xref: /netbsd-src/sys/uvm/uvm_map.c (revision b8c616269f5ebf18ab2e35cb8099d683130a177c)
1 /*	$NetBSD: uvm_map.c,v 1.130 2003/02/01 06:23:55 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 1.130 2003/02/01 06:23:55 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 	UVMHIST_FUNC("uvm_map_findspace");
1085 	UVMHIST_CALLED(maphist);
1086 
1087 	UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)",
1088 		    map, hint, length, flags);
1089 	KASSERT((align & (align - 1)) == 0);
1090 	KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0);
1091 
1092 	/*
1093 	 * remember the original hint.  if we are aligning, then we
1094 	 * may have to try again with no alignment constraint if
1095 	 * we fail the first time.
1096 	 */
1097 
1098 	orig_hint = hint;
1099 	if (hint < map->min_offset) {	/* check ranges ... */
1100 		if (flags & UVM_FLAG_FIXED) {
1101 			UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
1102 			return(NULL);
1103 		}
1104 		hint = map->min_offset;
1105 	}
1106 	if (hint > map->max_offset) {
1107 		UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
1108 				hint, map->min_offset, map->max_offset, 0);
1109 		return(NULL);
1110 	}
1111 
1112 	/*
1113 	 * Look for the first possible address; if there's already
1114 	 * something at this address, we have to start after it.
1115 	 */
1116 
1117 	if ((flags & UVM_FLAG_FIXED) == 0 && hint == map->min_offset) {
1118 		if ((entry = map->first_free) != &map->header)
1119 			hint = entry->end;
1120 	} else {
1121 		if (uvm_map_lookup_entry(map, hint, &tmp)) {
1122 			/* "hint" address already in use ... */
1123 			if (flags & UVM_FLAG_FIXED) {
1124 				UVMHIST_LOG(maphist,"<- fixed & VA in use",
1125 				    0, 0, 0, 0);
1126 				return(NULL);
1127 			}
1128 			hint = tmp->end;
1129 		}
1130 		entry = tmp;
1131 	}
1132 
1133 	/*
1134 	 * Look through the rest of the map, trying to fit a new region in
1135 	 * the gap between existing regions, or after the very last region.
1136 	 * note: entry->end   = base VA of current gap,
1137 	 *	 next->start  = VA of end of current gap
1138 	 */
1139 
1140 	for (;; hint = (entry = next)->end) {
1141 
1142 		/*
1143 		 * Find the end of the proposed new region.  Be sure we didn't
1144 		 * go beyond the end of the map, or wrap around the address;
1145 		 * if so, we lose.  Otherwise, if this is the last entry, or
1146 		 * if the proposed new region fits before the next entry, we
1147 		 * win.
1148 		 */
1149 
1150 #ifdef PMAP_PREFER
1151 		/*
1152 		 * push hint forward as needed to avoid VAC alias problems.
1153 		 * we only do this if a valid offset is specified.
1154 		 */
1155 
1156 		if ((flags & UVM_FLAG_FIXED) == 0 &&
1157 		    uoffset != UVM_UNKNOWN_OFFSET)
1158 			PMAP_PREFER(uoffset, &hint);
1159 #endif
1160 		if (align != 0) {
1161 			if ((hint & (align - 1)) != 0)
1162 				hint = roundup(hint, align);
1163 			/*
1164 			 * XXX Should we PMAP_PREFER() here again?
1165 			 */
1166 		}
1167 		end = hint + length;
1168 		if (end > map->max_offset || end < hint) {
1169 			UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0);
1170 			if (align != 0) {
1171 				UVMHIST_LOG(maphist,
1172 				    "calling recursively, no align",
1173 				    0,0,0,0);
1174 				return (uvm_map_findspace(map, orig_hint,
1175 				    length, result, uobj, uoffset, 0, flags));
1176 			}
1177 			return (NULL);
1178 		}
1179 		next = entry->next;
1180 		if (next == &map->header || next->start >= end)
1181 			break;
1182 		if (flags & UVM_FLAG_FIXED) {
1183 			UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0);
1184 			return(NULL); /* only one shot at it ... */
1185 		}
1186 	}
1187 	SAVE_HINT(map, map->hint, entry);
1188 	*result = hint;
1189 	UVMHIST_LOG(maphist,"<- got it!  (result=0x%x)", hint, 0,0,0);
1190 	return (entry);
1191 }
1192 
1193 /*
1194  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
1195  */
1196 
1197 /*
1198  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
1199  *
1200  * => caller must check alignment and size
1201  * => map must be locked by caller
1202  * => we return a list of map entries that we've remove from the map
1203  *    in "entry_list"
1204  */
1205 
1206 void
1207 uvm_unmap_remove(map, start, end, entry_list)
1208 	struct vm_map *map;
1209 	vaddr_t start, end;
1210 	struct vm_map_entry **entry_list;	/* OUT */
1211 {
1212 	struct vm_map_entry *entry, *first_entry, *next;
1213 	vaddr_t len;
1214 	UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist);
1215 
1216 	UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
1217 	    map, start, end, 0);
1218 	VM_MAP_RANGE_CHECK(map, start, end);
1219 
1220 	/*
1221 	 * find first entry
1222 	 */
1223 
1224 	if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
1225 		/* clip and go... */
1226 		entry = first_entry;
1227 		UVM_MAP_CLIP_START(map, entry, start);
1228 		/* critical!  prevents stale hint */
1229 		SAVE_HINT(map, entry, entry->prev);
1230 	} else {
1231 		entry = first_entry->next;
1232 	}
1233 
1234 	/*
1235 	 * Save the free space hint
1236 	 */
1237 
1238 	if (map->first_free->start >= start)
1239 		map->first_free = entry->prev;
1240 
1241 	/*
1242 	 * note: we now re-use first_entry for a different task.  we remove
1243 	 * a number of map entries from the map and save them in a linked
1244 	 * list headed by "first_entry".  once we remove them from the map
1245 	 * the caller should unlock the map and drop the references to the
1246 	 * backing objects [c.f. uvm_unmap_detach].  the object is to
1247 	 * separate unmapping from reference dropping.  why?
1248 	 *   [1] the map has to be locked for unmapping
1249 	 *   [2] the map need not be locked for reference dropping
1250 	 *   [3] dropping references may trigger pager I/O, and if we hit
1251 	 *       a pager that does synchronous I/O we may have to wait for it.
1252 	 *   [4] we would like all waiting for I/O to occur with maps unlocked
1253 	 *       so that we don't block other threads.
1254 	 */
1255 
1256 	first_entry = NULL;
1257 	*entry_list = NULL;
1258 
1259 	/*
1260 	 * break up the area into map entry sized regions and unmap.  note
1261 	 * that all mappings have to be removed before we can even consider
1262 	 * dropping references to amaps or VM objects (otherwise we could end
1263 	 * up with a mapping to a page on the free list which would be very bad)
1264 	 */
1265 
1266 	while ((entry != &map->header) && (entry->start < end)) {
1267 		UVM_MAP_CLIP_END(map, entry, end);
1268 		next = entry->next;
1269 		len = entry->end - entry->start;
1270 
1271 		/*
1272 		 * unwire before removing addresses from the pmap; otherwise
1273 		 * unwiring will put the entries back into the pmap (XXX).
1274 		 */
1275 
1276 		if (VM_MAPENT_ISWIRED(entry)) {
1277 			uvm_map_entry_unwire(map, entry);
1278 		}
1279 		if ((map->flags & VM_MAP_PAGEABLE) == 0) {
1280 
1281 			/*
1282 			 * if the map is non-pageable, any pages mapped there
1283 			 * must be wired and entered with pmap_kenter_pa(),
1284 			 * and we should free any such pages immediately.
1285 			 * this is mostly used for kmem_map and mb_map.
1286 			 */
1287 
1288 			uvm_km_pgremove_intrsafe(entry->start, entry->end);
1289 			pmap_kremove(entry->start, len);
1290 		} else if (UVM_ET_ISOBJ(entry) &&
1291 			   UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
1292 			KASSERT(vm_map_pmap(map) == pmap_kernel());
1293 
1294 			/*
1295 			 * note: kernel object mappings are currently used in
1296 			 * two ways:
1297 			 *  [1] "normal" mappings of pages in the kernel object
1298 			 *  [2] uvm_km_valloc'd allocations in which we
1299 			 *      pmap_enter in some non-kernel-object page
1300 			 *      (e.g. vmapbuf).
1301 			 *
1302 			 * for case [1], we need to remove the mapping from
1303 			 * the pmap and then remove the page from the kernel
1304 			 * object (because, once pages in a kernel object are
1305 			 * unmapped they are no longer needed, unlike, say,
1306 			 * a vnode where you might want the data to persist
1307 			 * until flushed out of a queue).
1308 			 *
1309 			 * for case [2], we need to remove the mapping from
1310 			 * the pmap.  there shouldn't be any pages at the
1311 			 * specified offset in the kernel object [but it
1312 			 * doesn't hurt to call uvm_km_pgremove just to be
1313 			 * safe?]
1314 			 *
1315 			 * uvm_km_pgremove currently does the following:
1316 			 *   for pages in the kernel object in range:
1317 			 *     - drops the swap slot
1318 			 *     - uvm_pagefree the page
1319 			 */
1320 
1321 			/*
1322 			 * remove mappings from pmap and drop the pages
1323 			 * from the object.  offsets are always relative
1324 			 * to vm_map_min(kernel_map).
1325 			 */
1326 
1327 			pmap_remove(pmap_kernel(), entry->start,
1328 			    entry->start + len);
1329 			uvm_km_pgremove(entry->object.uvm_obj,
1330 			    entry->start - vm_map_min(kernel_map),
1331 			    entry->end - vm_map_min(kernel_map));
1332 
1333 			/*
1334 			 * null out kernel_object reference, we've just
1335 			 * dropped it
1336 			 */
1337 
1338 			entry->etype &= ~UVM_ET_OBJ;
1339 			entry->object.uvm_obj = NULL;
1340 		} else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) {
1341 
1342 			/*
1343 		 	 * remove mappings the standard way.
1344 		 	 */
1345 
1346 			pmap_remove(map->pmap, entry->start, entry->end);
1347 		}
1348 
1349 		/*
1350 		 * remove entry from map and put it on our list of entries
1351 		 * that we've nuked.  then go to next entry.
1352 		 */
1353 
1354 		UVMHIST_LOG(maphist, "  removed map entry 0x%x", entry, 0, 0,0);
1355 
1356 		/* critical!  prevents stale hint */
1357 		SAVE_HINT(map, entry, entry->prev);
1358 
1359 		uvm_map_entry_unlink(map, entry);
1360 		map->size -= len;
1361 		entry->next = first_entry;
1362 		first_entry = entry;
1363 		entry = next;
1364 	}
1365 	if ((map->flags & VM_MAP_DYING) == 0) {
1366 		pmap_update(vm_map_pmap(map));
1367 	}
1368 
1369 	/*
1370 	 * now we've cleaned up the map and are ready for the caller to drop
1371 	 * references to the mapped objects.
1372 	 */
1373 
1374 	*entry_list = first_entry;
1375 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1376 }
1377 
1378 /*
1379  * uvm_unmap_detach: drop references in a chain of map entries
1380  *
1381  * => we will free the map entries as we traverse the list.
1382  */
1383 
1384 void
1385 uvm_unmap_detach(first_entry, flags)
1386 	struct vm_map_entry *first_entry;
1387 	int flags;
1388 {
1389 	struct vm_map_entry *next_entry;
1390 	UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1391 
1392 	while (first_entry) {
1393 		KASSERT(!VM_MAPENT_ISWIRED(first_entry));
1394 		UVMHIST_LOG(maphist,
1395 		    "  detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
1396 		    first_entry, first_entry->aref.ar_amap,
1397 		    first_entry->object.uvm_obj,
1398 		    UVM_ET_ISSUBMAP(first_entry));
1399 
1400 		/*
1401 		 * drop reference to amap, if we've got one
1402 		 */
1403 
1404 		if (first_entry->aref.ar_amap)
1405 			uvm_map_unreference_amap(first_entry, flags);
1406 
1407 		/*
1408 		 * drop reference to our backing object, if we've got one
1409 		 */
1410 
1411 		KASSERT(!UVM_ET_ISSUBMAP(first_entry));
1412 		if (UVM_ET_ISOBJ(first_entry) &&
1413 		    first_entry->object.uvm_obj->pgops->pgo_detach) {
1414 			(*first_entry->object.uvm_obj->pgops->pgo_detach)
1415 				(first_entry->object.uvm_obj);
1416 		}
1417 		next_entry = first_entry->next;
1418 		uvm_mapent_free(first_entry);
1419 		first_entry = next_entry;
1420 	}
1421 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1422 }
1423 
1424 /*
1425  *   E X T R A C T I O N   F U N C T I O N S
1426  */
1427 
1428 /*
1429  * uvm_map_reserve: reserve space in a vm_map for future use.
1430  *
1431  * => we reserve space in a map by putting a dummy map entry in the
1432  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
1433  * => map should be unlocked (we will write lock it)
1434  * => we return true if we were able to reserve space
1435  * => XXXCDC: should be inline?
1436  */
1437 
1438 int
1439 uvm_map_reserve(map, size, offset, align, raddr)
1440 	struct vm_map *map;
1441 	vsize_t size;
1442 	vaddr_t offset;	/* hint for pmap_prefer */
1443 	vsize_t align;	/* alignment hint */
1444 	vaddr_t *raddr;	/* IN:hint, OUT: reserved VA */
1445 {
1446 	UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1447 
1448 	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1449 	      map,size,offset,raddr);
1450 
1451 	size = round_page(size);
1452 	if (*raddr < vm_map_min(map))
1453 		*raddr = vm_map_min(map);                /* hint */
1454 
1455 	/*
1456 	 * reserve some virtual space.
1457 	 */
1458 
1459 	if (uvm_map(map, raddr, size, NULL, offset, 0,
1460 	    UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1461 	    UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) {
1462 	    UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1463 		return (FALSE);
1464 	}
1465 
1466 	UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1467 	return (TRUE);
1468 }
1469 
1470 /*
1471  * uvm_map_replace: replace a reserved (blank) area of memory with
1472  * real mappings.
1473  *
1474  * => caller must WRITE-LOCK the map
1475  * => we return TRUE if replacement was a success
1476  * => we expect the newents chain to have nnewents entrys on it and
1477  *    we expect newents->prev to point to the last entry on the list
1478  * => note newents is allowed to be NULL
1479  */
1480 
1481 int
1482 uvm_map_replace(map, start, end, newents, nnewents)
1483 	struct vm_map *map;
1484 	vaddr_t start, end;
1485 	struct vm_map_entry *newents;
1486 	int nnewents;
1487 {
1488 	struct vm_map_entry *oldent, *last;
1489 
1490 	/*
1491 	 * first find the blank map entry at the specified address
1492 	 */
1493 
1494 	if (!uvm_map_lookup_entry(map, start, &oldent)) {
1495 		return(FALSE);
1496 	}
1497 
1498 	/*
1499 	 * check to make sure we have a proper blank entry
1500 	 */
1501 
1502 	if (oldent->start != start || oldent->end != end ||
1503 	    oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1504 		return (FALSE);
1505 	}
1506 
1507 #ifdef DIAGNOSTIC
1508 
1509 	/*
1510 	 * sanity check the newents chain
1511 	 */
1512 
1513 	{
1514 		struct vm_map_entry *tmpent = newents;
1515 		int nent = 0;
1516 		vaddr_t cur = start;
1517 
1518 		while (tmpent) {
1519 			nent++;
1520 			if (tmpent->start < cur)
1521 				panic("uvm_map_replace1");
1522 			if (tmpent->start > tmpent->end || tmpent->end > end) {
1523 		printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
1524 			    tmpent->start, tmpent->end, end);
1525 				panic("uvm_map_replace2");
1526 			}
1527 			cur = tmpent->end;
1528 			if (tmpent->next) {
1529 				if (tmpent->next->prev != tmpent)
1530 					panic("uvm_map_replace3");
1531 			} else {
1532 				if (newents->prev != tmpent)
1533 					panic("uvm_map_replace4");
1534 			}
1535 			tmpent = tmpent->next;
1536 		}
1537 		if (nent != nnewents)
1538 			panic("uvm_map_replace5");
1539 	}
1540 #endif
1541 
1542 	/*
1543 	 * map entry is a valid blank!   replace it.   (this does all the
1544 	 * work of map entry link/unlink...).
1545 	 */
1546 
1547 	if (newents) {
1548 		last = newents->prev;
1549 
1550 		/* critical: flush stale hints out of map */
1551 		SAVE_HINT(map, map->hint, newents);
1552 		if (map->first_free == oldent)
1553 			map->first_free = last;
1554 
1555 		last->next = oldent->next;
1556 		last->next->prev = last;
1557 		newents->prev = oldent->prev;
1558 		newents->prev->next = newents;
1559 		map->nentries = map->nentries + (nnewents - 1);
1560 
1561 	} else {
1562 
1563 		/* critical: flush stale hints out of map */
1564 		SAVE_HINT(map, map->hint, oldent->prev);
1565 		if (map->first_free == oldent)
1566 			map->first_free = oldent->prev;
1567 
1568 		/* NULL list of new entries: just remove the old one */
1569 		uvm_map_entry_unlink(map, oldent);
1570 	}
1571 
1572 
1573 	/*
1574 	 * now we can free the old blank entry, unlock the map and return.
1575 	 */
1576 
1577 	uvm_mapent_free(oldent);
1578 	return(TRUE);
1579 }
1580 
1581 /*
1582  * uvm_map_extract: extract a mapping from a map and put it somewhere
1583  *	(maybe removing the old mapping)
1584  *
1585  * => maps should be unlocked (we will write lock them)
1586  * => returns 0 on success, error code otherwise
1587  * => start must be page aligned
1588  * => len must be page sized
1589  * => flags:
1590  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
1591  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
1592  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
1593  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
1594  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
1595  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
1596  *             be used from within the kernel in a kernel level map <<<
1597  */
1598 
1599 int
1600 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags)
1601 	struct vm_map *srcmap, *dstmap;
1602 	vaddr_t start, *dstaddrp;
1603 	vsize_t len;
1604 	int flags;
1605 {
1606 	vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge,
1607 	    oldstart;
1608 	struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry,
1609 	    *deadentry, *oldentry;
1610 	vsize_t elen;
1611 	int nchain, error, copy_ok;
1612 	UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1613 
1614 	UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
1615 	    len,0);
1616 	UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
1617 
1618 	/*
1619 	 * step 0: sanity check: start must be on a page boundary, length
1620 	 * must be page sized.  can't ask for CONTIG/QREF if you asked for
1621 	 * REMOVE.
1622 	 */
1623 
1624 	KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0);
1625 	KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 ||
1626 		(flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0);
1627 
1628 	/*
1629 	 * step 1: reserve space in the target map for the extracted area
1630 	 */
1631 
1632 	dstaddr = vm_map_min(dstmap);
1633 	if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE)
1634 		return(ENOMEM);
1635 	*dstaddrp = dstaddr;	/* pass address back to caller */
1636 	UVMHIST_LOG(maphist, "  dstaddr=0x%x", dstaddr,0,0,0);
1637 
1638 	/*
1639 	 * step 2: setup for the extraction process loop by init'ing the
1640 	 * map entry chain, locking src map, and looking up the first useful
1641 	 * entry in the map.
1642 	 */
1643 
1644 	end = start + len;
1645 	newend = dstaddr + len;
1646 	chain = endchain = NULL;
1647 	nchain = 0;
1648 	vm_map_lock(srcmap);
1649 
1650 	if (uvm_map_lookup_entry(srcmap, start, &entry)) {
1651 
1652 		/* "start" is within an entry */
1653 		if (flags & UVM_EXTRACT_QREF) {
1654 
1655 			/*
1656 			 * for quick references we don't clip the entry, so
1657 			 * the entry may map space "before" the starting
1658 			 * virtual address... this is the "fudge" factor
1659 			 * (which can be non-zero only the first time
1660 			 * through the "while" loop in step 3).
1661 			 */
1662 
1663 			fudge = start - entry->start;
1664 		} else {
1665 
1666 			/*
1667 			 * normal reference: we clip the map to fit (thus
1668 			 * fudge is zero)
1669 			 */
1670 
1671 			UVM_MAP_CLIP_START(srcmap, entry, start);
1672 			SAVE_HINT(srcmap, srcmap->hint, entry->prev);
1673 			fudge = 0;
1674 		}
1675 	} else {
1676 
1677 		/* "start" is not within an entry ... skip to next entry */
1678 		if (flags & UVM_EXTRACT_CONTIG) {
1679 			error = EINVAL;
1680 			goto bad;    /* definite hole here ... */
1681 		}
1682 
1683 		entry = entry->next;
1684 		fudge = 0;
1685 	}
1686 
1687 	/* save values from srcmap for step 6 */
1688 	orig_entry = entry;
1689 	orig_fudge = fudge;
1690 
1691 	/*
1692 	 * step 3: now start looping through the map entries, extracting
1693 	 * as we go.
1694 	 */
1695 
1696 	while (entry->start < end && entry != &srcmap->header) {
1697 
1698 		/* if we are not doing a quick reference, clip it */
1699 		if ((flags & UVM_EXTRACT_QREF) == 0)
1700 			UVM_MAP_CLIP_END(srcmap, entry, end);
1701 
1702 		/* clear needs_copy (allow chunking) */
1703 		if (UVM_ET_ISNEEDSCOPY(entry)) {
1704 			if (fudge)
1705 				oldstart = entry->start;
1706 			else
1707 				oldstart = 0;	/* XXX: gcc */
1708 			amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
1709 			if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
1710 				error = ENOMEM;
1711 				goto bad;
1712 			}
1713 
1714 			/* amap_copy could clip (during chunk)!  update fudge */
1715 			if (fudge) {
1716 				fudge = fudge - (entry->start - oldstart);
1717 				orig_fudge = fudge;
1718 			}
1719 		}
1720 
1721 		/* calculate the offset of this from "start" */
1722 		oldoffset = (entry->start + fudge) - start;
1723 
1724 		/* allocate a new map entry */
1725 		newentry = uvm_mapent_alloc(dstmap, 0);
1726 		if (newentry == NULL) {
1727 			error = ENOMEM;
1728 			goto bad;
1729 		}
1730 
1731 		/* set up new map entry */
1732 		newentry->next = NULL;
1733 		newentry->prev = endchain;
1734 		newentry->start = dstaddr + oldoffset;
1735 		newentry->end =
1736 		    newentry->start + (entry->end - (entry->start + fudge));
1737 		if (newentry->end > newend || newentry->end < newentry->start)
1738 			newentry->end = newend;
1739 		newentry->object.uvm_obj = entry->object.uvm_obj;
1740 		if (newentry->object.uvm_obj) {
1741 			if (newentry->object.uvm_obj->pgops->pgo_reference)
1742 				newentry->object.uvm_obj->pgops->
1743 				    pgo_reference(newentry->object.uvm_obj);
1744 				newentry->offset = entry->offset + fudge;
1745 		} else {
1746 			newentry->offset = 0;
1747 		}
1748 		newentry->etype = entry->etype;
1749 		newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
1750 			entry->max_protection : entry->protection;
1751 		newentry->max_protection = entry->max_protection;
1752 		newentry->inheritance = entry->inheritance;
1753 		newentry->wired_count = 0;
1754 		newentry->aref.ar_amap = entry->aref.ar_amap;
1755 		if (newentry->aref.ar_amap) {
1756 			newentry->aref.ar_pageoff =
1757 			    entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1758 			uvm_map_reference_amap(newentry, AMAP_SHARED |
1759 			    ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
1760 		} else {
1761 			newentry->aref.ar_pageoff = 0;
1762 		}
1763 		newentry->advice = entry->advice;
1764 
1765 		/* now link it on the chain */
1766 		nchain++;
1767 		if (endchain == NULL) {
1768 			chain = endchain = newentry;
1769 		} else {
1770 			endchain->next = newentry;
1771 			endchain = newentry;
1772 		}
1773 
1774 		/* end of 'while' loop! */
1775 		if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1776 		    (entry->next == &srcmap->header ||
1777 		    entry->next->start != entry->end)) {
1778 			error = EINVAL;
1779 			goto bad;
1780 		}
1781 		entry = entry->next;
1782 		fudge = 0;
1783 	}
1784 
1785 	/*
1786 	 * step 4: close off chain (in format expected by uvm_map_replace)
1787 	 */
1788 
1789 	if (chain)
1790 		chain->prev = endchain;
1791 
1792 	/*
1793 	 * step 5: attempt to lock the dest map so we can pmap_copy.
1794 	 * note usage of copy_ok:
1795 	 *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
1796 	 *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
1797 	 */
1798 
1799 	if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
1800 		copy_ok = 1;
1801 		if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1802 		    nchain)) {
1803 			if (srcmap != dstmap)
1804 				vm_map_unlock(dstmap);
1805 			error = EIO;
1806 			goto bad;
1807 		}
1808 	} else {
1809 		copy_ok = 0;
1810 		/* replace defered until step 7 */
1811 	}
1812 
1813 	/*
1814 	 * step 6: traverse the srcmap a second time to do the following:
1815 	 *  - if we got a lock on the dstmap do pmap_copy
1816 	 *  - if UVM_EXTRACT_REMOVE remove the entries
1817 	 * we make use of orig_entry and orig_fudge (saved in step 2)
1818 	 */
1819 
1820 	if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
1821 
1822 		/* purge possible stale hints from srcmap */
1823 		if (flags & UVM_EXTRACT_REMOVE) {
1824 			SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev);
1825 			if (srcmap->first_free->start >= start)
1826 				srcmap->first_free = orig_entry->prev;
1827 		}
1828 
1829 		entry = orig_entry;
1830 		fudge = orig_fudge;
1831 		deadentry = NULL;	/* for UVM_EXTRACT_REMOVE */
1832 
1833 		while (entry->start < end && entry != &srcmap->header) {
1834 			if (copy_ok) {
1835 				oldoffset = (entry->start + fudge) - start;
1836 				elen = MIN(end, entry->end) -
1837 				    (entry->start + fudge);
1838 				pmap_copy(dstmap->pmap, srcmap->pmap,
1839 				    dstaddr + oldoffset, elen,
1840 				    entry->start + fudge);
1841 			}
1842 
1843 			/* we advance "entry" in the following if statement */
1844 			if (flags & UVM_EXTRACT_REMOVE) {
1845 				pmap_remove(srcmap->pmap, entry->start,
1846 						entry->end);
1847         			oldentry = entry;	/* save entry */
1848         			entry = entry->next;	/* advance */
1849 				uvm_map_entry_unlink(srcmap, oldentry);
1850 							/* add to dead list */
1851 				oldentry->next = deadentry;
1852 				deadentry = oldentry;
1853       			} else {
1854         			entry = entry->next;		/* advance */
1855 			}
1856 
1857 			/* end of 'while' loop */
1858 			fudge = 0;
1859 		}
1860 		pmap_update(srcmap->pmap);
1861 
1862 		/*
1863 		 * unlock dstmap.  we will dispose of deadentry in
1864 		 * step 7 if needed
1865 		 */
1866 
1867 		if (copy_ok && srcmap != dstmap)
1868 			vm_map_unlock(dstmap);
1869 
1870 	} else {
1871 		deadentry = NULL;
1872 	}
1873 
1874 	/*
1875 	 * step 7: we are done with the source map, unlock.   if copy_ok
1876 	 * is 0 then we have not replaced the dummy mapping in dstmap yet
1877 	 * and we need to do so now.
1878 	 */
1879 
1880 	vm_map_unlock(srcmap);
1881 	if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
1882 		uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
1883 
1884 	/* now do the replacement if we didn't do it in step 5 */
1885 	if (copy_ok == 0) {
1886 		vm_map_lock(dstmap);
1887 		error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1888 		    nchain);
1889 		vm_map_unlock(dstmap);
1890 
1891 		if (error == FALSE) {
1892 			error = EIO;
1893 			goto bad2;
1894 		}
1895 	}
1896 	return(0);
1897 
1898 	/*
1899 	 * bad: failure recovery
1900 	 */
1901 bad:
1902 	vm_map_unlock(srcmap);
1903 bad2:			/* src already unlocked */
1904 	if (chain)
1905 		uvm_unmap_detach(chain,
1906 		    (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1907 	uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
1908 	return(error);
1909 }
1910 
1911 /* end of extraction functions */
1912 
1913 /*
1914  * uvm_map_submap: punch down part of a map into a submap
1915  *
1916  * => only the kernel_map is allowed to be submapped
1917  * => the purpose of submapping is to break up the locking granularity
1918  *	of a larger map
1919  * => the range specified must have been mapped previously with a uvm_map()
1920  *	call [with uobj==NULL] to create a blank map entry in the main map.
1921  *	[And it had better still be blank!]
1922  * => maps which contain submaps should never be copied or forked.
1923  * => to remove a submap, use uvm_unmap() on the main map
1924  *	and then uvm_map_deallocate() the submap.
1925  * => main map must be unlocked.
1926  * => submap must have been init'd and have a zero reference count.
1927  *	[need not be locked as we don't actually reference it]
1928  */
1929 
1930 int
1931 uvm_map_submap(map, start, end, submap)
1932 	struct vm_map *map, *submap;
1933 	vaddr_t start, end;
1934 {
1935 	struct vm_map_entry *entry;
1936 	int error;
1937 
1938 	vm_map_lock(map);
1939 	VM_MAP_RANGE_CHECK(map, start, end);
1940 
1941 	if (uvm_map_lookup_entry(map, start, &entry)) {
1942 		UVM_MAP_CLIP_START(map, entry, start);
1943 		UVM_MAP_CLIP_END(map, entry, end);		/* to be safe */
1944 	} else {
1945 		entry = NULL;
1946 	}
1947 
1948 	if (entry != NULL &&
1949 	    entry->start == start && entry->end == end &&
1950 	    entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
1951 	    !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1952 		entry->etype |= UVM_ET_SUBMAP;
1953 		entry->object.sub_map = submap;
1954 		entry->offset = 0;
1955 		uvm_map_reference(submap);
1956 		error = 0;
1957 	} else {
1958 		error = EINVAL;
1959 	}
1960 	vm_map_unlock(map);
1961 	return error;
1962 }
1963 
1964 
1965 /*
1966  * uvm_map_protect: change map protection
1967  *
1968  * => set_max means set max_protection.
1969  * => map must be unlocked.
1970  */
1971 
1972 #define MASK(entry)     (UVM_ET_ISCOPYONWRITE(entry) ? \
1973 			 ~VM_PROT_WRITE : VM_PROT_ALL)
1974 
1975 int
1976 uvm_map_protect(map, start, end, new_prot, set_max)
1977 	struct vm_map *map;
1978 	vaddr_t start, end;
1979 	vm_prot_t new_prot;
1980 	boolean_t set_max;
1981 {
1982 	struct vm_map_entry *current, *entry;
1983 	int error = 0;
1984 	UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
1985 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
1986 		    map, start, end, new_prot);
1987 
1988 	vm_map_lock(map);
1989 	VM_MAP_RANGE_CHECK(map, start, end);
1990 	if (uvm_map_lookup_entry(map, start, &entry)) {
1991 		UVM_MAP_CLIP_START(map, entry, start);
1992 	} else {
1993 		entry = entry->next;
1994 	}
1995 
1996 	/*
1997 	 * make a first pass to check for protection violations.
1998 	 */
1999 
2000 	current = entry;
2001 	while ((current != &map->header) && (current->start < end)) {
2002 		if (UVM_ET_ISSUBMAP(current)) {
2003 			error = EINVAL;
2004 			goto out;
2005 		}
2006 		if ((new_prot & current->max_protection) != new_prot) {
2007 			error = EACCES;
2008 			goto out;
2009 		}
2010 		/*
2011 		 * Don't allow VM_PROT_EXECUTE to be set on entries that
2012 		 * point to vnodes that are associated with a NOEXEC file
2013 		 * system.
2014 		 */
2015 		if (UVM_ET_ISOBJ(current) &&
2016 		    UVM_OBJ_IS_VNODE(current->object.uvm_obj)) {
2017 			struct vnode *vp =
2018 			    (struct vnode *) current->object.uvm_obj;
2019 
2020 			if ((new_prot & VM_PROT_EXECUTE) != 0 &&
2021 			    (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
2022 				error = EACCES;
2023 				goto out;
2024 			}
2025 		}
2026 		current = current->next;
2027 	}
2028 
2029 	/* go back and fix up protections (no need to clip this time). */
2030 
2031 	current = entry;
2032 	while ((current != &map->header) && (current->start < end)) {
2033 		vm_prot_t old_prot;
2034 
2035 		UVM_MAP_CLIP_END(map, current, end);
2036 		old_prot = current->protection;
2037 		if (set_max)
2038 			current->protection =
2039 			    (current->max_protection = new_prot) & old_prot;
2040 		else
2041 			current->protection = new_prot;
2042 
2043 		/*
2044 		 * update physical map if necessary.  worry about copy-on-write
2045 		 * here -- CHECK THIS XXX
2046 		 */
2047 
2048 		if (current->protection != old_prot) {
2049 			/* update pmap! */
2050 			pmap_protect(map->pmap, current->start, current->end,
2051 			    current->protection & MASK(entry));
2052 
2053 			/*
2054 			 * If this entry points at a vnode, and the
2055 			 * protection includes VM_PROT_EXECUTE, mark
2056 			 * the vnode as VEXECMAP.
2057 			 */
2058 			if (UVM_ET_ISOBJ(current)) {
2059 				struct uvm_object *uobj =
2060 				    current->object.uvm_obj;
2061 
2062 				if (UVM_OBJ_IS_VNODE(uobj) &&
2063 				    (current->protection & VM_PROT_EXECUTE))
2064 					vn_markexec((struct vnode *) uobj);
2065 			}
2066 		}
2067 
2068 		/*
2069 		 * If the map is configured to lock any future mappings,
2070 		 * wire this entry now if the old protection was VM_PROT_NONE
2071 		 * and the new protection is not VM_PROT_NONE.
2072 		 */
2073 
2074 		if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
2075 		    VM_MAPENT_ISWIRED(entry) == 0 &&
2076 		    old_prot == VM_PROT_NONE &&
2077 		    new_prot != VM_PROT_NONE) {
2078 			if (uvm_map_pageable(map, entry->start,
2079 			    entry->end, FALSE,
2080 			    UVM_LK_ENTER|UVM_LK_EXIT) != 0) {
2081 
2082 				/*
2083 				 * If locking the entry fails, remember the
2084 				 * error if it's the first one.  Note we
2085 				 * still continue setting the protection in
2086 				 * the map, but will return the error
2087 				 * condition regardless.
2088 				 *
2089 				 * XXX Ignore what the actual error is,
2090 				 * XXX just call it a resource shortage
2091 				 * XXX so that it doesn't get confused
2092 				 * XXX what uvm_map_protect() itself would
2093 				 * XXX normally return.
2094 				 */
2095 
2096 				error = ENOMEM;
2097 			}
2098 		}
2099 		current = current->next;
2100 	}
2101 	pmap_update(map->pmap);
2102 
2103  out:
2104 	vm_map_unlock(map);
2105 	UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0);
2106 	return error;
2107 }
2108 
2109 #undef  MASK
2110 
2111 /*
2112  * uvm_map_inherit: set inheritance code for range of addrs in map.
2113  *
2114  * => map must be unlocked
2115  * => note that the inherit code is used during a "fork".  see fork
2116  *	code for details.
2117  */
2118 
2119 int
2120 uvm_map_inherit(map, start, end, new_inheritance)
2121 	struct vm_map *map;
2122 	vaddr_t start;
2123 	vaddr_t end;
2124 	vm_inherit_t new_inheritance;
2125 {
2126 	struct vm_map_entry *entry, *temp_entry;
2127 	UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
2128 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
2129 	    map, start, end, new_inheritance);
2130 
2131 	switch (new_inheritance) {
2132 	case MAP_INHERIT_NONE:
2133 	case MAP_INHERIT_COPY:
2134 	case MAP_INHERIT_SHARE:
2135 		break;
2136 	default:
2137 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2138 		return EINVAL;
2139 	}
2140 
2141 	vm_map_lock(map);
2142 	VM_MAP_RANGE_CHECK(map, start, end);
2143 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
2144 		entry = temp_entry;
2145 		UVM_MAP_CLIP_START(map, entry, start);
2146 	}  else {
2147 		entry = temp_entry->next;
2148 	}
2149 	while ((entry != &map->header) && (entry->start < end)) {
2150 		UVM_MAP_CLIP_END(map, entry, end);
2151 		entry->inheritance = new_inheritance;
2152 		entry = entry->next;
2153 	}
2154 	vm_map_unlock(map);
2155 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
2156 	return 0;
2157 }
2158 
2159 /*
2160  * uvm_map_advice: set advice code for range of addrs in map.
2161  *
2162  * => map must be unlocked
2163  */
2164 
2165 int
2166 uvm_map_advice(map, start, end, new_advice)
2167 	struct vm_map *map;
2168 	vaddr_t start;
2169 	vaddr_t end;
2170 	int new_advice;
2171 {
2172 	struct vm_map_entry *entry, *temp_entry;
2173 	UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
2174 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
2175 	    map, start, end, new_advice);
2176 
2177 	vm_map_lock(map);
2178 	VM_MAP_RANGE_CHECK(map, start, end);
2179 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
2180 		entry = temp_entry;
2181 		UVM_MAP_CLIP_START(map, entry, start);
2182 	} else {
2183 		entry = temp_entry->next;
2184 	}
2185 
2186 	/*
2187 	 * XXXJRT: disallow holes?
2188 	 */
2189 
2190 	while ((entry != &map->header) && (entry->start < end)) {
2191 		UVM_MAP_CLIP_END(map, entry, end);
2192 
2193 		switch (new_advice) {
2194 		case MADV_NORMAL:
2195 		case MADV_RANDOM:
2196 		case MADV_SEQUENTIAL:
2197 			/* nothing special here */
2198 			break;
2199 
2200 		default:
2201 			vm_map_unlock(map);
2202 			UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
2203 			return EINVAL;
2204 		}
2205 		entry->advice = new_advice;
2206 		entry = entry->next;
2207 	}
2208 
2209 	vm_map_unlock(map);
2210 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
2211 	return 0;
2212 }
2213 
2214 /*
2215  * uvm_map_pageable: sets the pageability of a range in a map.
2216  *
2217  * => wires map entries.  should not be used for transient page locking.
2218  *	for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
2219  * => regions sepcified as not pageable require lock-down (wired) memory
2220  *	and page tables.
2221  * => map must never be read-locked
2222  * => if islocked is TRUE, map is already write-locked
2223  * => we always unlock the map, since we must downgrade to a read-lock
2224  *	to call uvm_fault_wire()
2225  * => XXXCDC: check this and try and clean it up.
2226  */
2227 
2228 int
2229 uvm_map_pageable(map, start, end, new_pageable, lockflags)
2230 	struct vm_map *map;
2231 	vaddr_t start, end;
2232 	boolean_t new_pageable;
2233 	int lockflags;
2234 {
2235 	struct vm_map_entry *entry, *start_entry, *failed_entry;
2236 	int rv;
2237 #ifdef DIAGNOSTIC
2238 	u_int timestamp_save;
2239 #endif
2240 	UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
2241 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
2242 		    map, start, end, new_pageable);
2243 	KASSERT(map->flags & VM_MAP_PAGEABLE);
2244 
2245 	if ((lockflags & UVM_LK_ENTER) == 0)
2246 		vm_map_lock(map);
2247 	VM_MAP_RANGE_CHECK(map, start, end);
2248 
2249 	/*
2250 	 * only one pageability change may take place at one time, since
2251 	 * uvm_fault_wire assumes it will be called only once for each
2252 	 * wiring/unwiring.  therefore, we have to make sure we're actually
2253 	 * changing the pageability for the entire region.  we do so before
2254 	 * making any changes.
2255 	 */
2256 
2257 	if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
2258 		if ((lockflags & UVM_LK_EXIT) == 0)
2259 			vm_map_unlock(map);
2260 
2261 		UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0);
2262 		return EFAULT;
2263 	}
2264 	entry = start_entry;
2265 
2266 	/*
2267 	 * handle wiring and unwiring separately.
2268 	 */
2269 
2270 	if (new_pageable) {		/* unwire */
2271 		UVM_MAP_CLIP_START(map, entry, start);
2272 
2273 		/*
2274 		 * unwiring.  first ensure that the range to be unwired is
2275 		 * really wired down and that there are no holes.
2276 		 */
2277 
2278 		while ((entry != &map->header) && (entry->start < end)) {
2279 			if (entry->wired_count == 0 ||
2280 			    (entry->end < end &&
2281 			     (entry->next == &map->header ||
2282 			      entry->next->start > entry->end))) {
2283 				if ((lockflags & UVM_LK_EXIT) == 0)
2284 					vm_map_unlock(map);
2285 				UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0);
2286 				return EINVAL;
2287 			}
2288 			entry = entry->next;
2289 		}
2290 
2291 		/*
2292 		 * POSIX 1003.1b - a single munlock call unlocks a region,
2293 		 * regardless of the number of mlock calls made on that
2294 		 * region.
2295 		 */
2296 
2297 		entry = start_entry;
2298 		while ((entry != &map->header) && (entry->start < end)) {
2299 			UVM_MAP_CLIP_END(map, entry, end);
2300 			if (VM_MAPENT_ISWIRED(entry))
2301 				uvm_map_entry_unwire(map, entry);
2302 			entry = entry->next;
2303 		}
2304 		if ((lockflags & UVM_LK_EXIT) == 0)
2305 			vm_map_unlock(map);
2306 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2307 		return 0;
2308 	}
2309 
2310 	/*
2311 	 * wire case: in two passes [XXXCDC: ugly block of code here]
2312 	 *
2313 	 * 1: holding the write lock, we create any anonymous maps that need
2314 	 *    to be created.  then we clip each map entry to the region to
2315 	 *    be wired and increment its wiring count.
2316 	 *
2317 	 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
2318 	 *    in the pages for any newly wired area (wired_count == 1).
2319 	 *
2320 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
2321 	 *    deadlock with another thread that may have faulted on one of
2322 	 *    the pages to be wired (it would mark the page busy, blocking
2323 	 *    us, then in turn block on the map lock that we hold).  because
2324 	 *    of problems in the recursive lock package, we cannot upgrade
2325 	 *    to a write lock in vm_map_lookup.  thus, any actions that
2326 	 *    require the write lock must be done beforehand.  because we
2327 	 *    keep the read lock on the map, the copy-on-write status of the
2328 	 *    entries we modify here cannot change.
2329 	 */
2330 
2331 	while ((entry != &map->header) && (entry->start < end)) {
2332 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2333 
2334 			/*
2335 			 * perform actions of vm_map_lookup that need the
2336 			 * write lock on the map: create an anonymous map
2337 			 * for a copy-on-write region, or an anonymous map
2338 			 * for a zero-fill region.  (XXXCDC: submap case
2339 			 * ok?)
2340 			 */
2341 
2342 			if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
2343 				if (UVM_ET_ISNEEDSCOPY(entry) &&
2344 				    ((entry->max_protection & VM_PROT_WRITE) ||
2345 				     (entry->object.uvm_obj == NULL))) {
2346 					amap_copy(map, entry, M_WAITOK, TRUE,
2347 					    start, end);
2348 					/* XXXCDC: wait OK? */
2349 				}
2350 			}
2351 		}
2352 		UVM_MAP_CLIP_START(map, entry, start);
2353 		UVM_MAP_CLIP_END(map, entry, end);
2354 		entry->wired_count++;
2355 
2356 		/*
2357 		 * Check for holes
2358 		 */
2359 
2360 		if (entry->protection == VM_PROT_NONE ||
2361 		    (entry->end < end &&
2362 		     (entry->next == &map->header ||
2363 		      entry->next->start > entry->end))) {
2364 
2365 			/*
2366 			 * found one.  amap creation actions do not need to
2367 			 * be undone, but the wired counts need to be restored.
2368 			 */
2369 
2370 			while (entry != &map->header && entry->end > start) {
2371 				entry->wired_count--;
2372 				entry = entry->prev;
2373 			}
2374 			if ((lockflags & UVM_LK_EXIT) == 0)
2375 				vm_map_unlock(map);
2376 			UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
2377 			return EINVAL;
2378 		}
2379 		entry = entry->next;
2380 	}
2381 
2382 	/*
2383 	 * Pass 2.
2384 	 */
2385 
2386 #ifdef DIAGNOSTIC
2387 	timestamp_save = map->timestamp;
2388 #endif
2389 	vm_map_busy(map);
2390 	vm_map_downgrade(map);
2391 
2392 	rv = 0;
2393 	entry = start_entry;
2394 	while (entry != &map->header && entry->start < end) {
2395 		if (entry->wired_count == 1) {
2396 			rv = uvm_fault_wire(map, entry->start, entry->end,
2397 			    VM_FAULT_WIREMAX, entry->max_protection);
2398 			if (rv) {
2399 
2400 				/*
2401 				 * wiring failed.  break out of the loop.
2402 				 * we'll clean up the map below, once we
2403 				 * have a write lock again.
2404 				 */
2405 
2406 				break;
2407 			}
2408 		}
2409 		entry = entry->next;
2410 	}
2411 
2412 	if (rv) {        /* failed? */
2413 
2414 		/*
2415 		 * Get back to an exclusive (write) lock.
2416 		 */
2417 
2418 		vm_map_upgrade(map);
2419 		vm_map_unbusy(map);
2420 
2421 #ifdef DIAGNOSTIC
2422 		if (timestamp_save != map->timestamp)
2423 			panic("uvm_map_pageable: stale map");
2424 #endif
2425 
2426 		/*
2427 		 * first drop the wiring count on all the entries
2428 		 * which haven't actually been wired yet.
2429 		 */
2430 
2431 		failed_entry = entry;
2432 		while (entry != &map->header && entry->start < end) {
2433 			entry->wired_count--;
2434 			entry = entry->next;
2435 		}
2436 
2437 		/*
2438 		 * now, unwire all the entries that were successfully
2439 		 * wired above.
2440 		 */
2441 
2442 		entry = start_entry;
2443 		while (entry != failed_entry) {
2444 			entry->wired_count--;
2445 			if (VM_MAPENT_ISWIRED(entry) == 0)
2446 				uvm_map_entry_unwire(map, entry);
2447 			entry = entry->next;
2448 		}
2449 		if ((lockflags & UVM_LK_EXIT) == 0)
2450 			vm_map_unlock(map);
2451 		UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
2452 		return(rv);
2453 	}
2454 
2455 	/* We are holding a read lock here. */
2456 	if ((lockflags & UVM_LK_EXIT) == 0) {
2457 		vm_map_unbusy(map);
2458 		vm_map_unlock_read(map);
2459 	} else {
2460 
2461 		/*
2462 		 * Get back to an exclusive (write) lock.
2463 		 */
2464 
2465 		vm_map_upgrade(map);
2466 		vm_map_unbusy(map);
2467 	}
2468 
2469 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2470 	return 0;
2471 }
2472 
2473 /*
2474  * uvm_map_pageable_all: special case of uvm_map_pageable - affects
2475  * all mapped regions.
2476  *
2477  * => map must not be locked.
2478  * => if no flags are specified, all regions are unwired.
2479  * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
2480  */
2481 
2482 int
2483 uvm_map_pageable_all(map, flags, limit)
2484 	struct vm_map *map;
2485 	int flags;
2486 	vsize_t limit;
2487 {
2488 	struct vm_map_entry *entry, *failed_entry;
2489 	vsize_t size;
2490 	int rv;
2491 #ifdef DIAGNOSTIC
2492 	u_int timestamp_save;
2493 #endif
2494 	UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
2495 	UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
2496 
2497 	KASSERT(map->flags & VM_MAP_PAGEABLE);
2498 
2499 	vm_map_lock(map);
2500 
2501 	/*
2502 	 * handle wiring and unwiring separately.
2503 	 */
2504 
2505 	if (flags == 0) {			/* unwire */
2506 
2507 		/*
2508 		 * POSIX 1003.1b -- munlockall unlocks all regions,
2509 		 * regardless of how many times mlockall has been called.
2510 		 */
2511 
2512 		for (entry = map->header.next; entry != &map->header;
2513 		     entry = entry->next) {
2514 			if (VM_MAPENT_ISWIRED(entry))
2515 				uvm_map_entry_unwire(map, entry);
2516 		}
2517 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2518 		vm_map_unlock(map);
2519 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2520 		return 0;
2521 	}
2522 
2523 	if (flags & MCL_FUTURE) {
2524 
2525 		/*
2526 		 * must wire all future mappings; remember this.
2527 		 */
2528 
2529 		vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
2530 	}
2531 
2532 	if ((flags & MCL_CURRENT) == 0) {
2533 
2534 		/*
2535 		 * no more work to do!
2536 		 */
2537 
2538 		UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
2539 		vm_map_unlock(map);
2540 		return 0;
2541 	}
2542 
2543 	/*
2544 	 * wire case: in three passes [XXXCDC: ugly block of code here]
2545 	 *
2546 	 * 1: holding the write lock, count all pages mapped by non-wired
2547 	 *    entries.  if this would cause us to go over our limit, we fail.
2548 	 *
2549 	 * 2: still holding the write lock, we create any anonymous maps that
2550 	 *    need to be created.  then we increment its wiring count.
2551 	 *
2552 	 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
2553 	 *    in the pages for any newly wired area (wired_count == 1).
2554 	 *
2555 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
2556 	 *    deadlock with another thread that may have faulted on one of
2557 	 *    the pages to be wired (it would mark the page busy, blocking
2558 	 *    us, then in turn block on the map lock that we hold).  because
2559 	 *    of problems in the recursive lock package, we cannot upgrade
2560 	 *    to a write lock in vm_map_lookup.  thus, any actions that
2561 	 *    require the write lock must be done beforehand.  because we
2562 	 *    keep the read lock on the map, the copy-on-write status of the
2563 	 *    entries we modify here cannot change.
2564 	 */
2565 
2566 	for (size = 0, entry = map->header.next; entry != &map->header;
2567 	     entry = entry->next) {
2568 		if (entry->protection != VM_PROT_NONE &&
2569 		    VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2570 			size += entry->end - entry->start;
2571 		}
2572 	}
2573 
2574 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
2575 		vm_map_unlock(map);
2576 		return ENOMEM;
2577 	}
2578 
2579 	/* XXX non-pmap_wired_count case must be handled by caller */
2580 #ifdef pmap_wired_count
2581 	if (limit != 0 &&
2582 	    (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
2583 		vm_map_unlock(map);
2584 		return ENOMEM;
2585 	}
2586 #endif
2587 
2588 	/*
2589 	 * Pass 2.
2590 	 */
2591 
2592 	for (entry = map->header.next; entry != &map->header;
2593 	     entry = entry->next) {
2594 		if (entry->protection == VM_PROT_NONE)
2595 			continue;
2596 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2597 
2598 			/*
2599 			 * perform actions of vm_map_lookup that need the
2600 			 * write lock on the map: create an anonymous map
2601 			 * for a copy-on-write region, or an anonymous map
2602 			 * for a zero-fill region.  (XXXCDC: submap case
2603 			 * ok?)
2604 			 */
2605 
2606 			if (!UVM_ET_ISSUBMAP(entry)) {	/* not submap */
2607 				if (UVM_ET_ISNEEDSCOPY(entry) &&
2608 				    ((entry->max_protection & VM_PROT_WRITE) ||
2609 				     (entry->object.uvm_obj == NULL))) {
2610 					amap_copy(map, entry, M_WAITOK, TRUE,
2611 					    entry->start, entry->end);
2612 					/* XXXCDC: wait OK? */
2613 				}
2614 			}
2615 		}
2616 		entry->wired_count++;
2617 	}
2618 
2619 	/*
2620 	 * Pass 3.
2621 	 */
2622 
2623 #ifdef DIAGNOSTIC
2624 	timestamp_save = map->timestamp;
2625 #endif
2626 	vm_map_busy(map);
2627 	vm_map_downgrade(map);
2628 
2629 	rv = 0;
2630 	for (entry = map->header.next; entry != &map->header;
2631 	     entry = entry->next) {
2632 		if (entry->wired_count == 1) {
2633 			rv = uvm_fault_wire(map, entry->start, entry->end,
2634 			    VM_FAULT_WIREMAX, entry->max_protection);
2635 			if (rv) {
2636 
2637 				/*
2638 				 * wiring failed.  break out of the loop.
2639 				 * we'll clean up the map below, once we
2640 				 * have a write lock again.
2641 				 */
2642 
2643 				break;
2644 			}
2645 		}
2646 	}
2647 
2648 	if (rv) {
2649 
2650 		/*
2651 		 * Get back an exclusive (write) lock.
2652 		 */
2653 
2654 		vm_map_upgrade(map);
2655 		vm_map_unbusy(map);
2656 
2657 #ifdef DIAGNOSTIC
2658 		if (timestamp_save != map->timestamp)
2659 			panic("uvm_map_pageable_all: stale map");
2660 #endif
2661 
2662 		/*
2663 		 * first drop the wiring count on all the entries
2664 		 * which haven't actually been wired yet.
2665 		 *
2666 		 * Skip VM_PROT_NONE entries like we did above.
2667 		 */
2668 
2669 		failed_entry = entry;
2670 		for (/* nothing */; entry != &map->header;
2671 		     entry = entry->next) {
2672 			if (entry->protection == VM_PROT_NONE)
2673 				continue;
2674 			entry->wired_count--;
2675 		}
2676 
2677 		/*
2678 		 * now, unwire all the entries that were successfully
2679 		 * wired above.
2680 		 *
2681 		 * Skip VM_PROT_NONE entries like we did above.
2682 		 */
2683 
2684 		for (entry = map->header.next; entry != failed_entry;
2685 		     entry = entry->next) {
2686 			if (entry->protection == VM_PROT_NONE)
2687 				continue;
2688 			entry->wired_count--;
2689 			if (VM_MAPENT_ISWIRED(entry))
2690 				uvm_map_entry_unwire(map, entry);
2691 		}
2692 		vm_map_unlock(map);
2693 		UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
2694 		return (rv);
2695 	}
2696 
2697 	/* We are holding a read lock here. */
2698 	vm_map_unbusy(map);
2699 	vm_map_unlock_read(map);
2700 
2701 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2702 	return 0;
2703 }
2704 
2705 /*
2706  * uvm_map_clean: clean out a map range
2707  *
2708  * => valid flags:
2709  *   if (flags & PGO_CLEANIT): dirty pages are cleaned first
2710  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
2711  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
2712  *   if (flags & PGO_FREE): any cached pages are freed after clean
2713  * => returns an error if any part of the specified range isn't mapped
2714  * => never a need to flush amap layer since the anonymous memory has
2715  *	no permanent home, but may deactivate pages there
2716  * => called from sys_msync() and sys_madvise()
2717  * => caller must not write-lock map (read OK).
2718  * => we may sleep while cleaning if SYNCIO [with map read-locked]
2719  */
2720 
2721 int
2722 uvm_map_clean(map, start, end, flags)
2723 	struct vm_map *map;
2724 	vaddr_t start, end;
2725 	int flags;
2726 {
2727 	struct vm_map_entry *current, *entry;
2728 	struct uvm_object *uobj;
2729 	struct vm_amap *amap;
2730 	struct vm_anon *anon;
2731 	struct vm_page *pg;
2732 	vaddr_t offset;
2733 	vsize_t size;
2734 	int error, refs;
2735 	UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
2736 
2737 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
2738 		    map, start, end, flags);
2739 	KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) !=
2740 		(PGO_FREE|PGO_DEACTIVATE));
2741 
2742 	vm_map_lock_read(map);
2743 	VM_MAP_RANGE_CHECK(map, start, end);
2744 	if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
2745 		vm_map_unlock_read(map);
2746 		return EFAULT;
2747 	}
2748 
2749 	/*
2750 	 * Make a first pass to check for holes.
2751 	 */
2752 
2753 	for (current = entry; current->start < end; current = current->next) {
2754 		if (UVM_ET_ISSUBMAP(current)) {
2755 			vm_map_unlock_read(map);
2756 			return EINVAL;
2757 		}
2758 		if (end <= current->end) {
2759 			break;
2760 		}
2761 		if (current->end != current->next->start) {
2762 			vm_map_unlock_read(map);
2763 			return EFAULT;
2764 		}
2765 	}
2766 
2767 	error = 0;
2768 	for (current = entry; start < end; current = current->next) {
2769 		amap = current->aref.ar_amap;	/* top layer */
2770 		uobj = current->object.uvm_obj;	/* bottom layer */
2771 		KASSERT(start >= current->start);
2772 
2773 		/*
2774 		 * No amap cleaning necessary if:
2775 		 *
2776 		 *	(1) There's no amap.
2777 		 *
2778 		 *	(2) We're not deactivating or freeing pages.
2779 		 */
2780 
2781 		if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
2782 			goto flush_object;
2783 
2784 		amap_lock(amap);
2785 		offset = start - current->start;
2786 		size = MIN(end, current->end) - start;
2787 		for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) {
2788 			anon = amap_lookup(&current->aref, offset);
2789 			if (anon == NULL)
2790 				continue;
2791 
2792 			simple_lock(&anon->an_lock);
2793 			pg = anon->u.an_page;
2794 			if (pg == NULL) {
2795 				simple_unlock(&anon->an_lock);
2796 				continue;
2797 			}
2798 
2799 			switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
2800 
2801 			/*
2802 			 * In these first 3 cases, we just deactivate the page.
2803 			 */
2804 
2805 			case PGO_CLEANIT|PGO_FREE:
2806 			case PGO_CLEANIT|PGO_DEACTIVATE:
2807 			case PGO_DEACTIVATE:
2808  deactivate_it:
2809 				/*
2810 				 * skip the page if it's loaned or wired,
2811 				 * since it shouldn't be on a paging queue
2812 				 * at all in these cases.
2813 				 */
2814 
2815 				uvm_lock_pageq();
2816 				if (pg->loan_count != 0 ||
2817 				    pg->wire_count != 0) {
2818 					uvm_unlock_pageq();
2819 					simple_unlock(&anon->an_lock);
2820 					continue;
2821 				}
2822 				KASSERT(pg->uanon == anon);
2823 				pmap_clear_reference(pg);
2824 				uvm_pagedeactivate(pg);
2825 				uvm_unlock_pageq();
2826 				simple_unlock(&anon->an_lock);
2827 				continue;
2828 
2829 			case PGO_FREE:
2830 
2831 				/*
2832 				 * If there are multiple references to
2833 				 * the amap, just deactivate the page.
2834 				 */
2835 
2836 				if (amap_refs(amap) > 1)
2837 					goto deactivate_it;
2838 
2839 				/* skip the page if it's wired */
2840 				if (pg->wire_count != 0) {
2841 					simple_unlock(&anon->an_lock);
2842 					continue;
2843 				}
2844 				amap_unadd(&current->aref, offset);
2845 				refs = --anon->an_ref;
2846 				simple_unlock(&anon->an_lock);
2847 				if (refs == 0)
2848 					uvm_anfree(anon);
2849 				continue;
2850 			}
2851 		}
2852 		amap_unlock(amap);
2853 
2854  flush_object:
2855 		/*
2856 		 * flush pages if we've got a valid backing object.
2857 		 * note that we must always clean object pages before
2858 		 * freeing them since otherwise we could reveal stale
2859 		 * data from files.
2860 		 */
2861 
2862 		offset = current->offset + (start - current->start);
2863 		size = MIN(end, current->end) - start;
2864 		if (uobj != NULL) {
2865 			simple_lock(&uobj->vmobjlock);
2866 			error = (uobj->pgops->pgo_put)(uobj, offset,
2867 			    offset + size, flags | PGO_CLEANIT);
2868 		}
2869 		start += size;
2870 	}
2871 	vm_map_unlock_read(map);
2872 	return (error);
2873 }
2874 
2875 
2876 /*
2877  * uvm_map_checkprot: check protection in map
2878  *
2879  * => must allow specified protection in a fully allocated region.
2880  * => map must be read or write locked by caller.
2881  */
2882 
2883 boolean_t
2884 uvm_map_checkprot(map, start, end, protection)
2885 	struct vm_map * map;
2886 	vaddr_t start, end;
2887 	vm_prot_t protection;
2888 {
2889 	struct vm_map_entry *entry;
2890 	struct vm_map_entry *tmp_entry;
2891 
2892 	if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
2893 		return(FALSE);
2894 	}
2895 	entry = tmp_entry;
2896 	while (start < end) {
2897 		if (entry == &map->header) {
2898 			return(FALSE);
2899 		}
2900 
2901 		/*
2902 		 * no holes allowed
2903 		 */
2904 
2905 		if (start < entry->start) {
2906 			return(FALSE);
2907 		}
2908 
2909 		/*
2910 		 * check protection associated with entry
2911 		 */
2912 
2913 		if ((entry->protection & protection) != protection) {
2914 			return(FALSE);
2915 		}
2916 		start = entry->end;
2917 		entry = entry->next;
2918 	}
2919 	return(TRUE);
2920 }
2921 
2922 /*
2923  * uvmspace_alloc: allocate a vmspace structure.
2924  *
2925  * - structure includes vm_map and pmap
2926  * - XXX: no locking on this structure
2927  * - refcnt set to 1, rest must be init'd by caller
2928  */
2929 struct vmspace *
2930 uvmspace_alloc(min, max)
2931 	vaddr_t min, max;
2932 {
2933 	struct vmspace *vm;
2934 	UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
2935 
2936 	vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
2937 	uvmspace_init(vm, NULL, min, max);
2938 	UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
2939 	return (vm);
2940 }
2941 
2942 /*
2943  * uvmspace_init: initialize a vmspace structure.
2944  *
2945  * - XXX: no locking on this structure
2946  * - refcnt set to 1, rest must me init'd by caller
2947  */
2948 void
2949 uvmspace_init(vm, pmap, min, max)
2950 	struct vmspace *vm;
2951 	struct pmap *pmap;
2952 	vaddr_t min, max;
2953 {
2954 	UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
2955 
2956 	memset(vm, 0, sizeof(*vm));
2957 	uvm_map_setup(&vm->vm_map, min, max, VM_MAP_PAGEABLE);
2958 	if (pmap)
2959 		pmap_reference(pmap);
2960 	else
2961 		pmap = pmap_create();
2962 	vm->vm_map.pmap = pmap;
2963 	vm->vm_refcnt = 1;
2964 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
2965 }
2966 
2967 /*
2968  * uvmspace_share: share a vmspace between two proceses
2969  *
2970  * - XXX: no locking on vmspace
2971  * - used for vfork, threads(?)
2972  */
2973 
2974 void
2975 uvmspace_share(p1, p2)
2976 	struct proc *p1, *p2;
2977 {
2978 	p2->p_vmspace = p1->p_vmspace;
2979 	p1->p_vmspace->vm_refcnt++;
2980 }
2981 
2982 /*
2983  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
2984  *
2985  * - XXX: no locking on vmspace
2986  */
2987 
2988 void
2989 uvmspace_unshare(l)
2990 	struct lwp *l;
2991 {
2992 	struct proc *p = l->l_proc;
2993 	struct vmspace *nvm, *ovm = p->p_vmspace;
2994 
2995 	if (ovm->vm_refcnt == 1)
2996 		/* nothing to do: vmspace isn't shared in the first place */
2997 		return;
2998 
2999 	/* make a new vmspace, still holding old one */
3000 	nvm = uvmspace_fork(ovm);
3001 
3002 	pmap_deactivate(l);		/* unbind old vmspace */
3003 	p->p_vmspace = nvm;
3004 	pmap_activate(l);		/* switch to new vmspace */
3005 
3006 	uvmspace_free(ovm);		/* drop reference to old vmspace */
3007 }
3008 
3009 /*
3010  * uvmspace_exec: the process wants to exec a new program
3011  *
3012  * - XXX: no locking on vmspace
3013  */
3014 
3015 void
3016 uvmspace_exec(l, start, end)
3017 	struct lwp *l;
3018 	vaddr_t start, end;
3019 {
3020 	struct proc *p = l->l_proc;
3021 	struct vmspace *nvm, *ovm = p->p_vmspace;
3022 	struct vm_map *map = &ovm->vm_map;
3023 
3024 #ifdef __sparc__
3025 	/* XXX cgd 960926: the sparc #ifdef should be a MD hook */
3026 	kill_user_windows(l);   /* before stack addresses go away */
3027 #endif
3028 
3029 	/*
3030 	 * see if more than one process is using this vmspace...
3031 	 */
3032 
3033 	if (ovm->vm_refcnt == 1) {
3034 
3035 		/*
3036 		 * if p is the only process using its vmspace then we can safely
3037 		 * recycle that vmspace for the program that is being exec'd.
3038 		 */
3039 
3040 #ifdef SYSVSHM
3041 		/*
3042 		 * SYSV SHM semantics require us to kill all segments on an exec
3043 		 */
3044 
3045 		if (ovm->vm_shm)
3046 			shmexit(ovm);
3047 #endif
3048 
3049 		/*
3050 		 * POSIX 1003.1b -- "lock future mappings" is revoked
3051 		 * when a process execs another program image.
3052 		 */
3053 
3054 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
3055 
3056 		/*
3057 		 * now unmap the old program
3058 		 */
3059 
3060 		pmap_remove_all(map->pmap);
3061 		uvm_unmap(map, map->min_offset, map->max_offset);
3062 
3063 		/*
3064 		 * resize the map
3065 		 */
3066 
3067 		map->min_offset = start;
3068 		map->max_offset = end;
3069 	} else {
3070 
3071 		/*
3072 		 * p's vmspace is being shared, so we can't reuse it for p since
3073 		 * it is still being used for others.   allocate a new vmspace
3074 		 * for p
3075 		 */
3076 
3077 		nvm = uvmspace_alloc(start, end);
3078 
3079 		/*
3080 		 * install new vmspace and drop our ref to the old one.
3081 		 */
3082 
3083 		pmap_deactivate(l);
3084 		p->p_vmspace = nvm;
3085 		pmap_activate(l);
3086 
3087 		uvmspace_free(ovm);
3088 	}
3089 }
3090 
3091 /*
3092  * uvmspace_free: free a vmspace data structure
3093  *
3094  * - XXX: no locking on vmspace
3095  */
3096 
3097 void
3098 uvmspace_free(vm)
3099 	struct vmspace *vm;
3100 {
3101 	struct vm_map_entry *dead_entries;
3102 	struct vm_map *map;
3103 	UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
3104 
3105 	UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
3106 	if (--vm->vm_refcnt > 0) {
3107 		return;
3108 	}
3109 
3110 	/*
3111 	 * at this point, there should be no other references to the map.
3112 	 * delete all of the mappings, then destroy the pmap.
3113 	 */
3114 
3115 	map = &vm->vm_map;
3116 	map->flags |= VM_MAP_DYING;
3117 	pmap_remove_all(map->pmap);
3118 #ifdef SYSVSHM
3119 	/* Get rid of any SYSV shared memory segments. */
3120 	if (vm->vm_shm != NULL)
3121 		shmexit(vm);
3122 #endif
3123 	if (map->nentries) {
3124 		uvm_unmap_remove(map, map->min_offset, map->max_offset,
3125 		    &dead_entries);
3126 		if (dead_entries != NULL)
3127 			uvm_unmap_detach(dead_entries, 0);
3128 	}
3129 	pmap_destroy(map->pmap);
3130 	pool_put(&uvm_vmspace_pool, vm);
3131 }
3132 
3133 /*
3134  *   F O R K   -   m a i n   e n t r y   p o i n t
3135  */
3136 /*
3137  * uvmspace_fork: fork a process' main map
3138  *
3139  * => create a new vmspace for child process from parent.
3140  * => parent's map must not be locked.
3141  */
3142 
3143 struct vmspace *
3144 uvmspace_fork(vm1)
3145 	struct vmspace *vm1;
3146 {
3147 	struct vmspace *vm2;
3148 	struct vm_map *old_map = &vm1->vm_map;
3149 	struct vm_map *new_map;
3150 	struct vm_map_entry *old_entry;
3151 	struct vm_map_entry *new_entry;
3152 	pmap_t new_pmap;
3153 	UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
3154 
3155 	vm_map_lock(old_map);
3156 
3157 	vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset);
3158 	memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
3159 	(caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
3160 	new_map = &vm2->vm_map;		  /* XXX */
3161 	new_pmap = new_map->pmap;
3162 
3163 	old_entry = old_map->header.next;
3164 
3165 	/*
3166 	 * go entry-by-entry
3167 	 */
3168 
3169 	while (old_entry != &old_map->header) {
3170 
3171 		/*
3172 		 * first, some sanity checks on the old entry
3173 		 */
3174 
3175 		KASSERT(!UVM_ET_ISSUBMAP(old_entry));
3176 		KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) ||
3177 			!UVM_ET_ISNEEDSCOPY(old_entry));
3178 
3179 		switch (old_entry->inheritance) {
3180 		case MAP_INHERIT_NONE:
3181 
3182 			/*
3183 			 * drop the mapping
3184 			 */
3185 
3186 			break;
3187 
3188 		case MAP_INHERIT_SHARE:
3189 
3190 			/*
3191 			 * share the mapping: this means we want the old and
3192 			 * new entries to share amaps and backing objects.
3193 			 */
3194 			/*
3195 			 * if the old_entry needs a new amap (due to prev fork)
3196 			 * then we need to allocate it now so that we have
3197 			 * something we own to share with the new_entry.   [in
3198 			 * other words, we need to clear needs_copy]
3199 			 */
3200 
3201 			if (UVM_ET_ISNEEDSCOPY(old_entry)) {
3202 				/* get our own amap, clears needs_copy */
3203 				amap_copy(old_map, old_entry, M_WAITOK, FALSE,
3204 				    0, 0);
3205 				/* XXXCDC: WAITOK??? */
3206 			}
3207 
3208 			new_entry = uvm_mapent_alloc(new_map, 0);
3209 			/* old_entry -> new_entry */
3210 			uvm_mapent_copy(old_entry, new_entry);
3211 
3212 			/* new pmap has nothing wired in it */
3213 			new_entry->wired_count = 0;
3214 
3215 			/*
3216 			 * gain reference to object backing the map (can't
3217 			 * be a submap, already checked this case).
3218 			 */
3219 
3220 			if (new_entry->aref.ar_amap)
3221 				uvm_map_reference_amap(new_entry, AMAP_SHARED);
3222 
3223 			if (new_entry->object.uvm_obj &&
3224 			    new_entry->object.uvm_obj->pgops->pgo_reference)
3225 				new_entry->object.uvm_obj->
3226 				    pgops->pgo_reference(
3227 				        new_entry->object.uvm_obj);
3228 
3229 			/* insert entry at end of new_map's entry list */
3230 			uvm_map_entry_link(new_map, new_map->header.prev,
3231 			    new_entry);
3232 
3233 			break;
3234 
3235 		case MAP_INHERIT_COPY:
3236 
3237 			/*
3238 			 * copy-on-write the mapping (using mmap's
3239 			 * MAP_PRIVATE semantics)
3240 			 *
3241 			 * allocate new_entry, adjust reference counts.
3242 			 * (note that new references are read-only).
3243 			 */
3244 
3245 			new_entry = uvm_mapent_alloc(new_map, 0);
3246 			/* old_entry -> new_entry */
3247 			uvm_mapent_copy(old_entry, new_entry);
3248 
3249 			if (new_entry->aref.ar_amap)
3250 				uvm_map_reference_amap(new_entry, 0);
3251 
3252 			if (new_entry->object.uvm_obj &&
3253 			    new_entry->object.uvm_obj->pgops->pgo_reference)
3254 				new_entry->object.uvm_obj->pgops->pgo_reference
3255 				    (new_entry->object.uvm_obj);
3256 
3257 			/* new pmap has nothing wired in it */
3258 			new_entry->wired_count = 0;
3259 
3260 			new_entry->etype |=
3261 			    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
3262 			uvm_map_entry_link(new_map, new_map->header.prev,
3263 			    new_entry);
3264 
3265 			/*
3266 			 * the new entry will need an amap.  it will either
3267 			 * need to be copied from the old entry or created
3268 			 * from scratch (if the old entry does not have an
3269 			 * amap).  can we defer this process until later
3270 			 * (by setting "needs_copy") or do we need to copy
3271 			 * the amap now?
3272 			 *
3273 			 * we must copy the amap now if any of the following
3274 			 * conditions hold:
3275 			 * 1. the old entry has an amap and that amap is
3276 			 *    being shared.  this means that the old (parent)
3277 			 *    process is sharing the amap with another
3278 			 *    process.  if we do not clear needs_copy here
3279 			 *    we will end up in a situation where both the
3280 			 *    parent and child process are refering to the
3281 			 *    same amap with "needs_copy" set.  if the
3282 			 *    parent write-faults, the fault routine will
3283 			 *    clear "needs_copy" in the parent by allocating
3284 			 *    a new amap.   this is wrong because the
3285 			 *    parent is supposed to be sharing the old amap
3286 			 *    and the new amap will break that.
3287 			 *
3288 			 * 2. if the old entry has an amap and a non-zero
3289 			 *    wire count then we are going to have to call
3290 			 *    amap_cow_now to avoid page faults in the
3291 			 *    parent process.   since amap_cow_now requires
3292 			 *    "needs_copy" to be clear we might as well
3293 			 *    clear it here as well.
3294 			 *
3295 			 */
3296 
3297 			if (old_entry->aref.ar_amap != NULL) {
3298 				if ((amap_flags(old_entry->aref.ar_amap) &
3299 				     AMAP_SHARED) != 0 ||
3300 				    VM_MAPENT_ISWIRED(old_entry)) {
3301 
3302 					amap_copy(new_map, new_entry, M_WAITOK,
3303 					    FALSE, 0, 0);
3304 					/* XXXCDC: M_WAITOK ... ok? */
3305 				}
3306 			}
3307 
3308 			/*
3309 			 * if the parent's entry is wired down, then the
3310 			 * parent process does not want page faults on
3311 			 * access to that memory.  this means that we
3312 			 * cannot do copy-on-write because we can't write
3313 			 * protect the old entry.   in this case we
3314 			 * resolve all copy-on-write faults now, using
3315 			 * amap_cow_now.   note that we have already
3316 			 * allocated any needed amap (above).
3317 			 */
3318 
3319 			if (VM_MAPENT_ISWIRED(old_entry)) {
3320 
3321 			  /*
3322 			   * resolve all copy-on-write faults now
3323 			   * (note that there is nothing to do if
3324 			   * the old mapping does not have an amap).
3325 			   */
3326 			  if (old_entry->aref.ar_amap)
3327 			    amap_cow_now(new_map, new_entry);
3328 
3329 			} else {
3330 
3331 			  /*
3332 			   * setup mappings to trigger copy-on-write faults
3333 			   * we must write-protect the parent if it has
3334 			   * an amap and it is not already "needs_copy"...
3335 			   * if it is already "needs_copy" then the parent
3336 			   * has already been write-protected by a previous
3337 			   * fork operation.
3338 			   */
3339 
3340 			  if (old_entry->aref.ar_amap &&
3341 			      !UVM_ET_ISNEEDSCOPY(old_entry)) {
3342 			      if (old_entry->max_protection & VM_PROT_WRITE) {
3343 				pmap_protect(old_map->pmap,
3344 					     old_entry->start,
3345 					     old_entry->end,
3346 					     old_entry->protection &
3347 					     ~VM_PROT_WRITE);
3348 				pmap_update(old_map->pmap);
3349 			      }
3350 			      old_entry->etype |= UVM_ET_NEEDSCOPY;
3351 			  }
3352 			}
3353 			break;
3354 		}  /* end of switch statement */
3355 		old_entry = old_entry->next;
3356 	}
3357 
3358 	new_map->size = old_map->size;
3359 	vm_map_unlock(old_map);
3360 
3361 #ifdef SYSVSHM
3362 	if (vm1->vm_shm)
3363 		shmfork(vm1, vm2);
3364 #endif
3365 
3366 #ifdef PMAP_FORK
3367 	pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
3368 #endif
3369 
3370 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3371 	return(vm2);
3372 }
3373 
3374 
3375 #if defined(DDB)
3376 
3377 /*
3378  * DDB hooks
3379  */
3380 
3381 /*
3382  * uvm_map_printit: actually prints the map
3383  */
3384 
3385 void
3386 uvm_map_printit(map, full, pr)
3387 	struct vm_map *map;
3388 	boolean_t full;
3389 	void (*pr) __P((const char *, ...));
3390 {
3391 	struct vm_map_entry *entry;
3392 
3393 	(*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
3394 	(*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
3395 	    map->nentries, map->size, map->ref_count, map->timestamp,
3396 	    map->flags);
3397 	(*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
3398 	    pmap_resident_count(map->pmap));
3399 	if (!full)
3400 		return;
3401 	for (entry = map->header.next; entry != &map->header;
3402 	    entry = entry->next) {
3403 		(*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
3404 		    entry, entry->start, entry->end, entry->object.uvm_obj,
3405 		    (long long)entry->offset, entry->aref.ar_amap,
3406 		    entry->aref.ar_pageoff);
3407 		(*pr)(
3408 		    "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, "
3409 		    "wc=%d, adv=%d\n",
3410 		    (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
3411 		    (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
3412 		    (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
3413 		    entry->protection, entry->max_protection,
3414 		    entry->inheritance, entry->wired_count, entry->advice);
3415 	}
3416 }
3417 
3418 /*
3419  * uvm_object_printit: actually prints the object
3420  */
3421 
3422 void
3423 uvm_object_printit(uobj, full, pr)
3424 	struct uvm_object *uobj;
3425 	boolean_t full;
3426 	void (*pr) __P((const char *, ...));
3427 {
3428 	struct vm_page *pg;
3429 	int cnt = 0;
3430 
3431 	(*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
3432 	    uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
3433 	if (UVM_OBJ_IS_KERN_OBJECT(uobj))
3434 		(*pr)("refs=<SYSTEM>\n");
3435 	else
3436 		(*pr)("refs=%d\n", uobj->uo_refs);
3437 
3438 	if (!full) {
3439 		return;
3440 	}
3441 	(*pr)("  PAGES <pg,offset>:\n  ");
3442 	TAILQ_FOREACH(pg, &uobj->memq, listq) {
3443 		cnt++;
3444 		(*pr)("<%p,0x%llx> ", pg, (long long)pg->offset);
3445 		if ((cnt % 3) == 0) {
3446 			(*pr)("\n  ");
3447 		}
3448 	}
3449 	if ((cnt % 3) != 0) {
3450 		(*pr)("\n");
3451 	}
3452 }
3453 
3454 /*
3455  * uvm_page_printit: actually print the page
3456  */
3457 
3458 static const char page_flagbits[] =
3459 	"\20\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY"
3460 	"\11ZERO\15PAGER1";
3461 static const char page_pqflagbits[] =
3462 	"\20\1FREE\2INACTIVE\3ACTIVE\5ANON\6AOBJ";
3463 
3464 void
3465 uvm_page_printit(pg, full, pr)
3466 	struct vm_page *pg;
3467 	boolean_t full;
3468 	void (*pr) __P((const char *, ...));
3469 {
3470 	struct vm_page *tpg;
3471 	struct uvm_object *uobj;
3472 	struct pglist *pgl;
3473 	char pgbuf[128];
3474 	char pqbuf[128];
3475 
3476 	(*pr)("PAGE %p:\n", pg);
3477 	bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf));
3478 	bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
3479 	(*pr)("  flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n",
3480 	    pgbuf, pqbuf, pg->wire_count, (long)pg->phys_addr);
3481 	(*pr)("  uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
3482 	    pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
3483 #if defined(UVM_PAGE_TRKOWN)
3484 	if (pg->flags & PG_BUSY)
3485 		(*pr)("  owning process = %d, tag=%s\n",
3486 		    pg->owner, pg->owner_tag);
3487 	else
3488 		(*pr)("  page not busy, no owner\n");
3489 #else
3490 	(*pr)("  [page ownership tracking disabled]\n");
3491 #endif
3492 
3493 	if (!full)
3494 		return;
3495 
3496 	/* cross-verify object/anon */
3497 	if ((pg->pqflags & PQ_FREE) == 0) {
3498 		if (pg->pqflags & PQ_ANON) {
3499 			if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
3500 			    (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
3501 				(pg->uanon) ? pg->uanon->u.an_page : NULL);
3502 			else
3503 				(*pr)("  anon backpointer is OK\n");
3504 		} else {
3505 			uobj = pg->uobject;
3506 			if (uobj) {
3507 				(*pr)("  checking object list\n");
3508 				TAILQ_FOREACH(tpg, &uobj->memq, listq) {
3509 					if (tpg == pg) {
3510 						break;
3511 					}
3512 				}
3513 				if (tpg)
3514 					(*pr)("  page found on object list\n");
3515 				else
3516 			(*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
3517 			}
3518 		}
3519 	}
3520 
3521 	/* cross-verify page queue */
3522 	if (pg->pqflags & PQ_FREE) {
3523 		int fl = uvm_page_lookup_freelist(pg);
3524 		int color = VM_PGCOLOR_BUCKET(pg);
3525 		pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[
3526 		    ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN];
3527 	} else if (pg->pqflags & PQ_INACTIVE) {
3528 		pgl = &uvm.page_inactive;
3529 	} else if (pg->pqflags & PQ_ACTIVE) {
3530 		pgl = &uvm.page_active;
3531  	} else {
3532 		pgl = NULL;
3533 	}
3534 
3535 	if (pgl) {
3536 		(*pr)("  checking pageq list\n");
3537 		TAILQ_FOREACH(tpg, pgl, pageq) {
3538 			if (tpg == pg) {
3539 				break;
3540 			}
3541 		}
3542 		if (tpg)
3543 			(*pr)("  page found on pageq list\n");
3544 		else
3545 			(*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
3546 	}
3547 }
3548 #endif
3549