->start;
2383 				orig_fudge = fudge;
2384 			}
2385 		}
2386 
2387 		/* calculate the offset of this from "start" */
2388 		oldoffset = (entry->start + fudge) - start;
2389 
2390 		/* allocate a new map entry */
2391 		newentry = uvm_mapent_alloc(dstmap, 0);
2392 		if (newentry == NULL) {
2393 			error = ENOMEM;
2394 			goto bad;
2395 		}
2396 
2397 		/* set up new map entry */
2398 		newentry->next = NULL;
2399 		newentry->prev = endchain;
2400 		newentry->start = dstaddr + oldoffset;
2401 		newentry->end =
2402 		    newentry->start + (entry->end - (entry->start + fudge));
2403 		if (newentry->end > newend || newentry->end < newentry->start)
2404 			newentry->end = newend;
2405 		newentry->object.uvm_obj = entry->object.uvm_obj;
2406 		if (newentry->object.uvm_obj) {
2407 			if (newentry->object.uvm_obj->pgops->pgo_reference)
2408 				newentry->object.uvm_obj->pgops->
2409 				    pgo_reference(newentry->object.uvm_obj);
2410 				newentry->offset = entry->offset + fudge;
2411 		} else {
2412 			newentry->offset = 0;
2413 		}
2414 		newentry->etype = entry->etype;
2415 		newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
2416 			entry->max_protection : entry->protection;
2417 		newentry->max_protection = entry->max_protection;
2418 		newentry->inheritance = entry->inheritance;
2419 		newentry->wired_count = 0;
2420 		newentry->aref.ar_amap = entry->aref.ar_amap;
2421 		if (newentry->aref.ar_amap) {
2422 			newentry->aref.ar_pageoff =
2423 			    entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
2424 			uvm_map_reference_amap(newentry, AMAP_SHARED |
2425 			    ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
2426 		} else {
2427 			newentry->aref.ar_pageoff = 0;
2428 		}
2429 		newentry->advice = entry->advice;
2430 
2431 		/* now link it on the chain */
2432 		nchain++;
2433 		if (endchain == NULL) {
2434 			chain = endchain = newentry;
2435 		} else {
2436 			endchain->next = newentry;
2437 			endchain = newentry;
2438 		}
2439 
2440 		/* end of 'while' loop! */
2441 		if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
2442 		    (entry->next == &srcmap->header ||
2443 		    entry->next->start != entry->end)) {
2444 			error = EINVAL;
2445 			goto bad;
2446 		}
2447 		entry = entry->next;
2448 		fudge = 0;
2449 	}
2450 
2451 	/*
2452 	 * step 4: close off chain (in format expected by uvm_map_replace)
2453 	 */
2454 
2455 	if (chain)
2456 		chain->prev = endchain;
2457 
2458 	/*
2459 	 * step 5: attempt to lock the dest map so we can pmap_copy.
2460 	 * note usage of copy_ok:
2461 	 *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
2462 	 *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
2463 	 */
2464 
2465 	if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
2466 		copy_ok = 1;
2467 		if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
2468 		    nchain)) {
2469 			if (srcmap != dstmap)
2470 				vm_map_unlock(dstmap);
2471 			error = EIO;
2472 			goto bad;
2473 		}
2474 	} else {
2475 		copy_ok = 0;
2476 		/* replace defered until step 7 */
2477 	}
2478 
2479 	/*
2480 	 * step 6: traverse the srcmap a second time to do the following:
2481 	 *  - if we got a lock on the dstmap do pmap_copy
2482 	 *  - if UVM_EXTRACT_REMOVE remove the entries
2483 	 * we make use of orig_entry and orig_fudge (saved in step 2)
2484 	 */
2485 
2486 	if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
2487 
2488 		/* purge possible stale hints from srcmap */
2489 		if (flags & UVM_EXTRACT_REMOVE) {
2490 			SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev);
2491 			if (srcmap->first_free->start >= start)
2492 				srcmap->first_free = orig_entry->prev;
2493 		}
2494 
2495 		entry = orig_entry;
2496 		fudge = orig_fudge;
2497 		deadentry = NULL;	/* for UVM_EXTRACT_REMOVE */
2498 
2499 		while (entry->start < end && entry != &srcmap->header) {
2500 			if (copy_ok) {
2501 				oldoffset = (entry->start + fudge) - start;
2502 				elen = MIN(end, entry->end) -
2503 				    (entry->start + fudge);
2504 				pmap_copy(dstmap->pmap, srcmap->pmap,
2505 				    dstaddr + oldoffset, elen,
2506 				    entry->start + fudge);
2507 			}
2508 
2509 			/* we advance "entry" in the following if statement */
2510 			if (flags & UVM_EXTRACT_REMOVE) {
2511 				pmap_remove(srcmap->pmap, entry->start,
2512 						entry->end);
2513 				oldentry = entry;	/* save entry */
2514 				entry = entry->next;	/* advance */
2515 				uvm_map_entry_unlink(srcmap, oldentry);
2516 							/* add to dead list */
2517 				oldentry->next = deadentry;
2518 				deadentry = oldentry;
2519 			} else {
2520 				entry = entry->next;		/* advance */
2521 			}
2522 
2523 			/* end of 'while' loop */
2524 			fudge = 0;
2525 		}
2526 		pmap_update(srcmap->pmap);
2527 
2528 		/*
2529 		 * unlock dstmap.  we will dispose of deadentry in
2530 		 * step 7 if needed
2531 		 */
2532 
2533 		if (copy_ok && srcmap != dstmap)
2534 			vm_map_unlock(dstmap);
2535 
2536 	} else {
2537 		deadentry = NULL;
2538 	}
2539 
2540 	/*
2541 	 * step 7: we are done with the source map, unlock.   if copy_ok
2542 	 * is 0 then we have not replaced the dummy mapping in dstmap yet
2543 	 * and we need to do so now.
2544 	 */
2545 
2546 	vm_map_unlock(srcmap);
2547 	if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
2548 		uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
2549 
2550 	/* now do the replacement if we didn't do it in step 5 */
2551 	if (copy_ok == 0) {
2552 		vm_map_lock(dstmap);
2553 		error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
2554 		    nchain);
2555 		vm_map_unlock(dstmap);
2556 
2557 		if (error == FALSE) {
2558 			error = EIO;
2559 			goto bad2;
2560 		}
2561 	}
2562 
2563 	uvm_tree_sanity(srcmap, "map_extract src leave");
2564 	uvm_tree_sanity(dstmap, "map_extract dst leave");
2565 
2566 	return (0);
2567 
2568 	/*
2569 	 * bad: failure recovery
2570 	 */
2571 bad:
2572 	vm_map_unlock(srcmap);
2573 bad2:			/* src already unlocked */
2574 	if (chain)
2575 		uvm_unmap_detach(chain,
2576 		    (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
2577 
2578 	uvm_tree_sanity(srcmap, "map_extract src err leave");
2579 	uvm_tree_sanity(dstmap, "map_extract dst err leave");
2580 
2581 	uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
2582 	return (error);
2583 }
2584 
2585 /* end of extraction functions */
2586 
2587 /*
2588  * uvm_map_submap: punch down part of a map into a submap
2589  *
2590  * => only the kernel_map is allowed to be submapped
2591  * => the purpose of submapping is to break up the locking granularity
2592  *	of a larger map
2593  * => the range specified must have been mapped previously with a uvm_map()
2594  *	call [with uobj==NULL] to create a blank map entry in the main map.
2595  *	[And it had better still be blank!]
2596  * => maps which contain submaps should never be copied or forked.
2597  * => to remove a submap, use uvm_unmap() on the main map
2598  *	and then uvm_map_deallocate() the submap.
2599  * => main map must be unlocked.
2600  * => submap must have been init'd and have a zero reference count.
2601  *	[need not be locked as we don't actually reference it]
2602  */
2603 
2604 int
2605 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end,
2606     struct vm_map *submap)
2607 {
2608 	struct vm_map_entry *entry;
2609 	struct uvm_mapent_reservation umr;
2610 	int error;
2611 
2612 	uvm_mapent_reserve(map, &umr, 2, 0);
2613 
2614 	vm_map_lock(map);
2615 	VM_MAP_RANGE_CHECK(map, start, end);
2616 
2617 	if (uvm_map_lookup_entry(map, start, &entry)) {
2618 		UVM_MAP_CLIP_START(map, entry, start, &umr);
2619 		UVM_MAP_CLIP_END(map, entry, end, &umr);	/* to be safe */
2620 	} else {
2621 		entry = NULL;
2622 	}
2623 
2624 	if (entry != NULL &&
2625 	    entry->start == start && entry->end == end &&
2626 	    entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
2627 	    !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
2628 		entry->etype |= UVM_ET_SUBMAP;
2629 		entry->object.sub_map = submap;
2630 		entry->offset = 0;
2631 		uvm_map_reference(submap);
2632 		error = 0;
2633 	} else {
2634 		error = EINVAL;
2635 	}
2636 	vm_map_unlock(map);
2637 
2638 	uvm_mapent_unreserve(map, &umr);
2639 
2640 	return error;
2641 }
2642 
2643 /*
2644  * uvm_map_setup_kernel: init in-kernel map
2645  *
2646  * => map must not be in service yet.
2647  */
2648 
2649 void
2650 uvm_map_setup_kernel(struct vm_map_kernel *map,
2651     vaddr_t vmin, vaddr_t vmax, int flags)
2652 {
2653 
2654 	uvm_map_setup(&map->vmk_map, vmin, vmax, flags);
2655 
2656 	LIST_INIT(&map->vmk_kentry_free);
2657 	map->vmk_merged_entries = NULL;
2658 }
2659 
2660 
2661 /*
2662  * uvm_map_protect: change map protection
2663  *
2664  * => set_max means set max_protection.
2665  * => map must be unlocked.
2666  */
2667 
2668 #define MASK(entry)	(UVM_ET_ISCOPYONWRITE(entry) ? \
2669 			 ~VM_PROT_WRITE : VM_PROT_ALL)
2670 
2671 int
2672 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end,
2673     vm_prot_t new_prot, boolean_t set_max)
2674 {
2675 	struct vm_map_entry *current, *entry;
2676 	int error = 0;
2677 	UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
2678 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
2679 		    map, start, end, new_prot);
2680 
2681 	vm_map_lock(map);
2682 	VM_MAP_RANGE_CHECK(map, start, end);
2683 	if (uvm_map_lookup_entry(map, start, &entry)) {
2684 		UVM_MAP_CLIP_START(map, entry, start, NULL);
2685 	} else {
2686 		entry = entry->next;
2687 	}
2688 
2689 	/*
2690 	 * make a first pass to check for protection violations.
2691 	 */
2692 
2693 	current = entry;
2694 	while ((current != &map->header) && (current->start < end)) {
2695 		if (UVM_ET_ISSUBMAP(current)) {
2696 			error = EINVAL;
2697 			goto out;
2698 		}
2699 		if ((new_prot & current->max_protection) != new_prot) {
2700 			error = EACCES;
2701 			goto out;
2702 		}
2703 		/*
2704 		 * Don't allow VM_PROT_EXECUTE to be set on entries that
2705 		 * point to vnodes that are associated with a NOEXEC file
2706 		 * system.
2707 		 */
2708 		if (UVM_ET_ISOBJ(current) &&
2709 		    UVM_OBJ_IS_VNODE(current->object.uvm_obj)) {
2710 			struct vnode *vp =
2711 			    (struct vnode *) current->object.uvm_obj;
2712 
2713 			if ((new_prot & VM_PROT_EXECUTE) != 0 &&
2714 			    (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
2715 				error = EACCES;
2716 				goto out;
2717 			}
2718 		}
2719 		current = current->next;
2720 	}
2721 
2722 	/* go back and fix up protections (no need to clip this time). */
2723 
2724 	current = entry;
2725 	while ((current != &map->header) && (current->start < end)) {
2726 		vm_prot_t old_prot;
2727 
2728 		UVM_MAP_CLIP_END(map, current, end, NULL);
2729 		old_prot = current->protection;
2730 		if (set_max)
2731 			current->protection =
2732 			    (current->max_protection = new_prot) & old_prot;
2733 		else
2734 			current->protection = new_prot;
2735 
2736 		/*
2737 		 * update physical map if necessary.  worry about copy-on-write
2738 		 * here -- CHECK THIS XXX
2739 		 */
2740 
2741 		if (current->protection != old_prot) {
2742 			/* update pmap! */
2743 			pmap_protect(map->pmap, current->start, current->end,
2744 			    current->protection & MASK(entry));
2745 
2746 			/*
2747 			 * If this entry points at a vnode, and the
2748 			 * protection includes VM_PROT_EXECUTE, mark
2749 			 * the vnode as VEXECMAP.
2750 			 */
2751 			if (UVM_ET_ISOBJ(current)) {
2752 				struct uvm_object *uobj =
2753 				    current->object.uvm_obj;
2754 
2755 				if (UVM_OBJ_IS_VNODE(uobj) &&
2756 				    (current->protection & VM_PROT_EXECUTE))
2757 					vn_markexec((struct vnode *) uobj);
2758 			}
2759 		}
2760 
2761 		/*
2762 		 * If the map is configured to lock any future mappings,
2763 		 * wire this entry now if the old protection was VM_PROT_NONE
2764 		 * and the new protection is not VM_PROT_NONE.
2765 		 */
2766 
2767 		if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
2768 		    VM_MAPENT_ISWIRED(entry) == 0 &&
2769 		    old_prot == VM_PROT_NONE &&
2770 		    new_prot != VM_PROT_NONE) {
2771 			if (uvm_map_pageable(map, entry->start,
2772 			    entry->end, FALSE,
2773 			    UVM_LK_ENTER|UVM_LK_EXIT) != 0) {
2774 
2775 				/*
2776 				 * If locking the entry fails, remember the
2777 				 * error if it's the first one.  Note we
2778 				 * still continue setting the protection in
2779 				 * the map, but will return the error
2780 				 * condition regardless.
2781 				 *
2782 				 * XXX Ignore what the actual error is,
2783 				 * XXX just call it a resource shortage
2784 				 * XXX so that it doesn't get confused
2785 				 * XXX what uvm_map_protect() itself would
2786 				 * XXX normally return.
2787 				 */
2788 
2789 				error = ENOMEM;
2790 			}
2791 		}
2792 		current = current->next;
2793 	}
2794 	pmap_update(map->pmap);
2795 
2796  out:
2797 	vm_map_unlock(map);
2798 
2799 	UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0);
2800 	return error;
2801 }
2802 
2803 #undef  MASK
2804 
2805 /*
2806  * uvm_map_inherit: set inheritance code for range of addrs in map.
2807  *
2808  * => map must be unlocked
2809  * => note that the inherit code is used during a "fork".  see fork
2810  *	code for details.
2811  */
2812 
2813 int
2814 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end,
2815     vm_inherit_t new_inheritance)
2816 {
2817 	struct vm_map_entry *entry, *temp_entry;
2818 	UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
2819 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
2820 	    map, start, end, new_inheritance);
2821 
2822 	switch (new_inheritance) {
2823 	case MAP_INHERIT_NONE:
2824 	case MAP_INHERIT_COPY:
2825 	case MAP_INHERIT_SHARE:
2826 		break;
2827 	default:
2828 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2829 		return EINVAL;
2830 	}
2831 
2832 	vm_map_lock(map);
2833 	VM_MAP_RANGE_CHECK(map, start, end);
2834 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
2835 		entry = temp_entry;
2836 		UVM_MAP_CLIP_START(map, entry, start, NULL);
2837 	}  else {
2838 		entry = temp_entry->next;
2839 	}
2840 	while ((entry != &map->header) && (entry->start < end)) {
2841 		UVM_MAP_CLIP_END(map, entry, end, NULL);
2842 		entry->inheritance = new_inheritance;
2843 		entry = entry->next;
2844 	}
2845 	vm_map_unlock(map);
2846 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
2847 	return 0;
2848 }
2849 
2850 /*
2851  * uvm_map_advice: set advice code for range of addrs in map.
2852  *
2853  * => map must be unlocked
2854  */
2855 
2856 int
2857 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice)
2858 {
2859 	struct vm_map_entry *entry, *temp_entry;
2860 	UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
2861 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
2862 	    map, start, end, new_advice);
2863 
2864 	vm_map_lock(map);
2865 	VM_MAP_RANGE_CHECK(map, start, end);
2866 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
2867 		entry = temp_entry;
2868 		UVM_MAP_CLIP_START(map, entry, start, NULL);
2869 	} else {
2870 		entry = temp_entry->next;
2871 	}
2872 
2873 	/*
2874 	 * XXXJRT: disallow holes?
2875 	 */
2876 
2877 	while ((entry != &map->header) && (entry->start < end)) {
2878 		UVM_MAP_CLIP_END(map, entry, end, NULL);
2879 
2880 		switch (new_advice) {
2881 		case MADV_NORMAL:
2882 		case MADV_RANDOM:
2883 		case MADV_SEQUENTIAL:
2884 			/* nothing special here */
2885 			break;
2886 
2887 		default:
2888 			vm_map_unlock(map);
2889 			UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2890 			return EINVAL;
2891 		}
2892 		entry->advice = new_advice;
2893 		entry = entry->next;
2894 	}
2895 
2896 	vm_map_unlock(map);
2897 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
2898 	return 0;
2899 }
2900 
2901 /*
2902  * uvm_map_pageable: sets the pageability of a range in a map.
2903  *
2904  * => wires map entries.  should not be used for transient page locking.
2905  *	for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
2906  * => regions sepcified as not pageable require lock-down (wired) memory
2907  *	and page tables.
2908  * => map must never be read-locked
2909  * => if islocked is TRUE, map is already write-locked
2910  * => we always unlock the map, since we must downgrade to a read-lock
2911  *	to call uvm_fault_wire()
2912  * => XXXCDC: check this and try and clean it up.
2913  */
2914 
2915 int
2916 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end,
2917     boolean_t new_pageable, int lockflags)
2918 {
2919 	struct vm_map_entry *entry, *start_entry, *failed_entry;
2920 	int rv;
2921 #ifdef DIAGNOSTIC
2922 	u_int timestamp_save;
2923 #endif
2924 	UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
2925 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
2926 		    map, start, end, new_pageable);
2927 	KASSERT(map->flags & VM_MAP_PAGEABLE);
2928 
2929 	if ((lockflags & UVM_LK_ENTER) == 0)
2930 		vm_map_lock(map);
2931 	VM_MAP_RANGE_CHECK(map, start, end);
2932 
2933 	/*
2934 	 * only one pageability change may take place at one time, since
2935 	 * uvm_fault_wire assumes it will be called only once for each
2936 	 * wiring/unwiring.  therefore, we have to make sure we're actually
2937 	 * changing the pageability for the entire region.  we do so before
2938 	 * making any changes.
2939 	 */
2940 
2941 	if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
2942 		if ((lockflags & UVM_LK_EXIT) == 0)
2943 			vm_map_unlock(map);
2944 
2945 		UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0);
2946 		return EFAULT;
2947 	}
2948 	entry = start_entry;
2949 
2950 	/*
2951 	 * handle wiring and unwiring separately.
2952 	 */
2953 
2954 	if (new_pageable) {		/* unwire */
2955 		UVM_MAP_CLIP_START(map, entry, start, NULL);
2956 
2957 		/*
2958 		 * unwiring.  first ensure that the range to be unwired is
2959 		 * really wired down and that there are no holes.
2960 		 */
2961 
2962 		while ((entry != &map->header) && (entry->start < end)) {
2963 			if (entry->wired_count == 0 ||
2964 			    (entry->end < end &&
2965 			     (entry->next == &map->header ||
2966 			      entry->next->start > entry->end))) {
2967 				if ((lockflags & UVM_LK_EXIT) == 0)
2968 					vm_map_unlock(map);
2969 				UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0);
2970 				return EINVAL;
2971 			}
2972 			entry = entry->next;
2973 		}
2974 
2975 		/*
2976 		 * POSIX 1003.1b - a single munlock call unlocks a region,
2977 		 * regardless of the number of mlock calls made on that
2978 		 * region.
2979 		 */
2980 
2981 		entry = start_entry;
2982 		while ((entry != &map->header) && (entry->start < end)) {
2983 			UVM_MAP_CLIP_END(map, entry, end, NULL);
2984 			if (VM_MAPENT_ISWIRED(entry))
2985 				uvm_map_entry_unwire(map, entry);
2986 			entry = entry->next;
2987 		}
2988 		if ((lockflags & UVM_LK_EXIT) == 0)
2989 			vm_map_unlock(map);
2990 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2991 		return 0;
2992 	}
2993 
2994 	/*
2995 	 * wire case: in two passes [XXXCDC: ugly block of code here]
2996 	 *
2997 	 * 1: holding the write lock, we create any anonymous maps that need
2998 	 *    to be created.  then we clip each map entry to the region to
2999 	 *    be wired and increment its wiring count.
3000 	 *
3001 	 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
3002 	 *    in the pages for any newly wired area (wired_count == 1).
3003 	 *
3004 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
3005 	 *    deadlock with another thread that may have faulted on one of
3006 	 *    the pages to be wired (it would mark the page busy, blocking
3007 	 *    us, then in turn block on the map lock that we hold).  because
3008 	 *    of problems in the recursive lock package, we cannot upgrade
3009 	 *    to a write lock in vm_map_lookup.  thus, any actions that
3010 	 *    require the write lock must be done beforehand.  because we
3011 	 *    keep the read lock on the map, the copy-on-write status of the
3012 	 *    entries we modify here cannot change.
3013 	 */
3014 
3015 	while ((entry != &map->header) && (entry->start < end)) {
3016 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3017 
3018 			/*
3019 			 * perform actions of vm_map_lookup that need the
3020 			 * write lock on the map: create an anonymous map
3021 			 * for a copy-on-write region, or an anonymous map
3022 			 * for a zero-fill region.  (XXXCDC: submap case
3023 			 * ok?)
3024 			 */
3025 
3026 			if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
3027 				if (UVM_ET_ISNEEDSCOPY(entry) &&
3028 				    ((entry->max_protection & VM_PROT_WRITE) ||
3029 				     (entry->object.uvm_obj == NULL))) {
3030 					amap_copy(map, entry, M_WAITOK, TRUE,
3031 					    start, end);
3032 					/* XXXCDC: wait OK? */
3033 				}
3034 			}
3035 		}
3036 		UVM_MAP_CLIP_START(map, entry, start, NULL);
3037 		UVM_MAP_CLIP_END(map, entry, end, NULL);
3038 		entry->wired_count++;
3039 
3040 		/*
3041 		 * Check for holes
3042 		 */
3043 
3044 		if (entry->protection == VM_PROT_NONE ||
3045 		    (entry->end < end &&
3046 		     (entry->next == &map->header ||
3047 		      entry->next->start > entry->end))) {
3048 
3049 			/*
3050 			 * found one.  amap creation actions do not need to
3051 			 * be undone, but the wired counts need to be restored.
3052 			 */
3053 
3054 			while (entry != &map->header && entry->end > start) {
3055 				entry->wired_count--;
3056 				entry = entry->prev;
3057 			}
3058 			if ((lockflags & UVM_LK_EXIT) == 0)
3059 				vm_map_unlock(map);
3060 			UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
3061 			return EINVAL;
3062 		}
3063 		entry = entry->next;
3064 	}
3065 
3066 	/*
3067 	 * Pass 2.
3068 	 */
3069 
3070 #ifdef DIAGNOSTIC
3071 	timestamp_save = map->timestamp;
3072 #endif
3073 	vm_map_busy(map);
3074 	vm_map_downgrade(map);
3075 
3076 	rv = 0;
3077 	entry = start_entry;
3078 	while (entry != &map->header && entry->start < end) {
3079 		if (entry->wired_count == 1) {
3080 			rv = uvm_fault_wire(map, entry->start, entry->end,
3081 			    VM_FAULT_WIREMAX, entry->max_protection);
3082 			if (rv) {
3083 
3084 				/*
3085 				 * wiring failed.  break out of the loop.
3086 				 * we'll clean up the map below, once we
3087 				 * have a write lock again.
3088 				 */
3089 
3090 				break;
3091 			}
3092 		}
3093 		entry = entry->next;
3094 	}
3095 
3096 	if (rv) {	/* failed? */
3097 
3098 		/*
3099 		 * Get back to an exclusive (write) lock.
3100 		 */
3101 
3102 		vm_map_upgrade(map);
3103 		vm_map_unbusy(map);
3104 
3105 #ifdef DIAGNOSTIC
3106 		if (timestamp_save != map->timestamp)
3107 			panic("uvm_map_pageable: stale map");
3108 #endif
3109 
3110 		/*
3111 		 * first drop the wiring count on all the entries
3112 		 * which haven't actually been wired yet.
3113 		 */
3114 
3115 		failed_entry = entry;
3116 		while (entry != &map->header && entry->start < end) {
3117 			entry->wired_count--;
3118 			entry = entry->next;
3119 		}
3120 
3121 		/*
3122 		 * now, unwire all the entries that were successfully
3123 		 * wired above.
3124 		 */
3125 
3126 		entry = start_entry;
3127 		while (entry != failed_entry) {
3128 			entry->wired_count--;
3129 			if (VM_MAPENT_ISWIRED(entry) == 0)
3130 				uvm_map_entry_unwire(map, entry);
3131 			entry = entry->next;
3132 		}
3133 		if ((lockflags & UVM_LK_EXIT) == 0)
3134 			vm_map_unlock(map);
3135 		UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
3136 		return (rv);
3137 	}
3138 
3139 	/* We are holding a read lock here. */
3140 	if ((lockflags & UVM_LK_EXIT) == 0) {
3141 		vm_map_unbusy(map);
3142 		vm_map_unlock_read(map);
3143 	} else {
3144 
3145 		/*
3146 		 * Get back to an exclusive (write) lock.
3147 		 */
3148 
3149 		vm_map_upgrade(map);
3150 		vm_map_unbusy(map);
3151 	}
3152 
3153 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
3154 	return 0;
3155 }
3156 
3157 /*
3158  * uvm_map_pageable_all: special case of uvm_map_pageable - affects
3159  * all mapped regions.
3160  *
3161  * => map must not be locked.
3162  * => if no flags are specified, all regions are unwired.
3163  * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
3164  */
3165 
3166 int
3167 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit)
3168 {
3169 	struct vm_map_entry *entry, *failed_entry;
3170 	vsize_t size;
3171 	int rv;
3172 #ifdef DIAGNOSTIC
3173 	u_int timestamp_save;
3174 #endif
3175 	UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
3176 	UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
3177 
3178 	KASSERT(map->flags & VM_MAP_PAGEABLE);
3179 
3180 	vm_map_lock(map);
3181 
3182 	/*
3183 	 * handle wiring and unwiring separately.
3184 	 */
3185 
3186 	if (flags == 0) {			/* unwire */
3187 
3188 		/*
3189 		 * POSIX 1003.1b -- munlockall unlocks all regions,
3190 		 * regardless of how many times mlockall has been called.
3191 		 */
3192 
3193 		for (entry = map->header.next; entry != &map->header;
3194 		     entry = entry->next) {
3195 			if (VM_MAPENT_ISWIRED(entry))
3196 				uvm_map_entry_unwire(map, entry);
3197 		}
3198 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
3199 		vm_map_unlock(map);
3200 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
3201 		return 0;
3202 	}
3203 
3204 	if (flags & MCL_FUTURE) {
3205 
3206 		/*
3207 		 * must wire all future mappings; remember this.
3208 		 */
3209 
3210 		vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
3211 	}
3212 
3213 	if ((flags & MCL_CURRENT) == 0) {
3214 
3215 		/*
3216 		 * no more work to do!
3217 		 */
3218 
3219 		UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
3220 		vm_map_unlock(map);
3221 		return 0;
3222 	}
3223 
3224 	/*
3225 	 * wire case: in three passes [XXXCDC: ugly block of code here]
3226 	 *
3227 	 * 1: holding the write lock, count all pages mapped by non-wired
3228 	 *    entries.  if this would cause us to go over our limit, we fail.
3229 	 *
3230 	 * 2: still holding the write lock, we create any anonymous maps that
3231 	 *    need to be created.  then we increment its wiring count.
3232 	 *
3233 	 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
3234 	 *    in the pages for any newly wired area (wired_count == 1).
3235 	 *
3236 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
3237 	 *    deadlock with another thread that may have faulted on one of
3238 	 *    the pages to be wired (it would mark the page busy, blocking
3239 	 *    us, then in turn block on the map lock that we hold).  because
3240 	 *    of problems in the recursive lock package, we cannot upgrade
3241 	 *    to a write lock in vm_map_lookup.  thus, any actions that
3242 	 *    require the write lock must be done beforehand.  because we
3243 	 *    keep the read lock on the map, the copy-on-write status of the
3244 	 *    entries we modify here cannot change.
3245 	 */
3246 
3247 	for (size = 0, entry = map->header.next; entry != &map->header;
3248 	     entry = entry->next) {
3249 		if (entry->protection != VM_PROT_NONE &&
3250 		    VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3251 			size += entry->end - entry->start;
3252 		}
3253 	}
3254 
3255 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
3256 		vm_map_unlock(map);
3257 		return ENOMEM;
3258 	}
3259 
3260 	if (limit != 0 &&
3261 	    (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
3262 		vm_map_unlock(map);
3263 		return ENOMEM;
3264 	}
3265 
3266 	/*
3267 	 * Pass 2.
3268 	 */
3269 
3270 	for (entry = map->header.next; entry != &map->header;
3271 	     entry = entry->next) {
3272 		if (entry->protection == VM_PROT_NONE)
3273 			continue;
3274 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3275 
3276 			/*
3277 			 * perform actions of vm_map_lookup that need the
3278 			 * write lock on the map: create an anonymous map
3279 			 * for a copy-on-write region, or an anonymous map
3280 			 * for a zero-fill region.  (XXXCDC: submap case
3281 			 * ok?)
3282 			 */
3283 
3284 			if (!UVM_ET_ISSUBMAP(entry)) {	/* not submap */
3285 				if (UVM_ET_ISNEEDSCOPY(entry) &&
3286 				    ((entry->max_protection & VM_PROT_WRITE) ||
3287 				     (entry->object.uvm_obj == NULL))) {
3288 					amap_copy(map, entry, M_WAITOK, TRUE,
3289 					    entry->start, entry->end);
3290 					/* XXXCDC: wait OK? */
3291 				}
3292 			}
3293 		}
3294 		entry->wired_count++;
3295 	}
3296 
3297 	/*
3298 	 * Pass 3.
3299 	 */
3300 
3301 #ifdef DIAGNOSTIC
3302 	timestamp_save = map->timestamp;
3303 #endif
3304 	vm_map_busy(map);
3305 	vm_map_downgrade(map);
3306 
3307 	rv = 0;
3308 	for (entry = map->header.next; entry != &map->header;
3309 	     entry = entry->next) {
3310 		if (entry->wired_count == 1) {
3311 			rv = uvm_fault_wire(map, entry->start, entry->end,
3312 			    VM_FAULT_WIREMAX, entry->max_protection);
3313 			if (rv) {
3314 
3315 				/*
3316 				 * wiring failed.  break out of the loop.
3317 				 * we'll clean up the map below, once we
3318 				 * have a write lock again.
3319 				 */
3320 
3321 				break;
3322 			}
3323 		}
3324 	}
3325 
3326 	if (rv) {
3327 
3328 		/*
3329 		 * Get back an exclusive (write) lock.
3330 		 */
3331 
3332 		vm_map_upgrade(map);
3333 		vm_map_unbusy(map);
3334 
3335 #ifdef DIAGNOSTIC
3336 		if (timestamp_save != map->timestamp)
3337 			panic("uvm_map_pageable_all: stale map");
3338 #endif
3339 
3340 		/*
3341 		 * first drop the wiring count on all the entries
3342 		 * which haven't actually been wired yet.
3343 		 *
3344 		 * Skip VM_PROT_NONE entries like we did above.
3345 		 */
3346 
3347 		failed_entry = entry;
3348 		for (/* nothing */; entry != &map->header;
3349 		     entry = entry->next) {
3350 			if (entry->protection == VM_PROT_NONE)
3351 				continue;
3352 			entry->wired_count--;
3353 		}
3354 
3355 		/*
3356 		 * now, unwire all the entries that were successfully
3357 		 * wired above.
3358 		 *
3359 		 * Skip VM_PROT_NONE entries like we did above.
3360 		 */
3361 
3362 		for (entry = map->header.next; entry != failed_entry;
3363 		     entry = entry->next) {
3364 			if (entry->protection == VM_PROT_NONE)
3365 				continue;
3366 			entry->wired_count--;
3367 			if (VM_MAPENT_ISWIRED(entry))
3368 				uvm_map_entry_unwire(map, entry);
3369 		}
3370 		vm_map_unlock(map);
3371 		UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
3372 		return (rv);
3373 	}
3374 
3375 	/* We are holding a read lock here. */
3376 	vm_map_unbusy(map);
3377 	vm_map_unlock_read(map);
3378 
3379 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
3380 	return 0;
3381 }
3382 
3383 /*
3384  * uvm_map_clean: clean out a map range
3385  *
3386  * => valid flags:
3387  *   if (flags & PGO_CLEANIT): dirty pages are cleaned first
3388  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
3389  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
3390  *   if (flags & PGO_FREE): any cached pages are freed after clean
3391  * => returns an error if any part of the specified range isn't mapped
3392  * => never a need to flush amap layer since the anonymous memory has
3393  *	no permanent home, but may deactivate pages there
3394  * => called from sys_msync() and sys_madvise()
3395  * => caller must not write-lock map (read OK).
3396  * => we may sleep while cleaning if SYNCIO [with map read-locked]
3397  */
3398 
3399 int
3400 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
3401 {
3402 	struct vm_map_entry *current, *entry;
3403 	struct uvm_object *uobj;
3404 	struct vm_amap *amap;
3405 	struct vm_anon *anon;
3406 	struct vm_page *pg;
3407 	vaddr_t offset;
3408 	vsize_t size;
3409 	voff_t uoff;
3410 	int error, refs;
3411 	UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
3412 
3413 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
3414 		    map, start, end, flags);
3415 	KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) !=
3416 		(PGO_FREE|PGO_DEACTIVATE));
3417 
3418 	vm_map_lock_read(map);
3419 	VM_MAP_RANGE_CHECK(map, start, end);
3420 	if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
3421 		vm_map_unlock_read(map);
3422 		return EFAULT;
3423 	}
3424 
3425 	/*
3426 	 * Make a first pass to check for holes and wiring problems.
3427 	 */
3428 
3429 	for (current = entry; current->start < end; current = current->next) {
3430 		if (UVM_ET_ISSUBMAP(current)) {
3431 			vm_map_unlock_read(map);
3432 			return EINVAL;
3433 		}
3434 		if ((flags & PGO_FREE) != 0 && VM_MAPENT_ISWIRED(entry)) {
3435 			vm_map_unlock_read(map);
3436 			return EBUSY;
3437 		}
3438 		if (end <= current->end) {
3439 			break;
3440 		}
3441 		if (current->end != current->next->start) {
3442 			vm_map_unlock_read(map);
3443 			return EFAULT;
3444 		}
3445 	}
3446 
3447 	error = 0;
3448 	for (current = entry; start < end; current = current->next) {
3449 		amap = current->aref.ar_amap;	/* top layer */
3450 		uobj = current->object.uvm_obj;	/* bottom layer */
3451 		KASSERT(start >= current->start);
3452 
3453 		/*
3454 		 * No amap cleaning necessary if:
3455 		 *
3456 		 *	(1) There's no amap.
3457 		 *
3458 		 *	(2) We're not deactivating or freeing pages.
3459 		 */
3460 
3461 		if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
3462 			goto flush_object;
3463 
3464 		amap_lock(amap);
3465 		offset = start - current->start;
3466 		size = MIN(end, current->end) - start;
3467 		for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) {
3468 			anon = amap_lookup(&current->aref, offset);
3469 			if (anon == NULL)
3470 				continue;
3471 
3472 			simple_lock(&anon->an_lock);
3473 			pg = anon->an_page;
3474 			if (pg == NULL) {
3475 				simple_unlock(&anon->an_lock);
3476 				continue;
3477 			}
3478 
3479 			switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
3480 
3481 			/*
3482 			 * In these first 3 cases, we just deactivate the page.
3483 			 */
3484 
3485 			case PGO_CLEANIT|PGO_FREE:
3486 			case PGO_CLEANIT|PGO_DEACTIVATE:
3487 			case PGO_DEACTIVATE:
3488  deactivate_it:
3489 				/*
3490 				 * skip the page if it's loaned or wired,
3491 				 * since it shouldn't be on a paging queue
3492 				 * at all in these cases.
3493 				 */
3494 
3495 				uvm_lock_pageq();
3496 				if (pg->loan_count != 0 ||
3497 				    pg->wire_count != 0) {
3498 					uvm_unlock_pageq();
3499 					simple_unlock(&anon->an_lock);
3500 					continue;
3501 				}
3502 				KASSERT(pg->uanon == anon);
3503 				pmap_clear_reference(pg);
3504 				uvm_pagedeactivate(pg);
3505 				uvm_unlock_pageq();
3506 				simple_unlock(&anon->an_lock);
3507 				continue;
3508 
3509 			case PGO_FREE:
3510 
3511 				/*
3512 				 * If there are multiple references to
3513 				 * the amap, just deactivate the page.
3514 				 */
3515 
3516 				if (amap_refs(amap) > 1)
3517 					goto deactivate_it;
3518 
3519 				/* skip the page if it's wired */
3520 				if (pg->wire_count != 0) {
3521 					simple_unlock(&anon->an_lock);
3522 					continue;
3523 				}
3524 				amap_unadd(&current->aref, offset);
3525 				refs = --anon->an_ref;
3526 				simple_unlock(&anon->an_lock);
3527 				if (refs == 0)
3528 					uvm_anfree(anon);
3529 				continue;
3530 			}
3531 		}
3532 		amap_unlock(amap);
3533 
3534  flush_object:
3535 		/*
3536 		 * flush pages if we've got a valid backing object.
3537 		 * note that we must always clean object pages before
3538 		 * freeing them since otherwise we could reveal stale
3539 		 * data from files.
3540 		 */
3541 
3542 		uoff = current->offset + (start - current->start);
3543 		size = MIN(end, current->end) - start;
3544 		if (uobj != NULL) {
3545 			simple_lock(&uobj->vmobjlock);
3546 			if (uobj->pgops->pgo_put != NULL)
3547 				error = (uobj->pgops->pgo_put)(uobj, uoff,
3548 				    uoff + size, flags | PGO_CLEANIT);
3549 			else
3550 				error = 0;
3551 		}
3552 		start += size;
3553 	}
3554 	vm_map_unlock_read(map);
3555 	return (error);
3556 }
3557 
3558 
3559 /*
3560  * uvm_map_checkprot: check protection in map
3561  *
3562  * => must allow specified protection in a fully allocated region.
3563  * => map must be read or write locked by caller.
3564  */
3565 
3566 boolean_t
3567 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end,
3568     vm_prot_t protection)
3569 {
3570 	struct vm_map_entry *entry;
3571 	struct vm_map_entry *tmp_entry;
3572 
3573 	if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
3574 		return (FALSE);
3575 	}
3576 	entry = tmp_entry;
3577 	while (start < end) {
3578 		if (entry == &map->header) {
3579 			return (FALSE);
3580 		}
3581 
3582 		/*
3583 		 * no holes allowed
3584 		 */
3585 
3586 		if (start < entry->start) {
3587 			return (FALSE);
3588 		}
3589 
3590 		/*
3591 		 * check protection associated with entry
3592 		 */
3593 
3594 		if ((entry->protection & protection) != protection) {
3595 			return (FALSE);
3596 		}
3597 		start = entry->end;
3598 		entry = entry->next;
3599 	}
3600 	return (TRUE);
3601 }
3602 
3603 /*
3604  * uvmspace_alloc: allocate a vmspace structure.
3605  *
3606  * - structure includes vm_map and pmap
3607  * - XXX: no locking on this structure
3608  * - refcnt set to 1, rest must be init'd by caller
3609  */
3610 struct vmspace *
3611 uvmspace_alloc(vaddr_t vmin, vaddr_t vmax)
3612 {
3613 	struct vmspace *vm;
3614 	UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
3615 
3616 	vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
3617 	uvmspace_init(vm, NULL, vmin, vmax);
3618 	UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
3619 	return (vm);
3620 }
3621 
3622 /*
3623  * uvmspace_init: initialize a vmspace structure.
3624  *
3625  * - XXX: no locking on this structure
3626  * - refcnt set to 1, rest must be init'd by caller
3627  */
3628 void
3629 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t vmin, vaddr_t vmax)
3630 {
3631 	UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
3632 
3633 	memset(vm, 0, sizeof(*vm));
3634 	uvm_map_setup(&vm->vm_map, vmin, vmax, VM_MAP_PAGEABLE
3635 #ifdef __USING_TOPDOWN_VM
3636 	    | VM_MAP_TOPDOWN
3637 #endif
3638 	    );
3639 	if (pmap)
3640 		pmap_reference(pmap);
3641 	else
3642 		pmap = pmap_create();
3643 	vm->vm_map.pmap = pmap;
3644 	vm->vm_refcnt = 1;
3645 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3646 }
3647 
3648 /*
3649  * uvmspace_share: share a vmspace between two processes
3650  *
3651  * - used for vfork, threads(?)
3652  */
3653 
3654 void
3655 uvmspace_share(struct proc *p1, struct proc *p2)
3656 {
3657 	struct simplelock *slock = &p1->p_vmspace->vm_map.ref_lock;
3658 
3659 	p2->p_vmspace = p1->p_vmspace;
3660 	simple_lock(slock);
3661 	p1->p_vmspace->vm_refcnt++;
3662 	simple_unlock(slock);
3663 }
3664 
3665 /*
3666  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
3667  *
3668  * - XXX: no locking on vmspace
3669  */
3670 
3671 void
3672 uvmspace_unshare(struct lwp *l)
3673 {
3674 	struct proc *p = l->l_proc;
3675 	struct vmspace *nvm, *ovm = p->p_vmspace;
3676 
3677 	if (ovm->vm_refcnt == 1)
3678 		/* nothing to do: vmspace isn't shared in the first place */
3679 		return;
3680 
3681 	/* make a new vmspace, still holding old one */
3682 	nvm = uvmspace_fork(ovm);
3683 
3684 	pmap_deactivate(l);		/* unbind old vmspace */
3685 	p->p_vmspace = nvm;
3686 	pmap_activate(l);		/* switch to new vmspace */
3687 
3688 	uvmspace_free(ovm);		/* drop reference to old vmspace */
3689 }
3690 
3691 /*
3692  * uvmspace_exec: the process wants to exec a new program
3693  */
3694 
3695 void
3696 uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end)
3697 {
3698 	struct proc *p = l->l_proc;
3699 	struct vmspace *nvm, *ovm = p->p_vmspace;
3700 	struct vm_map *map = &ovm->vm_map;
3701 
3702 #ifdef __sparc__
3703 	/* XXX cgd 960926: the sparc #ifdef should be a MD hook */
3704 	kill_user_windows(l);   /* before stack addresses go away */
3705 #endif
3706 
3707 	/*
3708 	 * see if more than one process is using this vmspace...
3709 	 */
3710 
3711 	if (ovm->vm_refcnt == 1) {
3712 
3713 		/*
3714 		 * if p is the only process using its vmspace then we can safely
3715 		 * recycle that vmspace for the program that is being exec'd.
3716 		 */
3717 
3718 #ifdef SYSVSHM
3719 		/*
3720 		 * SYSV SHM semantics require us to kill all segments on an exec
3721 		 */
3722 
3723 		if (ovm->vm_shm)
3724 			shmexit(ovm);
3725 #endif
3726 
3727 		/*
3728 		 * POSIX 1003.1b -- "lock future mappings" is revoked
3729 		 * when a process execs another program image.
3730 		 */
3731 
3732 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
3733 
3734 		/*
3735 		 * now unmap the old program
3736 		 */
3737 
3738 		pmap_remove_all(map->pmap);
3739 		uvm_unmap(map, vm_map_min(map), vm_map_max(map));
3740 		KASSERT(map->header.prev == &map->header);
3741 		KASSERT(map->nentries == 0);
3742 
3743 		/*
3744 		 * resize the map
3745 		 */
3746 
3747 		vm_map_setmin(map, start);
3748 		vm_map_setmax(map, end);
3749 	} else {
3750 
3751 		/*
3752 		 * p's vmspace is being shared, so we can't reuse it for p since
3753 		 * it is still being used for others.   allocate a new vmspace
3754 		 * for p
3755 		 */
3756 
3757 		nvm = uvmspace_alloc(start, end);
3758 
3759 		/*
3760 		 * install new vmspace and drop our ref to the old one.
3761 		 */
3762 
3763 		pmap_deactivate(l);
3764 		p->p_vmspace = nvm;
3765 		pmap_activate(l);
3766 
3767 		uvmspace_free(ovm);
3768 	}
3769 }
3770 
3771 /*
3772  * uvmspace_free: free a vmspace data structure
3773  */
3774 
3775 void
3776 uvmspace_free(struct vmspace *vm)
3777 {
3778 	struct vm_map_entry *dead_entries;
3779 	struct vm_map *map = &vm->vm_map;
3780 	int n;
3781 
3782 	UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
3783 
3784 	UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
3785 	simple_lock(&map->ref_lock);
3786 	n = --vm->vm_refcnt;
3787 	simple_unlock(&map->ref_lock);
3788 	if (n > 0)
3789 		return;
3790 
3791 	/*
3792 	 * at this point, there should be no other references to the map.
3793 	 * delete all of the mappings, then destroy the pmap.
3794 	 */
3795 
3796 	map->flags |= VM_MAP_DYING;
3797 	pmap_remove_all(map->pmap);
3798 #ifdef SYSVSHM
3799 	/* Get rid of any SYSV shared memory segments. */
3800 	if (vm->vm_shm != NULL)
3801 		shmexit(vm);
3802 #endif
3803 	if (map->nentries) {
3804 		uvm_unmap_remove(map, vm_map_min(map), vm_map_max(map),
3805 		    &dead_entries, NULL, 0);
3806 		if (dead_entries != NULL)
3807 			uvm_unmap_detach(dead_entries, 0);
3808 	}
3809 	KASSERT(map->nentries == 0);
3810 	KASSERT(map->size == 0);
3811 	pmap_destroy(map->pmap);
3812 	pool_put(&uvm_vmspace_pool, vm);
3813 }
3814 
3815 /*
3816  *   F O R K   -   m a i n   e n t r y   p o i n t
3817  */
3818 /*
3819  * uvmspace_fork: fork a process' main map
3820  *
3821  * => create a new vmspace for child process from parent.
3822  * => parent's map must not be locked.
3823  */
3824 
3825 struct vmspace *
3826 uvmspace_fork(struct vmspace *vm1)
3827 {
3828 	struct vmspace *vm2;
3829 	struct vm_map *old_map = &vm1->vm_map;
3830 	struct vm_map *new_map;
3831 	struct vm_map_entry *old_entry;
3832 	struct vm_map_entry *new_entry;
3833 	UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
3834 
3835 	vm_map_lock(old_map);
3836 
3837 	vm2 = uvmspace_alloc(vm_map_min(old_map), vm_map_max(old_map));
3838 	memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
3839 	    (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
3840 	new_map = &vm2->vm_map;		  /* XXX */
3841 
3842 	old_entry = old_map->header.next;
3843 	new_map->size = old_map->size;
3844 
3845 	/*
3846 	 * go entry-by-entry
3847 	 */
3848 
3849 	while (old_entry != &old_map->header) {
3850 
3851 		/*
3852 		 * first, some sanity checks on the old entry
3853 		 */
3854 
3855 		KASSERT(!UVM_ET_ISSUBMAP(old_entry));
3856 		KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) ||
3857 			!UVM_ET_ISNEEDSCOPY(old_entry));
3858 
3859 		switch (old_entry->inheritance) {
3860 		case MAP_INHERIT_NONE:
3861 
3862 			/*
3863 			 * drop the mapping, modify size
3864 			 */
3865 			new_map->size -= old_entry->end - old_entry->start;
3866 			break;
3867 
3868 		case MAP_INHERIT_SHARE:
3869 
3870 			/*
3871 			 * share the mapping: this means we want the old and
3872 			 * new entries to share amaps and backing objects.
3873 			 */
3874 			/*
3875 			 * if the old_entry needs a new amap (due to prev fork)
3876 			 * then we need to allocate it now so that we have
3877 			 * something we own to share with the new_entry.   [in
3878 			 * other words, we need to clear needs_copy]
3879 			 */
3880 
3881 			if (UVM_ET_ISNEEDSCOPY(old_entry)) {
3882 				/* get our own amap, clears needs_copy */
3883 				amap_copy(old_map, old_entry, M_WAITOK, FALSE,
3884 				    0, 0);
3885 				/* XXXCDC: WAITOK??? */
3886 			}
3887 
3888 			new_entry = uvm_mapent_alloc(new_map, 0);
3889 			/* old_entry -> new_entry */
3890 			uvm_mapent_copy(old_entry, new_entry);
3891 
3892 			/* new pmap has nothing wired in it */
3893 			new_entry->wired_count = 0;
3894 
3895 			/*
3896 			 * gain reference to object backing the map (can't
3897 			 * be a submap, already checked this case).
3898 			 */
3899 
3900 			if (new_entry->aref.ar_amap)
3901 				uvm_map_reference_amap(new_entry, AMAP_SHARED);
3902 
3903 			if (new_entry->object.uvm_obj &&
3904 			    new_entry->object.uvm_obj->pgops->pgo_reference)
3905 				new_entry->object.uvm_obj->
3906 				    pgops->pgo_reference(
3907 				        new_entry->object.uvm_obj);
3908 
3909 			/* insert entry at end of new_map's entry list */
3910 			uvm_map_entry_link(new_map, new_map->header.prev,
3911 			    new_entry);
3912 
3913 			break;
3914 
3915 		case MAP_INHERIT_COPY:
3916 
3917 			/*
3918 			 * copy-on-write the mapping (using mmap's
3919 			 * MAP_PRIVATE semantics)
3920 			 *
3921 			 * allocate new_entry, adjust reference counts.
3922 			 * (note that new references are read-only).
3923 			 */
3924 
3925 			new_entry = uvm_mapent_alloc(new_map, 0);
3926 			/* old_entry -> new_entry */
3927 			uvm_mapent_copy(old_entry, new_entry);
3928 
3929 			if (new_entry->aref.ar_amap)
3930 				uvm_map_reference_amap(new_entry, 0);
3931 
3932 			if (new_entry->object.uvm_obj &&
3933 			    new_entry->object.uvm_obj->pgops->pgo_reference)
3934 				new_entry->object.uvm_obj->pgops->pgo_reference
3935 				    (new_entry->object.uvm_obj);
3936 
3937 			/* new pmap has nothing wired in it */
3938 			new_entry->wired_count = 0;
3939 
3940 			new_entry->etype |=
3941 			    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
3942 			uvm_map_entry_link(new_map, new_map->header.prev,
3943 			    new_entry);
3944 
3945 			/*
3946 			 * the new entry will need an amap.  it will either
3947 			 * need to be copied from the old entry or created
3948 			 * from scratch (if the old entry does not have an
3949 			 * amap).  can we defer this process until later
3950 			 * (by setting "needs_copy") or do we need to copy
3951 			 * the amap now?
3952 			 *
3953 			 * we must copy the amap now if any of the following
3954 			 * conditions hold:
3955 			 * 1. the old entry has an amap and that amap is
3956 			 *    being shared.  this means that the old (parent)
3957 			 *    process is sharing the amap with another
3958 			 *    process.  if we do not clear needs_copy here
3959 			 *    we will end up in a situation where both the
3960 			 *    parent and child process are refering to the
3961 			 *    same amap with "needs_copy" set.  if the
3962 			 *    parent write-faults, the fault routine will
3963 			 *    clear "needs_copy" in the parent by allocating
3964 			 *    a new amap.   this is wrong because the
3965 			 *    parent is supposed to be sharing the old amap
3966 			 *    and the new amap will break that.
3967 			 *
3968 			 * 2. if the old entry has an amap and a non-zero
3969 			 *    wire count then we are going to have to call
3970 			 *    amap_cow_now to avoid page faults in the
3971 			 *    parent process.   since amap_cow_now requires
3972 			 *    "needs_copy" to be clear we might as well
3973 			 *    clear it here as well.
3974 			 *
3975 			 */
3976 
3977 			if (old_entry->aref.ar_amap != NULL) {
3978 				if ((amap_flags(old_entry->aref.ar_amap) &
3979 				     AMAP_SHARED) != 0 ||
3980 				    VM_MAPENT_ISWIRED(old_entry)) {
3981 
3982 					amap_copy(new_map, new_entry, M_WAITOK,
3983 					    FALSE, 0, 0);
3984 					/* XXXCDC: M_WAITOK ... ok? */
3985 				}
3986 			}
3987 
3988 			/*
3989 			 * if the parent's entry is wired down, then the
3990 			 * parent process does not want page faults on
3991 			 * access to that memory.  this means that we
3992 			 * cannot do copy-on-write because we can't write
3993 			 * protect the old entry.   in this case we
3994 			 * resolve all copy-on-write faults now, using
3995 			 * amap_cow_now.   note that we have already
3996 			 * allocated any needed amap (above).
3997 			 */
3998 
3999 			if (VM_MAPENT_ISWIRED(old_entry)) {
4000 
4001 			  /*
4002 			   * resolve all copy-on-write faults now
4003 			   * (note that there is nothing to do if
4004 			   * the old mapping does not have an amap).
4005 			   */
4006 			  if (old_entry->aref.ar_amap)
4007 			    amap_cow_now(new_map, new_entry);
4008 
4009 			} else {
4010 
4011 			  /*
4012 			   * setup mappings to trigger copy-on-write faults
4013 			   * we must write-protect the parent if it has
4014 			   * an amap and it is not already "needs_copy"...
4015 			   * if it is already "needs_copy" then the parent
4016 			   * has already been write-protected by a previous
4017 			   * fork operation.
4018 			   */
4019 
4020 			  if (old_entry->aref.ar_amap &&
4021 			      !UVM_ET_ISNEEDSCOPY(old_entry)) {
4022 			      if (old_entry->max_protection & VM_PROT_WRITE) {
4023 				pmap_protect(old_map->pmap,
4024 					     old_entry->start,
4025 					     old_entry->end,
4026 					     old_entry->protection &
4027 					     ~VM_PROT_WRITE);
4028 				pmap_update(old_map->pmap);
4029 			      }
4030 			      old_entry->etype |= UVM_ET_NEEDSCOPY;
4031 			  }
4032 			}
4033 			break;
4034 		}  /* end of switch statement */
4035 		old_entry = old_entry->next;
4036 	}
4037 
4038 	vm_map_unlock(old_map);
4039 
4040 #ifdef SYSVSHM
4041 	if (vm1->vm_shm)
4042 		shmfork(vm1, vm2);
4043 #endif
4044 
4045 #ifdef PMAP_FORK
4046 	pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
4047 #endif
4048 
4049 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
4050 	return (vm2);
4051 }
4052 
4053 
4054 /*
4055  * in-kernel map entry allocation.
4056  */
4057 
4058 int ukh_alloc, ukh_free;
4059 int uke_alloc, uke_free;
4060 
4061 struct uvm_kmapent_hdr {
4062 	LIST_ENTRY(uvm_kmapent_hdr) ukh_listq;
4063 	int ukh_nused;
4064 	struct vm_map_entry *ukh_freelist;
4065 	struct vm_map *ukh_map;
4066 	struct vm_map_entry ukh_entries[0];
4067 };
4068 
4069 #define	UVM_KMAPENT_CHUNK				\
4070 	((PAGE_SIZE - sizeof(struct uvm_kmapent_hdr))	\
4071 	/ sizeof(struct vm_map_entry))
4072 
4073 #define	UVM_KHDR_FIND(entry)	\
4074 	((struct uvm_kmapent_hdr *)(((vaddr_t)entry) & ~PAGE_MASK))
4075 
4076 
4077 #ifdef DIAGNOSTIC
4078 static struct vm_map *
4079 uvm_kmapent_map(struct vm_map_entry *entry)
4080 {
4081 	const struct uvm_kmapent_hdr *ukh;
4082 
4083 	ukh = UVM_KHDR_FIND(entry);
4084 	return ukh->ukh_map;
4085 }
4086 #endif
4087 
4088 static __inline struct vm_map_entry *
4089 uvm_kmapent_get(struct uvm_kmapent_hdr *ukh)
4090 {
4091 	struct vm_map_entry *entry;
4092 
4093 	KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
4094 	KASSERT(ukh->ukh_nused >= 0);
4095 
4096 	entry = ukh->ukh_freelist;
4097 	if (entry) {
4098 		KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT))
4099 		    == UVM_MAP_KERNEL);
4100 		ukh->ukh_freelist = entry->next;
4101 		ukh->ukh_nused++;
4102 		KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
4103 	} else {
4104 		KASSERT(ukh->ukh_nused == UVM_KMAPENT_CHUNK);
4105 	}
4106 
4107 	return entry;
4108 }
4109 
4110 static __inline void
4111 uvm_kmapent_put(struct uvm_kmapent_hdr *ukh, struct vm_map_entry *entry)
4112 {
4113 
4114 	KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT))
4115 	    == UVM_MAP_KERNEL);
4116 	KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
4117 	KASSERT(ukh->ukh_nused > 0);
4118 	KASSERT(ukh->ukh_freelist != NULL ||
4119 	    ukh->ukh_nused == UVM_KMAPENT_CHUNK);
4120 	KASSERT(ukh->ukh_freelist == NULL ||
4121 	    ukh->ukh_nused < UVM_KMAPENT_CHUNK);
4122 
4123 	ukh->ukh_nused--;
4124 	entry->next = ukh->ukh_freelist;
4125 	ukh->ukh_freelist = entry;
4126 }
4127 
4128 /*
4129  * uvm_kmapent_alloc: allocate a map entry for in-kernel map
4130  */
4131 
4132 static struct vm_map_entry *
4133 uvm_kmapent_alloc(struct vm_map *map, int flags)
4134 {
4135 	struct vm_page *pg;
4136 	struct uvm_map_args args;
4137 	struct uvm_kmapent_hdr *ukh;
4138 	struct vm_map_entry *entry;
4139 	uvm_flag_t mapflags = UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL,
4140 	    UVM_INH_NONE, UVM_ADV_RANDOM, flags | UVM_FLAG_NOMERGE);
4141 	vaddr_t va;
4142 	int error;
4143 	int i;
4144 	int s;
4145 
4146 	KDASSERT(UVM_KMAPENT_CHUNK > 2);
4147 	KDASSERT(kernel_map != NULL);
4148 	KASSERT(vm_map_pmap(map) == pmap_kernel());
4149 
4150 	uke_alloc++;
4151 	entry = NULL;
4152 again:
4153 	/*
4154 	 * try to grab an entry from freelist.
4155 	 */
4156 	s = splvm();
4157 	simple_lock(&uvm.kentry_lock);
4158 	ukh = LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free);
4159 	if (ukh) {
4160 		entry = uvm_kmapent_get(ukh);
4161 		if (ukh->ukh_nused == UVM_KMAPENT_CHUNK)
4162 			LIST_REMOVE(ukh, ukh_listq);
4163 	}
4164 	simple_unlock(&uvm.kentry_lock);
4165 	splx(s);
4166 
4167 	if (entry)
4168 		return entry;
4169 
4170 	/*
4171 	 * there's no free entry for this vm_map.
4172 	 * now we need to allocate some vm_map_entry.
4173 	 * for simplicity, always allocate one page chunk of them at once.
4174 	 */
4175 
4176 	pg = uvm_pagealloc(NULL, 0, NULL, 0);
4177 	if (__predict_false(pg == NULL)) {
4178 		if (flags & UVM_FLAG_NOWAIT)
4179 			return NULL;
4180 		uvm_wait("kme_alloc");
4181 		goto again;
4182 	}
4183 
4184 	error = uvm_map_prepare(map, 0, PAGE_SIZE, NULL, 0, 0, mapflags, &args);
4185 	if (error) {
4186 		uvm_pagefree(pg);
4187 		return NULL;
4188 	}
4189 
4190 	va = args.uma_start;
4191 
4192 	pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), VM_PROT_READ|VM_PROT_WRITE);
4193 	pmap_update(vm_map_pmap(map));
4194 
4195 	ukh = (void *)va;
4196 
4197 	/*
4198 	 * use the first entry for ukh itsself.
4199 	 */
4200 
4201 	entry = &ukh->ukh_entries[0];
4202 	entry->flags = UVM_MAP_KERNEL | UVM_MAP_KMAPENT;
4203 	error = uvm_map_enter(map, &args, entry);
4204 	KASSERT(error == 0);
4205 
4206 	ukh->ukh_nused = UVM_KMAPENT_CHUNK;
4207 	ukh->ukh_map = map;
4208 	ukh->ukh_freelist = NULL;
4209 	for (i = UVM_KMAPENT_CHUNK - 1; i >= 2; i--) {
4210 		struct vm_map_entry *xentry = &ukh->ukh_entries[i];
4211 
4212 		xentry->flags = UVM_MAP_KERNEL;
4213 		uvm_kmapent_put(ukh, xentry);
4214 	}
4215 	KASSERT(ukh->ukh_nused == 2);
4216 
4217 	s = splvm();
4218 	simple_lock(&uvm.kentry_lock);
4219 	LIST_INSERT_HEAD(&vm_map_to_kernel(map)->vmk_kentry_free,
4220 	    ukh, ukh_listq);
4221 	simple_unlock(&uvm.kentry_lock);
4222 	splx(s);
4223 
4224 	/*
4225 	 * return second entry.
4226 	 */
4227 
4228 	entry = &ukh->ukh_entries[1];
4229 	entry->flags = UVM_MAP_KERNEL;
4230 	ukh_alloc++;
4231 	return entry;
4232 }
4233 
4234 /*
4235  * uvm_mapent_free: free map entry for in-kernel map
4236  */
4237 
4238 static void
4239 uvm_kmapent_free(struct vm_map_entry *entry)
4240 {
4241 	struct uvm_kmapent_hdr *ukh;
4242 	struct vm_page *pg;
4243 	struct vm_map *map;
4244 	struct pmap *pmap;
4245 	vaddr_t va;
4246 	paddr_t pa;
4247 	struct vm_map_entry *deadentry;
4248 	int s;
4249 
4250 	uke_free++;
4251 	ukh = UVM_KHDR_FIND(entry);
4252 	map = ukh->ukh_map;
4253 
4254 	s = splvm();
4255 	simple_lock(&uvm.kentry_lock);
4256 	uvm_kmapent_put(ukh, entry);
4257 	if (ukh->ukh_nused > 1) {
4258 		if (ukh->ukh_nused == UVM_KMAPENT_CHUNK - 1)
4259 			LIST_INSERT_HEAD(
4260 			    &vm_map_to_kernel(map)->vmk_kentry_free,
4261 			    ukh, ukh_listq);
4262 		simple_unlock(&uvm.kentry_lock);
4263 		splx(s);
4264 		return;
4265 	}
4266 
4267 	/*
4268 	 * now we can free this ukh.
4269 	 *
4270 	 * however, keep an empty ukh to avoid ping-pong.
4271 	 */
4272 
4273 	if (LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free) == ukh &&
4274 	    LIST_NEXT(ukh, ukh_listq) == NULL) {
4275 		simple_unlock(&uvm.kentry_lock);
4276 		splx(s);
4277 		return;
4278 	}
4279 	LIST_REMOVE(ukh, ukh_listq);
4280 	simple_unlock(&uvm.kentry_lock);
4281 	splx(s);
4282 
4283 	KASSERT(ukh->ukh_nused == 1);
4284 
4285 	/*
4286 	 * remove map entry for ukh itsself.
4287 	 */
4288 
4289 	va = (vaddr_t)ukh;
4290 	KASSERT((va & PAGE_MASK) == 0);
4291 	vm_map_lock(map);
4292 	uvm_unmap_remove(map, va, va + PAGE_SIZE, &deadentry, NULL, 0);
4293 	KASSERT(deadentry->flags & UVM_MAP_KERNEL);
4294 	KASSERT(deadentry->flags & UVM_MAP_KMAPENT);
4295 	KASSERT(deadentry->next == NULL);
4296 	KASSERT(deadentry == &ukh->ukh_entries[0]);
4297 
4298 	/*
4299 	 * unmap the page from pmap and free it.
4300 	 */
4301 
4302 	pmap = vm_map_pmap(map);
4303 	KASSERT(pmap == pmap_kernel());
4304 	if (!pmap_extract(pmap, va, &pa))
4305 		panic("%s: no mapping", __func__);
4306 	pmap_kremove(va, PAGE_SIZE);
4307 	vm_map_unlock(map);
4308 	pg = PHYS_TO_VM_PAGE(pa);
4309 	uvm_pagefree(pg);
4310 	ukh_free++;
4311 }
4312 
4313 /*
4314  * map entry reservation
4315  */
4316 
4317 /*
4318  * uvm_mapent_reserve: reserve map entries for clipping before locking map.
4319  *
4320  * => needed when unmapping entries allocated without UVM_FLAG_QUANTUM.
4321  * => caller shouldn't hold map locked.
4322  */
4323 int
4324 uvm_mapent_reserve(struct vm_map *map, struct uvm_mapent_reservation *umr,
4325     int nentries, int flags)
4326 {
4327 
4328 	umr->umr_nentries = 0;
4329 
4330 	if ((flags & UVM_FLAG_QUANTUM) != 0)
4331 		return 0;
4332 
4333 	if (!VM_MAP_USE_KMAPENT(map))
4334 		return 0;
4335 
4336 	while (nentries--) {
4337 		struct vm_map_entry *ent;
4338 		ent = uvm_kmapent_alloc(map, flags);
4339 		if (!ent) {
4340 			uvm_mapent_unreserve(map, umr);
4341 			return ENOMEM;
4342 		}
4343 		UMR_PUTENTRY(umr, ent);
4344 	}
4345 
4346 	return 0;
4347 }
4348 
4349 /*
4350  * uvm_mapent_unreserve:
4351  *
4352  * => caller shouldn't hold map locked.
4353  * => never fail or sleep.
4354  */
4355 void
4356 uvm_mapent_unreserve(struct vm_map *map, struct uvm_mapent_reservation *umr)
4357 {
4358 
4359 	while (!UMR_EMPTY(umr))
4360 		uvm_kmapent_free(UMR_GETENTRY(umr));
4361 }
4362 
4363 /*
4364  * uvm_mapent_trymerge: try to merge an entry with its neighbors.
4365  *
4366  * => called with map locked.
4367  * => return non zero if successfully merged.
4368  */
4369 
4370 int
4371 uvm_mapent_trymerge(struct vm_map *map, struct vm_map_entry *entry, int flags)
4372 {
4373 	struct uvm_object *uobj;
4374 	struct vm_map_entry *next;
4375 	struct vm_map_entry *prev;
4376 	vsize_t size;
4377 	int merged = 0;
4378 	boolean_t copying;
4379 	int newetype;
4380 
4381 	if (VM_MAP_USE_KMAPENT(map)) {
4382 		return 0;
4383 	}
4384 	if (entry->aref.ar_amap != NULL) {
4385 		return 0;
4386 	}
4387 	if ((entry->flags & UVM_MAP_NOMERGE) != 0) {
4388 		return 0;
4389 	}
4390 
4391 	uobj = entry->object.uvm_obj;
4392 	size = entry->end - entry->start;
4393 	copying = (flags & UVM_MERGE_COPYING) != 0;
4394 	newetype = copying ? (entry->etype & ~UVM_ET_NEEDSCOPY) : entry->etype;
4395 
4396 	next = entry->next;
4397 	if (next != &map->header &&
4398 	    next->start == entry->end &&
4399 	    ((copying && next->aref.ar_amap != NULL &&
4400 	    amap_refs(next->aref.ar_amap) == 1) ||
4401 	    (!copying && next->aref.ar_amap == NULL)) &&
4402 	    UVM_ET_ISCOMPATIBLE(next, newetype,
4403 	    uobj, entry->flags, entry->protection,
4404 	    entry->max_protection, entry->inheritance, entry->advice,
4405 	    entry->wired_count) &&
4406 	    (uobj == NULL || entry->offset + size == next->offset)) {
4407 		int error;
4408 
4409 		if (copying) {
4410 			error = amap_extend(next, size,
4411 			    AMAP_EXTEND_NOWAIT|AMAP_EXTEND_BACKWARDS);
4412 		} else {
4413 			error = 0;
4414 		}
4415 		if (error == 0) {
4416 			if (uobj) {
4417 				if (uobj->pgops->pgo_detach) {
4418 					uobj->pgops->pgo_detach(uobj);
4419 				}
4420 			}
4421 
4422 			entry->end = next->end;
4423 			uvm_map_entry_unlink(map, next);
4424 			if (copying) {
4425 				entry->aref = next->aref;
4426 				entry->etype &= ~UVM_ET_NEEDSCOPY;
4427 			}
4428 			uvm_tree_sanity(map, "trymerge forwardmerge");
4429 			uvm_mapent_free_merged(map, next);
4430 			merged++;
4431 		}
4432 	}
4433 
4434 	prev = entry->prev;
4435 	if (prev != &map->header &&
4436 	    prev->end == entry->start &&
4437 	    ((copying && !merged && prev->aref.ar_amap != NULL &&
4438 	    amap_refs(prev->aref.ar_amap) == 1) ||
4439 	    (!copying && prev->aref.ar_amap == NULL)) &&
4440 	    UVM_ET_ISCOMPATIBLE(prev, newetype,
4441 	    uobj, entry->flags, entry->protection,
4442 	    entry->max_protection, entry->inheritance, entry->advice,
4443 	    entry->wired_count) &&
4444 	    (uobj == NULL ||
4445 	    prev->offset + prev->end - prev->start == entry->offset)) {
4446 		int error;
4447 
4448 		if (copying) {
4449 			error = amap_extend(prev, size,
4450 			    AMAP_EXTEND_NOWAIT|AMAP_EXTEND_FORWARDS);
4451 		} else {
4452 			error = 0;
4453 		}
4454 		if (error == 0) {
4455 			if (uobj) {
4456 				if (uobj->pgops->pgo_detach) {
4457 					uobj->pgops->pgo_detach(uobj);
4458 				}
4459 				entry->offset = prev->offset;
4460 			}
4461 
4462 			entry->start = prev->start;
4463 			uvm_map_entry_unlink(map, prev);
4464 			if (copying) {
4465 				entry->aref = prev->aref;
4466 				entry->etype &= ~UVM_ET_NEEDSCOPY;
4467 			}
4468 			uvm_tree_sanity(map, "trymerge backmerge");
4469 			uvm_mapent_free_merged(map, prev);
4470 			merged++;
4471 		}
4472 	}
4473 
4474 	return merged;
4475 }
4476 
4477 #if defined(DDB)
4478 
4479 /*
4480  * DDB hooks
4481  */
4482 
4483 /*
4484  * uvm_map_printit: actually prints the map
4485  */
4486 
4487 void
4488 uvm_map_printit(struct vm_map *map, boolean_t full,
4489     void (*pr)(const char *, ...))
4490 {
4491 	struct vm_map_entry *entry;
4492 
4493 	(*pr)("MAP %p: [0x%lx->0x%lx]\n", map, vm_map_min(map),
4494 	    vm_map_max(map));
4495 	(*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
4496 	    map->nentries, map->size, map->ref_count, map->timestamp,
4497 	    map->flags);
4498 	(*pr)("\tpmap=%p(resident=%ld, wired=%ld)\n", map->pmap,
4499 	    pmap_resident_count(map->pmap), pmap_wired_count(map->pmap));
4500 	if (!full)
4501 		return;
4502 	for (entry = map->header.next; entry != &map->header;
4503 	    entry = entry->next) {
4504 		(*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
4505 		    entry, entry->start, entry->end, entry->object.uvm_obj,
4506 		    (long long)entry->offset, entry->aref.ar_amap,
4507 		    entry->aref.ar_pageoff);
4508 		(*pr)(
4509 		    "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, "
4510 		    "wc=%d, adv=%d\n",
4511 		    (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
4512 		    (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
4513 		    (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
4514 		    entry->protection, entry->max_protection,
4515 		    entry->inheritance, entry->wired_count, entry->advice);
4516 	}
4517 }
4518 
4519 /*
4520  * uvm_object_printit: actually prints the object
4521  */
4522 
4523 void
4524 uvm_object_printit(struct uvm_object *uobj, boolean_t full,
4525     void (*pr)(const char *, ...))
4526 {
4527 	struct vm_page *pg;
4528 	int cnt = 0;
4529 
4530 	(*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
4531 	    uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
4532 	if (UVM_OBJ_IS_KERN_OBJECT(uobj))
4533 		(*pr)("refs=<SYSTEM>\n");
4534 	else
4535 		(*pr)("refs=%d\n", uobj->uo_refs);
4536 
4537 	if (!full) {
4538 		return;
4539 	}
4540 	(*pr)("  PAGES <pg,offset>:\n  ");
4541 	TAILQ_FOREACH(pg, &uobj->memq, listq) {
4542 		cnt++;
4543 		(*pr)("<%p,0x%llx> ", pg, (long long)pg->offset);
4544 		if ((cnt % 3) == 0) {
4545 			(*pr)("\n  ");
4546 		}
4547 	}
4548 	if ((cnt % 3) != 0) {
4549 		(*pr)("\n");
4550 	}
4551 }
4552 
4553 /*
4554  * uvm_page_printit: actually print the page
4555  */
4556 
4557 static const char page_flagbits[] =
4558 	"\20\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY"
4559 	"\11ZERO\15PAGER1";
4560 static const char page_pqflagbits[] =
4561 	"\20\1FREE\2INACTIVE\3ACTIVE\5ANON\6AOBJ";
4562 
4563 void
4564 uvm_page_printit(struct vm_page *pg, boolean_t full,
4565     void (*pr)(const char *, ...))
4566 {
4567 	struct vm_page *tpg;
4568 	struct uvm_object *uobj;
4569 	struct pglist *pgl;
4570 	char pgbuf[128];
4571 	char pqbuf[128];
4572 
4573 	(*pr)("PAGE %p:\n", pg);
4574 	bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf));
4575 	bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
4576 	(*pr)("  flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n",
4577 	    pgbuf, pqbuf, pg->wire_count, (long)VM_PAGE_TO_PHYS(pg));
4578 	(*pr)("  uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
4579 	    pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
4580 #if defined(UVM_PAGE_TRKOWN)
4581 	if (pg->flags & PG_BUSY)
4582 		(*pr)("  owning process = %d, tag=%s\n",
4583 		    pg->owner, pg->owner_tag);
4584 	else
4585 		(*pr)("  page not busy, no owner\n");
4586 #else
4587 	(*pr)("  [page ownership tracking disabled]\n");
4588 #endif
4589 
4590 	if (!full)
4591 		return;
4592 
4593 	/* cross-verify object/anon */
4594 	if ((pg->pqflags & PQ_FREE) == 0) {
4595 		if (pg->pqflags & PQ_ANON) {
4596 			if (pg->uanon == NULL || pg->uanon->an_page != pg)
4597 			    (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
4598 				(pg->uanon) ? pg->uanon->an_page : NULL);
4599 			else
4600 				(*pr)("  anon backpointer is OK\n");
4601 		} else {
4602 			uobj = pg->uobject;
4603 			if (uobj) {
4604 				(*pr)("  checking object list\n");
4605 				TAILQ_FOREACH(tpg, &uobj->memq, listq) {
4606 					if (tpg == pg) {
4607 						break;
4608 					}
4609 				}
4610 				if (tpg)
4611 					(*pr)("  page found on object list\n");
4612 				else
4613 			(*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
4614 			}
4615 		}
4616 	}
4617 
4618 	/* cross-verify page queue */
4619 	if (pg->pqflags & PQ_FREE) {
4620 		int fl = uvm_page_lookup_freelist(pg);
4621 		int color = VM_PGCOLOR_BUCKET(pg);
4622 		pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[
4623 		    ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN];
4624 	} else if (pg->pqflags & PQ_INACTIVE) {
4625 		pgl = &uvm.page_inactive;
4626 	} else if (pg->pqflags & PQ_ACTIVE) {
4627 		pgl = &uvm.page_active;
4628 	} else {
4629 		pgl = NULL;
4630 	}
4631 
4632 	if (pgl) {
4633 		(*pr)("  checking pageq list\n");
4634 		TAILQ_FOREACH(tpg, pgl, pageq) {
4635 			if (tpg == pg) {
4636 				break;
4637 			}
4638 		}
4639 		if (tpg)
4640 			(*pr)("  page found on pageq list\n");
4641 		else
4642 			(*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
4643 	}
4644 }
4645 #endif
4646