xref: /netbsd-src/sys/uvm/uvm_map.c (revision 001c68bd94f75ce9270b69227c4199fbf34ee396)
1 /*	$NetBSD: uvm_map.c,v 1.136 2003/04/09 21:39:29 thorpej 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 1.136 2003/04/09 21:39:29 thorpej Exp $");
75 
76 #include "opt_ddb.h"
77 #include "opt_uvmhist.h"
78 #include "opt_sysv.h"
79 
80 #include <sys/param.h>
81 #include <sys/systm.h>
82 #include <sys/mman.h>
83 #include <sys/proc.h>
84 #include <sys/malloc.h>
85 #include <sys/pool.h>
86 #include <sys/kernel.h>
87 #include <sys/mount.h>
88 #include <sys/vnode.h>
89 
90 #ifdef SYSVSHM
91 #include <sys/shm.h>
92 #endif
93 
94 #define UVM_MAP
95 #include <uvm/uvm.h>
96 
97 #ifdef DDB
98 #include <uvm/uvm_ddb.h>
99 #endif
100 
101 extern struct vm_map *pager_map;
102 
103 struct uvm_cnt map_ubackmerge, map_uforwmerge;
104 struct uvm_cnt map_ubimerge, map_unomerge;
105 struct uvm_cnt map_kbackmerge, map_kforwmerge;
106 struct uvm_cnt map_kbimerge, map_knomerge;
107 struct uvm_cnt uvm_map_call, uvm_mlk_call, uvm_mlk_hint;
108 const char vmmapbsy[] = "vmmapbsy";
109 
110 /*
111  * pool for vmspace structures.
112  */
113 
114 struct pool uvm_vmspace_pool;
115 
116 /*
117  * pool for dynamically-allocated map entries.
118  */
119 
120 struct pool uvm_map_entry_pool;
121 struct pool uvm_map_entry_kmem_pool;
122 
123 MALLOC_DEFINE(M_VMMAP, "VM map", "VM map structures");
124 MALLOC_DEFINE(M_VMPMAP, "VM pmap", "VM pmap");
125 
126 #ifdef PMAP_GROWKERNEL
127 /*
128  * This global represents the end of the kernel virtual address
129  * space.  If we want to exceed this, we must grow the kernel
130  * virtual address space dynamically.
131  *
132  * Note, this variable is locked by kernel_map's lock.
133  */
134 vaddr_t uvm_maxkaddr;
135 #endif
136 
137 /*
138  * macros
139  */
140 
141 /*
142  * uvm_map_entry_link: insert entry into a map
143  *
144  * => map must be locked
145  */
146 #define uvm_map_entry_link(map, after_where, entry) do { \
147 	(map)->nentries++; \
148 	(entry)->prev = (after_where); \
149 	(entry)->next = (after_where)->next; \
150 	(entry)->prev->next = (entry); \
151 	(entry)->next->prev = (entry); \
152 } while (/*CONSTCOND*/ 0)
153 
154 /*
155  * uvm_map_entry_unlink: remove entry from a map
156  *
157  * => map must be locked
158  */
159 #define uvm_map_entry_unlink(map, entry) do { \
160 	(map)->nentries--; \
161 	(entry)->next->prev = (entry)->prev; \
162 	(entry)->prev->next = (entry)->next; \
163 } while (/*CONSTCOND*/ 0)
164 
165 /*
166  * SAVE_HINT: saves the specified entry as the hint for future lookups.
167  *
168  * => map need not be locked (protected by hint_lock).
169  */
170 #define SAVE_HINT(map,check,value) do { \
171 	simple_lock(&(map)->hint_lock); \
172 	if ((map)->hint == (check)) \
173 		(map)->hint = (value); \
174 	simple_unlock(&(map)->hint_lock); \
175 } while (/*CONSTCOND*/ 0)
176 
177 /*
178  * VM_MAP_RANGE_CHECK: check and correct range
179  *
180  * => map must at least be read locked
181  */
182 
183 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
184 	if (start < vm_map_min(map)) 		\
185 		start = vm_map_min(map);        \
186 	if (end > vm_map_max(map))              \
187 		end = vm_map_max(map);          \
188 	if (start > end)                        \
189 		start = end;                    \
190 } while (/*CONSTCOND*/ 0)
191 
192 /*
193  * local prototypes
194  */
195 
196 static struct vm_map_entry *uvm_mapent_alloc __P((struct vm_map *, int));
197 static void uvm_mapent_copy __P((struct vm_map_entry *, struct vm_map_entry *));
198 static void uvm_mapent_free __P((struct vm_map_entry *));
199 static void uvm_map_entry_unwire __P((struct vm_map *, struct vm_map_entry *));
200 static void uvm_map_reference_amap __P((struct vm_map_entry *, int));
201 static void uvm_map_unreference_amap __P((struct vm_map_entry *, int));
202 
203 /*
204  * local inlines
205  */
206 
207 /*
208  * uvm_mapent_alloc: allocate a map entry
209  */
210 
211 static __inline struct vm_map_entry *
212 uvm_mapent_alloc(map, flags)
213 	struct vm_map *map;
214 	int flags;
215 {
216 	struct vm_map_entry *me;
217 	int s;
218 	int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK;
219 	UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist);
220 
221 	if (map->flags & VM_MAP_INTRSAFE || cold) {
222 		s = splvm();
223 		simple_lock(&uvm.kentry_lock);
224 		me = uvm.kentry_free;
225 		if (me) uvm.kentry_free = me->next;
226 		simple_unlock(&uvm.kentry_lock);
227 		splx(s);
228 		if (__predict_false(me == NULL)) {
229 			panic("uvm_mapent_alloc: out of static map entries, "
230 			      "check MAX_KMAPENT (currently %d)",
231 			      MAX_KMAPENT);
232 		}
233 		me->flags = UVM_MAP_STATIC;
234 	} else if (map == kernel_map) {
235 		me = pool_get(&uvm_map_entry_kmem_pool, pflags);
236 		if (__predict_false(me == NULL))
237 			return NULL;
238 		me->flags = UVM_MAP_KMEM;
239 	} else {
240 		me = pool_get(&uvm_map_entry_pool, pflags);
241 		if (__predict_false(me == NULL))
242 			return NULL;
243 		me->flags = 0;
244 	}
245 
246 	UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me,
247 	    ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0);
248 	return(me);
249 }
250 
251 /*
252  * uvm_mapent_free: free map entry
253  */
254 
255 static __inline void
256 uvm_mapent_free(me)
257 	struct vm_map_entry *me;
258 {
259 	int s;
260 	UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist);
261 
262 	UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
263 		me, me->flags, 0, 0);
264 	if (me->flags & UVM_MAP_STATIC) {
265 		s = splvm();
266 		simple_lock(&uvm.kentry_lock);
267 		me->next = uvm.kentry_free;
268 		uvm.kentry_free = me;
269 		simple_unlock(&uvm.kentry_lock);
270 		splx(s);
271 	} else if (me->flags & UVM_MAP_KMEM) {
272 		pool_put(&uvm_map_entry_kmem_pool, me);
273 	} else {
274 		pool_put(&uvm_map_entry_pool, me);
275 	}
276 }
277 
278 /*
279  * uvm_mapent_copy: copy a map entry, preserving flags
280  */
281 
282 static __inline void
283 uvm_mapent_copy(src, dst)
284 	struct vm_map_entry *src;
285 	struct vm_map_entry *dst;
286 {
287 	memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) -
288 	   ((char *)src));
289 }
290 
291 /*
292  * uvm_map_entry_unwire: unwire a map entry
293  *
294  * => map should be locked by caller
295  */
296 
297 static __inline void
298 uvm_map_entry_unwire(map, entry)
299 	struct vm_map *map;
300 	struct vm_map_entry *entry;
301 {
302 	entry->wired_count = 0;
303 	uvm_fault_unwire_locked(map, entry->start, entry->end);
304 }
305 
306 
307 /*
308  * wrapper for calling amap_ref()
309  */
310 static __inline void
311 uvm_map_reference_amap(entry, flags)
312 	struct vm_map_entry *entry;
313 	int flags;
314 {
315 	amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff,
316 	     (entry->end - entry->start) >> PAGE_SHIFT, flags);
317 }
318 
319 
320 /*
321  * wrapper for calling amap_unref()
322  */
323 static __inline void
324 uvm_map_unreference_amap(entry, flags)
325 	struct vm_map_entry *entry;
326 	int flags;
327 {
328 	amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff,
329 	     (entry->end - entry->start) >> PAGE_SHIFT, flags);
330 }
331 
332 
333 /*
334  * uvm_map_init: init mapping system at boot time.   note that we allocate
335  * and init the static pool of struct vm_map_entry *'s for the kernel here.
336  */
337 
338 void
339 uvm_map_init()
340 {
341 	static struct vm_map_entry kernel_map_entry[MAX_KMAPENT];
342 #if defined(UVMHIST)
343 	static struct uvm_history_ent maphistbuf[100];
344 	static struct uvm_history_ent pdhistbuf[100];
345 #endif
346 	int lcv;
347 
348 	/*
349 	 * first, init logging system.
350 	 */
351 
352 	UVMHIST_FUNC("uvm_map_init");
353 	UVMHIST_INIT_STATIC(maphist, maphistbuf);
354 	UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
355 	UVMHIST_CALLED(maphist);
356 	UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
357 	UVMCNT_INIT(uvm_map_call,  UVMCNT_CNT, 0,
358 	    "# uvm_map() successful calls", 0);
359 
360 	UVMCNT_INIT(map_ubackmerge, UVMCNT_CNT, 0,
361 	    "# uvm_map() back umerges", 0);
362 	UVMCNT_INIT(map_uforwmerge, UVMCNT_CNT, 0,
363 	    "# uvm_map() forward umerges", 0);
364 	UVMCNT_INIT(map_ubimerge, UVMCNT_CNT, 0,
365 	    "# uvm_map() dual umerge", 0);
366 	UVMCNT_INIT(map_unomerge, UVMCNT_CNT, 0,
367 	    "# uvm_map() no umerge", 0);
368 
369 	UVMCNT_INIT(map_kbackmerge, UVMCNT_CNT, 0,
370 	    "# uvm_map() back kmerges", 0);
371 	UVMCNT_INIT(map_kforwmerge, UVMCNT_CNT, 0,
372 	    "# uvm_map() forward kmerges", 0);
373 	UVMCNT_INIT(map_kbimerge, UVMCNT_CNT, 0,
374 	    "# uvm_map() dual kmerge", 0);
375 	UVMCNT_INIT(map_knomerge, UVMCNT_CNT, 0,
376 	    "# uvm_map() no kmerge", 0);
377 
378 	UVMCNT_INIT(uvm_mlk_call,  UVMCNT_CNT, 0, "# map lookup calls", 0);
379 	UVMCNT_INIT(uvm_mlk_hint,  UVMCNT_CNT, 0, "# map lookup hint hits", 0);
380 
381 	/*
382 	 * now set up static pool of kernel map entrys ...
383 	 */
384 
385 	simple_lock_init(&uvm.kentry_lock);
386 	uvm.kentry_free = NULL;
387 	for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) {
388 		kernel_map_entry[lcv].next = uvm.kentry_free;
389 		uvm.kentry_free = &kernel_map_entry[lcv];
390 	}
391 
392 	/*
393 	 * initialize the map-related pools.
394 	 */
395 	pool_init(&uvm_vmspace_pool, sizeof(struct vmspace),
396 	    0, 0, 0, "vmsppl", &pool_allocator_nointr);
397 	pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry),
398 	    0, 0, 0, "vmmpepl", &pool_allocator_nointr);
399 	pool_init(&uvm_map_entry_kmem_pool, sizeof(struct vm_map_entry),
400 	    0, 0, 0, "vmmpekpl", NULL);
401 }
402 
403 /*
404  * clippers
405  */
406 
407 /*
408  * uvm_map_clip_start: ensure that the entry begins at or after
409  *	the starting address, if it doesn't we split the entry.
410  *
411  * => caller should use UVM_MAP_CLIP_START macro rather than calling
412  *    this directly
413  * => map must be locked by caller
414  */
415 
416 void
417 uvm_map_clip_start(map, entry, start)
418 	struct vm_map *map;
419 	struct vm_map_entry *entry;
420 	vaddr_t start;
421 {
422 	struct vm_map_entry *new_entry;
423 	vaddr_t new_adj;
424 
425 	/* uvm_map_simplify_entry(map, entry); */ /* XXX */
426 
427 	/*
428 	 * Split off the front portion.  note that we must insert the new
429 	 * entry BEFORE this one, so that this entry has the specified
430 	 * starting address.
431 	 */
432 
433 	new_entry = uvm_mapent_alloc(map, 0);
434 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
435 
436 	new_entry->end = start;
437 	new_adj = start - new_entry->start;
438 	if (entry->object.uvm_obj)
439 		entry->offset += new_adj;	/* shift start over */
440 	entry->start = start;
441 
442 	if (new_entry->aref.ar_amap) {
443 		amap_splitref(&new_entry->aref, &entry->aref, new_adj);
444 	}
445 
446 	uvm_map_entry_link(map, entry->prev, new_entry);
447 
448 	if (UVM_ET_ISSUBMAP(entry)) {
449 		/* ... unlikely to happen, but play it safe */
450 		 uvm_map_reference(new_entry->object.sub_map);
451 	} else {
452 		if (UVM_ET_ISOBJ(entry) &&
453 		    entry->object.uvm_obj->pgops &&
454 		    entry->object.uvm_obj->pgops->pgo_reference)
455 			entry->object.uvm_obj->pgops->pgo_reference(
456 			    entry->object.uvm_obj);
457 	}
458 }
459 
460 /*
461  * uvm_map_clip_end: ensure that the entry ends at or before
462  *	the ending address, if it does't we split the reference
463  *
464  * => caller should use UVM_MAP_CLIP_END macro rather than calling
465  *    this directly
466  * => map must be locked by caller
467  */
468 
469 void
470 uvm_map_clip_end(map, entry, end)
471 	struct vm_map *map;
472 	struct vm_map_entry *entry;
473 	vaddr_t	end;
474 {
475 	struct vm_map_entry *	new_entry;
476 	vaddr_t new_adj; /* #bytes we move start forward */
477 
478 	/*
479 	 *	Create a new entry and insert it
480 	 *	AFTER the specified entry
481 	 */
482 
483 	new_entry = uvm_mapent_alloc(map, 0);
484 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
485 
486 	new_entry->start = entry->end = end;
487 	new_adj = end - entry->start;
488 	if (new_entry->object.uvm_obj)
489 		new_entry->offset += new_adj;
490 
491 	if (entry->aref.ar_amap)
492 		amap_splitref(&entry->aref, &new_entry->aref, new_adj);
493 
494 	uvm_map_entry_link(map, entry, new_entry);
495 
496 	if (UVM_ET_ISSUBMAP(entry)) {
497 		/* ... unlikely to happen, but play it safe */
498 	 	uvm_map_reference(new_entry->object.sub_map);
499 	} else {
500 		if (UVM_ET_ISOBJ(entry) &&
501 		    entry->object.uvm_obj->pgops &&
502 		    entry->object.uvm_obj->pgops->pgo_reference)
503 			entry->object.uvm_obj->pgops->pgo_reference(
504 			    entry->object.uvm_obj);
505 	}
506 }
507 
508 
509 /*
510  *   M A P   -   m a i n   e n t r y   p o i n t
511  */
512 /*
513  * uvm_map: establish a valid mapping in a map
514  *
515  * => assume startp is page aligned.
516  * => assume size is a multiple of PAGE_SIZE.
517  * => assume sys_mmap provides enough of a "hint" to have us skip
518  *	over text/data/bss area.
519  * => map must be unlocked (we will lock it)
520  * => <uobj,uoffset> value meanings (4 cases):
521  *	 [1] <NULL,uoffset> 		== uoffset is a hint for PMAP_PREFER
522  *	 [2] <NULL,UVM_UNKNOWN_OFFSET>	== don't PMAP_PREFER
523  *	 [3] <uobj,uoffset>		== normal mapping
524  *	 [4] <uobj,UVM_UNKNOWN_OFFSET>	== uvm_map finds offset based on VA
525  *
526  *    case [4] is for kernel mappings where we don't know the offset until
527  *    we've found a virtual address.   note that kernel object offsets are
528  *    always relative to vm_map_min(kernel_map).
529  *
530  * => if `align' is non-zero, we try to align the virtual address to
531  *	the specified alignment.  this is only a hint; if we can't
532  *	do it, the address will be unaligned.  this is provided as
533  *	a mechanism for large pages.
534  *
535  * => XXXCDC: need way to map in external amap?
536  */
537 
538 int
539 uvm_map(map, startp, size, uobj, uoffset, align, flags)
540 	struct vm_map *map;
541 	vaddr_t *startp;	/* IN/OUT */
542 	vsize_t size;
543 	struct uvm_object *uobj;
544 	voff_t uoffset;
545 	vsize_t align;
546 	uvm_flag_t flags;
547 {
548 	struct vm_map_entry *prev_entry, *new_entry;
549 	const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ?
550 	    AMAP_EXTEND_NOWAIT : 0;
551 	vm_prot_t prot = UVM_PROTECTION(flags), maxprot =
552 	    UVM_MAXPROTECTION(flags);
553 	vm_inherit_t inherit = UVM_INHERIT(flags);
554 	int advice = UVM_ADVICE(flags);
555 	int error, merged = 0, kmap = (vm_map_pmap(map) == pmap_kernel());
556 	UVMHIST_FUNC("uvm_map");
557 	UVMHIST_CALLED(maphist);
558 
559 	UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)",
560 	    map, *startp, size, flags);
561 	UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
562 
563 	/*
564 	 * detect a popular device driver bug.
565 	 */
566 
567 	KASSERT(doing_shutdown || curlwp != NULL ||
568 	    (map->flags & VM_MAP_INTRSAFE));
569 
570 	/*
571 	 * check sanity of protection code
572 	 */
573 
574 	if ((prot & maxprot) != prot) {
575 		UVMHIST_LOG(maphist, "<- prot. failure:  prot=0x%x, max=0x%x",
576 		prot, maxprot,0,0);
577 		return EACCES;
578 	}
579 
580 	/*
581 	 * for pager_map, allocate the new entry first to avoid sleeping
582 	 * for memory while we have the map locked.
583 	 */
584 
585 	new_entry = NULL;
586 	if (map == pager_map) {
587 		new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT));
588 		 if (__predict_false(new_entry == NULL))
589 			return ENOMEM;
590 	}
591 
592 	/*
593 	 * figure out where to put new VM range
594 	 */
595 
596 	if (vm_map_lock_try(map) == FALSE) {
597 		if (flags & UVM_FLAG_TRYLOCK) {
598 			if (new_entry) {
599 				uvm_mapent_free(new_entry);
600 			}
601 			return EAGAIN;
602 		}
603 		vm_map_lock(map); /* could sleep here */
604 	}
605 	if ((prev_entry = uvm_map_findspace(map, *startp, size, startp,
606 	    uobj, uoffset, align, flags)) == NULL) {
607 		UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0);
608 		vm_map_unlock(map);
609 		if (new_entry) {
610 			uvm_mapent_free(new_entry);
611 		}
612 		return ENOMEM;
613 	}
614 
615 #ifdef PMAP_GROWKERNEL
616 	{
617 		/*
618 		 * If the kernel pmap can't map the requested space,
619 		 * then allocate more resources for it.
620 		 */
621 		if (map == kernel_map && uvm_maxkaddr < (*startp + size))
622 			uvm_maxkaddr = pmap_growkernel(*startp + size);
623 	}
624 #endif
625 
626 	UVMCNT_INCR(uvm_map_call);
627 
628 	/*
629 	 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
630 	 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET.   in
631 	 * either case we want to zero it  before storing it in the map entry
632 	 * (because it looks strange and confusing when debugging...)
633 	 *
634 	 * if uobj is not null
635 	 *   if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
636 	 *      and we do not need to change uoffset.
637 	 *   if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
638 	 *      now (based on the starting address of the map).   this case is
639 	 *      for kernel object mappings where we don't know the offset until
640 	 *      the virtual address is found (with uvm_map_findspace).   the
641 	 *      offset is the distance we are from the start of the map.
642 	 */
643 
644 	if (uobj == NULL) {
645 		uoffset = 0;
646 	} else {
647 		if (uoffset == UVM_UNKNOWN_OFFSET) {
648 			KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj));
649 			uoffset = *startp - vm_map_min(kernel_map);
650 		}
651 	}
652 
653 	/*
654 	 * try and insert in map by extending previous entry, if possible.
655 	 * XXX: we don't try and pull back the next entry.   might be useful
656 	 * for a stack, but we are currently allocating our stack in advance.
657 	 */
658 
659 	if (flags & UVM_FLAG_NOMERGE)
660 		goto nomerge;
661 
662 	if (prev_entry->end == *startp &&
663 	    prev_entry != &map->header &&
664 	    prev_entry->object.uvm_obj == uobj) {
665 
666 		if (uobj && prev_entry->offset +
667 		    (prev_entry->end - prev_entry->start) != uoffset)
668 			goto forwardmerge;
669 
670 		if (UVM_ET_ISSUBMAP(prev_entry))
671 			goto forwardmerge;
672 
673 		if (prev_entry->protection != prot ||
674 		    prev_entry->max_protection != maxprot)
675 			goto forwardmerge;
676 
677 		if (prev_entry->inheritance != inherit ||
678 		    prev_entry->advice != advice)
679 			goto forwardmerge;
680 
681 		/* wiring status must match (new area is unwired) */
682 		if (VM_MAPENT_ISWIRED(prev_entry))
683 			goto forwardmerge;
684 
685 		/*
686 		 * can't extend a shared amap.  note: no need to lock amap to
687 		 * look at refs since we don't care about its exact value.
688 		 * if it is one (i.e. we have only reference) it will stay there
689 		 */
690 
691 		if (prev_entry->aref.ar_amap &&
692 		    amap_refs(prev_entry->aref.ar_amap) != 1) {
693 			goto forwardmerge;
694 		}
695 
696 		if (prev_entry->aref.ar_amap) {
697 			error = amap_extend(prev_entry, size,
698 			    amapwaitflag | AMAP_EXTEND_FORWARDS);
699 			if (error) {
700 				vm_map_unlock(map);
701 				if (new_entry) {
702 					uvm_mapent_free(new_entry);
703 				}
704 				return error;
705 			}
706 		}
707 
708 		if (kmap)
709 			UVMCNT_INCR(map_kbackmerge);
710 		else
711 			UVMCNT_INCR(map_ubackmerge);
712 		UVMHIST_LOG(maphist,"  starting back merge", 0, 0, 0, 0);
713 
714 		/*
715 		 * drop our reference to uobj since we are extending a reference
716 		 * that we already have (the ref count can not drop to zero).
717 		 */
718 
719 		if (uobj && uobj->pgops->pgo_detach)
720 			uobj->pgops->pgo_detach(uobj);
721 
722 		prev_entry->end += size;
723 		map->size += size;
724 
725 		UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
726 		if (new_entry) {
727 			uvm_mapent_free(new_entry);
728 			new_entry = NULL;
729 		}
730 		merged++;
731 	}
732 
733 forwardmerge:
734 	if (prev_entry->next->start == (*startp + size) &&
735 	    prev_entry->next != &map->header &&
736 	    prev_entry->next->object.uvm_obj == uobj) {
737 
738 		if (uobj && prev_entry->next->offset != uoffset + size)
739 			goto nomerge;
740 
741 		if (UVM_ET_ISSUBMAP(prev_entry->next))
742 			goto nomerge;
743 
744 		if (prev_entry->next->protection != prot ||
745 		    prev_entry->next->max_protection != maxprot)
746 			goto nomerge;
747 
748 		if (prev_entry->next->inheritance != inherit ||
749 		    prev_entry->next->advice != advice)
750 			goto nomerge;
751 
752 		/* wiring status must match (new area is unwired) */
753 		if (VM_MAPENT_ISWIRED(prev_entry->next))
754 			goto nomerge;
755 
756 		/*
757 		 * can't extend a shared amap.  note: no need to lock amap to
758 		 * look at refs since we don't care about its exact value.
759 		 * if it is one (i.e. we have only reference) it will stay there.
760 		 *
761 		 * note that we also can't merge two amaps, so if we
762 		 * merged with the previous entry which has an amap,
763 		 * and the next entry also has an amap, we give up.
764 		 *
765 		 * Interesting cases:
766 		 * amap, new, amap -> give up second merge (single fwd extend)
767 		 * amap, new, none -> double forward extend (extend again here)
768 		 * none, new, amap -> double backward extend (done here)
769 		 * uobj, new, amap -> single backward extend (done here)
770 		 *
771 		 * XXX should we attempt to deal with someone refilling
772 		 * the deallocated region between two entries that are
773 		 * backed by the same amap (ie, arefs is 2, "prev" and
774 		 * "next" refer to it, and adding this allocation will
775 		 * close the hole, thus restoring arefs to 1 and
776 		 * deallocating the "next" vm_map_entry)?  -- @@@
777 		 */
778 
779 		if (prev_entry->next->aref.ar_amap &&
780 		    (amap_refs(prev_entry->next->aref.ar_amap) != 1 ||
781 		     (merged && prev_entry->aref.ar_amap))) {
782 			goto nomerge;
783 		}
784 
785 		if (merged) {
786 			/*
787 			 * Try to extend the amap of the previous entry to
788 			 * cover the next entry as well.  If it doesn't work
789 			 * just skip on, don't actually give up, since we've
790 			 * already completed the back merge.
791 			 */
792 			if (prev_entry->aref.ar_amap) {
793 				if (amap_extend(prev_entry,
794 				    prev_entry->next->end -
795 				    prev_entry->next->start,
796 				    amapwaitflag | AMAP_EXTEND_FORWARDS))
797 				goto nomerge;
798 			}
799 
800 			/*
801 			 * Try to extend the amap of the *next* entry
802 			 * back to cover the new allocation *and* the
803 			 * previous entry as well (the previous merge
804 			 * didn't have an amap already otherwise we
805 			 * wouldn't be checking here for an amap).  If
806 			 * it doesn't work just skip on, again, don't
807 			 * actually give up, since we've already
808 			 * completed the back merge.
809 			 */
810 			else if (prev_entry->next->aref.ar_amap) {
811 				if (amap_extend(prev_entry->next,
812 				    prev_entry->end -
813 				    prev_entry->start + size,
814 				    amapwaitflag | AMAP_EXTEND_BACKWARDS))
815 				goto nomerge;
816 			}
817 		} else {
818 			/*
819 			 * Pull the next entry's amap backwards to cover this
820 			 * new allocation.
821 			 */
822 			if (prev_entry->next->aref.ar_amap) {
823 				error = amap_extend(prev_entry->next, size,
824 				    amapwaitflag | AMAP_EXTEND_BACKWARDS);
825 				if (error) {
826 					vm_map_unlock(map);
827 					if (new_entry) {
828 						uvm_mapent_free(new_entry);
829 					}
830 					return error;
831 				}
832 			}
833 		}
834 
835 		if (merged) {
836 			if (kmap) {
837 				UVMCNT_DECR(map_kbackmerge);
838 				UVMCNT_INCR(map_kbimerge);
839 			} else {
840 				UVMCNT_DECR(map_ubackmerge);
841 				UVMCNT_INCR(map_ubimerge);
842 			}
843 		} else {
844 			if (kmap)
845 				UVMCNT_INCR(map_kforwmerge);
846 			else
847 				UVMCNT_INCR(map_uforwmerge);
848 		}
849 		UVMHIST_LOG(maphist,"  starting forward merge", 0, 0, 0, 0);
850 
851 		/*
852 		 * drop our reference to uobj since we are extending a reference
853 		 * that we already have (the ref count can not drop to zero).
854 		 * (if merged, we've already detached)
855 		 */
856 		if (uobj && uobj->pgops->pgo_detach && !merged)
857 			uobj->pgops->pgo_detach(uobj);
858 
859 		if (merged) {
860 			struct vm_map_entry *dead = prev_entry->next;
861 			prev_entry->end = dead->end;
862 			uvm_map_entry_unlink(map, dead);
863 			if (dead->aref.ar_amap != NULL) {
864 				prev_entry->aref = dead->aref;
865 				dead->aref.ar_amap = NULL;
866 			}
867 			uvm_mapent_free(dead);
868 		} else {
869 			prev_entry->next->start -= size;
870 			map->size += size;
871 			if (uobj)
872 				prev_entry->next->offset = uoffset;
873 		}
874 
875 		UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0);
876 		if (new_entry) {
877 			uvm_mapent_free(new_entry);
878 			new_entry = NULL;
879 		}
880 		merged++;
881 	}
882 
883 nomerge:
884 	if (!merged) {
885 		UVMHIST_LOG(maphist,"  allocating new map entry", 0, 0, 0, 0);
886 		if (kmap)
887 			UVMCNT_INCR(map_knomerge);
888 		else
889 			UVMCNT_INCR(map_unomerge);
890 
891 		/*
892 		 * allocate new entry and link it in.
893 		 */
894 
895 		if (new_entry == NULL) {
896 			new_entry = uvm_mapent_alloc(map,
897 				(flags & UVM_FLAG_NOWAIT));
898 			if (__predict_false(new_entry == NULL)) {
899 				vm_map_unlock(map);
900 				return ENOMEM;
901 			}
902 		}
903 		new_entry->start = *startp;
904 		new_entry->end = new_entry->start + size;
905 		new_entry->object.uvm_obj = uobj;
906 		new_entry->offset = uoffset;
907 
908 		if (uobj)
909 			new_entry->etype = UVM_ET_OBJ;
910 		else
911 			new_entry->etype = 0;
912 
913 		if (flags & UVM_FLAG_COPYONW) {
914 			new_entry->etype |= UVM_ET_COPYONWRITE;
915 			if ((flags & UVM_FLAG_OVERLAY) == 0)
916 				new_entry->etype |= UVM_ET_NEEDSCOPY;
917 		}
918 
919 		new_entry->protection = prot;
920 		new_entry->max_protection = maxprot;
921 		new_entry->inheritance = inherit;
922 		new_entry->wired_count = 0;
923 		new_entry->advice = advice;
924 		if (flags & UVM_FLAG_OVERLAY) {
925 
926 			/*
927 			 * to_add: for BSS we overallocate a little since we
928 			 * are likely to extend
929 			 */
930 
931 			vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
932 				UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
933 			struct vm_amap *amap = amap_alloc(size, to_add,
934 			    (flags & UVM_FLAG_NOWAIT) ? M_NOWAIT : M_WAITOK);
935 			if (__predict_false(amap == NULL)) {
936 				vm_map_unlock(map);
937 				uvm_mapent_free(new_entry);
938 				return ENOMEM;
939 			}
940 			new_entry->aref.ar_pageoff = 0;
941 			new_entry->aref.ar_amap = amap;
942 		} else {
943 			new_entry->aref.ar_pageoff = 0;
944 			new_entry->aref.ar_amap = NULL;
945 		}
946 		uvm_map_entry_link(map, prev_entry, new_entry);
947 		map->size += size;
948 
949 		/*
950 		 * Update the free space hint
951 		 */
952 
953 		if ((map->first_free == prev_entry) &&
954 		    (prev_entry->end >= new_entry->start))
955 			map->first_free = new_entry;
956 	}
957 
958 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
959 	vm_map_unlock(map);
960 	return 0;
961 }
962 
963 /*
964  * uvm_map_lookup_entry: find map entry at or before an address
965  *
966  * => map must at least be read-locked by caller
967  * => entry is returned in "entry"
968  * => return value is true if address is in the returned entry
969  */
970 
971 boolean_t
972 uvm_map_lookup_entry(map, address, entry)
973 	struct vm_map *map;
974 	vaddr_t	address;
975 	struct vm_map_entry **entry;		/* OUT */
976 {
977 	struct vm_map_entry *cur;
978 	struct vm_map_entry *last;
979 	UVMHIST_FUNC("uvm_map_lookup_entry");
980 	UVMHIST_CALLED(maphist);
981 
982 	UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
983 	    map, address, entry, 0);
984 
985 	/*
986 	 * start looking either from the head of the
987 	 * list, or from the hint.
988 	 */
989 
990 	simple_lock(&map->hint_lock);
991 	cur = map->hint;
992 	simple_unlock(&map->hint_lock);
993 
994 	if (cur == &map->header)
995 		cur = cur->next;
996 
997 	UVMCNT_INCR(uvm_mlk_call);
998 	if (address >= cur->start) {
999 
1000 	    	/*
1001 		 * go from hint to end of list.
1002 		 *
1003 		 * but first, make a quick check to see if
1004 		 * we are already looking at the entry we
1005 		 * want (which is usually the case).
1006 		 * note also that we don't need to save the hint
1007 		 * here... it is the same hint (unless we are
1008 		 * at the header, in which case the hint didn't
1009 		 * buy us anything anyway).
1010 		 */
1011 
1012 		last = &map->header;
1013 		if ((cur != last) && (cur->end > address)) {
1014 			UVMCNT_INCR(uvm_mlk_hint);
1015 			*entry = cur;
1016 			UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
1017 			    cur, 0, 0, 0);
1018 			return (TRUE);
1019 		}
1020 	} else {
1021 
1022 	    	/*
1023 		 * go from start to hint, *inclusively*
1024 		 */
1025 
1026 		last = cur->next;
1027 		cur = map->header.next;
1028 	}
1029 
1030 	/*
1031 	 * search linearly
1032 	 */
1033 
1034 	while (cur != last) {
1035 		if (cur->end > address) {
1036 			if (address >= cur->start) {
1037 			    	/*
1038 				 * save this lookup for future
1039 				 * hints, and return
1040 				 */
1041 
1042 				*entry = cur;
1043 				SAVE_HINT(map, map->hint, cur);
1044 				UVMHIST_LOG(maphist,"<- search got it (0x%x)",
1045 					cur, 0, 0, 0);
1046 				return (TRUE);
1047 			}
1048 			break;
1049 		}
1050 		cur = cur->next;
1051 	}
1052 	*entry = cur->prev;
1053 	SAVE_HINT(map, map->hint, *entry);
1054 	UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
1055 	return (FALSE);
1056 }
1057 
1058 /*
1059  * uvm_map_findspace: find "length" sized space in "map".
1060  *
1061  * => "hint" is a hint about where we want it, unless FINDSPACE_FIXED is
1062  *	set (in which case we insist on using "hint").
1063  * => "result" is VA returned
1064  * => uobj/uoffset are to be used to handle VAC alignment, if required
1065  * => if `align' is non-zero, we attempt to align to that value.
1066  * => caller must at least have read-locked map
1067  * => returns NULL on failure, or pointer to prev. map entry if success
1068  * => note this is a cross between the old vm_map_findspace and vm_map_find
1069  */
1070 
1071 struct vm_map_entry *
1072 uvm_map_findspace(map, hint, length, result, uobj, uoffset, align, flags)
1073 	struct vm_map *map;
1074 	vaddr_t hint;
1075 	vsize_t length;
1076 	vaddr_t *result; /* OUT */
1077 	struct uvm_object *uobj;
1078 	voff_t uoffset;
1079 	vsize_t align;
1080 	int flags;
1081 {
1082 	struct vm_map_entry *entry, *next, *tmp;
1083 	vaddr_t end, orig_hint;
1084 	const int topdown = map->flags & VM_MAP_TOPDOWN;
1085 	UVMHIST_FUNC("uvm_map_findspace");
1086 	UVMHIST_CALLED(maphist);
1087 
1088 	UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)",
1089 		    map, hint, length, flags);
1090 	KASSERT((align & (align - 1)) == 0);
1091 	KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0);
1092 
1093 	/*
1094 	 * remember the original hint.  if we are aligning, then we
1095 	 * may have to try again with no alignment constraint if
1096 	 * we fail the first time.
1097 	 */
1098 
1099 	orig_hint = hint;
1100 	if (hint < map->min_offset) {	/* check ranges ... */
1101 		if (flags & UVM_FLAG_FIXED) {
1102 			UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
1103 			return(NULL);
1104 		}
1105 		hint = map->min_offset;
1106 	}
1107 	if (hint > map->max_offset) {
1108 		UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
1109 				hint, map->min_offset, map->max_offset, 0);
1110 		return(NULL);
1111 	}
1112 
1113 	/*
1114 	 * Look for the first possible address; if there's already
1115 	 * something at this address, we have to start after it.
1116 	 */
1117 
1118 	/*
1119 	 * @@@: there are four, no, eight cases to consider.
1120 	 *
1121 	 * 0: found,     fixed,     bottom up -> fail
1122 	 * 1: found,     fixed,     top down  -> fail
1123 	 * 2: found,     not fixed, bottom up -> start after tmp->end, loop up
1124 	 * 3: found,     not fixed, top down  -> start before tmp->start, loop down
1125 	 * 4: not found, fixed,     bottom up -> check tmp->next->start, fail
1126 	 * 5: not found, fixed,     top down  -> check tmp->next->start, fail
1127 	 * 6: not found, not fixed, bottom up -> check tmp->next->start, loop up
1128 	 * 7: not found, not fixed, top down  -> check tmp->next->start, loop down
1129 	 *
1130 	 * as you can see, it reduces to roughly five cases, and that
1131 	 * adding top down mapping only adds one unique case (without
1132 	 * it, there would be four cases).
1133 	 */
1134 
1135 	if ((flags & UVM_FLAG_FIXED) == 0 && hint == map->min_offset) {
1136 		if ((entry = map->first_free) != &map->header)
1137 			hint = entry->end;
1138 	} else {
1139 		if (uvm_map_lookup_entry(map, hint, &tmp)) {
1140 			/* "hint" address already in use ... */
1141 			if (flags & UVM_FLAG_FIXED) {
1142 				UVMHIST_LOG(maphist,"<- fixed & VA in use",
1143 				    0, 0, 0, 0);
1144 				return(NULL);
1145 			}
1146 			hint = topdown ? tmp->start - length : tmp->end;
1147 		} else if ((tmp->next == &map->header ||
1148 			    tmp->next->start >= hint + length)) {
1149 			entry = tmp;
1150 			goto quickfind;
1151 		}
1152 		entry = tmp;
1153 	}
1154 
1155 	/*
1156 	 * Look through the rest of the map, trying to fit a new region in
1157 	 * the gap between existing regions, or after the very last region.
1158 	 * note: entry->end   = base VA of current gap,
1159 	 *	 next->start  = VA of end of current gap
1160 	 */
1161 
1162 	for (next = NULL;; hint = topdown ?
1163 		(entry = next)->start - length :
1164 		(entry = next)->end) {
1165 
1166 		/*
1167 		 * Find the end of the proposed new region.  Be sure we didn't
1168 		 * go beyond the end of the map, or wrap around the address;
1169 		 * if so, we lose.  Otherwise, if this is the last entry, or
1170 		 * if the proposed new region fits before the next entry, we
1171 		 * win.
1172 		 */
1173 
1174 #ifdef PMAP_PREFER
1175 		/*
1176 		 * push hint forward as needed to avoid VAC alias problems.
1177 		 * we only do this if a valid offset is specified.
1178 		 */
1179 
1180 		if ((flags & UVM_FLAG_FIXED) == 0 &&
1181 		    uoffset != UVM_UNKNOWN_OFFSET)
1182 			PMAP_PREFER(uoffset, &hint);
1183 #endif
1184 		if (align != 0) {
1185 			if ((hint & (align - 1)) != 0) {
1186 				if (topdown)
1187 					hint &= ~(align-1);
1188 				else
1189 					hint = roundup(hint, align);
1190 			}
1191 			/*
1192 			 * XXX Should we PMAP_PREFER() here again?
1193 			 */
1194 		}
1195 		end = hint + length;
1196 		if (end > map->max_offset || end < hint) {
1197 			UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0);
1198 			if (align != 0) {
1199 				UVMHIST_LOG(maphist,
1200 				    "calling recursively, no align",
1201 				    0,0,0,0);
1202 				return (uvm_map_findspace(map, orig_hint,
1203 				    length, result, uobj, uoffset, 0, flags));
1204 			}
1205 			return (NULL);
1206 		}
1207 		if (!topdown || next == NULL) {
1208 			next = entry->next;
1209 		} else
1210 			next = entry->prev;
1211 		if (next == &map->header ||
1212 		    (!topdown && next->start >= end) ||
1213 		    ( topdown && next->end   <= hint))
1214 			break;
1215 		if (flags & UVM_FLAG_FIXED) {
1216 			UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0);
1217 			return(NULL); /* only one shot at it ... */
1218 		}
1219 	}
1220  quickfind:
1221 	SAVE_HINT(map, map->hint, entry);
1222 	if (topdown && entry->start >= hint + length)
1223 		entry = entry->prev;
1224 	*result = hint;
1225 	UVMHIST_LOG(maphist,"<- got it!  (result=0x%x)", hint, 0,0,0);
1226 	return (entry);
1227 }
1228 
1229 /*
1230  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
1231  */
1232 
1233 /*
1234  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
1235  *
1236  * => caller must check alignment and size
1237  * => map must be locked by caller
1238  * => we return a list of map entries that we've remove from the map
1239  *    in "entry_list"
1240  */
1241 
1242 void
1243 uvm_unmap_remove(map, start, end, entry_list)
1244 	struct vm_map *map;
1245 	vaddr_t start, end;
1246 	struct vm_map_entry **entry_list;	/* OUT */
1247 {
1248 	struct vm_map_entry *entry, *first_entry, *next;
1249 	vaddr_t len;
1250 	UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist);
1251 
1252 	UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
1253 	    map, start, end, 0);
1254 	VM_MAP_RANGE_CHECK(map, start, end);
1255 
1256 	/*
1257 	 * find first entry
1258 	 */
1259 
1260 	if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
1261 		/* clip and go... */
1262 		entry = first_entry;
1263 		UVM_MAP_CLIP_START(map, entry, start);
1264 		/* critical!  prevents stale hint */
1265 		SAVE_HINT(map, entry, entry->prev);
1266 	} else {
1267 		entry = first_entry->next;
1268 	}
1269 
1270 	/*
1271 	 * Save the free space hint
1272 	 */
1273 
1274 	if (map->first_free->start >= start)
1275 		map->first_free = entry->prev;
1276 
1277 	/*
1278 	 * note: we now re-use first_entry for a different task.  we remove
1279 	 * a number of map entries from the map and save them in a linked
1280 	 * list headed by "first_entry".  once we remove them from the map
1281 	 * the caller should unlock the map and drop the references to the
1282 	 * backing objects [c.f. uvm_unmap_detach].  the object is to
1283 	 * separate unmapping from reference dropping.  why?
1284 	 *   [1] the map has to be locked for unmapping
1285 	 *   [2] the map need not be locked for reference dropping
1286 	 *   [3] dropping references may trigger pager I/O, and if we hit
1287 	 *       a pager that does synchronous I/O we may have to wait for it.
1288 	 *   [4] we would like all waiting for I/O to occur with maps unlocked
1289 	 *       so that we don't block other threads.
1290 	 */
1291 
1292 	first_entry = NULL;
1293 	*entry_list = NULL;
1294 
1295 	/*
1296 	 * break up the area into map entry sized regions and unmap.  note
1297 	 * that all mappings have to be removed before we can even consider
1298 	 * dropping references to amaps or VM objects (otherwise we could end
1299 	 * up with a mapping to a page on the free list which would be very bad)
1300 	 */
1301 
1302 	while ((entry != &map->header) && (entry->start < end)) {
1303 		UVM_MAP_CLIP_END(map, entry, end);
1304 		next = entry->next;
1305 		len = entry->end - entry->start;
1306 
1307 		/*
1308 		 * unwire before removing addresses from the pmap; otherwise
1309 		 * unwiring will put the entries back into the pmap (XXX).
1310 		 */
1311 
1312 		if (VM_MAPENT_ISWIRED(entry)) {
1313 			uvm_map_entry_unwire(map, entry);
1314 		}
1315 		if ((map->flags & VM_MAP_PAGEABLE) == 0) {
1316 
1317 			/*
1318 			 * if the map is non-pageable, any pages mapped there
1319 			 * must be wired and entered with pmap_kenter_pa(),
1320 			 * and we should free any such pages immediately.
1321 			 * this is mostly used for kmem_map and mb_map.
1322 			 */
1323 
1324 			uvm_km_pgremove_intrsafe(entry->start, entry->end);
1325 			pmap_kremove(entry->start, len);
1326 		} else if (UVM_ET_ISOBJ(entry) &&
1327 			   UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
1328 			KASSERT(vm_map_pmap(map) == pmap_kernel());
1329 
1330 			/*
1331 			 * note: kernel object mappings are currently used in
1332 			 * two ways:
1333 			 *  [1] "normal" mappings of pages in the kernel object
1334 			 *  [2] uvm_km_valloc'd allocations in which we
1335 			 *      pmap_enter in some non-kernel-object page
1336 			 *      (e.g. vmapbuf).
1337 			 *
1338 			 * for case [1], we need to remove the mapping from
1339 			 * the pmap and then remove the page from the kernel
1340 			 * object (because, once pages in a kernel object are
1341 			 * unmapped they are no longer needed, unlike, say,
1342 			 * a vnode where you might want the data to persist
1343 			 * until flushed out of a queue).
1344 			 *
1345 			 * for case [2], we need to remove the mapping from
1346 			 * the pmap.  there shouldn't be any pages at the
1347 			 * specified offset in the kernel object [but it
1348 			 * doesn't hurt to call uvm_km_pgremove just to be
1349 			 * safe?]
1350 			 *
1351 			 * uvm_km_pgremove currently does the following:
1352 			 *   for pages in the kernel object in range:
1353 			 *     - drops the swap slot
1354 			 *     - uvm_pagefree the page
1355 			 */
1356 
1357 			/*
1358 			 * remove mappings from pmap and drop the pages
1359 			 * from the object.  offsets are always relative
1360 			 * to vm_map_min(kernel_map).
1361 			 */
1362 
1363 			pmap_remove(pmap_kernel(), entry->start,
1364 			    entry->start + len);
1365 			uvm_km_pgremove(entry->object.uvm_obj,
1366 			    entry->start - vm_map_min(kernel_map),
1367 			    entry->end - vm_map_min(kernel_map));
1368 
1369 			/*
1370 			 * null out kernel_object reference, we've just
1371 			 * dropped it
1372 			 */
1373 
1374 			entry->etype &= ~UVM_ET_OBJ;
1375 			entry->object.uvm_obj = NULL;
1376 		} else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) {
1377 
1378 			/*
1379 		 	 * remove mappings the standard way.
1380 		 	 */
1381 
1382 			pmap_remove(map->pmap, entry->start, entry->end);
1383 		}
1384 
1385 		/*
1386 		 * remove entry from map and put it on our list of entries
1387 		 * that we've nuked.  then go to next entry.
1388 		 */
1389 
1390 		UVMHIST_LOG(maphist, "  removed map entry 0x%x", entry, 0, 0,0);
1391 
1392 		/* critical!  prevents stale hint */
1393 		SAVE_HINT(map, entry, entry->prev);
1394 
1395 		uvm_map_entry_unlink(map, entry);
1396 		map->size -= len;
1397 		entry->prev = NULL;
1398 		entry->next = first_entry;
1399 		first_entry = entry;
1400 		entry = next;
1401 	}
1402 	if ((map->flags & VM_MAP_DYING) == 0) {
1403 		pmap_update(vm_map_pmap(map));
1404 	}
1405 
1406 	/*
1407 	 * now we've cleaned up the map and are ready for the caller to drop
1408 	 * references to the mapped objects.
1409 	 */
1410 
1411 	*entry_list = first_entry;
1412 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1413 }
1414 
1415 /*
1416  * uvm_unmap_detach: drop references in a chain of map entries
1417  *
1418  * => we will free the map entries as we traverse the list.
1419  */
1420 
1421 void
1422 uvm_unmap_detach(first_entry, flags)
1423 	struct vm_map_entry *first_entry;
1424 	int flags;
1425 {
1426 	struct vm_map_entry *next_entry;
1427 	UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1428 
1429 	while (first_entry) {
1430 		KASSERT(!VM_MAPENT_ISWIRED(first_entry));
1431 		UVMHIST_LOG(maphist,
1432 		    "  detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
1433 		    first_entry, first_entry->aref.ar_amap,
1434 		    first_entry->object.uvm_obj,
1435 		    UVM_ET_ISSUBMAP(first_entry));
1436 
1437 		/*
1438 		 * drop reference to amap, if we've got one
1439 		 */
1440 
1441 		if (first_entry->aref.ar_amap)
1442 			uvm_map_unreference_amap(first_entry, flags);
1443 
1444 		/*
1445 		 * drop reference to our backing object, if we've got one
1446 		 */
1447 
1448 		KASSERT(!UVM_ET_ISSUBMAP(first_entry));
1449 		if (UVM_ET_ISOBJ(first_entry) &&
1450 		    first_entry->object.uvm_obj->pgops->pgo_detach) {
1451 			(*first_entry->object.uvm_obj->pgops->pgo_detach)
1452 				(first_entry->object.uvm_obj);
1453 		}
1454 		next_entry = first_entry->next;
1455 		uvm_mapent_free(first_entry);
1456 		first_entry = next_entry;
1457 	}
1458 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1459 }
1460 
1461 /*
1462  *   E X T R A C T I O N   F U N C T I O N S
1463  */
1464 
1465 /*
1466  * uvm_map_reserve: reserve space in a vm_map for future use.
1467  *
1468  * => we reserve space in a map by putting a dummy map entry in the
1469  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
1470  * => map should be unlocked (we will write lock it)
1471  * => we return true if we were able to reserve space
1472  * => XXXCDC: should be inline?
1473  */
1474 
1475 int
1476 uvm_map_reserve(map, size, offset, align, raddr)
1477 	struct vm_map *map;
1478 	vsize_t size;
1479 	vaddr_t offset;	/* hint for pmap_prefer */
1480 	vsize_t align;	/* alignment hint */
1481 	vaddr_t *raddr;	/* IN:hint, OUT: reserved VA */
1482 {
1483 	UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1484 
1485 	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1486 	      map,size,offset,raddr);
1487 
1488 	size = round_page(size);
1489 	if (*raddr < vm_map_min(map))
1490 		*raddr = vm_map_min(map);                /* hint */
1491 
1492 	/*
1493 	 * reserve some virtual space.
1494 	 */
1495 
1496 	if (uvm_map(map, raddr, size, NULL, offset, 0,
1497 	    UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1498 	    UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) {
1499 	    UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1500 		return (FALSE);
1501 	}
1502 
1503 	UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1504 	return (TRUE);
1505 }
1506 
1507 /*
1508  * uvm_map_replace: replace a reserved (blank) area of memory with
1509  * real mappings.
1510  *
1511  * => caller must WRITE-LOCK the map
1512  * => we return TRUE if replacement was a success
1513  * => we expect the newents chain to have nnewents entrys on it and
1514  *    we expect newents->prev to point to the last entry on the list
1515  * => note newents is allowed to be NULL
1516  */
1517 
1518 int
1519 uvm_map_replace(map, start, end, newents, nnewents)
1520 	struct vm_map *map;
1521 	vaddr_t start, end;
1522 	struct vm_map_entry *newents;
1523 	int nnewents;
1524 {
1525 	struct vm_map_entry *oldent, *last;
1526 
1527 	/*
1528 	 * first find the blank map entry at the specified address
1529 	 */
1530 
1531 	if (!uvm_map_lookup_entry(map, start, &oldent)) {
1532 		return(FALSE);
1533 	}
1534 
1535 	/*
1536 	 * check to make sure we have a proper blank entry
1537 	 */
1538 
1539 	if (oldent->start != start || oldent->end != end ||
1540 	    oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1541 		return (FALSE);
1542 	}
1543 
1544 #ifdef DIAGNOSTIC
1545 
1546 	/*
1547 	 * sanity check the newents chain
1548 	 */
1549 
1550 	{
1551 		struct vm_map_entry *tmpent = newents;
1552 		int nent = 0;
1553 		vaddr_t cur = start;
1554 
1555 		while (tmpent) {
1556 			nent++;
1557 			if (tmpent->start < cur)
1558 				panic("uvm_map_replace1");
1559 			if (tmpent->start > tmpent->end || tmpent->end > end) {
1560 		printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
1561 			    tmpent->start, tmpent->end, end);
1562 				panic("uvm_map_replace2");
1563 			}
1564 			cur = tmpent->end;
1565 			if (tmpent->next) {
1566 				if (tmpent->next->prev != tmpent)
1567 					panic("uvm_map_replace3");
1568 			} else {
1569 				if (newents->prev != tmpent)
1570 					panic("uvm_map_replace4");
1571 			}
1572 			tmpent = tmpent->next;
1573 		}
1574 		if (nent != nnewents)
1575 			panic("uvm_map_replace5");
1576 	}
1577 #endif
1578 
1579 	/*
1580 	 * map entry is a valid blank!   replace it.   (this does all the
1581 	 * work of map entry link/unlink...).
1582 	 */
1583 
1584 	if (newents) {
1585 		last = newents->prev;
1586 
1587 		/* critical: flush stale hints out of map */
1588 		SAVE_HINT(map, map->hint, newents);
1589 		if (map->first_free == oldent)
1590 			map->first_free = last;
1591 
1592 		last->next = oldent->next;
1593 		last->next->prev = last;
1594 		newents->prev = oldent->prev;
1595 		newents->prev->next = newents;
1596 		map->nentries = map->nentries + (nnewents - 1);
1597 
1598 	} else {
1599 
1600 		/* critical: flush stale hints out of map */
1601 		SAVE_HINT(map, map->hint, oldent->prev);
1602 		if (map->first_free == oldent)
1603 			map->first_free = oldent->prev;
1604 
1605 		/* NULL list of new entries: just remove the old one */
1606 		uvm_map_entry_unlink(map, oldent);
1607 	}
1608 
1609 
1610 	/*
1611 	 * now we can free the old blank entry, unlock the map and return.
1612 	 */
1613 
1614 	uvm_mapent_free(oldent);
1615 	return(TRUE);
1616 }
1617 
1618 /*
1619  * uvm_map_extract: extract a mapping from a map and put it somewhere
1620  *	(maybe removing the old mapping)
1621  *
1622  * => maps should be unlocked (we will write lock them)
1623  * => returns 0 on success, error code otherwise
1624  * => start must be page aligned
1625  * => len must be page sized
1626  * => flags:
1627  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
1628  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
1629  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
1630  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
1631  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
1632  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
1633  *             be used from within the kernel in a kernel level map <<<
1634  */
1635 
1636 int
1637 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags)
1638 	struct vm_map *srcmap, *dstmap;
1639 	vaddr_t start, *dstaddrp;
1640 	vsize_t len;
1641 	int flags;
1642 {
1643 	vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge,
1644 	    oldstart;
1645 	struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry,
1646 	    *deadentry, *oldentry;
1647 	vsize_t elen;
1648 	int nchain, error, copy_ok;
1649 	UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1650 
1651 	UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
1652 	    len,0);
1653 	UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
1654 
1655 	/*
1656 	 * step 0: sanity check: start must be on a page boundary, length
1657 	 * must be page sized.  can't ask for CONTIG/QREF if you asked for
1658 	 * REMOVE.
1659 	 */
1660 
1661 	KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0);
1662 	KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 ||
1663 		(flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0);
1664 
1665 	/*
1666 	 * step 1: reserve space in the target map for the extracted area
1667 	 */
1668 
1669 	dstaddr = vm_map_min(dstmap);
1670 	if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE)
1671 		return(ENOMEM);
1672 	*dstaddrp = dstaddr;	/* pass address back to caller */
1673 	UVMHIST_LOG(maphist, "  dstaddr=0x%x", dstaddr,0,0,0);
1674 
1675 	/*
1676 	 * step 2: setup for the extraction process loop by init'ing the
1677 	 * map entry chain, locking src map, and looking up the first useful
1678 	 * entry in the map.
1679 	 */
1680 
1681 	end = start + len;
1682 	newend = dstaddr + len;
1683 	chain = endchain = NULL;
1684 	nchain = 0;
1685 	vm_map_lock(srcmap);
1686 
1687 	if (uvm_map_lookup_entry(srcmap, start, &entry)) {
1688 
1689 		/* "start" is within an entry */
1690 		if (flags & UVM_EXTRACT_QREF) {
1691 
1692 			/*
1693 			 * for quick references we don't clip the entry, so
1694 			 * the entry may map space "before" the starting
1695 			 * virtual address... this is the "fudge" factor
1696 			 * (which can be non-zero only the first time
1697 			 * through the "while" loop in step 3).
1698 			 */
1699 
1700 			fudge = start - entry->start;
1701 		} else {
1702 
1703 			/*
1704 			 * normal reference: we clip the map to fit (thus
1705 			 * fudge is zero)
1706 			 */
1707 
1708 			UVM_MAP_CLIP_START(srcmap, entry, start);
1709 			SAVE_HINT(srcmap, srcmap->hint, entry->prev);
1710 			fudge = 0;
1711 		}
1712 	} else {
1713 
1714 		/* "start" is not within an entry ... skip to next entry */
1715 		if (flags & UVM_EXTRACT_CONTIG) {
1716 			error = EINVAL;
1717 			goto bad;    /* definite hole here ... */
1718 		}
1719 
1720 		entry = entry->next;
1721 		fudge = 0;
1722 	}
1723 
1724 	/* save values from srcmap for step 6 */
1725 	orig_entry = entry;
1726 	orig_fudge = fudge;
1727 
1728 	/*
1729 	 * step 3: now start looping through the map entries, extracting
1730 	 * as we go.
1731 	 */
1732 
1733 	while (entry->start < end && entry != &srcmap->header) {
1734 
1735 		/* if we are not doing a quick reference, clip it */
1736 		if ((flags & UVM_EXTRACT_QREF) == 0)
1737 			UVM_MAP_CLIP_END(srcmap, entry, end);
1738 
1739 		/* clear needs_copy (allow chunking) */
1740 		if (UVM_ET_ISNEEDSCOPY(entry)) {
1741 			if (fudge)
1742 				oldstart = entry->start;
1743 			else
1744 				oldstart = 0;	/* XXX: gcc */
1745 			amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
1746 			if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
1747 				error = ENOMEM;
1748 				goto bad;
1749 			}
1750 
1751 			/* amap_copy could clip (during chunk)!  update fudge */
1752 			if (fudge) {
1753 				fudge = fudge - (entry->start - oldstart);
1754 				orig_fudge = fudge;
1755 			}
1756 		}
1757 
1758 		/* calculate the offset of this from "start" */
1759 		oldoffset = (entry->start + fudge) - start;
1760 
1761 		/* allocate a new map entry */
1762 		newentry = uvm_mapent_alloc(dstmap, 0);
1763 		if (newentry == NULL) {
1764 			error = ENOMEM;
1765 			goto bad;
1766 		}
1767 
1768 		/* set up new map entry */
1769 		newentry->next = NULL;
1770 		newentry->prev = endchain;
1771 		newentry->start = dstaddr + oldoffset;
1772 		newentry->end =
1773 		    newentry->start + (entry->end - (entry->start + fudge));
1774 		if (newentry->end > newend || newentry->end < newentry->start)
1775 			newentry->end = newend;
1776 		newentry->object.uvm_obj = entry->object.uvm_obj;
1777 		if (newentry->object.uvm_obj) {
1778 			if (newentry->object.uvm_obj->pgops->pgo_reference)
1779 				newentry->object.uvm_obj->pgops->
1780 				    pgo_reference(newentry->object.uvm_obj);
1781 				newentry->offset = entry->offset + fudge;
1782 		} else {
1783 			newentry->offset = 0;
1784 		}
1785 		newentry->etype = entry->etype;
1786 		newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
1787 			entry->max_protection : entry->protection;
1788 		newentry->max_protection = entry->max_protection;
1789 		newentry->inheritance = entry->inheritance;
1790 		newentry->wired_count = 0;
1791 		newentry->aref.ar_amap = entry->aref.ar_amap;
1792 		if (newentry->aref.ar_amap) {
1793 			newentry->aref.ar_pageoff =
1794 			    entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1795 			uvm_map_reference_amap(newentry, AMAP_SHARED |
1796 			    ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
1797 		} else {
1798 			newentry->aref.ar_pageoff = 0;
1799 		}
1800 		newentry->advice = entry->advice;
1801 
1802 		/* now link it on the chain */
1803 		nchain++;
1804 		if (endchain == NULL) {
1805 			chain = endchain = newentry;
1806 		} else {
1807 			endchain->next = newentry;
1808 			endchain = newentry;
1809 		}
1810 
1811 		/* end of 'while' loop! */
1812 		if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1813 		    (entry->next == &srcmap->header ||
1814 		    entry->next->start != entry->end)) {
1815 			error = EINVAL;
1816 			goto bad;
1817 		}
1818 		entry = entry->next;
1819 		fudge = 0;
1820 	}
1821 
1822 	/*
1823 	 * step 4: close off chain (in format expected by uvm_map_replace)
1824 	 */
1825 
1826 	if (chain)
1827 		chain->prev = endchain;
1828 
1829 	/*
1830 	 * step 5: attempt to lock the dest map so we can pmap_copy.
1831 	 * note usage of copy_ok:
1832 	 *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
1833 	 *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
1834 	 */
1835 
1836 	if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
1837 		copy_ok = 1;
1838 		if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1839 		    nchain)) {
1840 			if (srcmap != dstmap)
1841 				vm_map_unlock(dstmap);
1842 			error = EIO;
1843 			goto bad;
1844 		}
1845 	} else {
1846 		copy_ok = 0;
1847 		/* replace defered until step 7 */
1848 	}
1849 
1850 	/*
1851 	 * step 6: traverse the srcmap a second time to do the following:
1852 	 *  - if we got a lock on the dstmap do pmap_copy
1853 	 *  - if UVM_EXTRACT_REMOVE remove the entries
1854 	 * we make use of orig_entry and orig_fudge (saved in step 2)
1855 	 */
1856 
1857 	if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
1858 
1859 		/* purge possible stale hints from srcmap */
1860 		if (flags & UVM_EXTRACT_REMOVE) {
1861 			SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev);
1862 			if (srcmap->first_free->start >= start)
1863 				srcmap->first_free = orig_entry->prev;
1864 		}
1865 
1866 		entry = orig_entry;
1867 		fudge = orig_fudge;
1868 		deadentry = NULL;	/* for UVM_EXTRACT_REMOVE */
1869 
1870 		while (entry->start < end && entry != &srcmap->header) {
1871 			if (copy_ok) {
1872 				oldoffset = (entry->start + fudge) - start;
1873 				elen = MIN(end, entry->end) -
1874 				    (entry->start + fudge);
1875 				pmap_copy(dstmap->pmap, srcmap->pmap,
1876 				    dstaddr + oldoffset, elen,
1877 				    entry->start + fudge);
1878 			}
1879 
1880 			/* we advance "entry" in the following if statement */
1881 			if (flags & UVM_EXTRACT_REMOVE) {
1882 				pmap_remove(srcmap->pmap, entry->start,
1883 						entry->end);
1884         			oldentry = entry;	/* save entry */
1885         			entry = entry->next;	/* advance */
1886 				uvm_map_entry_unlink(srcmap, oldentry);
1887 							/* add to dead list */
1888 				oldentry->next = deadentry;
1889 				deadentry = oldentry;
1890       			} else {
1891         			entry = entry->next;		/* advance */
1892 			}
1893 
1894 			/* end of 'while' loop */
1895 			fudge = 0;
1896 		}
1897 		pmap_update(srcmap->pmap);
1898 
1899 		/*
1900 		 * unlock dstmap.  we will dispose of deadentry in
1901 		 * step 7 if needed
1902 		 */
1903 
1904 		if (copy_ok && srcmap != dstmap)
1905 			vm_map_unlock(dstmap);
1906 
1907 	} else {
1908 		deadentry = NULL;
1909 	}
1910 
1911 	/*
1912 	 * step 7: we are done with the source map, unlock.   if copy_ok
1913 	 * is 0 then we have not replaced the dummy mapping in dstmap yet
1914 	 * and we need to do so now.
1915 	 */
1916 
1917 	vm_map_unlock(srcmap);
1918 	if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
1919 		uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
1920 
1921 	/* now do the replacement if we didn't do it in step 5 */
1922 	if (copy_ok == 0) {
1923 		vm_map_lock(dstmap);
1924 		error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1925 		    nchain);
1926 		vm_map_unlock(dstmap);
1927 
1928 		if (error == FALSE) {
1929 			error = EIO;
1930 			goto bad2;
1931 		}
1932 	}
1933 	return(0);
1934 
1935 	/*
1936 	 * bad: failure recovery
1937 	 */
1938 bad:
1939 	vm_map_unlock(srcmap);
1940 bad2:			/* src already unlocked */
1941 	if (chain)
1942 		uvm_unmap_detach(chain,
1943 		    (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1944 	uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
1945 	return(error);
1946 }
1947 
1948 /* end of extraction functions */
1949 
1950 /*
1951  * uvm_map_submap: punch down part of a map into a submap
1952  *
1953  * => only the kernel_map is allowed to be submapped
1954  * => the purpose of submapping is to break up the locking granularity
1955  *	of a larger map
1956  * => the range specified must have been mapped previously with a uvm_map()
1957  *	call [with uobj==NULL] to create a blank map entry in the main map.
1958  *	[And it had better still be blank!]
1959  * => maps which contain submaps should never be copied or forked.
1960  * => to remove a submap, use uvm_unmap() on the main map
1961  *	and then uvm_map_deallocate() the submap.
1962  * => main map must be unlocked.
1963  * => submap must have been init'd and have a zero reference count.
1964  *	[need not be locked as we don't actually reference it]
1965  */
1966 
1967 int
1968 uvm_map_submap(map, start, end, submap)
1969 	struct vm_map *map, *submap;
1970 	vaddr_t start, end;
1971 {
1972 	struct vm_map_entry *entry;
1973 	int error;
1974 
1975 	vm_map_lock(map);
1976 	VM_MAP_RANGE_CHECK(map, start, end);
1977 
1978 	if (uvm_map_lookup_entry(map, start, &entry)) {
1979 		UVM_MAP_CLIP_START(map, entry, start);
1980 		UVM_MAP_CLIP_END(map, entry, end);		/* to be safe */
1981 	} else {
1982 		entry = NULL;
1983 	}
1984 
1985 	if (entry != NULL &&
1986 	    entry->start == start && entry->end == end &&
1987 	    entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
1988 	    !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1989 		entry->etype |= UVM_ET_SUBMAP;
1990 		entry->object.sub_map = submap;
1991 		entry->offset = 0;
1992 		uvm_map_reference(submap);
1993 		error = 0;
1994 	} else {
1995 		error = EINVAL;
1996 	}
1997 	vm_map_unlock(map);
1998 	return error;
1999 }
2000 
2001 
2002 /*
2003  * uvm_map_protect: change map protection
2004  *
2005  * => set_max means set max_protection.
2006  * => map must be unlocked.
2007  */
2008 
2009 #define MASK(entry)     (UVM_ET_ISCOPYONWRITE(entry) ? \
2010 			 ~VM_PROT_WRITE : VM_PROT_ALL)
2011 
2012 int
2013 uvm_map_protect(map, start, end, new_prot, set_max)
2014 	struct vm_map *map;
2015 	vaddr_t start, end;
2016 	vm_prot_t new_prot;
2017 	boolean_t set_max;
2018 {
2019 	struct vm_map_entry *current, *entry;
2020 	int error = 0;
2021 	UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
2022 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
2023 		    map, start, end, new_prot);
2024 
2025 	vm_map_lock(map);
2026 	VM_MAP_RANGE_CHECK(map, start, end);
2027 	if (uvm_map_lookup_entry(map, start, &entry)) {
2028 		UVM_MAP_CLIP_START(map, entry, start);
2029 	} else {
2030 		entry = entry->next;
2031 	}
2032 
2033 	/*
2034 	 * make a first pass to check for protection violations.
2035 	 */
2036 
2037 	current = entry;
2038 	while ((current != &map->header) && (current->start < end)) {
2039 		if (UVM_ET_ISSUBMAP(current)) {
2040 			error = EINVAL;
2041 			goto out;
2042 		}
2043 		if ((new_prot & current->max_protection) != new_prot) {
2044 			error = EACCES;
2045 			goto out;
2046 		}
2047 		/*
2048 		 * Don't allow VM_PROT_EXECUTE to be set on entries that
2049 		 * point to vnodes that are associated with a NOEXEC file
2050 		 * system.
2051 		 */
2052 		if (UVM_ET_ISOBJ(current) &&
2053 		    UVM_OBJ_IS_VNODE(current->object.uvm_obj)) {
2054 			struct vnode *vp =
2055 			    (struct vnode *) current->object.uvm_obj;
2056 
2057 			if ((new_prot & VM_PROT_EXECUTE) != 0 &&
2058 			    (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
2059 				error = EACCES;
2060 				goto out;
2061 			}
2062 		}
2063 		current = current->next;
2064 	}
2065 
2066 	/* go back and fix up protections (no need to clip this time). */
2067 
2068 	current = entry;
2069 	while ((current != &map->header) && (current->start < end)) {
2070 		vm_prot_t old_prot;
2071 
2072 		UVM_MAP_CLIP_END(map, current, end);
2073 		old_prot = current->protection;
2074 		if (set_max)
2075 			current->protection =
2076 			    (current->max_protection = new_prot) & old_prot;
2077 		else
2078 			current->protection = new_prot;
2079 
2080 		/*
2081 		 * update physical map if necessary.  worry about copy-on-write
2082 		 * here -- CHECK THIS XXX
2083 		 */
2084 
2085 		if (current->protection != old_prot) {
2086 			/* update pmap! */
2087 			pmap_protect(map->pmap, current->start, current->end,
2088 			    current->protection & MASK(entry));
2089 
2090 			/*
2091 			 * If this entry points at a vnode, and the
2092 			 * protection includes VM_PROT_EXECUTE, mark
2093 			 * the vnode as VEXECMAP.
2094 			 */
2095 			if (UVM_ET_ISOBJ(current)) {
2096 				struct uvm_object *uobj =
2097 				    current->object.uvm_obj;
2098 
2099 				if (UVM_OBJ_IS_VNODE(uobj) &&
2100 				    (current->protection & VM_PROT_EXECUTE))
2101 					vn_markexec((struct vnode *) uobj);
2102 			}
2103 		}
2104 
2105 		/*
2106 		 * If the map is configured to lock any future mappings,
2107 		 * wire this entry now if the old protection was VM_PROT_NONE
2108 		 * and the new protection is not VM_PROT_NONE.
2109 		 */
2110 
2111 		if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
2112 		    VM_MAPENT_ISWIRED(entry) == 0 &&
2113 		    old_prot == VM_PROT_NONE &&
2114 		    new_prot != VM_PROT_NONE) {
2115 			if (uvm_map_pageable(map, entry->start,
2116 			    entry->end, FALSE,
2117 			    UVM_LK_ENTER|UVM_LK_EXIT) != 0) {
2118 
2119 				/*
2120 				 * If locking the entry fails, remember the
2121 				 * error if it's the first one.  Note we
2122 				 * still continue setting the protection in
2123 				 * the map, but will return the error
2124 				 * condition regardless.
2125 				 *
2126 				 * XXX Ignore what the actual error is,
2127 				 * XXX just call it a resource shortage
2128 				 * XXX so that it doesn't get confused
2129 				 * XXX what uvm_map_protect() itself would
2130 				 * XXX normally return.
2131 				 */
2132 
2133 				error = ENOMEM;
2134 			}
2135 		}
2136 		current = current->next;
2137 	}
2138 	pmap_update(map->pmap);
2139 
2140  out:
2141 	vm_map_unlock(map);
2142 	UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0);
2143 	return error;
2144 }
2145 
2146 #undef  MASK
2147 
2148 /*
2149  * uvm_map_inherit: set inheritance code for range of addrs in map.
2150  *
2151  * => map must be unlocked
2152  * => note that the inherit code is used during a "fork".  see fork
2153  *	code for details.
2154  */
2155 
2156 int
2157 uvm_map_inherit(map, start, end, new_inheritance)
2158 	struct vm_map *map;
2159 	vaddr_t start;
2160 	vaddr_t end;
2161 	vm_inherit_t new_inheritance;
2162 {
2163 	struct vm_map_entry *entry, *temp_entry;
2164 	UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
2165 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
2166 	    map, start, end, new_inheritance);
2167 
2168 	switch (new_inheritance) {
2169 	case MAP_INHERIT_NONE:
2170 	case MAP_INHERIT_COPY:
2171 	case MAP_INHERIT_SHARE:
2172 		break;
2173 	default:
2174 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2175 		return EINVAL;
2176 	}
2177 
2178 	vm_map_lock(map);
2179 	VM_MAP_RANGE_CHECK(map, start, end);
2180 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
2181 		entry = temp_entry;
2182 		UVM_MAP_CLIP_START(map, entry, start);
2183 	}  else {
2184 		entry = temp_entry->next;
2185 	}
2186 	while ((entry != &map->header) && (entry->start < end)) {
2187 		UVM_MAP_CLIP_END(map, entry, end);
2188 		entry->inheritance = new_inheritance;
2189 		entry = entry->next;
2190 	}
2191 	vm_map_unlock(map);
2192 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
2193 	return 0;
2194 }
2195 
2196 /*
2197  * uvm_map_advice: set advice code for range of addrs in map.
2198  *
2199  * => map must be unlocked
2200  */
2201 
2202 int
2203 uvm_map_advice(map, start, end, new_advice)
2204 	struct vm_map *map;
2205 	vaddr_t start;
2206 	vaddr_t end;
2207 	int new_advice;
2208 {
2209 	struct vm_map_entry *entry, *temp_entry;
2210 	UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
2211 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
2212 	    map, start, end, new_advice);
2213 
2214 	vm_map_lock(map);
2215 	VM_MAP_RANGE_CHECK(map, start, end);
2216 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
2217 		entry = temp_entry;
2218 		UVM_MAP_CLIP_START(map, entry, start);
2219 	} else {
2220 		entry = temp_entry->next;
2221 	}
2222 
2223 	/*
2224 	 * XXXJRT: disallow holes?
2225 	 */
2226 
2227 	while ((entry != &map->header) && (entry->start < end)) {
2228 		UVM_MAP_CLIP_END(map, entry, end);
2229 
2230 		switch (new_advice) {
2231 		case MADV_NORMAL:
2232 		case MADV_RANDOM:
2233 		case MADV_SEQUENTIAL:
2234 			/* nothing special here */
2235 			break;
2236 
2237 		default:
2238 			vm_map_unlock(map);
2239 			UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2240 			return EINVAL;
2241 		}
2242 		entry->advice = new_advice;
2243 		entry = entry->next;
2244 	}
2245 
2246 	vm_map_unlock(map);
2247 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
2248 	return 0;
2249 }
2250 
2251 /*
2252  * uvm_map_pageable: sets the pageability of a range in a map.
2253  *
2254  * => wires map entries.  should not be used for transient page locking.
2255  *	for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
2256  * => regions sepcified as not pageable require lock-down (wired) memory
2257  *	and page tables.
2258  * => map must never be read-locked
2259  * => if islocked is TRUE, map is already write-locked
2260  * => we always unlock the map, since we must downgrade to a read-lock
2261  *	to call uvm_fault_wire()
2262  * => XXXCDC: check this and try and clean it up.
2263  */
2264 
2265 int
2266 uvm_map_pageable(map, start, end, new_pageable, lockflags)
2267 	struct vm_map *map;
2268 	vaddr_t start, end;
2269 	boolean_t new_pageable;
2270 	int lockflags;
2271 {
2272 	struct vm_map_entry *entry, *start_entry, *failed_entry;
2273 	int rv;
2274 #ifdef DIAGNOSTIC
2275 	u_int timestamp_save;
2276 #endif
2277 	UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
2278 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
2279 		    map, start, end, new_pageable);
2280 	KASSERT(map->flags & VM_MAP_PAGEABLE);
2281 
2282 	if ((lockflags & UVM_LK_ENTER) == 0)
2283 		vm_map_lock(map);
2284 	VM_MAP_RANGE_CHECK(map, start, end);
2285 
2286 	/*
2287 	 * only one pageability change may take place at one time, since
2288 	 * uvm_fault_wire assumes it will be called only once for each
2289 	 * wiring/unwiring.  therefore, we have to make sure we're actually
2290 	 * changing the pageability for the entire region.  we do so before
2291 	 * making any changes.
2292 	 */
2293 
2294 	if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
2295 		if ((lockflags & UVM_LK_EXIT) == 0)
2296 			vm_map_unlock(map);
2297 
2298 		UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0);
2299 		return EFAULT;
2300 	}
2301 	entry = start_entry;
2302 
2303 	/*
2304 	 * handle wiring and unwiring separately.
2305 	 */
2306 
2307 	if (new_pageable) {		/* unwire */
2308 		UVM_MAP_CLIP_START(map, entry, start);
2309 
2310 		/*
2311 		 * unwiring.  first ensure that the range to be unwired is
2312 		 * really wired down and that there are no holes.
2313 		 */
2314 
2315 		while ((entry != &map->header) && (entry->start < end)) {
2316 			if (entry->wired_count == 0 ||
2317 			    (entry->end < end &&
2318 			     (entry->next == &map->header ||
2319 			      entry->next->start > entry->end))) {
2320 				if ((lockflags & UVM_LK_EXIT) == 0)
2321 					vm_map_unlock(map);
2322 				UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0);
2323 				return EINVAL;
2324 			}
2325 			entry = entry->next;
2326 		}
2327 
2328 		/*
2329 		 * POSIX 1003.1b - a single munlock call unlocks a region,
2330 		 * regardless of the number of mlock calls made on that
2331 		 * region.
2332 		 */
2333 
2334 		entry = start_entry;
2335 		while ((entry != &map->header) && (entry->start < end)) {
2336 			UVM_MAP_CLIP_END(map, entry, end);
2337 			if (VM_MAPENT_ISWIRED(entry))
2338 				uvm_map_entry_unwire(map, entry);
2339 			entry = entry->next;
2340 		}
2341 		if ((lockflags & UVM_LK_EXIT) == 0)
2342 			vm_map_unlock(map);
2343 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2344 		return 0;
2345 	}
2346 
2347 	/*
2348 	 * wire case: in two passes [XXXCDC: ugly block of code here]
2349 	 *
2350 	 * 1: holding the write lock, we create any anonymous maps that need
2351 	 *    to be created.  then we clip each map entry to the region to
2352 	 *    be wired and increment its wiring count.
2353 	 *
2354 	 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
2355 	 *    in the pages for any newly wired area (wired_count == 1).
2356 	 *
2357 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
2358 	 *    deadlock with another thread that may have faulted on one of
2359 	 *    the pages to be wired (it would mark the page busy, blocking
2360 	 *    us, then in turn block on the map lock that we hold).  because
2361 	 *    of problems in the recursive lock package, we cannot upgrade
2362 	 *    to a write lock in vm_map_lookup.  thus, any actions that
2363 	 *    require the write lock must be done beforehand.  because we
2364 	 *    keep the read lock on the map, the copy-on-write status of the
2365 	 *    entries we modify here cannot change.
2366 	 */
2367 
2368 	while ((entry != &map->header) && (entry->start < end)) {
2369 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2370 
2371 			/*
2372 			 * perform actions of vm_map_lookup that need the
2373 			 * write lock on the map: create an anonymous map
2374 			 * for a copy-on-write region, or an anonymous map
2375 			 * for a zero-fill region.  (XXXCDC: submap case
2376 			 * ok?)
2377 			 */
2378 
2379 			if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
2380 				if (UVM_ET_ISNEEDSCOPY(entry) &&
2381 				    ((entry->max_protection & VM_PROT_WRITE) ||
2382 				     (entry->object.uvm_obj == NULL))) {
2383 					amap_copy(map, entry, M_WAITOK, TRUE,
2384 					    start, end);
2385 					/* XXXCDC: wait OK? */
2386 				}
2387 			}
2388 		}
2389 		UVM_MAP_CLIP_START(map, entry, start);
2390 		UVM_MAP_CLIP_END(map, entry, end);
2391 		entry->wired_count++;
2392 
2393 		/*
2394 		 * Check for holes
2395 		 */
2396 
2397 		if (entry->protection == VM_PROT_NONE ||
2398 		    (entry->end < end &&
2399 		     (entry->next == &map->header ||
2400 		      entry->next->start > entry->end))) {
2401 
2402 			/*
2403 			 * found one.  amap creation actions do not need to
2404 			 * be undone, but the wired counts need to be restored.
2405 			 */
2406 
2407 			while (entry != &map->header && entry->end > start) {
2408 				entry->wired_count--;
2409 				entry = entry->prev;
2410 			}
2411 			if ((lockflags & UVM_LK_EXIT) == 0)
2412 				vm_map_unlock(map);
2413 			UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
2414 			return EINVAL;
2415 		}
2416 		entry = entry->next;
2417 	}
2418 
2419 	/*
2420 	 * Pass 2.
2421 	 */
2422 
2423 #ifdef DIAGNOSTIC
2424 	timestamp_save = map->timestamp;
2425 #endif
2426 	vm_map_busy(map);
2427 	vm_map_downgrade(map);
2428 
2429 	rv = 0;
2430 	entry = start_entry;
2431 	while (entry != &map->header && entry->start < end) {
2432 		if (entry->wired_count == 1) {
2433 			rv = uvm_fault_wire(map, entry->start, entry->end,
2434 			    VM_FAULT_WIREMAX, entry->max_protection);
2435 			if (rv) {
2436 
2437 				/*
2438 				 * wiring failed.  break out of the loop.
2439 				 * we'll clean up the map below, once we
2440 				 * have a write lock again.
2441 				 */
2442 
2443 				break;
2444 			}
2445 		}
2446 		entry = entry->next;
2447 	}
2448 
2449 	if (rv) {        /* failed? */
2450 
2451 		/*
2452 		 * Get back to an exclusive (write) lock.
2453 		 */
2454 
2455 		vm_map_upgrade(map);
2456 		vm_map_unbusy(map);
2457 
2458 #ifdef DIAGNOSTIC
2459 		if (timestamp_save != map->timestamp)
2460 			panic("uvm_map_pageable: stale map");
2461 #endif
2462 
2463 		/*
2464 		 * first drop the wiring count on all the entries
2465 		 * which haven't actually been wired yet.
2466 		 */
2467 
2468 		failed_entry = entry;
2469 		while (entry != &map->header && entry->start < end) {
2470 			entry->wired_count--;
2471 			entry = entry->next;
2472 		}
2473 
2474 		/*
2475 		 * now, unwire all the entries that were successfully
2476 		 * wired above.
2477 		 */
2478 
2479 		entry = start_entry;
2480 		while (entry != failed_entry) {
2481 			entry->wired_count--;
2482 			if (VM_MAPENT_ISWIRED(entry) == 0)
2483 				uvm_map_entry_unwire(map, entry);
2484 			entry = entry->next;
2485 		}
2486 		if ((lockflags & UVM_LK_EXIT) == 0)
2487 			vm_map_unlock(map);
2488 		UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
2489 		return(rv);
2490 	}
2491 
2492 	/* We are holding a read lock here. */
2493 	if ((lockflags & UVM_LK_EXIT) == 0) {
2494 		vm_map_unbusy(map);
2495 		vm_map_unlock_read(map);
2496 	} else {
2497 
2498 		/*
2499 		 * Get back to an exclusive (write) lock.
2500 		 */
2501 
2502 		vm_map_upgrade(map);
2503 		vm_map_unbusy(map);
2504 	}
2505 
2506 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2507 	return 0;
2508 }
2509 
2510 /*
2511  * uvm_map_pageable_all: special case of uvm_map_pageable - affects
2512  * all mapped regions.
2513  *
2514  * => map must not be locked.
2515  * => if no flags are specified, all regions are unwired.
2516  * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
2517  */
2518 
2519 int
2520 uvm_map_pageable_all(map, flags, limit)
2521 	struct vm_map *map;
2522 	int flags;
2523 	vsize_t limit;
2524 {
2525 	struct vm_map_entry *entry, *failed_entry;
2526 	vsize_t size;
2527 	int rv;
2528 #ifdef DIAGNOSTIC
2529 	u_int timestamp_save;
2530 #endif
2531 	UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
2532 	UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
2533 
2534 	KASSERT(map->flags & VM_MAP_PAGEABLE);
2535 
2536 	vm_map_lock(map);
2537 
2538 	/*
2539 	 * handle wiring and unwiring separately.
2540 	 */
2541 
2542 	if (flags == 0) {			/* unwire */
2543 
2544 		/*
2545 		 * POSIX 1003.1b -- munlockall unlocks all regions,
2546 		 * regardless of how many times mlockall has been called.
2547 		 */
2548 
2549 		for (entry = map->header.next; entry != &map->header;
2550 		     entry = entry->next) {
2551 			if (VM_MAPENT_ISWIRED(entry))
2552 				uvm_map_entry_unwire(map, entry);
2553 		}
2554 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2555 		vm_map_unlock(map);
2556 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2557 		return 0;
2558 	}
2559 
2560 	if (flags & MCL_FUTURE) {
2561 
2562 		/*
2563 		 * must wire all future mappings; remember this.
2564 		 */
2565 
2566 		vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
2567 	}
2568 
2569 	if ((flags & MCL_CURRENT) == 0) {
2570 
2571 		/*
2572 		 * no more work to do!
2573 		 */
2574 
2575 		UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
2576 		vm_map_unlock(map);
2577 		return 0;
2578 	}
2579 
2580 	/*
2581 	 * wire case: in three passes [XXXCDC: ugly block of code here]
2582 	 *
2583 	 * 1: holding the write lock, count all pages mapped by non-wired
2584 	 *    entries.  if this would cause us to go over our limit, we fail.
2585 	 *
2586 	 * 2: still holding the write lock, we create any anonymous maps that
2587 	 *    need to be created.  then we increment its wiring count.
2588 	 *
2589 	 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
2590 	 *    in the pages for any newly wired area (wired_count == 1).
2591 	 *
2592 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
2593 	 *    deadlock with another thread that may have faulted on one of
2594 	 *    the pages to be wired (it would mark the page busy, blocking
2595 	 *    us, then in turn block on the map lock that we hold).  because
2596 	 *    of problems in the recursive lock package, we cannot upgrade
2597 	 *    to a write lock in vm_map_lookup.  thus, any actions that
2598 	 *    require the write lock must be done beforehand.  because we
2599 	 *    keep the read lock on the map, the copy-on-write status of the
2600 	 *    entries we modify here cannot change.
2601 	 */
2602 
2603 	for (size = 0, entry = map->header.next; entry != &map->header;
2604 	     entry = entry->next) {
2605 		if (entry->protection != VM_PROT_NONE &&
2606 		    VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2607 			size += entry->end - entry->start;
2608 		}
2609 	}
2610 
2611 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
2612 		vm_map_unlock(map);
2613 		return ENOMEM;
2614 	}
2615 
2616 	/* XXX non-pmap_wired_count case must be handled by caller */
2617 #ifdef pmap_wired_count
2618 	if (limit != 0 &&
2619 	    (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
2620 		vm_map_unlock(map);
2621 		return ENOMEM;
2622 	}
2623 #endif
2624 
2625 	/*
2626 	 * Pass 2.
2627 	 */
2628 
2629 	for (entry = map->header.next; entry != &map->header;
2630 	     entry = entry->next) {
2631 		if (entry->protection == VM_PROT_NONE)
2632 			continue;
2633 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2634 
2635 			/*
2636 			 * perform actions of vm_map_lookup that need the
2637 			 * write lock on the map: create an anonymous map
2638 			 * for a copy-on-write region, or an anonymous map
2639 			 * for a zero-fill region.  (XXXCDC: submap case
2640 			 * ok?)
2641 			 */
2642 
2643 			if (!UVM_ET_ISSUBMAP(entry)) {	/* not submap */
2644 				if (UVM_ET_ISNEEDSCOPY(entry) &&
2645 				    ((entry->max_protection & VM_PROT_WRITE) ||
2646 				     (entry->object.uvm_obj == NULL))) {
2647 					amap_copy(map, entry, M_WAITOK, TRUE,
2648 					    entry->start, entry->end);
2649 					/* XXXCDC: wait OK? */
2650 				}
2651 			}
2652 		}
2653 		entry->wired_count++;
2654 	}
2655 
2656 	/*
2657 	 * Pass 3.
2658 	 */
2659 
2660 #ifdef DIAGNOSTIC
2661 	timestamp_save = map->timestamp;
2662 #endif
2663 	vm_map_busy(map);
2664 	vm_map_downgrade(map);
2665 
2666 	rv = 0;
2667 	for (entry = map->header.next; entry != &map->header;
2668 	     entry = entry->next) {
2669 		if (entry->wired_count == 1) {
2670 			rv = uvm_fault_wire(map, entry->start, entry->end,
2671 			    VM_FAULT_WIREMAX, entry->max_protection);
2672 			if (rv) {
2673 
2674 				/*
2675 				 * wiring failed.  break out of the loop.
2676 				 * we'll clean up the map below, once we
2677 				 * have a write lock again.
2678 				 */
2679 
2680 				break;
2681 			}
2682 		}
2683 	}
2684 
2685 	if (rv) {
2686 
2687 		/*
2688 		 * Get back an exclusive (write) lock.
2689 		 */
2690 
2691 		vm_map_upgrade(map);
2692 		vm_map_unbusy(map);
2693 
2694 #ifdef DIAGNOSTIC
2695 		if (timestamp_save != map->timestamp)
2696 			panic("uvm_map_pageable_all: stale map");
2697 #endif
2698 
2699 		/*
2700 		 * first drop the wiring count on all the entries
2701 		 * which haven't actually been wired yet.
2702 		 *
2703 		 * Skip VM_PROT_NONE entries like we did above.
2704 		 */
2705 
2706 		failed_entry = entry;
2707 		for (/* nothing */; entry != &map->header;
2708 		     entry = entry->next) {
2709 			if (entry->protection == VM_PROT_NONE)
2710 				continue;
2711 			entry->wired_count--;
2712 		}
2713 
2714 		/*
2715 		 * now, unwire all the entries that were successfully
2716 		 * wired above.
2717 		 *
2718 		 * Skip VM_PROT_NONE entries like we did above.
2719 		 */
2720 
2721 		for (entry = map->header.next; entry != failed_entry;
2722 		     entry = entry->next) {
2723 			if (entry->protection == VM_PROT_NONE)
2724 				continue;
2725 			entry->wired_count--;
2726 			if (VM_MAPENT_ISWIRED(entry))
2727 				uvm_map_entry_unwire(map, entry);
2728 		}
2729 		vm_map_unlock(map);
2730 		UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
2731 		return (rv);
2732 	}
2733 
2734 	/* We are holding a read lock here. */
2735 	vm_map_unbusy(map);
2736 	vm_map_unlock_read(map);
2737 
2738 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2739 	return 0;
2740 }
2741 
2742 /*
2743  * uvm_map_clean: clean out a map range
2744  *
2745  * => valid flags:
2746  *   if (flags & PGO_CLEANIT): dirty pages are cleaned first
2747  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
2748  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
2749  *   if (flags & PGO_FREE): any cached pages are freed after clean
2750  * => returns an error if any part of the specified range isn't mapped
2751  * => never a need to flush amap layer since the anonymous memory has
2752  *	no permanent home, but may deactivate pages there
2753  * => called from sys_msync() and sys_madvise()
2754  * => caller must not write-lock map (read OK).
2755  * => we may sleep while cleaning if SYNCIO [with map read-locked]
2756  */
2757 
2758 int
2759 uvm_map_clean(map, start, end, flags)
2760 	struct vm_map *map;
2761 	vaddr_t start, end;
2762 	int flags;
2763 {
2764 	struct vm_map_entry *current, *entry;
2765 	struct uvm_object *uobj;
2766 	struct vm_amap *amap;
2767 	struct vm_anon *anon;
2768 	struct vm_page *pg;
2769 	vaddr_t offset;
2770 	vsize_t size;
2771 	int error, refs;
2772 	UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
2773 
2774 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
2775 		    map, start, end, flags);
2776 	KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) !=
2777 		(PGO_FREE|PGO_DEACTIVATE));
2778 
2779 	vm_map_lock_read(map);
2780 	VM_MAP_RANGE_CHECK(map, start, end);
2781 	if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
2782 		vm_map_unlock_read(map);
2783 		return EFAULT;
2784 	}
2785 
2786 	/*
2787 	 * Make a first pass to check for holes.
2788 	 */
2789 
2790 	for (current = entry; current->start < end; current = current->next) {
2791 		if (UVM_ET_ISSUBMAP(current)) {
2792 			vm_map_unlock_read(map);
2793 			return EINVAL;
2794 		}
2795 		if (end <= current->end) {
2796 			break;
2797 		}
2798 		if (current->end != current->next->start) {
2799 			vm_map_unlock_read(map);
2800 			return EFAULT;
2801 		}
2802 	}
2803 
2804 	error = 0;
2805 	for (current = entry; start < end; current = current->next) {
2806 		amap = current->aref.ar_amap;	/* top layer */
2807 		uobj = current->object.uvm_obj;	/* bottom layer */
2808 		KASSERT(start >= current->start);
2809 
2810 		/*
2811 		 * No amap cleaning necessary if:
2812 		 *
2813 		 *	(1) There's no amap.
2814 		 *
2815 		 *	(2) We're not deactivating or freeing pages.
2816 		 */
2817 
2818 		if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
2819 			goto flush_object;
2820 
2821 		amap_lock(amap);
2822 		offset = start - current->start;
2823 		size = MIN(end, current->end) - start;
2824 		for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) {
2825 			anon = amap_lookup(&current->aref, offset);
2826 			if (anon == NULL)
2827 				continue;
2828 
2829 			simple_lock(&anon->an_lock);
2830 			pg = anon->u.an_page;
2831 			if (pg == NULL) {
2832 				simple_unlock(&anon->an_lock);
2833 				continue;
2834 			}
2835 
2836 			switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
2837 
2838 			/*
2839 			 * In these first 3 cases, we just deactivate the page.
2840 			 */
2841 
2842 			case PGO_CLEANIT|PGO_FREE:
2843 			case PGO_CLEANIT|PGO_DEACTIVATE:
2844 			case PGO_DEACTIVATE:
2845  deactivate_it:
2846 				/*
2847 				 * skip the page if it's loaned or wired,
2848 				 * since it shouldn't be on a paging queue
2849 				 * at all in these cases.
2850 				 */
2851 
2852 				uvm_lock_pageq();
2853 				if (pg->loan_count != 0 ||
2854 				    pg->wire_count != 0) {
2855 					uvm_unlock_pageq();
2856 					simple_unlock(&anon->an_lock);
2857 					continue;
2858 				}
2859 				KASSERT(pg->uanon == anon);
2860 				pmap_clear_reference(pg);
2861 				uvm_pagedeactivate(pg);
2862 				uvm_unlock_pageq();
2863 				simple_unlock(&anon->an_lock);
2864 				continue;
2865 
2866 			case PGO_FREE:
2867 
2868 				/*
2869 				 * If there are multiple references to
2870 				 * the amap, just deactivate the page.
2871 				 */
2872 
2873 				if (amap_refs(amap) > 1)
2874 					goto deactivate_it;
2875 
2876 				/* skip the page if it's wired */
2877 				if (pg->wire_count != 0) {
2878 					simple_unlock(&anon->an_lock);
2879 					continue;
2880 				}
2881 				amap_unadd(&current->aref, offset);
2882 				refs = --anon->an_ref;
2883 				simple_unlock(&anon->an_lock);
2884 				if (refs == 0)
2885 					uvm_anfree(anon);
2886 				continue;
2887 			}
2888 		}
2889 		amap_unlock(amap);
2890 
2891  flush_object:
2892 		/*
2893 		 * flush pages if we've got a valid backing object.
2894 		 * note that we must always clean object pages before
2895 		 * freeing them since otherwise we could reveal stale
2896 		 * data from files.
2897 		 */
2898 
2899 		offset = current->offset + (start - current->start);
2900 		size = MIN(end, current->end) - start;
2901 		if (uobj != NULL) {
2902 			simple_lock(&uobj->vmobjlock);
2903 			if (uobj->pgops->pgo_put != NULL)
2904 				error = (uobj->pgops->pgo_put)(uobj, offset,
2905 				    offset + size, flags | PGO_CLEANIT);
2906 			else
2907 				error = 0;
2908 		}
2909 		start += size;
2910 	}
2911 	vm_map_unlock_read(map);
2912 	return (error);
2913 }
2914 
2915 
2916 /*
2917  * uvm_map_checkprot: check protection in map
2918  *
2919  * => must allow specified protection in a fully allocated region.
2920  * => map must be read or write locked by caller.
2921  */
2922 
2923 boolean_t
2924 uvm_map_checkprot(map, start, end, protection)
2925 	struct vm_map * map;
2926 	vaddr_t start, end;
2927 	vm_prot_t protection;
2928 {
2929 	struct vm_map_entry *entry;
2930 	struct vm_map_entry *tmp_entry;
2931 
2932 	if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
2933 		return(FALSE);
2934 	}
2935 	entry = tmp_entry;
2936 	while (start < end) {
2937 		if (entry == &map->header) {
2938 			return(FALSE);
2939 		}
2940 
2941 		/*
2942 		 * no holes allowed
2943 		 */
2944 
2945 		if (start < entry->start) {
2946 			return(FALSE);
2947 		}
2948 
2949 		/*
2950 		 * check protection associated with entry
2951 		 */
2952 
2953 		if ((entry->protection & protection) != protection) {
2954 			return(FALSE);
2955 		}
2956 		start = entry->end;
2957 		entry = entry->next;
2958 	}
2959 	return(TRUE);
2960 }
2961 
2962 /*
2963  * uvmspace_alloc: allocate a vmspace structure.
2964  *
2965  * - structure includes vm_map and pmap
2966  * - XXX: no locking on this structure
2967  * - refcnt set to 1, rest must be init'd by caller
2968  */
2969 struct vmspace *
2970 uvmspace_alloc(min, max)
2971 	vaddr_t min, max;
2972 {
2973 	struct vmspace *vm;
2974 	UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
2975 
2976 	vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
2977 	uvmspace_init(vm, NULL, min, max);
2978 	UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
2979 	return (vm);
2980 }
2981 
2982 /*
2983  * uvmspace_init: initialize a vmspace structure.
2984  *
2985  * - XXX: no locking on this structure
2986  * - refcnt set to 1, rest must be init'd by caller
2987  */
2988 void
2989 uvmspace_init(vm, pmap, min, max)
2990 	struct vmspace *vm;
2991 	struct pmap *pmap;
2992 	vaddr_t min, max;
2993 {
2994 	UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
2995 
2996 	memset(vm, 0, sizeof(*vm));
2997 	uvm_map_setup(&vm->vm_map, min, max, VM_MAP_PAGEABLE
2998 #ifdef __USING_TOPDOWN_VM
2999 	    | VM_MAP_TOPDOWN
3000 #endif
3001 	    );
3002 	if (pmap)
3003 		pmap_reference(pmap);
3004 	else
3005 		pmap = pmap_create();
3006 	vm->vm_map.pmap = pmap;
3007 	vm->vm_refcnt = 1;
3008 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3009 }
3010 
3011 /*
3012  * uvmspace_share: share a vmspace between two proceses
3013  *
3014  * - XXX: no locking on vmspace
3015  * - used for vfork, threads(?)
3016  */
3017 
3018 void
3019 uvmspace_share(p1, p2)
3020 	struct proc *p1, *p2;
3021 {
3022 	p2->p_vmspace = p1->p_vmspace;
3023 	p1->p_vmspace->vm_refcnt++;
3024 }
3025 
3026 /*
3027  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
3028  *
3029  * - XXX: no locking on vmspace
3030  */
3031 
3032 void
3033 uvmspace_unshare(l)
3034 	struct lwp *l;
3035 {
3036 	struct proc *p = l->l_proc;
3037 	struct vmspace *nvm, *ovm = p->p_vmspace;
3038 
3039 	if (ovm->vm_refcnt == 1)
3040 		/* nothing to do: vmspace isn't shared in the first place */
3041 		return;
3042 
3043 	/* make a new vmspace, still holding old one */
3044 	nvm = uvmspace_fork(ovm);
3045 
3046 	pmap_deactivate(l);		/* unbind old vmspace */
3047 	p->p_vmspace = nvm;
3048 	pmap_activate(l);		/* switch to new vmspace */
3049 
3050 	uvmspace_free(ovm);		/* drop reference to old vmspace */
3051 }
3052 
3053 /*
3054  * uvmspace_exec: the process wants to exec a new program
3055  *
3056  * - XXX: no locking on vmspace
3057  */
3058 
3059 void
3060 uvmspace_exec(l, start, end)
3061 	struct lwp *l;
3062 	vaddr_t start, end;
3063 {
3064 	struct proc *p = l->l_proc;
3065 	struct vmspace *nvm, *ovm = p->p_vmspace;
3066 	struct vm_map *map = &ovm->vm_map;
3067 
3068 #ifdef __sparc__
3069 	/* XXX cgd 960926: the sparc #ifdef should be a MD hook */
3070 	kill_user_windows(l);   /* before stack addresses go away */
3071 #endif
3072 
3073 	/*
3074 	 * see if more than one process is using this vmspace...
3075 	 */
3076 
3077 	if (ovm->vm_refcnt == 1) {
3078 
3079 		/*
3080 		 * if p is the only process using its vmspace then we can safely
3081 		 * recycle that vmspace for the program that is being exec'd.
3082 		 */
3083 
3084 #ifdef SYSVSHM
3085 		/*
3086 		 * SYSV SHM semantics require us to kill all segments on an exec
3087 		 */
3088 
3089 		if (ovm->vm_shm)
3090 			shmexit(ovm);
3091 #endif
3092 
3093 		/*
3094 		 * POSIX 1003.1b -- "lock future mappings" is revoked
3095 		 * when a process execs another program image.
3096 		 */
3097 
3098 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
3099 
3100 		/*
3101 		 * now unmap the old program
3102 		 */
3103 
3104 		pmap_remove_all(map->pmap);
3105 		uvm_unmap(map, map->min_offset, map->max_offset);
3106 
3107 		/*
3108 		 * resize the map
3109 		 */
3110 
3111 		map->min_offset = start;
3112 		map->max_offset = end;
3113 	} else {
3114 
3115 		/*
3116 		 * p's vmspace is being shared, so we can't reuse it for p since
3117 		 * it is still being used for others.   allocate a new vmspace
3118 		 * for p
3119 		 */
3120 
3121 		nvm = uvmspace_alloc(start, end);
3122 
3123 		/*
3124 		 * install new vmspace and drop our ref to the old one.
3125 		 */
3126 
3127 		pmap_deactivate(l);
3128 		p->p_vmspace = nvm;
3129 		pmap_activate(l);
3130 
3131 		uvmspace_free(ovm);
3132 	}
3133 }
3134 
3135 /*
3136  * uvmspace_free: free a vmspace data structure
3137  *
3138  * - XXX: no locking on vmspace
3139  */
3140 
3141 void
3142 uvmspace_free(vm)
3143 	struct vmspace *vm;
3144 {
3145 	struct vm_map_entry *dead_entries;
3146 	struct vm_map *map;
3147 	UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
3148 
3149 	UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
3150 	if (--vm->vm_refcnt > 0) {
3151 		return;
3152 	}
3153 
3154 	/*
3155 	 * at this point, there should be no other references to the map.
3156 	 * delete all of the mappings, then destroy the pmap.
3157 	 */
3158 
3159 	map = &vm->vm_map;
3160 	map->flags |= VM_MAP_DYING;
3161 	pmap_remove_all(map->pmap);
3162 #ifdef SYSVSHM
3163 	/* Get rid of any SYSV shared memory segments. */
3164 	if (vm->vm_shm != NULL)
3165 		shmexit(vm);
3166 #endif
3167 	if (map->nentries) {
3168 		uvm_unmap_remove(map, map->min_offset, map->max_offset,
3169 		    &dead_entries);
3170 		if (dead_entries != NULL)
3171 			uvm_unmap_detach(dead_entries, 0);
3172 	}
3173 	pmap_destroy(map->pmap);
3174 	pool_put(&uvm_vmspace_pool, vm);
3175 }
3176 
3177 /*
3178  *   F O R K   -   m a i n   e n t r y   p o i n t
3179  */
3180 /*
3181  * uvmspace_fork: fork a process' main map
3182  *
3183  * => create a new vmspace for child process from parent.
3184  * => parent's map must not be locked.
3185  */
3186 
3187 struct vmspace *
3188 uvmspace_fork(vm1)
3189 	struct vmspace *vm1;
3190 {
3191 	struct vmspace *vm2;
3192 	struct vm_map *old_map = &vm1->vm_map;
3193 	struct vm_map *new_map;
3194 	struct vm_map_entry *old_entry;
3195 	struct vm_map_entry *new_entry;
3196 	UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
3197 
3198 	vm_map_lock(old_map);
3199 
3200 	vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset);
3201 	memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
3202 	(caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
3203 	new_map = &vm2->vm_map;		  /* XXX */
3204 
3205 	old_entry = old_map->header.next;
3206 
3207 	/*
3208 	 * go entry-by-entry
3209 	 */
3210 
3211 	while (old_entry != &old_map->header) {
3212 
3213 		/*
3214 		 * first, some sanity checks on the old entry
3215 		 */
3216 
3217 		KASSERT(!UVM_ET_ISSUBMAP(old_entry));
3218 		KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) ||
3219 			!UVM_ET_ISNEEDSCOPY(old_entry));
3220 
3221 		switch (old_entry->inheritance) {
3222 		case MAP_INHERIT_NONE:
3223 
3224 			/*
3225 			 * drop the mapping
3226 			 */
3227 
3228 			break;
3229 
3230 		case MAP_INHERIT_SHARE:
3231 
3232 			/*
3233 			 * share the mapping: this means we want the old and
3234 			 * new entries to share amaps and backing objects.
3235 			 */
3236 			/*
3237 			 * if the old_entry needs a new amap (due to prev fork)
3238 			 * then we need to allocate it now so that we have
3239 			 * something we own to share with the new_entry.   [in
3240 			 * other words, we need to clear needs_copy]
3241 			 */
3242 
3243 			if (UVM_ET_ISNEEDSCOPY(old_entry)) {
3244 				/* get our own amap, clears needs_copy */
3245 				amap_copy(old_map, old_entry, M_WAITOK, FALSE,
3246 				    0, 0);
3247 				/* XXXCDC: WAITOK??? */
3248 			}
3249 
3250 			new_entry = uvm_mapent_alloc(new_map, 0);
3251 			/* old_entry -> new_entry */
3252 			uvm_mapent_copy(old_entry, new_entry);
3253 
3254 			/* new pmap has nothing wired in it */
3255 			new_entry->wired_count = 0;
3256 
3257 			/*
3258 			 * gain reference to object backing the map (can't
3259 			 * be a submap, already checked this case).
3260 			 */
3261 
3262 			if (new_entry->aref.ar_amap)
3263 				uvm_map_reference_amap(new_entry, AMAP_SHARED);
3264 
3265 			if (new_entry->object.uvm_obj &&
3266 			    new_entry->object.uvm_obj->pgops->pgo_reference)
3267 				new_entry->object.uvm_obj->
3268 				    pgops->pgo_reference(
3269 				        new_entry->object.uvm_obj);
3270 
3271 			/* insert entry at end of new_map's entry list */
3272 			uvm_map_entry_link(new_map, new_map->header.prev,
3273 			    new_entry);
3274 
3275 			break;
3276 
3277 		case MAP_INHERIT_COPY:
3278 
3279 			/*
3280 			 * copy-on-write the mapping (using mmap's
3281 			 * MAP_PRIVATE semantics)
3282 			 *
3283 			 * allocate new_entry, adjust reference counts.
3284 			 * (note that new references are read-only).
3285 			 */
3286 
3287 			new_entry = uvm_mapent_alloc(new_map, 0);
3288 			/* old_entry -> new_entry */
3289 			uvm_mapent_copy(old_entry, new_entry);
3290 
3291 			if (new_entry->aref.ar_amap)
3292 				uvm_map_reference_amap(new_entry, 0);
3293 
3294 			if (new_entry->object.uvm_obj &&
3295 			    new_entry->object.uvm_obj->pgops->pgo_reference)
3296 				new_entry->object.uvm_obj->pgops->pgo_reference
3297 				    (new_entry->object.uvm_obj);
3298 
3299 			/* new pmap has nothing wired in it */
3300 			new_entry->wired_count = 0;
3301 
3302 			new_entry->etype |=
3303 			    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
3304 			uvm_map_entry_link(new_map, new_map->header.prev,
3305 			    new_entry);
3306 
3307 			/*
3308 			 * the new entry will need an amap.  it will either
3309 			 * need to be copied from the old entry or created
3310 			 * from scratch (if the old entry does not have an
3311 			 * amap).  can we defer this process until later
3312 			 * (by setting "needs_copy") or do we need to copy
3313 			 * the amap now?
3314 			 *
3315 			 * we must copy the amap now if any of the following
3316 			 * conditions hold:
3317 			 * 1. the old entry has an amap and that amap is
3318 			 *    being shared.  this means that the old (parent)
3319 			 *    process is sharing the amap with another
3320 			 *    process.  if we do not clear needs_copy here
3321 			 *    we will end up in a situation where both the
3322 			 *    parent and child process are refering to the
3323 			 *    same amap with "needs_copy" set.  if the
3324 			 *    parent write-faults, the fault routine will
3325 			 *    clear "needs_copy" in the parent by allocating
3326 			 *    a new amap.   this is wrong because the
3327 			 *    parent is supposed to be sharing the old amap
3328 			 *    and the new amap will break that.
3329 			 *
3330 			 * 2. if the old entry has an amap and a non-zero
3331 			 *    wire count then we are going to have to call
3332 			 *    amap_cow_now to avoid page faults in the
3333 			 *    parent process.   since amap_cow_now requires
3334 			 *    "needs_copy" to be clear we might as well
3335 			 *    clear it here as well.
3336 			 *
3337 			 */
3338 
3339 			if (old_entry->aref.ar_amap != NULL) {
3340 				if ((amap_flags(old_entry->aref.ar_amap) &
3341 				     AMAP_SHARED) != 0 ||
3342 				    VM_MAPENT_ISWIRED(old_entry)) {
3343 
3344 					amap_copy(new_map, new_entry, M_WAITOK,
3345 					    FALSE, 0, 0);
3346 					/* XXXCDC: M_WAITOK ... ok? */
3347 				}
3348 			}
3349 
3350 			/*
3351 			 * if the parent's entry is wired down, then the
3352 			 * parent process does not want page faults on
3353 			 * access to that memory.  this means that we
3354 			 * cannot do copy-on-write because we can't write
3355 			 * protect the old entry.   in this case we
3356 			 * resolve all copy-on-write faults now, using
3357 			 * amap_cow_now.   note that we have already
3358 			 * allocated any needed amap (above).
3359 			 */
3360 
3361 			if (VM_MAPENT_ISWIRED(old_entry)) {
3362 
3363 			  /*
3364 			   * resolve all copy-on-write faults now
3365 			   * (note that there is nothing to do if
3366 			   * the old mapping does not have an amap).
3367 			   */
3368 			  if (old_entry->aref.ar_amap)
3369 			    amap_cow_now(new_map, new_entry);
3370 
3371 			} else {
3372 
3373 			  /*
3374 			   * setup mappings to trigger copy-on-write faults
3375 			   * we must write-protect the parent if it has
3376 			   * an amap and it is not already "needs_copy"...
3377 			   * if it is already "needs_copy" then the parent
3378 			   * has already been write-protected by a previous
3379 			   * fork operation.
3380 			   */
3381 
3382 			  if (old_entry->aref.ar_amap &&
3383 			      !UVM_ET_ISNEEDSCOPY(old_entry)) {
3384 			      if (old_entry->max_protection & VM_PROT_WRITE) {
3385 				pmap_protect(old_map->pmap,
3386 					     old_entry->start,
3387 					     old_entry->end,
3388 					     old_entry->protection &
3389 					     ~VM_PROT_WRITE);
3390 				pmap_update(old_map->pmap);
3391 			      }
3392 			      old_entry->etype |= UVM_ET_NEEDSCOPY;
3393 			  }
3394 			}
3395 			break;
3396 		}  /* end of switch statement */
3397 		old_entry = old_entry->next;
3398 	}
3399 
3400 	new_map->size = old_map->size;
3401 	vm_map_unlock(old_map);
3402 
3403 #ifdef SYSVSHM
3404 	if (vm1->vm_shm)
3405 		shmfork(vm1, vm2);
3406 #endif
3407 
3408 #ifdef PMAP_FORK
3409 	pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
3410 #endif
3411 
3412 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3413 	return(vm2);
3414 }
3415 
3416 
3417 #if defined(DDB)
3418 
3419 /*
3420  * DDB hooks
3421  */
3422 
3423 /*
3424  * uvm_map_printit: actually prints the map
3425  */
3426 
3427 void
3428 uvm_map_printit(map, full, pr)
3429 	struct vm_map *map;
3430 	boolean_t full;
3431 	void (*pr) __P((const char *, ...));
3432 {
3433 	struct vm_map_entry *entry;
3434 
3435 	(*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
3436 	(*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
3437 	    map->nentries, map->size, map->ref_count, map->timestamp,
3438 	    map->flags);
3439 	(*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
3440 	    pmap_resident_count(map->pmap));
3441 	if (!full)
3442 		return;
3443 	for (entry = map->header.next; entry != &map->header;
3444 	    entry = entry->next) {
3445 		(*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
3446 		    entry, entry->start, entry->end, entry->object.uvm_obj,
3447 		    (long long)entry->offset, entry->aref.ar_amap,
3448 		    entry->aref.ar_pageoff);
3449 		(*pr)(
3450 		    "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, "
3451 		    "wc=%d, adv=%d\n",
3452 		    (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
3453 		    (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
3454 		    (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
3455 		    entry->protection, entry->max_protection,
3456 		    entry->inheritance, entry->wired_count, entry->advice);
3457 	}
3458 }
3459 
3460 /*
3461  * uvm_object_printit: actually prints the object
3462  */
3463 
3464 void
3465 uvm_object_printit(uobj, full, pr)
3466 	struct uvm_object *uobj;
3467 	boolean_t full;
3468 	void (*pr) __P((const char *, ...));
3469 {
3470 	struct vm_page *pg;
3471 	int cnt = 0;
3472 
3473 	(*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
3474 	    uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
3475 	if (UVM_OBJ_IS_KERN_OBJECT(uobj))
3476 		(*pr)("refs=<SYSTEM>\n");
3477 	else
3478 		(*pr)("refs=%d\n", uobj->uo_refs);
3479 
3480 	if (!full) {
3481 		return;
3482 	}
3483 	(*pr)("  PAGES <pg,offset>:\n  ");
3484 	TAILQ_FOREACH(pg, &uobj->memq, listq) {
3485 		cnt++;
3486 		(*pr)("<%p,0x%llx> ", pg, (long long)pg->offset);
3487 		if ((cnt % 3) == 0) {
3488 			(*pr)("\n  ");
3489 		}
3490 	}
3491 	if ((cnt % 3) != 0) {
3492 		(*pr)("\n");
3493 	}
3494 }
3495 
3496 /*
3497  * uvm_page_printit: actually print the page
3498  */
3499 
3500 static const char page_flagbits[] =
3501 	"\20\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY"
3502 	"\11ZERO\15PAGER1";
3503 static const char page_pqflagbits[] =
3504 	"\20\1FREE\2INACTIVE\3ACTIVE\5ANON\6AOBJ";
3505 
3506 void
3507 uvm_page_printit(pg, full, pr)
3508 	struct vm_page *pg;
3509 	boolean_t full;
3510 	void (*pr) __P((const char *, ...));
3511 {
3512 	struct vm_page *tpg;
3513 	struct uvm_object *uobj;
3514 	struct pglist *pgl;
3515 	char pgbuf[128];
3516 	char pqbuf[128];
3517 
3518 	(*pr)("PAGE %p:\n", pg);
3519 	bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf));
3520 	bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
3521 	(*pr)("  flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n",
3522 	    pgbuf, pqbuf, pg->wire_count, (long)pg->phys_addr);
3523 	(*pr)("  uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
3524 	    pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
3525 #if defined(UVM_PAGE_TRKOWN)
3526 	if (pg->flags & PG_BUSY)
3527 		(*pr)("  owning process = %d, tag=%s\n",
3528 		    pg->owner, pg->owner_tag);
3529 	else
3530 		(*pr)("  page not busy, no owner\n");
3531 #else
3532 	(*pr)("  [page ownership tracking disabled]\n");
3533 #endif
3534 
3535 	if (!full)
3536 		return;
3537 
3538 	/* cross-verify object/anon */
3539 	if ((pg->pqflags & PQ_FREE) == 0) {
3540 		if (pg->pqflags & PQ_ANON) {
3541 			if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
3542 			    (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
3543 				(pg->uanon) ? pg->uanon->u.an_page : NULL);
3544 			else
3545 				(*pr)("  anon backpointer is OK\n");
3546 		} else {
3547 			uobj = pg->uobject;
3548 			if (uobj) {
3549 				(*pr)("  checking object list\n");
3550 				TAILQ_FOREACH(tpg, &uobj->memq, listq) {
3551 					if (tpg == pg) {
3552 						break;
3553 					}
3554 				}
3555 				if (tpg)
3556 					(*pr)("  page found on object list\n");
3557 				else
3558 			(*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
3559 			}
3560 		}
3561 	}
3562 
3563 	/* cross-verify page queue */
3564 	if (pg->pqflags & PQ_FREE) {
3565 		int fl = uvm_page_lookup_freelist(pg);
3566 		int color = VM_PGCOLOR_BUCKET(pg);
3567 		pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[
3568 		    ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN];
3569 	} else if (pg->pqflags & PQ_INACTIVE) {
3570 		pgl = &uvm.page_inactive;
3571 	} else if (pg->pqflags & PQ_ACTIVE) {
3572 		pgl = &uvm.page_active;
3573  	} else {
3574 		pgl = NULL;
3575 	}
3576 
3577 	if (pgl) {
3578 		(*pr)("  checking pageq list\n");
3579 		TAILQ_FOREACH(tpg, pgl, pageq) {
3580 			if (tpg == pg) {
3581 				break;
3582 			}
3583 		}
3584 		if (tpg)
3585 			(*pr)("  page found on pageq list\n");
3586 		else
3587 			(*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
3588 	}
3589 }
3590 #endif
3591