xref: /openbsd-src/sys/uvm/uvm_map.c (revision 8445c53715e7030056b779e8ab40efb7820981f2)
1 /*	$OpenBSD: uvm_map.c,v 1.24 2001/09/19 20:50:59 mickey Exp $	*/
2 /*	$NetBSD: uvm_map.c,v 1.77 2000/06/13 04:10:47 chs Exp $	*/
3 
4 /*
5  * Copyright (c) 1997 Charles D. Cranor and Washington University.
6  * Copyright (c) 1991, 1993, The Regents of the University of California.
7  *
8  * All rights reserved.
9  *
10  * This code is derived from software contributed to Berkeley by
11  * The Mach Operating System project at Carnegie-Mellon University.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by Charles D. Cranor,
24  *      Washington University, the University of California, Berkeley and
25  *      its contributors.
26  * 4. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)vm_map.c    8.3 (Berkeley) 1/12/94
43  * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp
44  *
45  *
46  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
47  * All rights reserved.
48  *
49  * Permission to use, copy, modify and distribute this software and
50  * its documentation is hereby granted, provided that both the copyright
51  * notice and this permission notice appear in all copies of the
52  * software, derivative works or modified versions, and any portions
53  * thereof, and that both notices appear in supporting documentation.
54  *
55  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
56  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
57  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
58  *
59  * Carnegie Mellon requests users of this software to return to
60  *
61  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
62  *  School of Computer Science
63  *  Carnegie Mellon University
64  *  Pittsburgh PA 15213-3890
65  *
66  * any improvements or extensions that they make and grant Carnegie the
67  * rights to redistribute these changes.
68  */
69 
70 /*
71  * uvm_map.c: uvm map operations
72  */
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/mman.h>
77 #include <sys/proc.h>
78 #include <sys/malloc.h>
79 #include <sys/pool.h>
80 
81 #ifdef SYSVSHM
82 #include <sys/shm.h>
83 #endif
84 
85 #include <vm/vm.h>
86 #include <vm/vm_page.h>
87 
88 #define UVM_MAP
89 #include <uvm/uvm.h>
90 
91 #ifdef DDB
92 #include <uvm/uvm_ddb.h>
93 #endif
94 
95 
96 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge;
97 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint;
98 
99 /*
100  * pool for vmspace structures.
101  */
102 
103 struct pool uvm_vmspace_pool;
104 
105 /*
106  * pool for dynamically-allocated map entries.
107  */
108 
109 struct pool uvm_map_entry_pool;
110 
111 #ifdef PMAP_GROWKERNEL
112 /*
113  * This global represents the end of the kernel virtual address
114  * space.  If we want to exceed this, we must grow the kernel
115  * virtual address space dynamically.
116  *
117  * Note, this variable is locked by kernel_map's lock.
118  */
119 vaddr_t uvm_maxkaddr;
120 #endif
121 
122 /*
123  * macros
124  */
125 
126 /*
127  * uvm_map_entry_link: insert entry into a map
128  *
129  * => map must be locked
130  */
131 #define uvm_map_entry_link(map, after_where, entry) do { \
132 	(map)->nentries++; \
133 	(entry)->prev = (after_where); \
134 	(entry)->next = (after_where)->next; \
135 	(entry)->prev->next = (entry); \
136 	(entry)->next->prev = (entry); \
137 } while (0)
138 
139 /*
140  * uvm_map_entry_unlink: remove entry from a map
141  *
142  * => map must be locked
143  */
144 #define uvm_map_entry_unlink(map, entry) do { \
145 	(map)->nentries--; \
146 	(entry)->next->prev = (entry)->prev; \
147 	(entry)->prev->next = (entry)->next; \
148 } while (0)
149 
150 /*
151  * SAVE_HINT: saves the specified entry as the hint for future lookups.
152  *
153  * => map need not be locked (protected by hint_lock).
154  */
155 #define SAVE_HINT(map,value) do { \
156 	simple_lock(&(map)->hint_lock); \
157 	(map)->hint = (value); \
158 	simple_unlock(&(map)->hint_lock); \
159 } while (0)
160 
161 /*
162  * VM_MAP_RANGE_CHECK: check and correct range
163  *
164  * => map must at least be read locked
165  */
166 
167 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
168 	if (start < vm_map_min(map)) 		\
169 		start = vm_map_min(map);        \
170 	if (end > vm_map_max(map))              \
171 		end = vm_map_max(map);          \
172 	if (start > end)                        \
173 		start = end;                    \
174 } while (0)
175 
176 /*
177  * local prototypes
178  */
179 
180 static vm_map_entry_t	uvm_mapent_alloc __P((vm_map_t));
181 static void		uvm_mapent_copy __P((vm_map_entry_t,vm_map_entry_t));
182 static void		uvm_mapent_free __P((vm_map_entry_t));
183 static void		uvm_map_entry_unwire __P((vm_map_t, vm_map_entry_t));
184 
185 /*
186  * local inlines
187  */
188 
189 /*
190  * uvm_mapent_alloc: allocate a map entry
191  *
192  * => XXX: static pool for kernel map?
193  */
194 
195 static __inline vm_map_entry_t
196 uvm_mapent_alloc(map)
197 	vm_map_t map;
198 {
199 	vm_map_entry_t me;
200 	int s;
201 	UVMHIST_FUNC("uvm_mapent_alloc");
202 	UVMHIST_CALLED(maphist);
203 
204 	if ((map->flags & VM_MAP_INTRSAFE) == 0 &&
205 	    map != kernel_map && kernel_map != NULL /* XXX */) {
206 		me = pool_get(&uvm_map_entry_pool, PR_WAITOK);
207 		me->flags = 0;
208 		/* me can't be null, wait ok */
209 	} else {
210 		s = splimp();	/* protect kentry_free list with splimp */
211 		simple_lock(&uvm.kentry_lock);
212 		me = uvm.kentry_free;
213 		if (me) uvm.kentry_free = me->next;
214 		simple_unlock(&uvm.kentry_lock);
215 		splx(s);
216 		if (!me)
217 	panic("mapent_alloc: out of static map entries, check MAX_KMAPENT");
218 		me->flags = UVM_MAP_STATIC;
219 	}
220 
221 	UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]",
222 		me, ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map)
223 		? TRUE : FALSE, 0, 0);
224 	return(me);
225 }
226 
227 /*
228  * uvm_mapent_free: free map entry
229  *
230  * => XXX: static pool for kernel map?
231  */
232 
233 static __inline void
234 uvm_mapent_free(me)
235 	vm_map_entry_t me;
236 {
237 	int s;
238 	UVMHIST_FUNC("uvm_mapent_free");
239 	UVMHIST_CALLED(maphist);
240 	UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
241 		me, me->flags, 0, 0);
242 	if ((me->flags & UVM_MAP_STATIC) == 0) {
243 		pool_put(&uvm_map_entry_pool, me);
244 	} else {
245 		s = splimp();	/* protect kentry_free list with splimp */
246 		simple_lock(&uvm.kentry_lock);
247 		me->next = uvm.kentry_free;
248 		uvm.kentry_free = me;
249 		simple_unlock(&uvm.kentry_lock);
250 		splx(s);
251 	}
252 }
253 
254 /*
255  * uvm_mapent_copy: copy a map entry, preserving flags
256  */
257 
258 static __inline void
259 uvm_mapent_copy(src, dst)
260 	vm_map_entry_t src;
261 	vm_map_entry_t dst;
262 {
263 
264 	memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - ((char*)src));
265 }
266 
267 /*
268  * uvm_map_entry_unwire: unwire a map entry
269  *
270  * => map should be locked by caller
271  */
272 
273 static __inline void
274 uvm_map_entry_unwire(map, entry)
275 	vm_map_t map;
276 	vm_map_entry_t entry;
277 {
278 
279 	entry->wired_count = 0;
280 	uvm_fault_unwire_locked(map, entry->start, entry->end);
281 }
282 
283 /*
284  * uvm_map_init: init mapping system at boot time.   note that we allocate
285  * and init the static pool of vm_map_entry_t's for the kernel here.
286  */
287 
288 void
289 uvm_map_init()
290 {
291 	static struct vm_map_entry kernel_map_entry[MAX_KMAPENT];
292 #if defined(UVMHIST)
293 	static struct uvm_history_ent maphistbuf[100];
294 	static struct uvm_history_ent pdhistbuf[100];
295 #endif
296 	int lcv;
297 
298 	/*
299 	 * first, init logging system.
300 	 */
301 
302 	UVMHIST_FUNC("uvm_map_init");
303 	UVMHIST_INIT_STATIC(maphist, maphistbuf);
304 	UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
305 	UVMHIST_CALLED(maphist);
306 	UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
307 	UVMCNT_INIT(uvm_map_call,  UVMCNT_CNT, 0,
308 	    "# uvm_map() successful calls", 0);
309 	UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0);
310 	UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward",
311 	    0);
312 	UVMCNT_INIT(uvm_mlk_call,  UVMCNT_CNT, 0, "# map lookup calls", 0);
313 	UVMCNT_INIT(uvm_mlk_hint,  UVMCNT_CNT, 0, "# map lookup hint hits", 0);
314 
315 	/*
316 	 * now set up static pool of kernel map entrys ...
317 	 */
318 
319 	simple_lock_init(&uvm.kentry_lock);
320 	uvm.kentry_free = NULL;
321 	for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) {
322 		kernel_map_entry[lcv].next = uvm.kentry_free;
323 		uvm.kentry_free = &kernel_map_entry[lcv];
324 	}
325 
326 	/*
327 	 * initialize the map-related pools.
328 	 */
329 	pool_init(&uvm_vmspace_pool, sizeof(struct vmspace),
330 	    0, 0, 0, "vmsppl", 0,
331 	    pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
332 	pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry),
333 	    0, 0, 0, "vmmpepl", 0,
334 	    pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
335 }
336 
337 /*
338  * clippers
339  */
340 
341 /*
342  * uvm_map_clip_start: ensure that the entry begins at or after
343  *	the starting address, if it doesn't we split the entry.
344  *
345  * => caller should use UVM_MAP_CLIP_START macro rather than calling
346  *    this directly
347  * => map must be locked by caller
348  */
349 
350 void uvm_map_clip_start(map, entry, start)
351 	vm_map_t       map;
352 	vm_map_entry_t entry;
353 	vaddr_t    start;
354 {
355 	vm_map_entry_t new_entry;
356 	vaddr_t new_adj;
357 
358 	/* uvm_map_simplify_entry(map, entry); */ /* XXX */
359 
360 	/*
361 	 * Split off the front portion.  note that we must insert the new
362 	 * entry BEFORE this one, so that this entry has the specified
363 	 * starting address.
364 	 */
365 
366 	new_entry = uvm_mapent_alloc(map);
367 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
368 
369 	new_entry->end = start;
370 	new_adj = start - new_entry->start;
371 	if (entry->object.uvm_obj)
372 		entry->offset += new_adj;	/* shift start over */
373 	entry->start = start;
374 
375 	if (new_entry->aref.ar_amap) {
376 		amap_splitref(&new_entry->aref, &entry->aref, new_adj);
377 	}
378 
379 	uvm_map_entry_link(map, entry->prev, new_entry);
380 
381 	if (UVM_ET_ISSUBMAP(entry)) {
382 		/* ... unlikely to happen, but play it safe */
383 		 uvm_map_reference(new_entry->object.sub_map);
384 	} else {
385 		if (UVM_ET_ISOBJ(entry) &&
386 		    entry->object.uvm_obj->pgops &&
387 		    entry->object.uvm_obj->pgops->pgo_reference)
388 			entry->object.uvm_obj->pgops->pgo_reference(
389 			    entry->object.uvm_obj);
390 	}
391 }
392 
393 /*
394  * uvm_map_clip_end: ensure that the entry ends at or before
395  *	the ending address, if it does't we split the reference
396  *
397  * => caller should use UVM_MAP_CLIP_END macro rather than calling
398  *    this directly
399  * => map must be locked by caller
400  */
401 
402 void
403 uvm_map_clip_end(map, entry, end)
404 	vm_map_t	map;
405 	vm_map_entry_t	entry;
406 	vaddr_t	end;
407 {
408 	vm_map_entry_t	new_entry;
409 	vaddr_t new_adj; /* #bytes we move start forward */
410 
411 	/*
412 	 *	Create a new entry and insert it
413 	 *	AFTER the specified entry
414 	 */
415 
416 	new_entry = uvm_mapent_alloc(map);
417 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
418 
419 	new_entry->start = entry->end = end;
420 	new_adj = end - entry->start;
421 	if (new_entry->object.uvm_obj)
422 		new_entry->offset += new_adj;
423 
424 	if (entry->aref.ar_amap)
425 		amap_splitref(&entry->aref, &new_entry->aref, new_adj);
426 
427 	uvm_map_entry_link(map, entry, new_entry);
428 
429 	if (UVM_ET_ISSUBMAP(entry)) {
430 		/* ... unlikely to happen, but play it safe */
431 	 	uvm_map_reference(new_entry->object.sub_map);
432 	} else {
433 		if (UVM_ET_ISOBJ(entry) &&
434 		    entry->object.uvm_obj->pgops &&
435 		    entry->object.uvm_obj->pgops->pgo_reference)
436 			entry->object.uvm_obj->pgops->pgo_reference(
437 			    entry->object.uvm_obj);
438 	}
439 }
440 
441 
442 /*
443  *   M A P   -   m a i n   e n t r y   p o i n t
444  */
445 /*
446  * uvm_map: establish a valid mapping in a map
447  *
448  * => assume startp is page aligned.
449  * => assume size is a multiple of PAGE_SIZE.
450  * => assume sys_mmap provides enough of a "hint" to have us skip
451  *	over text/data/bss area.
452  * => map must be unlocked (we will lock it)
453  * => <uobj,uoffset> value meanings (4 cases):
454  *	 [1] <NULL,uoffset> 		== uoffset is a hint for PMAP_PREFER
455  *	 [2] <NULL,UVM_UNKNOWN_OFFSET>	== don't PMAP_PREFER
456  *	 [3] <uobj,uoffset>		== normal mapping
457  *	 [4] <uobj,UVM_UNKNOWN_OFFSET>	== uvm_map finds offset based on VA
458  *
459  *    case [4] is for kernel mappings where we don't know the offset until
460  *    we've found a virtual address.   note that kernel object offsets are
461  *    always relative to vm_map_min(kernel_map).
462  * => XXXCDC: need way to map in external amap?
463  */
464 
465 int
466 uvm_map(map, startp, size, uobj, uoffset, flags)
467 	vm_map_t map;
468 	vaddr_t *startp;	/* IN/OUT */
469 	vsize_t size;
470 	struct uvm_object *uobj;
471 	voff_t uoffset;
472 	uvm_flag_t flags;
473 {
474 	vm_map_entry_t prev_entry, new_entry;
475 	vm_prot_t prot = UVM_PROTECTION(flags), maxprot =
476 	    UVM_MAXPROTECTION(flags);
477 	vm_inherit_t inherit = UVM_INHERIT(flags);
478 	int advice = UVM_ADVICE(flags);
479 	UVMHIST_FUNC("uvm_map");
480 	UVMHIST_CALLED(maphist);
481 
482 	UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)",
483 	    map, *startp, size, flags);
484 	UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
485 
486 	/*
487 	 * step 0: sanity check of protection code
488 	 */
489 
490 	if ((prot & maxprot) != prot) {
491 		UVMHIST_LOG(maphist, "<- prot. failure:  prot=0x%x, max=0x%x",
492 		prot, maxprot,0,0);
493 		return(KERN_PROTECTION_FAILURE);
494 	}
495 
496 	/*
497 	 * step 1: figure out where to put new VM range
498 	 */
499 
500 	if (vm_map_lock_try(map) == FALSE) {
501 		if (flags & UVM_FLAG_TRYLOCK)
502 			return(KERN_FAILURE);
503 		vm_map_lock(map); /* could sleep here */
504 	}
505 	if ((prev_entry = uvm_map_findspace(map, *startp, size, startp,
506 	    uobj, uoffset, flags & UVM_FLAG_FIXED)) == NULL) {
507 		UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0);
508 		vm_map_unlock(map);
509 		return (KERN_NO_SPACE);
510 	}
511 
512 #ifdef PMAP_GROWKERNEL
513 	{
514 		/*
515 		 * If the kernel pmap can't map the requested space,
516 		 * then allocate more resources for it.
517 		 */
518 		if (map == kernel_map && uvm_maxkaddr < (*startp + size))
519 			uvm_maxkaddr = pmap_growkernel(*startp + size);
520 	}
521 #endif
522 
523 	UVMCNT_INCR(uvm_map_call);
524 
525 	/*
526 	 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
527 	 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET.   in
528 	 * either case we want to zero it  before storing it in the map entry
529 	 * (because it looks strange and confusing when debugging...)
530 	 *
531 	 * if uobj is not null
532 	 *   if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
533 	 *      and we do not need to change uoffset.
534 	 *   if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
535 	 *      now (based on the starting address of the map).   this case is
536 	 *      for kernel object mappings where we don't know the offset until
537 	 *      the virtual address is found (with uvm_map_findspace).   the
538 	 *      offset is the distance we are from the start of the map.
539 	 */
540 
541 	if (uobj == NULL) {
542 		uoffset = 0;
543 	} else {
544 		if (uoffset == UVM_UNKNOWN_OFFSET) {
545 #ifdef DIAGNOSTIC
546 			if (UVM_OBJ_IS_KERN_OBJECT(uobj) == 0)
547 				panic("uvm_map: unknown offset with "
548 				    "non-kernel object");
549 #endif
550 			uoffset = *startp - vm_map_min(kernel_map);
551 		}
552 	}
553 
554 	/*
555 	 * step 2: try and insert in map by extending previous entry, if
556 	 * possible
557 	 * XXX: we don't try and pull back the next entry.   might be useful
558 	 * for a stack, but we are currently allocating our stack in advance.
559 	 */
560 
561 	if ((flags & UVM_FLAG_NOMERGE) == 0 &&
562 	    prev_entry->end == *startp && prev_entry != &map->header &&
563 	    prev_entry->object.uvm_obj == uobj) {
564 
565 		if (uobj && prev_entry->offset +
566 		    (prev_entry->end - prev_entry->start) != uoffset)
567 			goto step3;
568 
569 		if (UVM_ET_ISSUBMAP(prev_entry))
570 			goto step3;
571 
572 		if (prev_entry->protection != prot ||
573 		    prev_entry->max_protection != maxprot)
574 			goto step3;
575 
576 		if (prev_entry->inheritance != inherit ||
577 		    prev_entry->advice != advice)
578 			goto step3;
579 
580 		/* wiring status must match (new area is unwired) */
581 		if (VM_MAPENT_ISWIRED(prev_entry))
582 			goto step3;
583 
584 		/*
585 		 * can't extend a shared amap.  note: no need to lock amap to
586 		 * look at refs since we don't care about its exact value.
587 		 * if it is one (i.e. we have only reference) it will stay there
588 		 */
589 
590 		if (prev_entry->aref.ar_amap &&
591 		    amap_refs(prev_entry->aref.ar_amap) != 1) {
592 			goto step3;
593 		}
594 
595 		/* got it! */
596 
597 		UVMCNT_INCR(map_backmerge);
598 		UVMHIST_LOG(maphist,"  starting back merge", 0, 0, 0, 0);
599 
600 		/*
601 		 * drop our reference to uobj since we are extending a reference
602 		 * that we already have (the ref count can not drop to zero).
603 		 */
604 		if (uobj && uobj->pgops->pgo_detach)
605 			uobj->pgops->pgo_detach(uobj);
606 
607 		if (prev_entry->aref.ar_amap) {
608 			amap_extend(prev_entry, size);
609 		}
610 
611 		prev_entry->end += size;
612 		map->size += size;
613 
614 		UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
615 		vm_map_unlock(map);
616 		return (KERN_SUCCESS);
617 
618 	}
619 step3:
620 	UVMHIST_LOG(maphist,"  allocating new map entry", 0, 0, 0, 0);
621 
622 	/*
623 	 * check for possible forward merge (which we don't do) and count
624 	 * the number of times we missed a *possible* chance to merge more
625 	 */
626 
627 	if ((flags & UVM_FLAG_NOMERGE) == 0 &&
628 	    prev_entry->next != &map->header &&
629 	    prev_entry->next->start == (*startp + size))
630 		UVMCNT_INCR(map_forwmerge);
631 
632 	/*
633 	 * step 3: allocate new entry and link it in
634 	 */
635 
636 	new_entry = uvm_mapent_alloc(map);
637 	new_entry->start = *startp;
638 	new_entry->end = new_entry->start + size;
639 	new_entry->object.uvm_obj = uobj;
640 	new_entry->offset = uoffset;
641 
642 	if (uobj)
643 		new_entry->etype = UVM_ET_OBJ;
644 	else
645 		new_entry->etype = 0;
646 
647 	if (flags & UVM_FLAG_COPYONW) {
648 		new_entry->etype |= UVM_ET_COPYONWRITE;
649 		if ((flags & UVM_FLAG_OVERLAY) == 0)
650 			new_entry->etype |= UVM_ET_NEEDSCOPY;
651 	}
652 
653 	new_entry->protection = prot;
654 	new_entry->max_protection = maxprot;
655 	new_entry->inheritance = inherit;
656 	new_entry->wired_count = 0;
657 	new_entry->advice = advice;
658 	if (flags & UVM_FLAG_OVERLAY) {
659 		/*
660 		 * to_add: for BSS we overallocate a little since we
661 		 * are likely to extend
662 		 */
663 		vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
664 			UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
665 		struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK);
666 		new_entry->aref.ar_pageoff = 0;
667 		new_entry->aref.ar_amap = amap;
668 	} else {
669 		new_entry->aref.ar_pageoff = 0;
670 		new_entry->aref.ar_amap = NULL;
671 	}
672 
673 	uvm_map_entry_link(map, prev_entry, new_entry);
674 
675 	map->size += size;
676 
677 	/*
678 	 *      Update the free space hint
679 	 */
680 
681 	if ((map->first_free == prev_entry) &&
682 	    (prev_entry->end >= new_entry->start))
683 		map->first_free = new_entry;
684 
685 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
686 	vm_map_unlock(map);
687 	return(KERN_SUCCESS);
688 }
689 
690 /*
691  * uvm_map_lookup_entry: find map entry at or before an address
692  *
693  * => map must at least be read-locked by caller
694  * => entry is returned in "entry"
695  * => return value is true if address is in the returned entry
696  */
697 
698 boolean_t
699 uvm_map_lookup_entry(map, address, entry)
700 	vm_map_t	map;
701 	vaddr_t	address;
702 	vm_map_entry_t		*entry;		/* OUT */
703 {
704 	vm_map_entry_t		cur;
705 	vm_map_entry_t		last;
706 	UVMHIST_FUNC("uvm_map_lookup_entry");
707 	UVMHIST_CALLED(maphist);
708 
709 	UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
710 	    map, address, entry, 0);
711 
712 	/*
713 	 * start looking either from the head of the
714 	 * list, or from the hint.
715 	 */
716 
717 	simple_lock(&map->hint_lock);
718 	cur = map->hint;
719 	simple_unlock(&map->hint_lock);
720 
721 	if (cur == &map->header)
722 		cur = cur->next;
723 
724 	UVMCNT_INCR(uvm_mlk_call);
725 	if (address >= cur->start) {
726 	    	/*
727 		 * go from hint to end of list.
728 		 *
729 		 * but first, make a quick check to see if
730 		 * we are already looking at the entry we
731 		 * want (which is usually the case).
732 		 * note also that we don't need to save the hint
733 		 * here... it is the same hint (unless we are
734 		 * at the header, in which case the hint didn't
735 		 * buy us anything anyway).
736 		 */
737 		last = &map->header;
738 		if ((cur != last) && (cur->end > address)) {
739 			UVMCNT_INCR(uvm_mlk_hint);
740 			*entry = cur;
741 			UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
742 			    cur, 0, 0, 0);
743 			return (TRUE);
744 		}
745 	} else {
746 	    	/*
747 		 * go from start to hint, *inclusively*
748 		 */
749 		last = cur->next;
750 		cur = map->header.next;
751 	}
752 
753 	/*
754 	 * search linearly
755 	 */
756 
757 	while (cur != last) {
758 		if (cur->end > address) {
759 			if (address >= cur->start) {
760 			    	/*
761 				 * save this lookup for future
762 				 * hints, and return
763 				 */
764 
765 				*entry = cur;
766 				SAVE_HINT(map, cur);
767 				UVMHIST_LOG(maphist,"<- search got it (0x%x)",
768 					cur, 0, 0, 0);
769 				return (TRUE);
770 			}
771 			break;
772 		}
773 		cur = cur->next;
774 	}
775 	*entry = cur->prev;
776 	SAVE_HINT(map, *entry);
777 	UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
778 	return (FALSE);
779 }
780 
781 
782 /*
783  * uvm_map_findspace: find "length" sized space in "map".
784  *
785  * => "hint" is a hint about where we want it, unless fixed is true
786  *	(in which case we insist on using "hint").
787  * => "result" is VA returned
788  * => uobj/uoffset are to be used to handle VAC alignment, if required
789  * => caller must at least have read-locked map
790  * => returns NULL on failure, or pointer to prev. map entry if success
791  * => note this is a cross between the old vm_map_findspace and vm_map_find
792  */
793 
794 vm_map_entry_t
795 uvm_map_findspace(map, hint, length, result, uobj, uoffset, fixed)
796 	vm_map_t map;
797 	vaddr_t hint;
798 	vsize_t length;
799 	vaddr_t *result; /* OUT */
800 	struct uvm_object *uobj;
801 	voff_t uoffset;
802 	boolean_t fixed;
803 {
804 	vm_map_entry_t entry, next, tmp;
805 	vaddr_t end;
806 	UVMHIST_FUNC("uvm_map_findspace");
807 	UVMHIST_CALLED(maphist);
808 
809 	UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, fixed=%d)",
810 		map, hint, length, fixed);
811 
812 	if (hint < map->min_offset) {	/* check ranges ... */
813 		if (fixed) {
814 			UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
815 			return(NULL);
816 		}
817 		hint = map->min_offset;
818 	}
819 	if (hint > map->max_offset) {
820 		UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
821 				hint, map->min_offset, map->max_offset, 0);
822 		return(NULL);
823 	}
824 
825 	/*
826 	 * Look for the first possible address; if there's already
827 	 * something at this address, we have to start after it.
828 	 */
829 
830 	if (!fixed && hint == map->min_offset) {
831 		if ((entry = map->first_free) != &map->header)
832 			hint = entry->end;
833 	} else {
834 		if (uvm_map_lookup_entry(map, hint, &tmp)) {
835 			/* "hint" address already in use ... */
836 			if (fixed) {
837 				UVMHIST_LOG(maphist,"<- fixed & VA in use",
838 				    0, 0, 0, 0);
839 				return(NULL);
840 			}
841 			hint = tmp->end;
842 		}
843 		entry = tmp;
844 	}
845 
846 	/*
847 	 * Look through the rest of the map, trying to fit a new region in
848 	 * the gap between existing regions, or after the very last region.
849 	 * note: entry->end   = base VA of current gap,
850 	 *	 next->start  = VA of end of current gap
851 	 */
852 	for (;; hint = (entry = next)->end) {
853 		/*
854 		 * Find the end of the proposed new region.  Be sure we didn't
855 		 * go beyond the end of the map, or wrap around the address;
856 		 * if so, we lose.  Otherwise, if this is the last entry, or
857 		 * if the proposed new region fits before the next entry, we
858 		 * win.
859 		 */
860 
861 #ifdef PMAP_PREFER
862 		/*
863 		 * push hint forward as needed to avoid VAC alias problems.
864 		 * we only do this if a valid offset is specified.
865 		 */
866 		if (!fixed && uoffset != UVM_UNKNOWN_OFFSET)
867 		  PMAP_PREFER(uoffset, &hint);
868 #endif
869 		end = hint + length;
870 		if (end > map->max_offset || end < hint) {
871 			UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0);
872 			return (NULL);
873 		}
874 		next = entry->next;
875 		if (next == &map->header || next->start >= end)
876 			break;
877 		if (fixed) {
878 			UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0);
879 			return(NULL); /* only one shot at it ... */
880 		}
881 	}
882 	SAVE_HINT(map, entry);
883 	*result = hint;
884 	UVMHIST_LOG(maphist,"<- got it!  (result=0x%x)", hint, 0,0,0);
885 	return (entry);
886 }
887 
888 /*
889  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
890  */
891 
892 /*
893  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
894  *
895  * => caller must check alignment and size
896  * => map must be locked by caller
897  * => we return a list of map entries that we've remove from the map
898  *    in "entry_list"
899  */
900 
901 int
902 uvm_unmap_remove(map, start, end, entry_list)
903 	vm_map_t map;
904 	vaddr_t start,end;
905 	vm_map_entry_t *entry_list;	/* OUT */
906 {
907 	vm_map_entry_t entry, first_entry, next;
908 	vaddr_t len;
909 	UVMHIST_FUNC("uvm_unmap_remove");
910 	UVMHIST_CALLED(maphist);
911 
912 	UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
913 	    map, start, end, 0);
914 
915 	VM_MAP_RANGE_CHECK(map, start, end);
916 
917 	/*
918 	 * find first entry
919 	 */
920 	if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
921 		/* clip and go... */
922 		entry = first_entry;
923 		UVM_MAP_CLIP_START(map, entry, start);
924 		/* critical!  prevents stale hint */
925 		SAVE_HINT(map, entry->prev);
926 
927 	} else {
928 		entry = first_entry->next;
929 	}
930 
931 	/*
932 	 * Save the free space hint
933 	 */
934 
935 	if (map->first_free->start >= start)
936 		map->first_free = entry->prev;
937 
938 	/*
939 	 * note: we now re-use first_entry for a different task.  we remove
940 	 * a number of map entries from the map and save them in a linked
941 	 * list headed by "first_entry".  once we remove them from the map
942 	 * the caller should unlock the map and drop the references to the
943 	 * backing objects [c.f. uvm_unmap_detach].  the object is to
944 	 * separate unmapping from reference dropping.  why?
945 	 *   [1] the map has to be locked for unmapping
946 	 *   [2] the map need not be locked for reference dropping
947 	 *   [3] dropping references may trigger pager I/O, and if we hit
948 	 *       a pager that does synchronous I/O we may have to wait for it.
949 	 *   [4] we would like all waiting for I/O to occur with maps unlocked
950 	 *       so that we don't block other threads.
951 	 */
952 	first_entry = NULL;
953 	*entry_list = NULL;		/* to be safe */
954 
955 	/*
956 	 * break up the area into map entry sized regions and unmap.  note
957 	 * that all mappings have to be removed before we can even consider
958 	 * dropping references to amaps or VM objects (otherwise we could end
959 	 * up with a mapping to a page on the free list which would be very bad)
960 	 */
961 
962 	while ((entry != &map->header) && (entry->start < end)) {
963 
964 		UVM_MAP_CLIP_END(map, entry, end);
965 		next = entry->next;
966 		len = entry->end - entry->start;
967 
968 		/*
969 		 * unwire before removing addresses from the pmap; otherwise
970 		 * unwiring will put the entries back into the pmap (XXX).
971 		 */
972 
973 		if (VM_MAPENT_ISWIRED(entry))
974 			uvm_map_entry_unwire(map, entry);
975 
976 		/*
977 		 * special case: handle mappings to anonymous kernel objects.
978 		 * we want to free these pages right away...
979 		 */
980 		if (UVM_ET_ISOBJ(entry) &&
981 		    UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
982 #ifdef DIAGNOSTIC
983 			if (vm_map_pmap(map) != pmap_kernel())
984 				panic("uvm_unmap_remove: kernel object "
985 				    "mapped by non-kernel map");
986 #endif
987 
988 			/*
989 			 * note: kernel object mappings are currently used in
990 			 * two ways:
991 			 *  [1] "normal" mappings of pages in the kernel object
992 			 *  [2] uvm_km_valloc'd allocations in which we
993 			 *      pmap_enter in some non-kernel-object page
994 			 *      (e.g. vmapbuf).
995 			 *
996 			 * for case [1], we need to remove the mapping from
997 			 * the pmap and then remove the page from the kernel
998 			 * object (because, once pages in a kernel object are
999 			 * unmapped they are no longer needed, unlike, say,
1000 			 * a vnode where you might want the data to persist
1001 			 * until flushed out of a queue).
1002 			 *
1003 			 * for case [2], we need to remove the mapping from
1004 			 * the pmap.  there shouldn't be any pages at the
1005 			 * specified offset in the kernel object [but it
1006 			 * doesn't hurt to call uvm_km_pgremove just to be
1007 			 * safe?]
1008 			 *
1009 			 * uvm_km_pgremove currently does the following:
1010 			 *   for pages in the kernel object in range:
1011 			 *     - drops the swap slot
1012 			 *     - uvm_pagefree the page
1013 			 *
1014 			 * note there is version of uvm_km_pgremove() that
1015 			 * is used for "intrsafe" objects.
1016 			 */
1017 
1018 			/*
1019 			 * remove mappings from pmap and drop the pages
1020 			 * from the object.  offsets are always relative
1021 			 * to vm_map_min(kernel_map).
1022 			 */
1023 			if (UVM_OBJ_IS_INTRSAFE_OBJECT(entry->object.uvm_obj)) {
1024 				pmap_kremove(entry->start, len);
1025 				uvm_km_pgremove_intrsafe(entry->object.uvm_obj,
1026 				    entry->start - vm_map_min(kernel_map),
1027 				    entry->end - vm_map_min(kernel_map));
1028 			} else {
1029 				pmap_remove(pmap_kernel(), entry->start,
1030 				    entry->start + len);
1031 				uvm_km_pgremove(entry->object.uvm_obj,
1032 				    entry->start - vm_map_min(kernel_map),
1033 				    entry->end - vm_map_min(kernel_map));
1034 			}
1035 
1036 			/*
1037 			 * null out kernel_object reference, we've just
1038 			 * dropped it
1039 			 */
1040 			entry->etype &= ~UVM_ET_OBJ;
1041 			entry->object.uvm_obj = NULL;	/* to be safe */
1042 
1043 		} else {
1044 			/*
1045 		 	 * remove mappings the standard way.
1046 		 	 */
1047 			pmap_remove(map->pmap, entry->start, entry->end);
1048 		}
1049 
1050 		/*
1051 		 * remove entry from map and put it on our list of entries
1052 		 * that we've nuked.  then go do next entry.
1053 		 */
1054 		UVMHIST_LOG(maphist, "  removed map entry 0x%x", entry, 0, 0,0);
1055 
1056 		/* critical! prevents stale hint */
1057 		/* XXX: need SAVE_HINT with three parms */
1058 		simple_lock(&map->hint_lock);
1059 		if (map->hint == entry)
1060 		    map->hint = entry->prev;
1061 		simple_unlock(&map->hint_lock);
1062 
1063 		uvm_map_entry_unlink(map, entry);
1064 		map->size -= len;
1065 		entry->next = first_entry;
1066 		first_entry = entry;
1067 		entry = next;		/* next entry, please */
1068 	}
1069 
1070 	/*
1071 	 * now we've cleaned up the map and are ready for the caller to drop
1072 	 * references to the mapped objects.
1073 	 */
1074 
1075 	*entry_list = first_entry;
1076 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1077 	return(KERN_SUCCESS);
1078 }
1079 
1080 /*
1081  * uvm_unmap_detach: drop references in a chain of map entries
1082  *
1083  * => we will free the map entries as we traverse the list.
1084  */
1085 
1086 void
1087 uvm_unmap_detach(first_entry, amap_unref_flags)
1088 	vm_map_entry_t first_entry;
1089 	int amap_unref_flags;
1090 {
1091 	vm_map_entry_t next_entry;
1092 	UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1093 
1094 	while (first_entry) {
1095 
1096 #ifdef DIAGNOSTIC
1097 		/*
1098 		 * sanity check
1099 		 */
1100 		/* was part of vm_map_entry_delete() */
1101 		if (VM_MAPENT_ISWIRED(first_entry))
1102 			panic("unmap: still wired!");
1103 #endif
1104 
1105 		UVMHIST_LOG(maphist,
1106 		    "  detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
1107 		    first_entry, first_entry->aref.ar_amap,
1108 		    first_entry->object.uvm_obj,
1109 		    UVM_ET_ISSUBMAP(first_entry));
1110 
1111 		/*
1112 		 * drop reference to amap, if we've got one
1113 		 */
1114 
1115 		if (first_entry->aref.ar_amap)
1116 			amap_unref(first_entry, amap_unref_flags);
1117 
1118 		/*
1119 		 * drop reference to our backing object, if we've got one
1120 		 */
1121 
1122 		if (UVM_ET_ISSUBMAP(first_entry)) {
1123 			/* ... unlikely to happen, but play it safe */
1124 			uvm_map_deallocate(first_entry->object.sub_map);
1125 		} else {
1126 			if (UVM_ET_ISOBJ(first_entry) &&
1127 			    first_entry->object.uvm_obj->pgops->pgo_detach)
1128 				first_entry->object.uvm_obj->pgops->
1129 				    pgo_detach(first_entry->object.uvm_obj);
1130 		}
1131 
1132 		/*
1133 		 * next entry
1134 		 */
1135 		next_entry = first_entry->next;
1136 		uvm_mapent_free(first_entry);
1137 		first_entry = next_entry;
1138 	}
1139 
1140 	/*
1141 	 * done!
1142 	 */
1143 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1144 	return;
1145 }
1146 
1147 /*
1148  *   E X T R A C T I O N   F U N C T I O N S
1149  */
1150 
1151 /*
1152  * uvm_map_reserve: reserve space in a vm_map for future use.
1153  *
1154  * => we reserve space in a map by putting a dummy map entry in the
1155  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
1156  * => map should be unlocked (we will write lock it)
1157  * => we return true if we were able to reserve space
1158  * => XXXCDC: should be inline?
1159  */
1160 
1161 int
1162 uvm_map_reserve(map, size, offset, raddr)
1163 	vm_map_t map;
1164 	vsize_t size;
1165 	vaddr_t offset;    /* hint for pmap_prefer */
1166 	vaddr_t *raddr;	/* IN:hint, OUT: reserved VA */
1167 {
1168 	UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1169 
1170 	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1171 	      map,size,offset,raddr);
1172 
1173 	size = round_page(size);
1174 	if (*raddr < vm_map_min(map))
1175 		*raddr = vm_map_min(map);                /* hint */
1176 
1177 	/*
1178 	 * reserve some virtual space.
1179 	 */
1180 
1181 	if (uvm_map(map, raddr, size, NULL, offset,
1182 	    UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1183 	    UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != KERN_SUCCESS) {
1184 	    UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1185 		return (FALSE);
1186 	}
1187 
1188 	UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1189 	return (TRUE);
1190 }
1191 
1192 /*
1193  * uvm_map_replace: replace a reserved (blank) area of memory with
1194  * real mappings.
1195  *
1196  * => caller must WRITE-LOCK the map
1197  * => we return TRUE if replacement was a success
1198  * => we expect the newents chain to have nnewents entrys on it and
1199  *    we expect newents->prev to point to the last entry on the list
1200  * => note newents is allowed to be NULL
1201  */
1202 
1203 int
1204 uvm_map_replace(map, start, end, newents, nnewents)
1205 	struct vm_map *map;
1206 	vaddr_t start, end;
1207 	vm_map_entry_t newents;
1208 	int nnewents;
1209 {
1210 	vm_map_entry_t oldent, last;
1211 	UVMHIST_FUNC("uvm_map_replace");
1212 	UVMHIST_CALLED(maphist);
1213 
1214 	/*
1215 	 * first find the blank map entry at the specified address
1216 	 */
1217 
1218 	if (!uvm_map_lookup_entry(map, start, &oldent)) {
1219 		return(FALSE);
1220 	}
1221 
1222 	/*
1223 	 * check to make sure we have a proper blank entry
1224 	 */
1225 
1226 	if (oldent->start != start || oldent->end != end ||
1227 	    oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1228 		return (FALSE);
1229 	}
1230 
1231 #ifdef DIAGNOSTIC
1232 	/*
1233 	 * sanity check the newents chain
1234 	 */
1235 	{
1236 		vm_map_entry_t tmpent = newents;
1237 		int nent = 0;
1238 		vaddr_t cur = start;
1239 
1240 		while (tmpent) {
1241 			nent++;
1242 			if (tmpent->start < cur)
1243 				panic("uvm_map_replace1");
1244 			if (tmpent->start > tmpent->end || tmpent->end > end) {
1245 		printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
1246 			    tmpent->start, tmpent->end, end);
1247 				panic("uvm_map_replace2");
1248 			}
1249 			cur = tmpent->end;
1250 			if (tmpent->next) {
1251 				if (tmpent->next->prev != tmpent)
1252 					panic("uvm_map_replace3");
1253 			} else {
1254 				if (newents->prev != tmpent)
1255 					panic("uvm_map_replace4");
1256 			}
1257 			tmpent = tmpent->next;
1258 		}
1259 		if (nent != nnewents)
1260 			panic("uvm_map_replace5");
1261 	}
1262 #endif
1263 
1264 	/*
1265 	 * map entry is a valid blank!   replace it.   (this does all the
1266 	 * work of map entry link/unlink...).
1267 	 */
1268 
1269 	if (newents) {
1270 
1271 		last = newents->prev;		/* we expect this */
1272 
1273 		/* critical: flush stale hints out of map */
1274 		SAVE_HINT(map, newents);
1275 		if (map->first_free == oldent)
1276 			map->first_free = last;
1277 
1278 		last->next = oldent->next;
1279 		last->next->prev = last;
1280 		newents->prev = oldent->prev;
1281 		newents->prev->next = newents;
1282 		map->nentries = map->nentries + (nnewents - 1);
1283 
1284 	} else {
1285 
1286 		/* critical: flush stale hints out of map */
1287 		SAVE_HINT(map, oldent->prev);
1288 		if (map->first_free == oldent)
1289 			map->first_free = oldent->prev;
1290 
1291 		/* NULL list of new entries: just remove the old one */
1292 		uvm_map_entry_unlink(map, oldent);
1293 	}
1294 
1295 
1296 	/*
1297 	 * now we can free the old blank entry, unlock the map and return.
1298 	 */
1299 
1300 	uvm_mapent_free(oldent);
1301 	return(TRUE);
1302 }
1303 
1304 /*
1305  * uvm_map_extract: extract a mapping from a map and put it somewhere
1306  *	(maybe removing the old mapping)
1307  *
1308  * => maps should be unlocked (we will write lock them)
1309  * => returns 0 on success, error code otherwise
1310  * => start must be page aligned
1311  * => len must be page sized
1312  * => flags:
1313  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
1314  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
1315  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
1316  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
1317  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
1318  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
1319  *             be used from within the kernel in a kernel level map <<<
1320  */
1321 
1322 int
1323 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags)
1324 	vm_map_t srcmap, dstmap;
1325 	vaddr_t start, *dstaddrp;
1326 	vsize_t len;
1327 	int flags;
1328 {
1329 	vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge,
1330 	    oldstart;
1331 	vm_map_entry_t chain, endchain, entry, orig_entry, newentry, deadentry;
1332 	vm_map_entry_t oldentry;
1333 	vsize_t elen;
1334 	int nchain, error, copy_ok;
1335 	UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1336 	UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
1337 	    len,0);
1338 	UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
1339 
1340 #ifdef DIAGNOSTIC
1341 	/*
1342 	 * step 0: sanity check: start must be on a page boundary, length
1343 	 * must be page sized.  can't ask for CONTIG/QREF if you asked for
1344 	 * REMOVE.
1345 	 */
1346 	if ((start & PAGE_MASK) || (len & PAGE_MASK))
1347 		panic("uvm_map_extract1");
1348 	if (flags & UVM_EXTRACT_REMOVE)
1349 		if (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF))
1350 			panic("uvm_map_extract2");
1351 #endif
1352 
1353 
1354 	/*
1355 	 * step 1: reserve space in the target map for the extracted area
1356 	 */
1357 
1358 	dstaddr = vm_map_min(dstmap);
1359 	if (uvm_map_reserve(dstmap, len, start, &dstaddr) == FALSE)
1360 		return(ENOMEM);
1361 	*dstaddrp = dstaddr;	/* pass address back to caller */
1362 	UVMHIST_LOG(maphist, "  dstaddr=0x%x", dstaddr,0,0,0);
1363 
1364 
1365 	/*
1366 	 * step 2: setup for the extraction process loop by init'ing the
1367 	 * map entry chain, locking src map, and looking up the first useful
1368 	 * entry in the map.
1369 	 */
1370 
1371 	end = start + len;
1372 	newend = dstaddr + len;
1373 	chain = endchain = NULL;
1374 	nchain = 0;
1375 	vm_map_lock(srcmap);
1376 
1377 	if (uvm_map_lookup_entry(srcmap, start, &entry)) {
1378 
1379 		/* "start" is within an entry */
1380 		if (flags & UVM_EXTRACT_QREF) {
1381 			/*
1382 			 * for quick references we don't clip the entry, so
1383 			 * the entry may map space "before" the starting
1384 			 * virtual address... this is the "fudge" factor
1385 			 * (which can be non-zero only the first time
1386 			 * through the "while" loop in step 3).
1387 			 */
1388 			fudge = start - entry->start;
1389 		} else {
1390 			/*
1391 			 * normal reference: we clip the map to fit (thus
1392 			 * fudge is zero)
1393 			 */
1394 			UVM_MAP_CLIP_START(srcmap, entry, start);
1395 			SAVE_HINT(srcmap, entry->prev);
1396 			fudge = 0;
1397 		}
1398 
1399 	} else {
1400 
1401 		/* "start" is not within an entry ... skip to next entry */
1402 		if (flags & UVM_EXTRACT_CONTIG) {
1403 			error = EINVAL;
1404 			goto bad;    /* definite hole here ... */
1405 		}
1406 
1407 		entry = entry->next;
1408 		fudge = 0;
1409 	}
1410 	/* save values from srcmap for step 6 */
1411 	orig_entry = entry;
1412 	orig_fudge = fudge;
1413 
1414 
1415 	/*
1416 	 * step 3: now start looping through the map entries, extracting
1417 	 * as we go.
1418 	 */
1419 
1420 	while (entry->start < end && entry != &srcmap->header) {
1421 
1422 		/* if we are not doing a quick reference, clip it */
1423 		if ((flags & UVM_EXTRACT_QREF) == 0)
1424 			UVM_MAP_CLIP_END(srcmap, entry, end);
1425 
1426 		/* clear needs_copy (allow chunking) */
1427 		if (UVM_ET_ISNEEDSCOPY(entry)) {
1428 			if (fudge)
1429 				oldstart = entry->start;
1430 			else
1431 				oldstart = 0;	/* XXX: gcc */
1432 			amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
1433 			if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
1434 				error = ENOMEM;
1435 				goto bad;
1436 			}
1437 			/* amap_copy could clip (during chunk)!  update fudge */
1438 			if (fudge) {
1439 				fudge = fudge - (entry->start - oldstart);
1440 				orig_fudge = fudge;
1441 			}
1442 		}
1443 
1444 		/* calculate the offset of this from "start" */
1445 		oldoffset = (entry->start + fudge) - start;
1446 
1447 		/* allocate a new map entry */
1448 		newentry = uvm_mapent_alloc(dstmap);
1449 		if (newentry == NULL) {
1450 			error = ENOMEM;
1451 			goto bad;
1452 		}
1453 
1454 		/* set up new map entry */
1455 		newentry->next = NULL;
1456 		newentry->prev = endchain;
1457 		newentry->start = dstaddr + oldoffset;
1458 		newentry->end =
1459 		    newentry->start + (entry->end - (entry->start + fudge));
1460 		if (newentry->end > newend || newentry->end < newentry->start)
1461 			newentry->end = newend;
1462 		newentry->object.uvm_obj = entry->object.uvm_obj;
1463 		if (newentry->object.uvm_obj) {
1464 			if (newentry->object.uvm_obj->pgops->pgo_reference)
1465 				newentry->object.uvm_obj->pgops->
1466 				    pgo_reference(newentry->object.uvm_obj);
1467 				newentry->offset = entry->offset + fudge;
1468 		} else {
1469 			newentry->offset = 0;
1470 		}
1471 		newentry->etype = entry->etype;
1472 		newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
1473 			entry->max_protection : entry->protection;
1474 		newentry->max_protection = entry->max_protection;
1475 		newentry->inheritance = entry->inheritance;
1476 		newentry->wired_count = 0;
1477 		newentry->aref.ar_amap = entry->aref.ar_amap;
1478 		if (newentry->aref.ar_amap) {
1479 			newentry->aref.ar_pageoff =
1480 			    entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1481 			amap_ref(newentry, AMAP_SHARED |
1482 			    ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
1483 		} else {
1484 			newentry->aref.ar_pageoff = 0;
1485 		}
1486 		newentry->advice = entry->advice;
1487 
1488 		/* now link it on the chain */
1489 		nchain++;
1490 		if (endchain == NULL) {
1491 			chain = endchain = newentry;
1492 		} else {
1493 			endchain->next = newentry;
1494 			endchain = newentry;
1495 		}
1496 
1497 		/* end of 'while' loop! */
1498 		if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1499 		    (entry->next == &srcmap->header ||
1500 		    entry->next->start != entry->end)) {
1501 			error = EINVAL;
1502 			goto bad;
1503 		}
1504 		entry = entry->next;
1505 		fudge = 0;
1506 	}
1507 
1508 
1509 	/*
1510 	 * step 4: close off chain (in format expected by uvm_map_replace)
1511 	 */
1512 
1513 	if (chain)
1514 		chain->prev = endchain;
1515 
1516 
1517 	/*
1518 	 * step 5: attempt to lock the dest map so we can pmap_copy.
1519 	 * note usage of copy_ok:
1520 	 *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
1521 	 *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
1522 	 */
1523 
1524 	if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
1525 
1526 		copy_ok = 1;
1527 		if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1528 		    nchain)) {
1529 			if (srcmap != dstmap)
1530 				vm_map_unlock(dstmap);
1531 			error = EIO;
1532 			goto bad;
1533 		}
1534 
1535 	} else {
1536 
1537 		copy_ok = 0;
1538 		/* replace defered until step 7 */
1539 
1540 	}
1541 
1542 
1543 	/*
1544 	 * step 6: traverse the srcmap a second time to do the following:
1545 	 *  - if we got a lock on the dstmap do pmap_copy
1546 	 *  - if UVM_EXTRACT_REMOVE remove the entries
1547 	 * we make use of orig_entry and orig_fudge (saved in step 2)
1548 	 */
1549 
1550 	if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
1551 
1552 		/* purge possible stale hints from srcmap */
1553 		if (flags & UVM_EXTRACT_REMOVE) {
1554 			SAVE_HINT(srcmap, orig_entry->prev);
1555 			if (srcmap->first_free->start >= start)
1556 				srcmap->first_free = orig_entry->prev;
1557 		}
1558 
1559 		entry = orig_entry;
1560 		fudge = orig_fudge;
1561 		deadentry = NULL;	/* for UVM_EXTRACT_REMOVE */
1562 
1563 		while (entry->start < end && entry != &srcmap->header) {
1564 
1565 			if (copy_ok) {
1566 				oldoffset = (entry->start + fudge) - start;
1567 				elen = min(end, entry->end) -
1568 				    (entry->start + fudge);
1569 				pmap_copy(dstmap->pmap, srcmap->pmap,
1570 				    dstaddr + oldoffset, elen,
1571 				    entry->start + fudge);
1572 			}
1573 
1574 			/* we advance "entry" in the following if statement */
1575 			if (flags & UVM_EXTRACT_REMOVE) {
1576 				pmap_remove(srcmap->pmap, entry->start,
1577 						entry->end);
1578         			oldentry = entry;	/* save entry */
1579         			entry = entry->next;	/* advance */
1580 				uvm_map_entry_unlink(srcmap, oldentry);
1581 							/* add to dead list */
1582 				oldentry->next = deadentry;
1583 				deadentry = oldentry;
1584       			} else {
1585         			entry = entry->next;		/* advance */
1586 			}
1587 
1588 			/* end of 'while' loop */
1589 			fudge = 0;
1590 		}
1591 
1592 		/*
1593 		 * unlock dstmap.  we will dispose of deadentry in
1594 		 * step 7 if needed
1595 		 */
1596 		if (copy_ok && srcmap != dstmap)
1597 			vm_map_unlock(dstmap);
1598 
1599 	}
1600 	else
1601 		deadentry = NULL; /* XXX: gcc */
1602 
1603 	/*
1604 	 * step 7: we are done with the source map, unlock.   if copy_ok
1605 	 * is 0 then we have not replaced the dummy mapping in dstmap yet
1606 	 * and we need to do so now.
1607 	 */
1608 
1609 	vm_map_unlock(srcmap);
1610 	if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
1611 		uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
1612 
1613 	/* now do the replacement if we didn't do it in step 5 */
1614 	if (copy_ok == 0) {
1615 		vm_map_lock(dstmap);
1616 		error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1617 		    nchain);
1618 		vm_map_unlock(dstmap);
1619 
1620 		if (error == FALSE) {
1621 			error = EIO;
1622 			goto bad2;
1623 		}
1624 	}
1625 
1626 	/*
1627 	 * done!
1628 	 */
1629 	return(0);
1630 
1631 	/*
1632 	 * bad: failure recovery
1633 	 */
1634 bad:
1635 	vm_map_unlock(srcmap);
1636 bad2:			/* src already unlocked */
1637 	if (chain)
1638 		uvm_unmap_detach(chain,
1639 		    (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1640 	uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
1641 	return(error);
1642 }
1643 
1644 /* end of extraction functions */
1645 
1646 /*
1647  * uvm_map_submap: punch down part of a map into a submap
1648  *
1649  * => only the kernel_map is allowed to be submapped
1650  * => the purpose of submapping is to break up the locking granularity
1651  *	of a larger map
1652  * => the range specified must have been mapped previously with a uvm_map()
1653  *	call [with uobj==NULL] to create a blank map entry in the main map.
1654  *	[And it had better still be blank!]
1655  * => maps which contain submaps should never be copied or forked.
1656  * => to remove a submap, use uvm_unmap() on the main map
1657  *	and then uvm_map_deallocate() the submap.
1658  * => main map must be unlocked.
1659  * => submap must have been init'd and have a zero reference count.
1660  *	[need not be locked as we don't actually reference it]
1661  */
1662 
1663 int
1664 uvm_map_submap(map, start, end, submap)
1665 	vm_map_t map, submap;
1666 	vaddr_t start, end;
1667 {
1668 	vm_map_entry_t entry;
1669 	int result;
1670 	UVMHIST_FUNC("uvm_map_submap"); UVMHIST_CALLED(maphist);
1671 
1672 	vm_map_lock(map);
1673 
1674 	VM_MAP_RANGE_CHECK(map, start, end);
1675 
1676 	if (uvm_map_lookup_entry(map, start, &entry)) {
1677 		UVM_MAP_CLIP_START(map, entry, start);
1678 		UVM_MAP_CLIP_END(map, entry, end);		/* to be safe */
1679 	}
1680 	else {
1681 		entry = NULL;
1682 	}
1683 
1684 	if (entry != NULL &&
1685 	    entry->start == start && entry->end == end &&
1686 	    entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
1687 	    !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1688 
1689 		/*
1690 		 * doit!
1691 		 */
1692 		entry->etype |= UVM_ET_SUBMAP;
1693 		entry->object.sub_map = submap;
1694 		entry->offset = 0;
1695 		uvm_map_reference(submap);
1696 		result = KERN_SUCCESS;
1697 	} else {
1698 		result = KERN_INVALID_ARGUMENT;
1699 	}
1700 	vm_map_unlock(map);
1701 
1702 	return(result);
1703 }
1704 
1705 
1706 /*
1707  * uvm_map_protect: change map protection
1708  *
1709  * => set_max means set max_protection.
1710  * => map must be unlocked.
1711  */
1712 
1713 #define MASK(entry)     (UVM_ET_ISCOPYONWRITE(entry) ? \
1714 			 ~VM_PROT_WRITE : VM_PROT_ALL)
1715 #define max(a,b)        ((a) > (b) ? (a) : (b))
1716 
1717 int
1718 uvm_map_protect(map, start, end, new_prot, set_max)
1719 	vm_map_t map;
1720 	vaddr_t start, end;
1721 	vm_prot_t new_prot;
1722 	boolean_t set_max;
1723 {
1724 	vm_map_entry_t current, entry;
1725 	int rv = KERN_SUCCESS;
1726 	UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
1727 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
1728 	map, start, end, new_prot);
1729 
1730 	vm_map_lock(map);
1731 
1732 	VM_MAP_RANGE_CHECK(map, start, end);
1733 
1734 	if (uvm_map_lookup_entry(map, start, &entry)) {
1735 		UVM_MAP_CLIP_START(map, entry, start);
1736 	} else {
1737 		entry = entry->next;
1738 	}
1739 
1740 	/*
1741 	 * make a first pass to check for protection violations.
1742 	 */
1743 
1744 	current = entry;
1745 	while ((current != &map->header) && (current->start < end)) {
1746 		if (UVM_ET_ISSUBMAP(current)) {
1747 			rv = KERN_INVALID_ARGUMENT;
1748 			goto out;
1749 		}
1750 		if ((new_prot & current->max_protection) != new_prot) {
1751 			rv = KERN_PROTECTION_FAILURE;
1752 			goto out;
1753 		}
1754 		current = current->next;
1755 	}
1756 
1757 	/* go back and fix up protections (no need to clip this time). */
1758 
1759 	current = entry;
1760 
1761 	while ((current != &map->header) && (current->start < end)) {
1762 		vm_prot_t old_prot;
1763 
1764 		UVM_MAP_CLIP_END(map, current, end);
1765 
1766 		old_prot = current->protection;
1767 		if (set_max)
1768 			current->protection =
1769 			    (current->max_protection = new_prot) & old_prot;
1770 		else
1771 			current->protection = new_prot;
1772 
1773 		/*
1774 		 * update physical map if necessary.  worry about copy-on-write
1775 		 * here -- CHECK THIS XXX
1776 		 */
1777 
1778 		if (current->protection != old_prot) {
1779 			/* update pmap! */
1780 			pmap_protect(map->pmap, current->start, current->end,
1781 			    current->protection & MASK(entry));
1782 		}
1783 
1784 		/*
1785 		 * If the map is configured to lock any future mappings,
1786 		 * wire this entry now if the old protection was VM_PROT_NONE
1787 		 * and the new protection is not VM_PROT_NONE.
1788 		 */
1789 
1790 		if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
1791 		    VM_MAPENT_ISWIRED(entry) == 0 &&
1792 		    old_prot == VM_PROT_NONE &&
1793 		    new_prot != VM_PROT_NONE) {
1794 			if (uvm_map_pageable(map, entry->start,
1795 			    entry->end, FALSE,
1796 			    UVM_LK_ENTER|UVM_LK_EXIT) != KERN_SUCCESS) {
1797 				/*
1798 				 * If locking the entry fails, remember the
1799 				 * error if it's the first one.  Note we
1800 				 * still continue setting the protection in
1801 				 * the map, but will return the resource
1802 				 * shortage condition regardless.
1803 				 *
1804 				 * XXX Ignore what the actual error is,
1805 				 * XXX just call it a resource shortage
1806 				 * XXX so that it doesn't get confused
1807 				 * XXX what uvm_map_protect() itself would
1808 				 * XXX normally return.
1809 				 */
1810 				rv = KERN_RESOURCE_SHORTAGE;
1811 			}
1812 		}
1813 
1814 		current = current->next;
1815 	}
1816 
1817  out:
1818 	vm_map_unlock(map);
1819 	UVMHIST_LOG(maphist, "<- done, rv=%d",rv,0,0,0);
1820 	return (rv);
1821 }
1822 
1823 #undef  max
1824 #undef  MASK
1825 
1826 /*
1827  * uvm_map_inherit: set inheritance code for range of addrs in map.
1828  *
1829  * => map must be unlocked
1830  * => note that the inherit code is used during a "fork".  see fork
1831  *	code for details.
1832  */
1833 
1834 int
1835 uvm_map_inherit(map, start, end, new_inheritance)
1836 	vm_map_t map;
1837 	vaddr_t start;
1838 	vaddr_t end;
1839 	vm_inherit_t new_inheritance;
1840 {
1841 	vm_map_entry_t entry, temp_entry;
1842 	UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
1843 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
1844 	    map, start, end, new_inheritance);
1845 
1846 	switch (new_inheritance) {
1847 	case VM_INHERIT_NONE:
1848 	case VM_INHERIT_COPY:
1849 	case VM_INHERIT_SHARE:
1850 		break;
1851 	default:
1852 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1853 		return (KERN_INVALID_ARGUMENT);
1854 	}
1855 
1856 	vm_map_lock(map);
1857 
1858 	VM_MAP_RANGE_CHECK(map, start, end);
1859 
1860 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1861 		entry = temp_entry;
1862 		UVM_MAP_CLIP_START(map, entry, start);
1863 	}  else {
1864 		entry = temp_entry->next;
1865 	}
1866 
1867 	while ((entry != &map->header) && (entry->start < end)) {
1868 		UVM_MAP_CLIP_END(map, entry, end);
1869 
1870 		entry->inheritance = new_inheritance;
1871 
1872 		entry = entry->next;
1873 	}
1874 
1875 	vm_map_unlock(map);
1876 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1877 	return(KERN_SUCCESS);
1878 }
1879 
1880 /*
1881  * uvm_map_advice: set advice code for range of addrs in map.
1882  *
1883  * => map must be unlocked
1884  */
1885 
1886 int
1887 uvm_map_advice(map, start, end, new_advice)
1888 	vm_map_t map;
1889 	vaddr_t start;
1890 	vaddr_t end;
1891 	int new_advice;
1892 {
1893 	vm_map_entry_t entry, temp_entry;
1894 	UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
1895 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
1896 	    map, start, end, new_advice);
1897 
1898 	vm_map_lock(map);
1899 
1900 	VM_MAP_RANGE_CHECK(map, start, end);
1901 
1902 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1903 		entry = temp_entry;
1904 		UVM_MAP_CLIP_START(map, entry, start);
1905 	} else {
1906 		entry = temp_entry->next;
1907 	}
1908 
1909 	/*
1910 	 * XXXJRT: disallow holes?
1911 	 */
1912 
1913 	while ((entry != &map->header) && (entry->start < end)) {
1914 		UVM_MAP_CLIP_END(map, entry, end);
1915 
1916 		switch (new_advice) {
1917 		case MADV_NORMAL:
1918 		case MADV_RANDOM:
1919 		case MADV_SEQUENTIAL:
1920 			/* nothing special here */
1921 			break;
1922 
1923 		default:
1924 			vm_map_unlock(map);
1925 			UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1926 			return (KERN_INVALID_ARGUMENT);
1927 		}
1928 
1929 
1930 		entry->advice = new_advice;
1931 
1932 		entry = entry->next;
1933 	}
1934 
1935 	vm_map_unlock(map);
1936 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1937 	return (KERN_SUCCESS);
1938 }
1939 
1940 /*
1941  * uvm_map_pageable: sets the pageability of a range in a map.
1942  *
1943  * => wires map entries.  should not be used for transient page locking.
1944  *	for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
1945  * => regions sepcified as not pageable require lock-down (wired) memory
1946  *	and page tables.
1947  * => map must never be read-locked
1948  * => if islocked is TRUE, map is already write-locked
1949  * => we always unlock the map, since we must downgrade to a read-lock
1950  *	to call uvm_fault_wire()
1951  * => XXXCDC: check this and try and clean it up.
1952  */
1953 
1954 int
1955 uvm_map_pageable(map, start, end, new_pageable, lockflags)
1956 	vm_map_t map;
1957 	vaddr_t start, end;
1958 	boolean_t new_pageable;
1959 	int lockflags;
1960 {
1961 	vm_map_entry_t entry, start_entry, failed_entry;
1962 	int rv;
1963 #ifdef DIAGNOSTIC
1964 	u_int timestamp_save;
1965 #endif
1966 	UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
1967 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
1968 	map, start, end, new_pageable);
1969 
1970 #ifdef DIAGNOSTIC
1971 	if ((map->flags & VM_MAP_PAGEABLE) == 0)
1972 		panic("uvm_map_pageable: map %p not pageable", map);
1973 #endif
1974 
1975 	if ((lockflags & UVM_LK_ENTER) == 0)
1976 		vm_map_lock(map);
1977 
1978 	VM_MAP_RANGE_CHECK(map, start, end);
1979 
1980 	/*
1981 	 * only one pageability change may take place at one time, since
1982 	 * uvm_fault_wire assumes it will be called only once for each
1983 	 * wiring/unwiring.  therefore, we have to make sure we're actually
1984 	 * changing the pageability for the entire region.  we do so before
1985 	 * making any changes.
1986 	 */
1987 
1988 	if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1989 		if ((lockflags & UVM_LK_EXIT) == 0)
1990 			vm_map_unlock(map);
1991 
1992 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1993 		return (KERN_INVALID_ADDRESS);
1994 	}
1995 	entry = start_entry;
1996 
1997 	/*
1998 	 * handle wiring and unwiring separately.
1999 	 */
2000 
2001 	if (new_pageable) {		/* unwire */
2002 		UVM_MAP_CLIP_START(map, entry, start);
2003 		/*
2004 		 * unwiring.  first ensure that the range to be unwired is
2005 		 * really wired down and that there are no holes.
2006 		 */
2007 		while ((entry != &map->header) && (entry->start < end)) {
2008 			if (entry->wired_count == 0 ||
2009 			    (entry->end < end &&
2010 			     (entry->next == &map->header ||
2011 			      entry->next->start > entry->end))) {
2012 				if ((lockflags & UVM_LK_EXIT) == 0)
2013 					vm_map_unlock(map);
2014 				UVMHIST_LOG(maphist,
2015 				    "<- done (INVALID UNWIRE ARG)",0,0,0,0);
2016 				return (KERN_INVALID_ARGUMENT);
2017 			}
2018 			entry = entry->next;
2019 		}
2020 
2021 		/*
2022 		 * POSIX 1003.1b - a single munlock call unlocks a region,
2023 		 * regardless of the number of mlock calls made on that
2024 		 * region.
2025 		 */
2026 		entry = start_entry;
2027 		while ((entry != &map->header) && (entry->start < end)) {
2028 			UVM_MAP_CLIP_END(map, entry, end);
2029 			if (VM_MAPENT_ISWIRED(entry))
2030 				uvm_map_entry_unwire(map, entry);
2031 			entry = entry->next;
2032 		}
2033 		if ((lockflags & UVM_LK_EXIT) == 0)
2034 			vm_map_unlock(map);
2035 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2036 		return(KERN_SUCCESS);
2037 
2038 		/*
2039 		 * end of unwire case!
2040 		 */
2041 	}
2042 
2043 	/*
2044 	 * wire case: in two passes [XXXCDC: ugly block of code here]
2045 	 *
2046 	 * 1: holding the write lock, we create any anonymous maps that need
2047 	 *    to be created.  then we clip each map entry to the region to
2048 	 *    be wired and increment its wiring count.
2049 	 *
2050 	 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
2051 	 *    in the pages for any newly wired area (wired_count == 1).
2052 	 *
2053 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
2054 	 *    deadlock with another thread that may have faulted on one of
2055 	 *    the pages to be wired (it would mark the page busy, blocking
2056 	 *    us, then in turn block on the map lock that we hold).  because
2057 	 *    of problems in the recursive lock package, we cannot upgrade
2058 	 *    to a write lock in vm_map_lookup.  thus, any actions that
2059 	 *    require the write lock must be done beforehand.  because we
2060 	 *    keep the read lock on the map, the copy-on-write status of the
2061 	 *    entries we modify here cannot change.
2062 	 */
2063 
2064 	while ((entry != &map->header) && (entry->start < end)) {
2065 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2066 			/*
2067 			 * perform actions of vm_map_lookup that need the
2068 			 * write lock on the map: create an anonymous map
2069 			 * for a copy-on-write region, or an anonymous map
2070 			 * for a zero-fill region.  (XXXCDC: submap case
2071 			 * ok?)
2072 			 */
2073 			if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
2074 				if (UVM_ET_ISNEEDSCOPY(entry) &&
2075 				    ((entry->protection & VM_PROT_WRITE) ||
2076 				     (entry->object.uvm_obj == NULL))) {
2077 					amap_copy(map, entry, M_WAITOK, TRUE,
2078 					    start, end);
2079 					/* XXXCDC: wait OK? */
2080 				}
2081 			}
2082 		}
2083 		UVM_MAP_CLIP_START(map, entry, start);
2084 		UVM_MAP_CLIP_END(map, entry, end);
2085 		entry->wired_count++;
2086 
2087 		/*
2088 		 * Check for holes
2089 		 */
2090 		if (entry->protection == VM_PROT_NONE ||
2091 		    (entry->end < end &&
2092 		     (entry->next == &map->header ||
2093 		      entry->next->start > entry->end))) {
2094 			/*
2095 			 * found one.  amap creation actions do not need to
2096 			 * be undone, but the wired counts need to be restored.
2097 			 */
2098 			while (entry != &map->header && entry->end > start) {
2099 				entry->wired_count--;
2100 				entry = entry->prev;
2101 			}
2102 			if ((lockflags & UVM_LK_EXIT) == 0)
2103 				vm_map_unlock(map);
2104 			UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
2105 			return (KERN_INVALID_ARGUMENT);
2106 		}
2107 		entry = entry->next;
2108 	}
2109 
2110 	/*
2111 	 * Pass 2.
2112 	 */
2113 
2114 #ifdef DIAGNOSTIC
2115 	timestamp_save = map->timestamp;
2116 #endif
2117 	vm_map_busy(map);
2118 	vm_map_downgrade(map);
2119 
2120 	rv = 0;
2121 	entry = start_entry;
2122 	while (entry != &map->header && entry->start < end) {
2123 		if (entry->wired_count == 1) {
2124 			rv = uvm_fault_wire(map, entry->start, entry->end,
2125 			    entry->protection);
2126 			if (rv) {
2127 				/*
2128 				 * wiring failed.  break out of the loop.
2129 				 * we'll clean up the map below, once we
2130 				 * have a write lock again.
2131 				 */
2132 				break;
2133 			}
2134 		}
2135 		entry = entry->next;
2136 	}
2137 
2138 	if (rv) {        /* failed? */
2139 		/*
2140 		 * Get back to an exclusive (write) lock.
2141 		 */
2142 		vm_map_upgrade(map);
2143 		vm_map_unbusy(map);
2144 
2145 #ifdef DIAGNOSTIC
2146 		if (timestamp_save != map->timestamp)
2147 			panic("uvm_map_pageable: stale map");
2148 #endif
2149 
2150 		/*
2151 		 * first drop the wiring count on all the entries
2152 		 * which haven't actually been wired yet.
2153 		 */
2154 		failed_entry = entry;
2155 		while (entry != &map->header && entry->start < end) {
2156 			entry->wired_count--;
2157 			entry = entry->next;
2158 		}
2159 
2160 		/*
2161 		 * now, unwire all the entries that were successfully
2162 		 * wired above.
2163 		 */
2164 		entry = start_entry;
2165 		while (entry != failed_entry) {
2166 			entry->wired_count--;
2167 			if (VM_MAPENT_ISWIRED(entry) == 0)
2168 				uvm_map_entry_unwire(map, entry);
2169 			entry = entry->next;
2170 		}
2171 		if ((lockflags & UVM_LK_EXIT) == 0)
2172 			vm_map_unlock(map);
2173 		UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
2174 		return(rv);
2175 	}
2176 
2177 	/* We are holding a read lock here. */
2178 	if ((lockflags & UVM_LK_EXIT) == 0) {
2179 		vm_map_unbusy(map);
2180 		vm_map_unlock_read(map);
2181 	} else {
2182 		/*
2183 		 * Get back to an exclusive (write) lock.
2184 		 */
2185 		vm_map_upgrade(map);
2186 		vm_map_unbusy(map);
2187 	}
2188 
2189 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2190 	return(KERN_SUCCESS);
2191 }
2192 
2193 /*
2194  * uvm_map_pageable_all: special case of uvm_map_pageable - affects
2195  * all mapped regions.
2196  *
2197  * => map must not be locked.
2198  * => if no flags are specified, all regions are unwired.
2199  * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
2200  */
2201 
2202 int
2203 uvm_map_pageable_all(map, flags, limit)
2204 	vm_map_t map;
2205 	int flags;
2206 	vsize_t limit;
2207 {
2208 	vm_map_entry_t entry, failed_entry;
2209 	vsize_t size;
2210 	int rv;
2211 #ifdef DIAGNOSTIC
2212 	u_int timestamp_save;
2213 #endif
2214 	UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
2215 	UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
2216 
2217 #ifdef DIAGNOSTIC
2218 	if ((map->flags & VM_MAP_PAGEABLE) == 0)
2219 		panic("uvm_map_pageable_all: map %p not pageable", map);
2220 #endif
2221 
2222 	vm_map_lock(map);
2223 
2224 	/*
2225 	 * handle wiring and unwiring separately.
2226 	 */
2227 
2228 	if (flags == 0) {			/* unwire */
2229 		/*
2230 		 * POSIX 1003.1b -- munlockall unlocks all regions,
2231 		 * regardless of how many times mlockall has been called.
2232 		 */
2233 		for (entry = map->header.next; entry != &map->header;
2234 		     entry = entry->next) {
2235 			if (VM_MAPENT_ISWIRED(entry))
2236 				uvm_map_entry_unwire(map, entry);
2237 		}
2238 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2239 		vm_map_unlock(map);
2240 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2241 		return (KERN_SUCCESS);
2242 
2243 		/*
2244 		 * end of unwire case!
2245 		 */
2246 	}
2247 
2248 	if (flags & MCL_FUTURE) {
2249 		/*
2250 		 * must wire all future mappings; remember this.
2251 		 */
2252 		vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
2253 	}
2254 
2255 	if ((flags & MCL_CURRENT) == 0) {
2256 		/*
2257 		 * no more work to do!
2258 		 */
2259 		UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
2260 		vm_map_unlock(map);
2261 		return (KERN_SUCCESS);
2262 	}
2263 
2264 	/*
2265 	 * wire case: in three passes [XXXCDC: ugly block of code here]
2266 	 *
2267 	 * 1: holding the write lock, count all pages mapped by non-wired
2268 	 *    entries.  if this would cause us to go over our limit, we fail.
2269 	 *
2270 	 * 2: still holding the write lock, we create any anonymous maps that
2271 	 *    need to be created.  then we increment its wiring count.
2272 	 *
2273 	 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
2274 	 *    in the pages for any newly wired area (wired_count == 1).
2275 	 *
2276 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
2277 	 *    deadlock with another thread that may have faulted on one of
2278 	 *    the pages to be wired (it would mark the page busy, blocking
2279 	 *    us, then in turn block on the map lock that we hold).  because
2280 	 *    of problems in the recursive lock package, we cannot upgrade
2281 	 *    to a write lock in vm_map_lookup.  thus, any actions that
2282 	 *    require the write lock must be done beforehand.  because we
2283 	 *    keep the read lock on the map, the copy-on-write status of the
2284 	 *    entries we modify here cannot change.
2285 	 */
2286 
2287 	for (size = 0, entry = map->header.next; entry != &map->header;
2288 	     entry = entry->next) {
2289 		if (entry->protection != VM_PROT_NONE &&
2290 		    VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2291 			size += entry->end - entry->start;
2292 		}
2293 	}
2294 
2295 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
2296 		vm_map_unlock(map);
2297 		return (KERN_NO_SPACE);		/* XXX overloaded */
2298 	}
2299 
2300 	/* XXX non-pmap_wired_count case must be handled by caller */
2301 #ifdef pmap_wired_count
2302 	if (limit != 0 &&
2303 	    (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
2304 		vm_map_unlock(map);
2305 		return (KERN_NO_SPACE);		/* XXX overloaded */
2306 	}
2307 #endif
2308 
2309 	/*
2310 	 * Pass 2.
2311 	 */
2312 
2313 	for (entry = map->header.next; entry != &map->header;
2314 	     entry = entry->next) {
2315 		if (entry->protection == VM_PROT_NONE)
2316 			continue;
2317 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2318 			/*
2319 			 * perform actions of vm_map_lookup that need the
2320 			 * write lock on the map: create an anonymous map
2321 			 * for a copy-on-write region, or an anonymous map
2322 			 * for a zero-fill region.  (XXXCDC: submap case
2323 			 * ok?)
2324 			 */
2325 			if (!UVM_ET_ISSUBMAP(entry)) {	/* not submap */
2326 				if (UVM_ET_ISNEEDSCOPY(entry) &&
2327 				    ((entry->protection & VM_PROT_WRITE) ||
2328 				     (entry->object.uvm_obj == NULL))) {
2329 					amap_copy(map, entry, M_WAITOK, TRUE,
2330 					    entry->start, entry->end);
2331 					/* XXXCDC: wait OK? */
2332 				}
2333 			}
2334 		}
2335 		entry->wired_count++;
2336 	}
2337 
2338 	/*
2339 	 * Pass 3.
2340 	 */
2341 
2342 #ifdef DIAGNOSTIC
2343 	timestamp_save = map->timestamp;
2344 #endif
2345 	vm_map_busy(map);
2346 	vm_map_downgrade(map);
2347 
2348 	rv = KERN_SUCCESS;
2349 	for (entry = map->header.next; entry != &map->header;
2350 	     entry = entry->next) {
2351 		if (entry->wired_count == 1) {
2352 			rv = uvm_fault_wire(map, entry->start, entry->end,
2353 			     entry->protection);
2354 			if (rv) {
2355 				/*
2356 				 * wiring failed.  break out of the loop.
2357 				 * we'll clean up the map below, once we
2358 				 * have a write lock again.
2359 				 */
2360 				break;
2361 			}
2362 		}
2363 	}
2364 
2365 	if (rv) {	/* failed? */
2366 		/*
2367 		 * Get back an exclusive (write) lock.
2368 		 */
2369 		vm_map_upgrade(map);
2370 		vm_map_unbusy(map);
2371 
2372 #ifdef DIAGNOSTIC
2373 		if (timestamp_save != map->timestamp)
2374 			panic("uvm_map_pageable_all: stale map");
2375 #endif
2376 
2377 		/*
2378 		 * first drop the wiring count on all the entries
2379 		 * which haven't actually been wired yet.
2380 		 *
2381 		 * Skip VM_PROT_NONE entries like we did above.
2382 		 */
2383 		failed_entry = entry;
2384 		for (/* nothing */; entry != &map->header;
2385 		     entry = entry->next) {
2386 			if (entry->protection == VM_PROT_NONE)
2387 				continue;
2388 			entry->wired_count--;
2389 		}
2390 
2391 		/*
2392 		 * now, unwire all the entries that were successfully
2393 		 * wired above.
2394 		 *
2395 		 * Skip VM_PROT_NONE entries like we did above.
2396 		 */
2397 		for (entry = map->header.next; entry != failed_entry;
2398 		     entry = entry->next) {
2399 			if (entry->protection == VM_PROT_NONE)
2400 				continue;
2401 			entry->wired_count--;
2402 			if (VM_MAPENT_ISWIRED(entry))
2403 				uvm_map_entry_unwire(map, entry);
2404 		}
2405 		vm_map_unlock(map);
2406 		UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
2407 		return (rv);
2408 	}
2409 
2410 	/* We are holding a read lock here. */
2411 	vm_map_unbusy(map);
2412 	vm_map_unlock_read(map);
2413 
2414 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2415 	return (KERN_SUCCESS);
2416 }
2417 
2418 /*
2419  * uvm_map_clean: clean out a map range
2420  *
2421  * => valid flags:
2422  *   if (flags & PGO_CLEANIT): dirty pages are cleaned first
2423  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
2424  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
2425  *   if (flags & PGO_FREE): any cached pages are freed after clean
2426  * => returns an error if any part of the specified range isn't mapped
2427  * => never a need to flush amap layer since the anonymous memory has
2428  *	no permanent home, but may deactivate pages there
2429  * => called from sys_msync() and sys_madvise()
2430  * => caller must not write-lock map (read OK).
2431  * => we may sleep while cleaning if SYNCIO [with map read-locked]
2432  */
2433 
2434 int	amap_clean_works = 1;	/* XXX for now, just in case... */
2435 
2436 int
2437 uvm_map_clean(map, start, end, flags)
2438 	vm_map_t map;
2439 	vaddr_t start, end;
2440 	int flags;
2441 {
2442 	vm_map_entry_t current, entry;
2443 	struct uvm_object *uobj;
2444 	struct vm_amap *amap;
2445 	struct vm_anon *anon;
2446 	struct vm_page *pg;
2447 	vaddr_t offset;
2448 	vsize_t size;
2449 	int rv, error, refs;
2450 	UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
2451 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
2452 	map, start, end, flags);
2453 
2454 #ifdef DIAGNOSTIC
2455 	if ((flags & (PGO_FREE|PGO_DEACTIVATE)) == (PGO_FREE|PGO_DEACTIVATE))
2456 		panic("uvm_map_clean: FREE and DEACTIVATE");
2457 #endif
2458 
2459 	vm_map_lock_read(map);
2460 	VM_MAP_RANGE_CHECK(map, start, end);
2461 	if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
2462 		vm_map_unlock_read(map);
2463 		return(KERN_INVALID_ADDRESS);
2464 	}
2465 
2466 	/*
2467 	 * Make a first pass to check for holes.
2468 	 */
2469 	for (current = entry; current->start < end; current = current->next) {
2470 		if (UVM_ET_ISSUBMAP(current)) {
2471 			vm_map_unlock_read(map);
2472 			return (KERN_INVALID_ARGUMENT);
2473 		}
2474 		if (end > current->end && (current->next == &map->header ||
2475 		    current->end != current->next->start)) {
2476 			vm_map_unlock_read(map);
2477 			return (KERN_INVALID_ADDRESS);
2478 		}
2479 	}
2480 
2481 	error = KERN_SUCCESS;
2482 
2483 	for (current = entry; current->start < end; current = current->next) {
2484 		amap = current->aref.ar_amap;	/* top layer */
2485 		uobj = current->object.uvm_obj;	/* bottom layer */
2486 
2487 #ifdef DIAGNOSTIC
2488 		if (start < current->start)
2489 			panic("uvm_map_clean: hole");
2490 #endif
2491 
2492 		/*
2493 		 * No amap cleaning necessary if:
2494 		 *
2495 		 *	(1) There's no amap.
2496 		 *
2497 		 *	(2) We're not deactivating or freeing pages.
2498 		 */
2499 		if (amap == NULL ||
2500 		    (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
2501 			goto flush_object;
2502 
2503 		/* XXX for now, just in case... */
2504 		if (amap_clean_works == 0)
2505 			goto flush_object;
2506 
2507 		amap_lock(amap);
2508 
2509 		offset = start - current->start;
2510 		size = (end <= current->end ? end : current->end) -
2511 		    start;
2512 
2513 		for (/* nothing */; size != 0; size -= PAGE_SIZE,
2514 		     offset += PAGE_SIZE) {
2515 			anon = amap_lookup(&current->aref, offset);
2516 			if (anon == NULL)
2517 				continue;
2518 
2519 			simple_lock(&anon->an_lock);
2520 
2521 			pg = anon->u.an_page;
2522 			if (pg == NULL) {
2523 				simple_unlock(&anon->an_lock);
2524 				continue;
2525 			}
2526 
2527 			switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
2528 			/*
2529 			 * XXX In these first 3 cases, we always just
2530 			 * XXX deactivate the page.  We may want to
2531 			 * XXX handle the different cases more
2532 			 * XXX specifically, in the future.
2533 			 */
2534 			case PGO_CLEANIT|PGO_FREE:
2535 			case PGO_CLEANIT|PGO_DEACTIVATE:
2536 			case PGO_DEACTIVATE:
2537  deactivate_it:
2538 				/* skip the page if it's loaned or wired */
2539 				if (pg->loan_count != 0 ||
2540 				    pg->wire_count != 0) {
2541 					simple_unlock(&anon->an_lock);
2542 					continue;
2543 				}
2544 
2545 				uvm_lock_pageq();
2546 
2547 				/*
2548 				 * skip the page if it's not actually owned
2549 				 * by the anon (may simply be loaned to the
2550 				 * anon).
2551 				 */
2552 				if ((pg->pqflags & PQ_ANON) == 0) {
2553 #ifdef DIAGNOSTIC
2554 					if (pg->uobject != NULL)
2555 						panic("uvm_map_clean: "
2556 						    "page anon vs. object "
2557 						    "inconsistency");
2558 #endif
2559 					uvm_unlock_pageq();
2560 					simple_unlock(&anon->an_lock);
2561 					continue;
2562 				}
2563 
2564 #ifdef DIAGNOSTIC
2565 				if (pg->uanon != anon)
2566 					panic("uvm_map_clean: anon "
2567 					    "inconsistency");
2568 #endif
2569 
2570 				/* zap all mappings for the page. */
2571 				pmap_page_protect(pg, VM_PROT_NONE);
2572 
2573 				/* ...and deactivate the page. */
2574 				uvm_pagedeactivate(pg);
2575 
2576 				uvm_unlock_pageq();
2577 				simple_unlock(&anon->an_lock);
2578 				continue;
2579 
2580 			case PGO_FREE:
2581 				/*
2582 				 * If there are multiple references to
2583 				 * the amap, just deactivate the page.
2584 				 */
2585 				if (amap_refs(amap) > 1)
2586 					goto deactivate_it;
2587 
2588 				/* XXX skip the page if it's wired */
2589 				if (pg->wire_count != 0) {
2590 					simple_unlock(&anon->an_lock);
2591 					continue;
2592 				}
2593 				amap_unadd(&current->aref, offset);
2594 				refs = --anon->an_ref;
2595 				simple_unlock(&anon->an_lock);
2596 				if (refs == 0)
2597 					uvm_anfree(anon);
2598 				continue;
2599 
2600 			default:
2601 				panic("uvm_map_clean: wierd flags");
2602 			}
2603 #ifdef DIAGNOSTIC
2604 			panic("uvm_map_clean: unreachable code");
2605 #endif
2606 		}
2607 
2608 		amap_unlock(amap);
2609 
2610  flush_object:
2611 		/*
2612 		 * flush pages if we've got a valid backing object.
2613 		 */
2614 
2615 		offset = current->offset + (start - current->start);
2616 		size = (end <= current->end ? end : current->end) - start;
2617 
2618 		if (uobj != NULL) {
2619 			simple_lock(&uobj->vmobjlock);
2620 			rv = uobj->pgops->pgo_flush(uobj, offset,
2621 			    offset + size, flags);
2622 			simple_unlock(&uobj->vmobjlock);
2623 
2624 			if (rv == FALSE)
2625 				error = KERN_FAILURE;
2626 		}
2627 		start += size;
2628 	}
2629 
2630 	vm_map_unlock_read(map);
2631 	return (error);
2632 }
2633 
2634 
2635 /*
2636  * uvm_map_checkprot: check protection in map
2637  *
2638  * => must allow specified protection in a fully allocated region.
2639  * => map must be read or write locked by caller.
2640  */
2641 
2642 boolean_t
2643 uvm_map_checkprot(map, start, end, protection)
2644 	vm_map_t       map;
2645 	vaddr_t    start, end;
2646 	vm_prot_t      protection;
2647 {
2648 	 vm_map_entry_t entry;
2649 	 vm_map_entry_t tmp_entry;
2650 
2651 	 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
2652 		 return(FALSE);
2653 	 }
2654 
2655 	 entry = tmp_entry;
2656 
2657 	 while (start < end) {
2658 		 if (entry == &map->header) {
2659 			 return(FALSE);
2660 		 }
2661 
2662 		/*
2663 		 * no holes allowed
2664 		 */
2665 
2666 		 if (start < entry->start) {
2667 			 return(FALSE);
2668 		 }
2669 
2670 		/*
2671 		 * check protection associated with entry
2672 		 */
2673 
2674 		 if ((entry->protection & protection) != protection) {
2675 			 return(FALSE);
2676 		 }
2677 
2678 		 /* go to next entry */
2679 
2680 		 start = entry->end;
2681 		 entry = entry->next;
2682 	 }
2683 	 return(TRUE);
2684 }
2685 
2686 /*
2687  * uvmspace_alloc: allocate a vmspace structure.
2688  *
2689  * - structure includes vm_map and pmap
2690  * - XXX: no locking on this structure
2691  * - refcnt set to 1, rest must be init'd by caller
2692  */
2693 struct vmspace *
2694 uvmspace_alloc(min, max, pageable)
2695 	vaddr_t min, max;
2696 	int pageable;
2697 {
2698 	struct vmspace *vm;
2699 	UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
2700 
2701 	vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
2702 	uvmspace_init(vm, NULL, min, max, pageable);
2703 	UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
2704 	return (vm);
2705 }
2706 
2707 /*
2708  * uvmspace_init: initialize a vmspace structure.
2709  *
2710  * - XXX: no locking on this structure
2711  * - refcnt set to 1, rest must me init'd by caller
2712  */
2713 void
2714 uvmspace_init(vm, pmap, min, max, pageable)
2715 	struct vmspace *vm;
2716 	struct pmap *pmap;
2717 	vaddr_t min, max;
2718 	boolean_t pageable;
2719 {
2720 	UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
2721 
2722 	memset(vm, 0, sizeof(*vm));
2723 
2724 	uvm_map_setup(&vm->vm_map, min, max, pageable ? VM_MAP_PAGEABLE : 0);
2725 
2726 	if (pmap)
2727 		pmap_reference(pmap);
2728 	else
2729 		pmap = pmap_create();
2730 	vm->vm_map.pmap = pmap;
2731 
2732 	vm->vm_refcnt = 1;
2733 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
2734 }
2735 
2736 /*
2737  * uvmspace_share: share a vmspace between two proceses
2738  *
2739  * - XXX: no locking on vmspace
2740  * - used for vfork, threads(?)
2741  */
2742 
2743 void
2744 uvmspace_share(p1, p2)
2745 	struct proc *p1, *p2;
2746 {
2747 	p2->p_vmspace = p1->p_vmspace;
2748 	p1->p_vmspace->vm_refcnt++;
2749 }
2750 
2751 /*
2752  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
2753  *
2754  * - XXX: no locking on vmspace
2755  */
2756 
2757 void
2758 uvmspace_unshare(p)
2759 	struct proc *p;
2760 {
2761 	struct vmspace *nvm, *ovm = p->p_vmspace;
2762 
2763 	if (ovm->vm_refcnt == 1)
2764 		/* nothing to do: vmspace isn't shared in the first place */
2765 		return;
2766 
2767 	/* make a new vmspace, still holding old one */
2768 	nvm = uvmspace_fork(ovm);
2769 
2770 	pmap_deactivate(p);		/* unbind old vmspace */
2771 	p->p_vmspace = nvm;
2772 	pmap_activate(p);		/* switch to new vmspace */
2773 
2774 	uvmspace_free(ovm);		/* drop reference to old vmspace */
2775 }
2776 
2777 /*
2778  * uvmspace_exec: the process wants to exec a new program
2779  *
2780  * - XXX: no locking on vmspace
2781  */
2782 
2783 void
2784 uvmspace_exec(p)
2785 	struct proc *p;
2786 {
2787 	struct vmspace *nvm, *ovm = p->p_vmspace;
2788 	vm_map_t map = &ovm->vm_map;
2789 
2790 #ifdef __sparc__
2791 	/* XXX cgd 960926: the sparc #ifdef should be a MD hook */
2792 	kill_user_windows(p);   /* before stack addresses go away */
2793 #endif
2794 
2795 	/*
2796 	 * see if more than one process is using this vmspace...
2797 	 */
2798 
2799 	if (ovm->vm_refcnt == 1) {
2800 
2801 		/*
2802 		 * if p is the only process using its vmspace then we can safely
2803 		 * recycle that vmspace for the program that is being exec'd.
2804 		 */
2805 
2806 #ifdef SYSVSHM
2807 		/*
2808 		 * SYSV SHM semantics require us to kill all segments on an exec
2809 		 */
2810 		if (ovm->vm_shm)
2811 			shmexit(ovm);
2812 #endif
2813 
2814 		/*
2815 		 * POSIX 1003.1b -- "lock future mappings" is revoked
2816 		 * when a process execs another program image.
2817 		 */
2818 		vm_map_lock(map);
2819 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2820 		vm_map_unlock(map);
2821 
2822 		/*
2823 		 * now unmap the old program
2824 		 */
2825 		uvm_unmap(map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
2826 
2827 	} else {
2828 
2829 		/*
2830 		 * p's vmspace is being shared, so we can't reuse it for p since
2831 		 * it is still being used for others.   allocate a new vmspace
2832 		 * for p
2833 		 */
2834 		nvm = uvmspace_alloc(map->min_offset, map->max_offset,
2835 			 (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2836 
2837 		/*
2838 		 * install new vmspace and drop our ref to the old one.
2839 		 */
2840 
2841 		pmap_deactivate(p);
2842 		p->p_vmspace = nvm;
2843 		pmap_activate(p);
2844 
2845 		uvmspace_free(ovm);
2846 	}
2847 }
2848 
2849 /*
2850  * uvmspace_free: free a vmspace data structure
2851  *
2852  * - XXX: no locking on vmspace
2853  */
2854 
2855 void
2856 uvmspace_free(vm)
2857 	struct vmspace *vm;
2858 {
2859 	vm_map_entry_t dead_entries;
2860 	UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
2861 
2862 	UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
2863 	if (--vm->vm_refcnt == 0) {
2864 		/*
2865 		 * lock the map, to wait out all other references to it.  delete
2866 		 * all of the mappings and pages they hold, then call the pmap
2867 		 * module to reclaim anything left.
2868 		 */
2869 		vm_map_lock(&vm->vm_map);
2870 		if (vm->vm_map.nentries) {
2871 			(void)uvm_unmap_remove(&vm->vm_map,
2872 			    vm->vm_map.min_offset, vm->vm_map.max_offset,
2873 			    &dead_entries);
2874 			if (dead_entries != NULL)
2875 				uvm_unmap_detach(dead_entries, 0);
2876 		}
2877 		pmap_destroy(vm->vm_map.pmap);
2878 		vm->vm_map.pmap = NULL;
2879 		pool_put(&uvm_vmspace_pool, vm);
2880 	}
2881 	UVMHIST_LOG(maphist,"<- done", 0,0,0,0);
2882 }
2883 
2884 /*
2885  *   F O R K   -   m a i n   e n t r y   p o i n t
2886  */
2887 /*
2888  * uvmspace_fork: fork a process' main map
2889  *
2890  * => create a new vmspace for child process from parent.
2891  * => parent's map must not be locked.
2892  */
2893 
2894 struct vmspace *
2895 uvmspace_fork(vm1)
2896 	struct vmspace *vm1;
2897 {
2898 	struct vmspace *vm2;
2899 	vm_map_t        old_map = &vm1->vm_map;
2900 	vm_map_t        new_map;
2901 	vm_map_entry_t  old_entry;
2902 	vm_map_entry_t  new_entry;
2903 	pmap_t          new_pmap;
2904 	boolean_t	protect_child;
2905 	UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
2906 
2907 	vm_map_lock(old_map);
2908 
2909 	vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset,
2910 		      (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2911 	memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
2912 	(caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
2913 	new_map = &vm2->vm_map;		  /* XXX */
2914 	new_pmap = new_map->pmap;
2915 
2916 	old_entry = old_map->header.next;
2917 
2918 	/*
2919 	 * go entry-by-entry
2920 	 */
2921 
2922 	while (old_entry != &old_map->header) {
2923 
2924 		/*
2925 		 * first, some sanity checks on the old entry
2926 		 */
2927 		if (UVM_ET_ISSUBMAP(old_entry))
2928 		    panic("fork: encountered a submap during fork (illegal)");
2929 
2930 		if (!UVM_ET_ISCOPYONWRITE(old_entry) &&
2931 			    UVM_ET_ISNEEDSCOPY(old_entry))
2932 	panic("fork: non-copy_on_write map entry marked needs_copy (illegal)");
2933 
2934 
2935 		switch (old_entry->inheritance) {
2936 		case VM_INHERIT_NONE:
2937 			/*
2938 			 * drop the mapping
2939 			 */
2940 			break;
2941 
2942 		case VM_INHERIT_SHARE:
2943 			/*
2944 			 * share the mapping: this means we want the old and
2945 			 * new entries to share amaps and backing objects.
2946 			 */
2947 
2948 			/*
2949 			 * if the old_entry needs a new amap (due to prev fork)
2950 			 * then we need to allocate it now so that we have
2951 			 * something we own to share with the new_entry.   [in
2952 			 * other words, we need to clear needs_copy]
2953 			 */
2954 
2955 			if (UVM_ET_ISNEEDSCOPY(old_entry)) {
2956 				/* get our own amap, clears needs_copy */
2957 				amap_copy(old_map, old_entry, M_WAITOK, FALSE,
2958 				    0, 0);
2959 				/* XXXCDC: WAITOK??? */
2960 			}
2961 
2962 			new_entry = uvm_mapent_alloc(new_map);
2963 			/* old_entry -> new_entry */
2964 			uvm_mapent_copy(old_entry, new_entry);
2965 
2966 			/* new pmap has nothing wired in it */
2967 			new_entry->wired_count = 0;
2968 
2969 			/*
2970 			 * gain reference to object backing the map (can't
2971 			 * be a submap, already checked this case).
2972 			 */
2973 			if (new_entry->aref.ar_amap)
2974 				/* share reference */
2975 				amap_ref(new_entry, AMAP_SHARED);
2976 
2977 			if (new_entry->object.uvm_obj &&
2978 			    new_entry->object.uvm_obj->pgops->pgo_reference)
2979 				new_entry->object.uvm_obj->
2980 				    pgops->pgo_reference(
2981 				        new_entry->object.uvm_obj);
2982 
2983 			/* insert entry at end of new_map's entry list */
2984 			uvm_map_entry_link(new_map, new_map->header.prev,
2985 			    new_entry);
2986 
2987 			/*
2988 			 * pmap_copy the mappings: this routine is optional
2989 			 * but if it is there it will reduce the number of
2990 			 * page faults in the new proc.
2991 			 */
2992 
2993 			pmap_copy(new_pmap, old_map->pmap, new_entry->start,
2994 			    (old_entry->end - old_entry->start),
2995 			    old_entry->start);
2996 
2997 			break;
2998 
2999 		case VM_INHERIT_COPY:
3000 
3001 			/*
3002 			 * copy-on-write the mapping (using mmap's
3003 			 * MAP_PRIVATE semantics)
3004 			 *
3005 			 * allocate new_entry, adjust reference counts.
3006 			 * (note that new references are read-only).
3007 			 */
3008 
3009 			new_entry = uvm_mapent_alloc(new_map);
3010 			/* old_entry -> new_entry */
3011 			uvm_mapent_copy(old_entry, new_entry);
3012 
3013 			if (new_entry->aref.ar_amap)
3014 				amap_ref(new_entry, 0);
3015 
3016 			if (new_entry->object.uvm_obj &&
3017 			    new_entry->object.uvm_obj->pgops->pgo_reference)
3018 				new_entry->object.uvm_obj->pgops->pgo_reference
3019 				    (new_entry->object.uvm_obj);
3020 
3021 			/* new pmap has nothing wired in it */
3022 			new_entry->wired_count = 0;
3023 
3024 			new_entry->etype |=
3025 			    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
3026 			uvm_map_entry_link(new_map, new_map->header.prev,
3027 			    new_entry);
3028 
3029 			/*
3030 			 * the new entry will need an amap.  it will either
3031 			 * need to be copied from the old entry or created
3032 			 * from scratch (if the old entry does not have an
3033 			 * amap).  can we defer this process until later
3034 			 * (by setting "needs_copy") or do we need to copy
3035 			 * the amap now?
3036 			 *
3037 			 * we must copy the amap now if any of the following
3038 			 * conditions hold:
3039 			 * 1. the old entry has an amap and that amap is
3040 			 *    being shared.  this means that the old (parent)
3041 			 *    process is sharing the amap with another
3042 			 *    process.  if we do not clear needs_copy here
3043 			 *    we will end up in a situation where both the
3044 			 *    parent and child process are refering to the
3045 			 *    same amap with "needs_copy" set.  if the
3046 			 *    parent write-faults, the fault routine will
3047 			 *    clear "needs_copy" in the parent by allocating
3048 			 *    a new amap.   this is wrong because the
3049 			 *    parent is supposed to be sharing the old amap
3050 			 *    and the new amap will break that.
3051 			 *
3052 			 * 2. if the old entry has an amap and a non-zero
3053 			 *    wire count then we are going to have to call
3054 			 *    amap_cow_now to avoid page faults in the
3055 			 *    parent process.   since amap_cow_now requires
3056 			 *    "needs_copy" to be clear we might as well
3057 			 *    clear it here as well.
3058 			 *
3059 			 */
3060 
3061 			if (old_entry->aref.ar_amap != NULL) {
3062 
3063 			  if ((amap_flags(old_entry->aref.ar_amap) &
3064 			       AMAP_SHARED) != 0 ||
3065 			      VM_MAPENT_ISWIRED(old_entry)) {
3066 
3067 			    amap_copy(new_map, new_entry, M_WAITOK, FALSE,
3068 				      0, 0);
3069 			    /* XXXCDC: M_WAITOK ... ok? */
3070 			  }
3071 			}
3072 
3073 			/*
3074 			 * if the parent's entry is wired down, then the
3075 			 * parent process does not want page faults on
3076 			 * access to that memory.  this means that we
3077 			 * cannot do copy-on-write because we can't write
3078 			 * protect the old entry.   in this case we
3079 			 * resolve all copy-on-write faults now, using
3080 			 * amap_cow_now.   note that we have already
3081 			 * allocated any needed amap (above).
3082 			 */
3083 
3084 			if (VM_MAPENT_ISWIRED(old_entry)) {
3085 
3086 			  /*
3087 			   * resolve all copy-on-write faults now
3088 			   * (note that there is nothing to do if
3089 			   * the old mapping does not have an amap).
3090 			   * XXX: is it worthwhile to bother with pmap_copy
3091 			   * in this case?
3092 			   */
3093 			  if (old_entry->aref.ar_amap)
3094 			    amap_cow_now(new_map, new_entry);
3095 
3096 			} else {
3097 
3098 			  /*
3099 			   * setup mappings to trigger copy-on-write faults
3100 			   * we must write-protect the parent if it has
3101 			   * an amap and it is not already "needs_copy"...
3102 			   * if it is already "needs_copy" then the parent
3103 			   * has already been write-protected by a previous
3104 			   * fork operation.
3105 			   *
3106 			   * if we do not write-protect the parent, then
3107 			   * we must be sure to write-protect the child
3108 			   * after the pmap_copy() operation.
3109 			   *
3110 			   * XXX: pmap_copy should have some way of telling
3111 			   * us that it didn't do anything so we can avoid
3112 			   * calling pmap_protect needlessly.
3113 			   */
3114 
3115 			  if (old_entry->aref.ar_amap) {
3116 
3117 			    if (!UVM_ET_ISNEEDSCOPY(old_entry)) {
3118 			      if (old_entry->max_protection & VM_PROT_WRITE) {
3119 				pmap_protect(old_map->pmap,
3120 					     old_entry->start,
3121 					     old_entry->end,
3122 					     old_entry->protection &
3123 					     ~VM_PROT_WRITE);
3124 			      }
3125 			      old_entry->etype |= UVM_ET_NEEDSCOPY;
3126 			    }
3127 
3128 			    /*
3129 			     * parent must now be write-protected
3130 			     */
3131 			    protect_child = FALSE;
3132 			  } else {
3133 
3134 			    /*
3135 			     * we only need to protect the child if the
3136 			     * parent has write access.
3137 			     */
3138 			    if (old_entry->max_protection & VM_PROT_WRITE)
3139 			      protect_child = TRUE;
3140 			    else
3141 			      protect_child = FALSE;
3142 
3143 			  }
3144 
3145 			  /*
3146 			   * copy the mappings
3147 			   * XXX: need a way to tell if this does anything
3148 			   */
3149 
3150 			  pmap_copy(new_pmap, old_map->pmap,
3151 				    new_entry->start,
3152 				    (old_entry->end - old_entry->start),
3153 				    old_entry->start);
3154 
3155 			  /*
3156 			   * protect the child's mappings if necessary
3157 			   */
3158 			  if (protect_child) {
3159 			    pmap_protect(new_pmap, new_entry->start,
3160 					 new_entry->end,
3161 					 new_entry->protection &
3162 					          ~VM_PROT_WRITE);
3163 			  }
3164 
3165 			}
3166 			break;
3167 		}  /* end of switch statement */
3168 		old_entry = old_entry->next;
3169 	}
3170 
3171 	new_map->size = old_map->size;
3172 	vm_map_unlock(old_map);
3173 
3174 #ifdef SYSVSHM
3175 	if (vm1->vm_shm)
3176 		shmfork(vm1, vm2);
3177 #endif
3178 
3179 #ifdef PMAP_FORK
3180 	pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
3181 #endif
3182 
3183 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3184 	return(vm2);
3185 }
3186 
3187 
3188 #if defined(DDB)
3189 
3190 /*
3191  * DDB hooks
3192  */
3193 
3194 /*
3195  * uvm_map_print: print out a map
3196  */
3197 
3198 void
3199 uvm_map_print(map, full)
3200 	vm_map_t map;
3201 	boolean_t full;
3202 {
3203 
3204 	uvm_map_printit(map, full, printf);
3205 }
3206 
3207 /*
3208  * uvm_map_printit: actually prints the map
3209  */
3210 
3211 void
3212 uvm_map_printit(map, full, pr)
3213 	vm_map_t map;
3214 	boolean_t full;
3215 	int (*pr) __P((const char *, ...));
3216 {
3217 	vm_map_entry_t entry;
3218 
3219 	(*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
3220 	(*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
3221 	    map->nentries, map->size, map->ref_count, map->timestamp,
3222 	    map->flags);
3223 #ifdef pmap_resident_count
3224 	(*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
3225 	    pmap_resident_count(map->pmap));
3226 #else
3227 	/* XXXCDC: this should be required ... */
3228 	(*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap);
3229 #endif
3230 	if (!full)
3231 		return;
3232 	for (entry = map->header.next; entry != &map->header;
3233 	    entry = entry->next) {
3234 		(*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
3235 		    entry, entry->start, entry->end, entry->object.uvm_obj,
3236 		    (long long)entry->offset, entry->aref.ar_amap, entry->aref.ar_pageoff);
3237 		(*pr)(
3238 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, wc=%d, adv=%d\n",
3239 		    (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
3240 		    (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
3241 		    (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
3242 		    entry->protection, entry->max_protection,
3243 		    entry->inheritance, entry->wired_count, entry->advice);
3244 	}
3245 }
3246 
3247 /*
3248  * uvm_object_print: print out an object
3249  */
3250 
3251 void
3252 uvm_object_print(uobj, full)
3253 	struct uvm_object *uobj;
3254 	boolean_t full;
3255 {
3256 
3257 	uvm_object_printit(uobj, full, printf);
3258 }
3259 
3260 /*
3261  * uvm_object_printit: actually prints the object
3262  */
3263 
3264 void
3265 uvm_object_printit(uobj, full, pr)
3266 	struct uvm_object *uobj;
3267 	boolean_t full;
3268 	int (*pr) __P((const char *, ...));
3269 {
3270 	struct vm_page *pg;
3271 	int cnt = 0;
3272 
3273 	(*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
3274 	    uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
3275 	if (UVM_OBJ_IS_KERN_OBJECT(uobj))
3276 		(*pr)("refs=<SYSTEM>\n");
3277 	else
3278 		(*pr)("refs=%d\n", uobj->uo_refs);
3279 
3280 	if (!full) {
3281 		return;
3282 	}
3283 	(*pr)("  PAGES <pg,offset>:\n  ");
3284 	for (pg = TAILQ_FIRST(&uobj->memq);
3285 	     pg != NULL;
3286 	     pg = TAILQ_NEXT(pg, listq), cnt++) {
3287 		(*pr)("<%p,0x%lx> ", pg, pg->offset);
3288 		if ((cnt % 3) == 2) {
3289 			(*pr)("\n  ");
3290 		}
3291 	}
3292 	if ((cnt % 3) != 2) {
3293 		(*pr)("\n");
3294 	}
3295 }
3296 
3297 const char page_flagbits[] =
3298 	"\20\4CLEAN\5BUSY\6WANTED\7TABLED\12FAKE\13FILLED\14DIRTY\15RELEASED"
3299 	"\16FAULTING\17CLEANCHK";
3300 const char page_pqflagbits[] =
3301 	"\20\1FREE\2INACTIVE\3ACTIVE\4LAUNDRY\5ANON\6AOBJ";
3302 
3303 /*
3304  * uvm_page_print: print out a page
3305  */
3306 
3307 void
3308 uvm_page_print(pg, full)
3309 	struct vm_page *pg;
3310 	boolean_t full;
3311 {
3312 
3313 	uvm_page_printit(pg, full, printf);
3314 }
3315 
3316 /*
3317  * uvm_page_printit: actually print the page
3318  */
3319 
3320 void
3321 uvm_page_printit(pg, full, pr)
3322 	struct vm_page *pg;
3323 	boolean_t full;
3324 	int (*pr) __P((const char *, ...));
3325 {
3326 	struct vm_page *lcv;
3327 	struct uvm_object *uobj;
3328 	struct pglist *pgl;
3329 	char pgbuf[128];
3330 	char pqbuf[128];
3331 
3332 	(*pr)("PAGE %p:\n", pg);
3333 	snprintf(pgbuf, sizeof(pgbuf), "%b", pg->flags, page_flagbits);
3334 	snprintf(pqbuf, sizeof(pqbuf), "%b", pg->pqflags, page_pqflagbits);
3335 	(*pr)("  flags=%s, pqflags=%s, vers=%d, wire_count=%d, pa=0x%lx\n",
3336 	    pgbuf, pqbuf, pg->version, pg->wire_count, (long)pg->phys_addr);
3337 	(*pr)("  uobject=%p, uanon=%p, offset=0x%lx loan_count=%d\n",
3338 	    pg->uobject, pg->uanon, pg->offset, pg->loan_count);
3339 #if defined(UVM_PAGE_TRKOWN)
3340 	if (pg->flags & PG_BUSY)
3341 		(*pr)("  owning process = %d, tag=%s\n",
3342 		    pg->owner, pg->owner_tag);
3343 	else
3344 		(*pr)("  page not busy, no owner\n");
3345 #else
3346 	(*pr)("  [page ownership tracking disabled]\n");
3347 #endif
3348 
3349 	if (!full)
3350 		return;
3351 
3352 	/* cross-verify object/anon */
3353 	if ((pg->pqflags & PQ_FREE) == 0) {
3354 		if (pg->pqflags & PQ_ANON) {
3355 			if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
3356 			    (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
3357 				(pg->uanon) ? pg->uanon->u.an_page : NULL);
3358 			else
3359 				(*pr)("  anon backpointer is OK\n");
3360 		} else {
3361 			uobj = pg->uobject;
3362 			if (uobj) {
3363 				(*pr)("  checking object list\n");
3364 				for (lcv = uobj->memq.tqh_first ; lcv ;
3365 				    lcv = lcv->listq.tqe_next) {
3366 					if (lcv == pg) break;
3367 				}
3368 				if (lcv)
3369 					(*pr)("  page found on object list\n");
3370 				else
3371 			(*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
3372 			}
3373 		}
3374 	}
3375 
3376 	/* cross-verify page queue */
3377 	if (pg->pqflags & PQ_FREE) {
3378 		int fl = uvm_page_lookup_freelist(pg);
3379 		pgl = &uvm.page_free[fl].pgfl_queues[((pg)->flags & PG_ZERO) ?
3380 		    PGFL_ZEROS : PGFL_UNKNOWN];
3381 	}
3382 	else if (pg->pqflags & PQ_INACTIVE)
3383 		pgl = (pg->pqflags & PQ_SWAPBACKED) ?
3384 		    &uvm.page_inactive_swp : &uvm.page_inactive_obj;
3385 	else if (pg->pqflags & PQ_ACTIVE)
3386 		pgl = &uvm.page_active;
3387 	else
3388 		pgl = NULL;
3389 
3390 	if (pgl) {
3391 		(*pr)("  checking pageq list\n");
3392 		for (lcv = pgl->tqh_first ; lcv ; lcv = lcv->pageq.tqe_next) {
3393 			if (lcv == pg) break;
3394 		}
3395 		if (lcv)
3396 			(*pr)("  page found on pageq list\n");
3397 		else
3398 			(*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
3399 	}
3400 }
3401 #endif
3402