xref: /openbsd-src/sys/uvm/uvm_map.c (revision b2ea75c1b17e1a9a339660e7ed45cd24946b230e)
1 /*	$OpenBSD: uvm_map.c,v 1.23 2001/08/11 10:57:22 art 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 #include <vm/vm_kern.h>
88 
89 #define UVM_MAP
90 #include <uvm/uvm.h>
91 
92 #ifdef DDB
93 #include <uvm/uvm_ddb.h>
94 #endif
95 
96 
97 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge;
98 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint;
99 
100 /*
101  * pool for vmspace structures.
102  */
103 
104 struct pool uvm_vmspace_pool;
105 
106 /*
107  * pool for dynamically-allocated map entries.
108  */
109 
110 struct pool uvm_map_entry_pool;
111 
112 #ifdef PMAP_GROWKERNEL
113 /*
114  * This global represents the end of the kernel virtual address
115  * space.  If we want to exceed this, we must grow the kernel
116  * virtual address space dynamically.
117  *
118  * Note, this variable is locked by kernel_map's lock.
119  */
120 vaddr_t uvm_maxkaddr;
121 #endif
122 
123 /*
124  * macros
125  */
126 
127 /*
128  * uvm_map_entry_link: insert entry into a map
129  *
130  * => map must be locked
131  */
132 #define uvm_map_entry_link(map, after_where, entry) do { \
133 	(map)->nentries++; \
134 	(entry)->prev = (after_where); \
135 	(entry)->next = (after_where)->next; \
136 	(entry)->prev->next = (entry); \
137 	(entry)->next->prev = (entry); \
138 } while (0)
139 
140 /*
141  * uvm_map_entry_unlink: remove entry from a map
142  *
143  * => map must be locked
144  */
145 #define uvm_map_entry_unlink(map, entry) do { \
146 	(map)->nentries--; \
147 	(entry)->next->prev = (entry)->prev; \
148 	(entry)->prev->next = (entry)->next; \
149 } while (0)
150 
151 /*
152  * SAVE_HINT: saves the specified entry as the hint for future lookups.
153  *
154  * => map need not be locked (protected by hint_lock).
155  */
156 #define SAVE_HINT(map,value) do { \
157 	simple_lock(&(map)->hint_lock); \
158 	(map)->hint = (value); \
159 	simple_unlock(&(map)->hint_lock); \
160 } while (0)
161 
162 /*
163  * VM_MAP_RANGE_CHECK: check and correct range
164  *
165  * => map must at least be read locked
166  */
167 
168 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
169 	if (start < vm_map_min(map)) 		\
170 		start = vm_map_min(map);        \
171 	if (end > vm_map_max(map))              \
172 		end = vm_map_max(map);          \
173 	if (start > end)                        \
174 		start = end;                    \
175 } while (0)
176 
177 /*
178  * local prototypes
179  */
180 
181 static vm_map_entry_t	uvm_mapent_alloc __P((vm_map_t));
182 static void		uvm_mapent_copy __P((vm_map_entry_t,vm_map_entry_t));
183 static void		uvm_mapent_free __P((vm_map_entry_t));
184 static void		uvm_map_entry_unwire __P((vm_map_t, vm_map_entry_t));
185 
186 /*
187  * local inlines
188  */
189 
190 /*
191  * uvm_mapent_alloc: allocate a map entry
192  *
193  * => XXX: static pool for kernel map?
194  */
195 
196 static __inline vm_map_entry_t
197 uvm_mapent_alloc(map)
198 	vm_map_t map;
199 {
200 	vm_map_entry_t me;
201 	int s;
202 	UVMHIST_FUNC("uvm_mapent_alloc");
203 	UVMHIST_CALLED(maphist);
204 
205 	if ((map->flags & VM_MAP_INTRSAFE) == 0 &&
206 	    map != kernel_map && kernel_map != NULL /* XXX */) {
207 		me = pool_get(&uvm_map_entry_pool, PR_WAITOK);
208 		me->flags = 0;
209 		/* me can't be null, wait ok */
210 	} else {
211 		s = splimp();	/* protect kentry_free list with splimp */
212 		simple_lock(&uvm.kentry_lock);
213 		me = uvm.kentry_free;
214 		if (me) uvm.kentry_free = me->next;
215 		simple_unlock(&uvm.kentry_lock);
216 		splx(s);
217 		if (!me)
218 	panic("mapent_alloc: out of static map entries, check MAX_KMAPENT");
219 		me->flags = UVM_MAP_STATIC;
220 	}
221 
222 	UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]",
223 		me, ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map)
224 		? TRUE : FALSE, 0, 0);
225 	return(me);
226 }
227 
228 /*
229  * uvm_mapent_free: free map entry
230  *
231  * => XXX: static pool for kernel map?
232  */
233 
234 static __inline void
235 uvm_mapent_free(me)
236 	vm_map_entry_t me;
237 {
238 	int s;
239 	UVMHIST_FUNC("uvm_mapent_free");
240 	UVMHIST_CALLED(maphist);
241 	UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
242 		me, me->flags, 0, 0);
243 	if ((me->flags & UVM_MAP_STATIC) == 0) {
244 		pool_put(&uvm_map_entry_pool, me);
245 	} else {
246 		s = splimp();	/* protect kentry_free list with splimp */
247 		simple_lock(&uvm.kentry_lock);
248 		me->next = uvm.kentry_free;
249 		uvm.kentry_free = me;
250 		simple_unlock(&uvm.kentry_lock);
251 		splx(s);
252 	}
253 }
254 
255 /*
256  * uvm_mapent_copy: copy a map entry, preserving flags
257  */
258 
259 static __inline void
260 uvm_mapent_copy(src, dst)
261 	vm_map_entry_t src;
262 	vm_map_entry_t dst;
263 {
264 
265 	memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - ((char*)src));
266 }
267 
268 /*
269  * uvm_map_entry_unwire: unwire a map entry
270  *
271  * => map should be locked by caller
272  */
273 
274 static __inline void
275 uvm_map_entry_unwire(map, entry)
276 	vm_map_t map;
277 	vm_map_entry_t entry;
278 {
279 
280 	entry->wired_count = 0;
281 	uvm_fault_unwire_locked(map, entry->start, entry->end);
282 }
283 
284 /*
285  * uvm_map_init: init mapping system at boot time.   note that we allocate
286  * and init the static pool of vm_map_entry_t's for the kernel here.
287  */
288 
289 void
290 uvm_map_init()
291 {
292 	static struct vm_map_entry kernel_map_entry[MAX_KMAPENT];
293 #if defined(UVMHIST)
294 	static struct uvm_history_ent maphistbuf[100];
295 	static struct uvm_history_ent pdhistbuf[100];
296 #endif
297 	int lcv;
298 
299 	/*
300 	 * first, init logging system.
301 	 */
302 
303 	UVMHIST_FUNC("uvm_map_init");
304 	UVMHIST_INIT_STATIC(maphist, maphistbuf);
305 	UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
306 	UVMHIST_CALLED(maphist);
307 	UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
308 	UVMCNT_INIT(uvm_map_call,  UVMCNT_CNT, 0,
309 	    "# uvm_map() successful calls", 0);
310 	UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0);
311 	UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward",
312 	    0);
313 	UVMCNT_INIT(uvm_mlk_call,  UVMCNT_CNT, 0, "# map lookup calls", 0);
314 	UVMCNT_INIT(uvm_mlk_hint,  UVMCNT_CNT, 0, "# map lookup hint hits", 0);
315 
316 	/*
317 	 * now set up static pool of kernel map entrys ...
318 	 */
319 
320 	simple_lock_init(&uvm.kentry_lock);
321 	uvm.kentry_free = NULL;
322 	for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) {
323 		kernel_map_entry[lcv].next = uvm.kentry_free;
324 		uvm.kentry_free = &kernel_map_entry[lcv];
325 	}
326 
327 	/*
328 	 * initialize the map-related pools.
329 	 */
330 	pool_init(&uvm_vmspace_pool, sizeof(struct vmspace),
331 	    0, 0, 0, "vmsppl", 0,
332 	    pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
333 	pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry),
334 	    0, 0, 0, "vmmpepl", 0,
335 	    pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
336 }
337 
338 /*
339  * clippers
340  */
341 
342 /*
343  * uvm_map_clip_start: ensure that the entry begins at or after
344  *	the starting address, if it doesn't we split the entry.
345  *
346  * => caller should use UVM_MAP_CLIP_START macro rather than calling
347  *    this directly
348  * => map must be locked by caller
349  */
350 
351 void uvm_map_clip_start(map, entry, start)
352 	vm_map_t       map;
353 	vm_map_entry_t entry;
354 	vaddr_t    start;
355 {
356 	vm_map_entry_t new_entry;
357 	vaddr_t new_adj;
358 
359 	/* uvm_map_simplify_entry(map, entry); */ /* XXX */
360 
361 	/*
362 	 * Split off the front portion.  note that we must insert the new
363 	 * entry BEFORE this one, so that this entry has the specified
364 	 * starting address.
365 	 */
366 
367 	new_entry = uvm_mapent_alloc(map);
368 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
369 
370 	new_entry->end = start;
371 	new_adj = start - new_entry->start;
372 	if (entry->object.uvm_obj)
373 		entry->offset += new_adj;	/* shift start over */
374 	entry->start = start;
375 
376 	if (new_entry->aref.ar_amap) {
377 		amap_splitref(&new_entry->aref, &entry->aref, new_adj);
378 	}
379 
380 	uvm_map_entry_link(map, entry->prev, new_entry);
381 
382 	if (UVM_ET_ISSUBMAP(entry)) {
383 		/* ... unlikely to happen, but play it safe */
384 		 uvm_map_reference(new_entry->object.sub_map);
385 	} else {
386 		if (UVM_ET_ISOBJ(entry) &&
387 		    entry->object.uvm_obj->pgops &&
388 		    entry->object.uvm_obj->pgops->pgo_reference)
389 			entry->object.uvm_obj->pgops->pgo_reference(
390 			    entry->object.uvm_obj);
391 	}
392 }
393 
394 /*
395  * uvm_map_clip_end: ensure that the entry ends at or before
396  *	the ending address, if it does't we split the reference
397  *
398  * => caller should use UVM_MAP_CLIP_END macro rather than calling
399  *    this directly
400  * => map must be locked by caller
401  */
402 
403 void
404 uvm_map_clip_end(map, entry, end)
405 	vm_map_t	map;
406 	vm_map_entry_t	entry;
407 	vaddr_t	end;
408 {
409 	vm_map_entry_t	new_entry;
410 	vaddr_t new_adj; /* #bytes we move start forward */
411 
412 	/*
413 	 *	Create a new entry and insert it
414 	 *	AFTER the specified entry
415 	 */
416 
417 	new_entry = uvm_mapent_alloc(map);
418 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
419 
420 	new_entry->start = entry->end = end;
421 	new_adj = end - entry->start;
422 	if (new_entry->object.uvm_obj)
423 		new_entry->offset += new_adj;
424 
425 	if (entry->aref.ar_amap)
426 		amap_splitref(&entry->aref, &new_entry->aref, new_adj);
427 
428 	uvm_map_entry_link(map, entry, new_entry);
429 
430 	if (UVM_ET_ISSUBMAP(entry)) {
431 		/* ... unlikely to happen, but play it safe */
432 	 	uvm_map_reference(new_entry->object.sub_map);
433 	} else {
434 		if (UVM_ET_ISOBJ(entry) &&
435 		    entry->object.uvm_obj->pgops &&
436 		    entry->object.uvm_obj->pgops->pgo_reference)
437 			entry->object.uvm_obj->pgops->pgo_reference(
438 			    entry->object.uvm_obj);
439 	}
440 }
441 
442 
443 /*
444  *   M A P   -   m a i n   e n t r y   p o i n t
445  */
446 /*
447  * uvm_map: establish a valid mapping in a map
448  *
449  * => assume startp is page aligned.
450  * => assume size is a multiple of PAGE_SIZE.
451  * => assume sys_mmap provides enough of a "hint" to have us skip
452  *	over text/data/bss area.
453  * => map must be unlocked (we will lock it)
454  * => <uobj,uoffset> value meanings (4 cases):
455  *	 [1] <NULL,uoffset> 		== uoffset is a hint for PMAP_PREFER
456  *	 [2] <NULL,UVM_UNKNOWN_OFFSET>	== don't PMAP_PREFER
457  *	 [3] <uobj,uoffset>		== normal mapping
458  *	 [4] <uobj,UVM_UNKNOWN_OFFSET>	== uvm_map finds offset based on VA
459  *
460  *    case [4] is for kernel mappings where we don't know the offset until
461  *    we've found a virtual address.   note that kernel object offsets are
462  *    always relative to vm_map_min(kernel_map).
463  * => XXXCDC: need way to map in external amap?
464  */
465 
466 int
467 uvm_map(map, startp, size, uobj, uoffset, flags)
468 	vm_map_t map;
469 	vaddr_t *startp;	/* IN/OUT */
470 	vsize_t size;
471 	struct uvm_object *uobj;
472 	voff_t uoffset;
473 	uvm_flag_t flags;
474 {
475 	vm_map_entry_t prev_entry, new_entry;
476 	vm_prot_t prot = UVM_PROTECTION(flags), maxprot =
477 	    UVM_MAXPROTECTION(flags);
478 	vm_inherit_t inherit = UVM_INHERIT(flags);
479 	int advice = UVM_ADVICE(flags);
480 	UVMHIST_FUNC("uvm_map");
481 	UVMHIST_CALLED(maphist);
482 
483 	UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)",
484 	    map, *startp, size, flags);
485 	UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
486 
487 	/*
488 	 * step 0: sanity check of protection code
489 	 */
490 
491 	if ((prot & maxprot) != prot) {
492 		UVMHIST_LOG(maphist, "<- prot. failure:  prot=0x%x, max=0x%x",
493 		prot, maxprot,0,0);
494 		return(KERN_PROTECTION_FAILURE);
495 	}
496 
497 	/*
498 	 * step 1: figure out where to put new VM range
499 	 */
500 
501 	if (vm_map_lock_try(map) == FALSE) {
502 		if (flags & UVM_FLAG_TRYLOCK)
503 			return(KERN_FAILURE);
504 		vm_map_lock(map); /* could sleep here */
505 	}
506 	if ((prev_entry = uvm_map_findspace(map, *startp, size, startp,
507 	    uobj, uoffset, flags & UVM_FLAG_FIXED)) == NULL) {
508 		UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0);
509 		vm_map_unlock(map);
510 		return (KERN_NO_SPACE);
511 	}
512 
513 #ifdef PMAP_GROWKERNEL
514 	{
515 		/*
516 		 * If the kernel pmap can't map the requested space,
517 		 * then allocate more resources for it.
518 		 */
519 		if (map == kernel_map && uvm_maxkaddr < (*startp + size))
520 			uvm_maxkaddr = pmap_growkernel(*startp + size);
521 	}
522 #endif
523 
524 	UVMCNT_INCR(uvm_map_call);
525 
526 	/*
527 	 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
528 	 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET.   in
529 	 * either case we want to zero it  before storing it in the map entry
530 	 * (because it looks strange and confusing when debugging...)
531 	 *
532 	 * if uobj is not null
533 	 *   if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
534 	 *      and we do not need to change uoffset.
535 	 *   if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
536 	 *      now (based on the starting address of the map).   this case is
537 	 *      for kernel object mappings where we don't know the offset until
538 	 *      the virtual address is found (with uvm_map_findspace).   the
539 	 *      offset is the distance we are from the start of the map.
540 	 */
541 
542 	if (uobj == NULL) {
543 		uoffset = 0;
544 	} else {
545 		if (uoffset == UVM_UNKNOWN_OFFSET) {
546 #ifdef DIAGNOSTIC
547 			if (UVM_OBJ_IS_KERN_OBJECT(uobj) == 0)
548 				panic("uvm_map: unknown offset with "
549 				    "non-kernel object");
550 #endif
551 			uoffset = *startp - vm_map_min(kernel_map);
552 		}
553 	}
554 
555 	/*
556 	 * step 2: try and insert in map by extending previous entry, if
557 	 * possible
558 	 * XXX: we don't try and pull back the next entry.   might be useful
559 	 * for a stack, but we are currently allocating our stack in advance.
560 	 */
561 
562 	if ((flags & UVM_FLAG_NOMERGE) == 0 &&
563 	    prev_entry->end == *startp && prev_entry != &map->header &&
564 	    prev_entry->object.uvm_obj == uobj) {
565 
566 		if (uobj && prev_entry->offset +
567 		    (prev_entry->end - prev_entry->start) != uoffset)
568 			goto step3;
569 
570 		if (UVM_ET_ISSUBMAP(prev_entry))
571 			goto step3;
572 
573 		if (prev_entry->protection != prot ||
574 		    prev_entry->max_protection != maxprot)
575 			goto step3;
576 
577 		if (prev_entry->inheritance != inherit ||
578 		    prev_entry->advice != advice)
579 			goto step3;
580 
581 		/* wiring status must match (new area is unwired) */
582 		if (VM_MAPENT_ISWIRED(prev_entry))
583 			goto step3;
584 
585 		/*
586 		 * can't extend a shared amap.  note: no need to lock amap to
587 		 * look at refs since we don't care about its exact value.
588 		 * if it is one (i.e. we have only reference) it will stay there
589 		 */
590 
591 		if (prev_entry->aref.ar_amap &&
592 		    amap_refs(prev_entry->aref.ar_amap) != 1) {
593 			goto step3;
594 		}
595 
596 		/* got it! */
597 
598 		UVMCNT_INCR(map_backmerge);
599 		UVMHIST_LOG(maphist,"  starting back merge", 0, 0, 0, 0);
600 
601 		/*
602 		 * drop our reference to uobj since we are extending a reference
603 		 * that we already have (the ref count can not drop to zero).
604 		 */
605 		if (uobj && uobj->pgops->pgo_detach)
606 			uobj->pgops->pgo_detach(uobj);
607 
608 		if (prev_entry->aref.ar_amap) {
609 			amap_extend(prev_entry, size);
610 		}
611 
612 		prev_entry->end += size;
613 		map->size += size;
614 
615 		UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
616 		vm_map_unlock(map);
617 		return (KERN_SUCCESS);
618 
619 	}
620 step3:
621 	UVMHIST_LOG(maphist,"  allocating new map entry", 0, 0, 0, 0);
622 
623 	/*
624 	 * check for possible forward merge (which we don't do) and count
625 	 * the number of times we missed a *possible* chance to merge more
626 	 */
627 
628 	if ((flags & UVM_FLAG_NOMERGE) == 0 &&
629 	    prev_entry->next != &map->header &&
630 	    prev_entry->next->start == (*startp + size))
631 		UVMCNT_INCR(map_forwmerge);
632 
633 	/*
634 	 * step 3: allocate new entry and link it in
635 	 */
636 
637 	new_entry = uvm_mapent_alloc(map);
638 	new_entry->start = *startp;
639 	new_entry->end = new_entry->start + size;
640 	new_entry->object.uvm_obj = uobj;
641 	new_entry->offset = uoffset;
642 
643 	if (uobj)
644 		new_entry->etype = UVM_ET_OBJ;
645 	else
646 		new_entry->etype = 0;
647 
648 	if (flags & UVM_FLAG_COPYONW) {
649 		new_entry->etype |= UVM_ET_COPYONWRITE;
650 		if ((flags & UVM_FLAG_OVERLAY) == 0)
651 			new_entry->etype |= UVM_ET_NEEDSCOPY;
652 	}
653 
654 	new_entry->protection = prot;
655 	new_entry->max_protection = maxprot;
656 	new_entry->inheritance = inherit;
657 	new_entry->wired_count = 0;
658 	new_entry->advice = advice;
659 	if (flags & UVM_FLAG_OVERLAY) {
660 		/*
661 		 * to_add: for BSS we overallocate a little since we
662 		 * are likely to extend
663 		 */
664 		vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
665 			UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
666 		struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK);
667 		new_entry->aref.ar_pageoff = 0;
668 		new_entry->aref.ar_amap = amap;
669 	} else {
670 		new_entry->aref.ar_pageoff = 0;
671 		new_entry->aref.ar_amap = NULL;
672 	}
673 
674 	uvm_map_entry_link(map, prev_entry, new_entry);
675 
676 	map->size += size;
677 
678 	/*
679 	 *      Update the free space hint
680 	 */
681 
682 	if ((map->first_free == prev_entry) &&
683 	    (prev_entry->end >= new_entry->start))
684 		map->first_free = new_entry;
685 
686 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
687 	vm_map_unlock(map);
688 	return(KERN_SUCCESS);
689 }
690 
691 /*
692  * uvm_map_lookup_entry: find map entry at or before an address
693  *
694  * => map must at least be read-locked by caller
695  * => entry is returned in "entry"
696  * => return value is true if address is in the returned entry
697  */
698 
699 boolean_t
700 uvm_map_lookup_entry(map, address, entry)
701 	vm_map_t	map;
702 	vaddr_t	address;
703 	vm_map_entry_t		*entry;		/* OUT */
704 {
705 	vm_map_entry_t		cur;
706 	vm_map_entry_t		last;
707 	UVMHIST_FUNC("uvm_map_lookup_entry");
708 	UVMHIST_CALLED(maphist);
709 
710 	UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
711 	    map, address, entry, 0);
712 
713 	/*
714 	 * start looking either from the head of the
715 	 * list, or from the hint.
716 	 */
717 
718 	simple_lock(&map->hint_lock);
719 	cur = map->hint;
720 	simple_unlock(&map->hint_lock);
721 
722 	if (cur == &map->header)
723 		cur = cur->next;
724 
725 	UVMCNT_INCR(uvm_mlk_call);
726 	if (address >= cur->start) {
727 	    	/*
728 		 * go from hint to end of list.
729 		 *
730 		 * but first, make a quick check to see if
731 		 * we are already looking at the entry we
732 		 * want (which is usually the case).
733 		 * note also that we don't need to save the hint
734 		 * here... it is the same hint (unless we are
735 		 * at the header, in which case the hint didn't
736 		 * buy us anything anyway).
737 		 */
738 		last = &map->header;
739 		if ((cur != last) && (cur->end > address)) {
740 			UVMCNT_INCR(uvm_mlk_hint);
741 			*entry = cur;
742 			UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
743 			    cur, 0, 0, 0);
744 			return (TRUE);
745 		}
746 	} else {
747 	    	/*
748 		 * go from start to hint, *inclusively*
749 		 */
750 		last = cur->next;
751 		cur = map->header.next;
752 	}
753 
754 	/*
755 	 * search linearly
756 	 */
757 
758 	while (cur != last) {
759 		if (cur->end > address) {
760 			if (address >= cur->start) {
761 			    	/*
762 				 * save this lookup for future
763 				 * hints, and return
764 				 */
765 
766 				*entry = cur;
767 				SAVE_HINT(map, cur);
768 				UVMHIST_LOG(maphist,"<- search got it (0x%x)",
769 					cur, 0, 0, 0);
770 				return (TRUE);
771 			}
772 			break;
773 		}
774 		cur = cur->next;
775 	}
776 	*entry = cur->prev;
777 	SAVE_HINT(map, *entry);
778 	UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
779 	return (FALSE);
780 }
781 
782 
783 /*
784  * uvm_map_findspace: find "length" sized space in "map".
785  *
786  * => "hint" is a hint about where we want it, unless fixed is true
787  *	(in which case we insist on using "hint").
788  * => "result" is VA returned
789  * => uobj/uoffset are to be used to handle VAC alignment, if required
790  * => caller must at least have read-locked map
791  * => returns NULL on failure, or pointer to prev. map entry if success
792  * => note this is a cross between the old vm_map_findspace and vm_map_find
793  */
794 
795 vm_map_entry_t
796 uvm_map_findspace(map, hint, length, result, uobj, uoffset, fixed)
797 	vm_map_t map;
798 	vaddr_t hint;
799 	vsize_t length;
800 	vaddr_t *result; /* OUT */
801 	struct uvm_object *uobj;
802 	voff_t uoffset;
803 	boolean_t fixed;
804 {
805 	vm_map_entry_t entry, next, tmp;
806 	vaddr_t end;
807 	UVMHIST_FUNC("uvm_map_findspace");
808 	UVMHIST_CALLED(maphist);
809 
810 	UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, fixed=%d)",
811 		map, hint, length, fixed);
812 
813 	if (hint < map->min_offset) {	/* check ranges ... */
814 		if (fixed) {
815 			UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
816 			return(NULL);
817 		}
818 		hint = map->min_offset;
819 	}
820 	if (hint > map->max_offset) {
821 		UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
822 				hint, map->min_offset, map->max_offset, 0);
823 		return(NULL);
824 	}
825 
826 	/*
827 	 * Look for the first possible address; if there's already
828 	 * something at this address, we have to start after it.
829 	 */
830 
831 	if (!fixed && hint == map->min_offset) {
832 		if ((entry = map->first_free) != &map->header)
833 			hint = entry->end;
834 	} else {
835 		if (uvm_map_lookup_entry(map, hint, &tmp)) {
836 			/* "hint" address already in use ... */
837 			if (fixed) {
838 				UVMHIST_LOG(maphist,"<- fixed & VA in use",
839 				    0, 0, 0, 0);
840 				return(NULL);
841 			}
842 			hint = tmp->end;
843 		}
844 		entry = tmp;
845 	}
846 
847 	/*
848 	 * Look through the rest of the map, trying to fit a new region in
849 	 * the gap between existing regions, or after the very last region.
850 	 * note: entry->end   = base VA of current gap,
851 	 *	 next->start  = VA of end of current gap
852 	 */
853 	for (;; hint = (entry = next)->end) {
854 		/*
855 		 * Find the end of the proposed new region.  Be sure we didn't
856 		 * go beyond the end of the map, or wrap around the address;
857 		 * if so, we lose.  Otherwise, if this is the last entry, or
858 		 * if the proposed new region fits before the next entry, we
859 		 * win.
860 		 */
861 
862 #ifdef PMAP_PREFER
863 		/*
864 		 * push hint forward as needed to avoid VAC alias problems.
865 		 * we only do this if a valid offset is specified.
866 		 */
867 		if (!fixed && uoffset != UVM_UNKNOWN_OFFSET)
868 		  PMAP_PREFER(uoffset, &hint);
869 #endif
870 		end = hint + length;
871 		if (end > map->max_offset || end < hint) {
872 			UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0);
873 			return (NULL);
874 		}
875 		next = entry->next;
876 		if (next == &map->header || next->start >= end)
877 			break;
878 		if (fixed) {
879 			UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0);
880 			return(NULL); /* only one shot at it ... */
881 		}
882 	}
883 	SAVE_HINT(map, entry);
884 	*result = hint;
885 	UVMHIST_LOG(maphist,"<- got it!  (result=0x%x)", hint, 0,0,0);
886 	return (entry);
887 }
888 
889 /*
890  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
891  */
892 
893 /*
894  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
895  *
896  * => caller must check alignment and size
897  * => map must be locked by caller
898  * => we return a list of map entries that we've remove from the map
899  *    in "entry_list"
900  */
901 
902 int
903 uvm_unmap_remove(map, start, end, entry_list)
904 	vm_map_t map;
905 	vaddr_t start,end;
906 	vm_map_entry_t *entry_list;	/* OUT */
907 {
908 	vm_map_entry_t entry, first_entry, next;
909 	vaddr_t len;
910 	UVMHIST_FUNC("uvm_unmap_remove");
911 	UVMHIST_CALLED(maphist);
912 
913 	UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
914 	    map, start, end, 0);
915 
916 	VM_MAP_RANGE_CHECK(map, start, end);
917 
918 	/*
919 	 * find first entry
920 	 */
921 	if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
922 		/* clip and go... */
923 		entry = first_entry;
924 		UVM_MAP_CLIP_START(map, entry, start);
925 		/* critical!  prevents stale hint */
926 		SAVE_HINT(map, entry->prev);
927 
928 	} else {
929 		entry = first_entry->next;
930 	}
931 
932 	/*
933 	 * Save the free space hint
934 	 */
935 
936 	if (map->first_free->start >= start)
937 		map->first_free = entry->prev;
938 
939 	/*
940 	 * note: we now re-use first_entry for a different task.  we remove
941 	 * a number of map entries from the map and save them in a linked
942 	 * list headed by "first_entry".  once we remove them from the map
943 	 * the caller should unlock the map and drop the references to the
944 	 * backing objects [c.f. uvm_unmap_detach].  the object is to
945 	 * separate unmapping from reference dropping.  why?
946 	 *   [1] the map has to be locked for unmapping
947 	 *   [2] the map need not be locked for reference dropping
948 	 *   [3] dropping references may trigger pager I/O, and if we hit
949 	 *       a pager that does synchronous I/O we may have to wait for it.
950 	 *   [4] we would like all waiting for I/O to occur with maps unlocked
951 	 *       so that we don't block other threads.
952 	 */
953 	first_entry = NULL;
954 	*entry_list = NULL;		/* to be safe */
955 
956 	/*
957 	 * break up the area into map entry sized regions and unmap.  note
958 	 * that all mappings have to be removed before we can even consider
959 	 * dropping references to amaps or VM objects (otherwise we could end
960 	 * up with a mapping to a page on the free list which would be very bad)
961 	 */
962 
963 	while ((entry != &map->header) && (entry->start < end)) {
964 
965 		UVM_MAP_CLIP_END(map, entry, end);
966 		next = entry->next;
967 		len = entry->end - entry->start;
968 
969 		/*
970 		 * unwire before removing addresses from the pmap; otherwise
971 		 * unwiring will put the entries back into the pmap (XXX).
972 		 */
973 
974 		if (VM_MAPENT_ISWIRED(entry))
975 			uvm_map_entry_unwire(map, entry);
976 
977 		/*
978 		 * special case: handle mappings to anonymous kernel objects.
979 		 * we want to free these pages right away...
980 		 */
981 		if (UVM_ET_ISOBJ(entry) &&
982 		    UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
983 #ifdef DIAGNOSTIC
984 			if (vm_map_pmap(map) != pmap_kernel())
985 				panic("uvm_unmap_remove: kernel object "
986 				    "mapped by non-kernel map");
987 #endif
988 
989 			/*
990 			 * note: kernel object mappings are currently used in
991 			 * two ways:
992 			 *  [1] "normal" mappings of pages in the kernel object
993 			 *  [2] uvm_km_valloc'd allocations in which we
994 			 *      pmap_enter in some non-kernel-object page
995 			 *      (e.g. vmapbuf).
996 			 *
997 			 * for case [1], we need to remove the mapping from
998 			 * the pmap and then remove the page from the kernel
999 			 * object (because, once pages in a kernel object are
1000 			 * unmapped they are no longer needed, unlike, say,
1001 			 * a vnode where you might want the data to persist
1002 			 * until flushed out of a queue).
1003 			 *
1004 			 * for case [2], we need to remove the mapping from
1005 			 * the pmap.  there shouldn't be any pages at the
1006 			 * specified offset in the kernel object [but it
1007 			 * doesn't hurt to call uvm_km_pgremove just to be
1008 			 * safe?]
1009 			 *
1010 			 * uvm_km_pgremove currently does the following:
1011 			 *   for pages in the kernel object in range:
1012 			 *     - drops the swap slot
1013 			 *     - uvm_pagefree the page
1014 			 *
1015 			 * note there is version of uvm_km_pgremove() that
1016 			 * is used for "intrsafe" objects.
1017 			 */
1018 
1019 			/*
1020 			 * remove mappings from pmap and drop the pages
1021 			 * from the object.  offsets are always relative
1022 			 * to vm_map_min(kernel_map).
1023 			 */
1024 			if (UVM_OBJ_IS_INTRSAFE_OBJECT(entry->object.uvm_obj)) {
1025 				pmap_kremove(entry->start, len);
1026 				uvm_km_pgremove_intrsafe(entry->object.uvm_obj,
1027 				    entry->start - vm_map_min(kernel_map),
1028 				    entry->end - vm_map_min(kernel_map));
1029 			} else {
1030 				pmap_remove(pmap_kernel(), entry->start,
1031 				    entry->start + len);
1032 				uvm_km_pgremove(entry->object.uvm_obj,
1033 				    entry->start - vm_map_min(kernel_map),
1034 				    entry->end - vm_map_min(kernel_map));
1035 			}
1036 
1037 			/*
1038 			 * null out kernel_object reference, we've just
1039 			 * dropped it
1040 			 */
1041 			entry->etype &= ~UVM_ET_OBJ;
1042 			entry->object.uvm_obj = NULL;	/* to be safe */
1043 
1044 		} else {
1045 			/*
1046 		 	 * remove mappings the standard way.
1047 		 	 */
1048 			pmap_remove(map->pmap, entry->start, entry->end);
1049 		}
1050 
1051 		/*
1052 		 * remove entry from map and put it on our list of entries
1053 		 * that we've nuked.  then go do next entry.
1054 		 */
1055 		UVMHIST_LOG(maphist, "  removed map entry 0x%x", entry, 0, 0,0);
1056 
1057 		/* critical! prevents stale hint */
1058 		/* XXX: need SAVE_HINT with three parms */
1059 		simple_lock(&map->hint_lock);
1060 		if (map->hint == entry)
1061 		    map->hint = entry->prev;
1062 		simple_unlock(&map->hint_lock);
1063 
1064 		uvm_map_entry_unlink(map, entry);
1065 		map->size -= len;
1066 		entry->next = first_entry;
1067 		first_entry = entry;
1068 		entry = next;		/* next entry, please */
1069 	}
1070 
1071 	/*
1072 	 * now we've cleaned up the map and are ready for the caller to drop
1073 	 * references to the mapped objects.
1074 	 */
1075 
1076 	*entry_list = first_entry;
1077 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1078 	return(KERN_SUCCESS);
1079 }
1080 
1081 /*
1082  * uvm_unmap_detach: drop references in a chain of map entries
1083  *
1084  * => we will free the map entries as we traverse the list.
1085  */
1086 
1087 void
1088 uvm_unmap_detach(first_entry, amap_unref_flags)
1089 	vm_map_entry_t first_entry;
1090 	int amap_unref_flags;
1091 {
1092 	vm_map_entry_t next_entry;
1093 	UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1094 
1095 	while (first_entry) {
1096 
1097 #ifdef DIAGNOSTIC
1098 		/*
1099 		 * sanity check
1100 		 */
1101 		/* was part of vm_map_entry_delete() */
1102 		if (VM_MAPENT_ISWIRED(first_entry))
1103 			panic("unmap: still wired!");
1104 #endif
1105 
1106 		UVMHIST_LOG(maphist,
1107 		    "  detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
1108 		    first_entry, first_entry->aref.ar_amap,
1109 		    first_entry->object.uvm_obj,
1110 		    UVM_ET_ISSUBMAP(first_entry));
1111 
1112 		/*
1113 		 * drop reference to amap, if we've got one
1114 		 */
1115 
1116 		if (first_entry->aref.ar_amap)
1117 			amap_unref(first_entry, amap_unref_flags);
1118 
1119 		/*
1120 		 * drop reference to our backing object, if we've got one
1121 		 */
1122 
1123 		if (UVM_ET_ISSUBMAP(first_entry)) {
1124 			/* ... unlikely to happen, but play it safe */
1125 			uvm_map_deallocate(first_entry->object.sub_map);
1126 		} else {
1127 			if (UVM_ET_ISOBJ(first_entry) &&
1128 			    first_entry->object.uvm_obj->pgops->pgo_detach)
1129 				first_entry->object.uvm_obj->pgops->
1130 				    pgo_detach(first_entry->object.uvm_obj);
1131 		}
1132 
1133 		/*
1134 		 * next entry
1135 		 */
1136 		next_entry = first_entry->next;
1137 		uvm_mapent_free(first_entry);
1138 		first_entry = next_entry;
1139 	}
1140 
1141 	/*
1142 	 * done!
1143 	 */
1144 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1145 	return;
1146 }
1147 
1148 /*
1149  *   E X T R A C T I O N   F U N C T I O N S
1150  */
1151 
1152 /*
1153  * uvm_map_reserve: reserve space in a vm_map for future use.
1154  *
1155  * => we reserve space in a map by putting a dummy map entry in the
1156  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
1157  * => map should be unlocked (we will write lock it)
1158  * => we return true if we were able to reserve space
1159  * => XXXCDC: should be inline?
1160  */
1161 
1162 int
1163 uvm_map_reserve(map, size, offset, raddr)
1164 	vm_map_t map;
1165 	vsize_t size;
1166 	vaddr_t offset;    /* hint for pmap_prefer */
1167 	vaddr_t *raddr;	/* IN:hint, OUT: reserved VA */
1168 {
1169 	UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1170 
1171 	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1172 	      map,size,offset,raddr);
1173 
1174 	size = round_page(size);
1175 	if (*raddr < vm_map_min(map))
1176 		*raddr = vm_map_min(map);                /* hint */
1177 
1178 	/*
1179 	 * reserve some virtual space.
1180 	 */
1181 
1182 	if (uvm_map(map, raddr, size, NULL, offset,
1183 	    UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1184 	    UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != KERN_SUCCESS) {
1185 	    UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1186 		return (FALSE);
1187 	}
1188 
1189 	UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1190 	return (TRUE);
1191 }
1192 
1193 /*
1194  * uvm_map_replace: replace a reserved (blank) area of memory with
1195  * real mappings.
1196  *
1197  * => caller must WRITE-LOCK the map
1198  * => we return TRUE if replacement was a success
1199  * => we expect the newents chain to have nnewents entrys on it and
1200  *    we expect newents->prev to point to the last entry on the list
1201  * => note newents is allowed to be NULL
1202  */
1203 
1204 int
1205 uvm_map_replace(map, start, end, newents, nnewents)
1206 	struct vm_map *map;
1207 	vaddr_t start, end;
1208 	vm_map_entry_t newents;
1209 	int nnewents;
1210 {
1211 	vm_map_entry_t oldent, last;
1212 	UVMHIST_FUNC("uvm_map_replace");
1213 	UVMHIST_CALLED(maphist);
1214 
1215 	/*
1216 	 * first find the blank map entry at the specified address
1217 	 */
1218 
1219 	if (!uvm_map_lookup_entry(map, start, &oldent)) {
1220 		return(FALSE);
1221 	}
1222 
1223 	/*
1224 	 * check to make sure we have a proper blank entry
1225 	 */
1226 
1227 	if (oldent->start != start || oldent->end != end ||
1228 	    oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1229 		return (FALSE);
1230 	}
1231 
1232 #ifdef DIAGNOSTIC
1233 	/*
1234 	 * sanity check the newents chain
1235 	 */
1236 	{
1237 		vm_map_entry_t tmpent = newents;
1238 		int nent = 0;
1239 		vaddr_t cur = start;
1240 
1241 		while (tmpent) {
1242 			nent++;
1243 			if (tmpent->start < cur)
1244 				panic("uvm_map_replace1");
1245 			if (tmpent->start > tmpent->end || tmpent->end > end) {
1246 		printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
1247 			    tmpent->start, tmpent->end, end);
1248 				panic("uvm_map_replace2");
1249 			}
1250 			cur = tmpent->end;
1251 			if (tmpent->next) {
1252 				if (tmpent->next->prev != tmpent)
1253 					panic("uvm_map_replace3");
1254 			} else {
1255 				if (newents->prev != tmpent)
1256 					panic("uvm_map_replace4");
1257 			}
1258 			tmpent = tmpent->next;
1259 		}
1260 		if (nent != nnewents)
1261 			panic("uvm_map_replace5");
1262 	}
1263 #endif
1264 
1265 	/*
1266 	 * map entry is a valid blank!   replace it.   (this does all the
1267 	 * work of map entry link/unlink...).
1268 	 */
1269 
1270 	if (newents) {
1271 
1272 		last = newents->prev;		/* we expect this */
1273 
1274 		/* critical: flush stale hints out of map */
1275 		SAVE_HINT(map, newents);
1276 		if (map->first_free == oldent)
1277 			map->first_free = last;
1278 
1279 		last->next = oldent->next;
1280 		last->next->prev = last;
1281 		newents->prev = oldent->prev;
1282 		newents->prev->next = newents;
1283 		map->nentries = map->nentries + (nnewents - 1);
1284 
1285 	} else {
1286 
1287 		/* critical: flush stale hints out of map */
1288 		SAVE_HINT(map, oldent->prev);
1289 		if (map->first_free == oldent)
1290 			map->first_free = oldent->prev;
1291 
1292 		/* NULL list of new entries: just remove the old one */
1293 		uvm_map_entry_unlink(map, oldent);
1294 	}
1295 
1296 
1297 	/*
1298 	 * now we can free the old blank entry, unlock the map and return.
1299 	 */
1300 
1301 	uvm_mapent_free(oldent);
1302 	return(TRUE);
1303 }
1304 
1305 /*
1306  * uvm_map_extract: extract a mapping from a map and put it somewhere
1307  *	(maybe removing the old mapping)
1308  *
1309  * => maps should be unlocked (we will write lock them)
1310  * => returns 0 on success, error code otherwise
1311  * => start must be page aligned
1312  * => len must be page sized
1313  * => flags:
1314  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
1315  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
1316  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
1317  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
1318  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
1319  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
1320  *             be used from within the kernel in a kernel level map <<<
1321  */
1322 
1323 int
1324 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags)
1325 	vm_map_t srcmap, dstmap;
1326 	vaddr_t start, *dstaddrp;
1327 	vsize_t len;
1328 	int flags;
1329 {
1330 	vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge,
1331 	    oldstart;
1332 	vm_map_entry_t chain, endchain, entry, orig_entry, newentry, deadentry;
1333 	vm_map_entry_t oldentry;
1334 	vsize_t elen;
1335 	int nchain, error, copy_ok;
1336 	UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1337 	UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
1338 	    len,0);
1339 	UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
1340 
1341 #ifdef DIAGNOSTIC
1342 	/*
1343 	 * step 0: sanity check: start must be on a page boundary, length
1344 	 * must be page sized.  can't ask for CONTIG/QREF if you asked for
1345 	 * REMOVE.
1346 	 */
1347 	if ((start & PAGE_MASK) || (len & PAGE_MASK))
1348 		panic("uvm_map_extract1");
1349 	if (flags & UVM_EXTRACT_REMOVE)
1350 		if (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF))
1351 			panic("uvm_map_extract2");
1352 #endif
1353 
1354 
1355 	/*
1356 	 * step 1: reserve space in the target map for the extracted area
1357 	 */
1358 
1359 	dstaddr = vm_map_min(dstmap);
1360 	if (uvm_map_reserve(dstmap, len, start, &dstaddr) == FALSE)
1361 		return(ENOMEM);
1362 	*dstaddrp = dstaddr;	/* pass address back to caller */
1363 	UVMHIST_LOG(maphist, "  dstaddr=0x%x", dstaddr,0,0,0);
1364 
1365 
1366 	/*
1367 	 * step 2: setup for the extraction process loop by init'ing the
1368 	 * map entry chain, locking src map, and looking up the first useful
1369 	 * entry in the map.
1370 	 */
1371 
1372 	end = start + len;
1373 	newend = dstaddr + len;
1374 	chain = endchain = NULL;
1375 	nchain = 0;
1376 	vm_map_lock(srcmap);
1377 
1378 	if (uvm_map_lookup_entry(srcmap, start, &entry)) {
1379 
1380 		/* "start" is within an entry */
1381 		if (flags & UVM_EXTRACT_QREF) {
1382 			/*
1383 			 * for quick references we don't clip the entry, so
1384 			 * the entry may map space "before" the starting
1385 			 * virtual address... this is the "fudge" factor
1386 			 * (which can be non-zero only the first time
1387 			 * through the "while" loop in step 3).
1388 			 */
1389 			fudge = start - entry->start;
1390 		} else {
1391 			/*
1392 			 * normal reference: we clip the map to fit (thus
1393 			 * fudge is zero)
1394 			 */
1395 			UVM_MAP_CLIP_START(srcmap, entry, start);
1396 			SAVE_HINT(srcmap, entry->prev);
1397 			fudge = 0;
1398 		}
1399 
1400 	} else {
1401 
1402 		/* "start" is not within an entry ... skip to next entry */
1403 		if (flags & UVM_EXTRACT_CONTIG) {
1404 			error = EINVAL;
1405 			goto bad;    /* definite hole here ... */
1406 		}
1407 
1408 		entry = entry->next;
1409 		fudge = 0;
1410 	}
1411 	/* save values from srcmap for step 6 */
1412 	orig_entry = entry;
1413 	orig_fudge = fudge;
1414 
1415 
1416 	/*
1417 	 * step 3: now start looping through the map entries, extracting
1418 	 * as we go.
1419 	 */
1420 
1421 	while (entry->start < end && entry != &srcmap->header) {
1422 
1423 		/* if we are not doing a quick reference, clip it */
1424 		if ((flags & UVM_EXTRACT_QREF) == 0)
1425 			UVM_MAP_CLIP_END(srcmap, entry, end);
1426 
1427 		/* clear needs_copy (allow chunking) */
1428 		if (UVM_ET_ISNEEDSCOPY(entry)) {
1429 			if (fudge)
1430 				oldstart = entry->start;
1431 			else
1432 				oldstart = 0;	/* XXX: gcc */
1433 			amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
1434 			if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
1435 				error = ENOMEM;
1436 				goto bad;
1437 			}
1438 			/* amap_copy could clip (during chunk)!  update fudge */
1439 			if (fudge) {
1440 				fudge = fudge - (entry->start - oldstart);
1441 				orig_fudge = fudge;
1442 			}
1443 		}
1444 
1445 		/* calculate the offset of this from "start" */
1446 		oldoffset = (entry->start + fudge) - start;
1447 
1448 		/* allocate a new map entry */
1449 		newentry = uvm_mapent_alloc(dstmap);
1450 		if (newentry == NULL) {
1451 			error = ENOMEM;
1452 			goto bad;
1453 		}
1454 
1455 		/* set up new map entry */
1456 		newentry->next = NULL;
1457 		newentry->prev = endchain;
1458 		newentry->start = dstaddr + oldoffset;
1459 		newentry->end =
1460 		    newentry->start + (entry->end - (entry->start + fudge));
1461 		if (newentry->end > newend || newentry->end < newentry->start)
1462 			newentry->end = newend;
1463 		newentry->object.uvm_obj = entry->object.uvm_obj;
1464 		if (newentry->object.uvm_obj) {
1465 			if (newentry->object.uvm_obj->pgops->pgo_reference)
1466 				newentry->object.uvm_obj->pgops->
1467 				    pgo_reference(newentry->object.uvm_obj);
1468 				newentry->offset = entry->offset + fudge;
1469 		} else {
1470 			newentry->offset = 0;
1471 		}
1472 		newentry->etype = entry->etype;
1473 		newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
1474 			entry->max_protection : entry->protection;
1475 		newentry->max_protection = entry->max_protection;
1476 		newentry->inheritance = entry->inheritance;
1477 		newentry->wired_count = 0;
1478 		newentry->aref.ar_amap = entry->aref.ar_amap;
1479 		if (newentry->aref.ar_amap) {
1480 			newentry->aref.ar_pageoff =
1481 			    entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1482 			amap_ref(newentry, AMAP_SHARED |
1483 			    ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
1484 		} else {
1485 			newentry->aref.ar_pageoff = 0;
1486 		}
1487 		newentry->advice = entry->advice;
1488 
1489 		/* now link it on the chain */
1490 		nchain++;
1491 		if (endchain == NULL) {
1492 			chain = endchain = newentry;
1493 		} else {
1494 			endchain->next = newentry;
1495 			endchain = newentry;
1496 		}
1497 
1498 		/* end of 'while' loop! */
1499 		if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1500 		    (entry->next == &srcmap->header ||
1501 		    entry->next->start != entry->end)) {
1502 			error = EINVAL;
1503 			goto bad;
1504 		}
1505 		entry = entry->next;
1506 		fudge = 0;
1507 	}
1508 
1509 
1510 	/*
1511 	 * step 4: close off chain (in format expected by uvm_map_replace)
1512 	 */
1513 
1514 	if (chain)
1515 		chain->prev = endchain;
1516 
1517 
1518 	/*
1519 	 * step 5: attempt to lock the dest map so we can pmap_copy.
1520 	 * note usage of copy_ok:
1521 	 *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
1522 	 *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
1523 	 */
1524 
1525 	if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
1526 
1527 		copy_ok = 1;
1528 		if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1529 		    nchain)) {
1530 			if (srcmap != dstmap)
1531 				vm_map_unlock(dstmap);
1532 			error = EIO;
1533 			goto bad;
1534 		}
1535 
1536 	} else {
1537 
1538 		copy_ok = 0;
1539 		/* replace defered until step 7 */
1540 
1541 	}
1542 
1543 
1544 	/*
1545 	 * step 6: traverse the srcmap a second time to do the following:
1546 	 *  - if we got a lock on the dstmap do pmap_copy
1547 	 *  - if UVM_EXTRACT_REMOVE remove the entries
1548 	 * we make use of orig_entry and orig_fudge (saved in step 2)
1549 	 */
1550 
1551 	if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
1552 
1553 		/* purge possible stale hints from srcmap */
1554 		if (flags & UVM_EXTRACT_REMOVE) {
1555 			SAVE_HINT(srcmap, orig_entry->prev);
1556 			if (srcmap->first_free->start >= start)
1557 				srcmap->first_free = orig_entry->prev;
1558 		}
1559 
1560 		entry = orig_entry;
1561 		fudge = orig_fudge;
1562 		deadentry = NULL;	/* for UVM_EXTRACT_REMOVE */
1563 
1564 		while (entry->start < end && entry != &srcmap->header) {
1565 
1566 			if (copy_ok) {
1567 				oldoffset = (entry->start + fudge) - start;
1568 				elen = min(end, entry->end) -
1569 				    (entry->start + fudge);
1570 				pmap_copy(dstmap->pmap, srcmap->pmap,
1571 				    dstaddr + oldoffset, elen,
1572 				    entry->start + fudge);
1573 			}
1574 
1575 			/* we advance "entry" in the following if statement */
1576 			if (flags & UVM_EXTRACT_REMOVE) {
1577 				pmap_remove(srcmap->pmap, entry->start,
1578 						entry->end);
1579         			oldentry = entry;	/* save entry */
1580         			entry = entry->next;	/* advance */
1581 				uvm_map_entry_unlink(srcmap, oldentry);
1582 							/* add to dead list */
1583 				oldentry->next = deadentry;
1584 				deadentry = oldentry;
1585       			} else {
1586         			entry = entry->next;		/* advance */
1587 			}
1588 
1589 			/* end of 'while' loop */
1590 			fudge = 0;
1591 		}
1592 
1593 		/*
1594 		 * unlock dstmap.  we will dispose of deadentry in
1595 		 * step 7 if needed
1596 		 */
1597 		if (copy_ok && srcmap != dstmap)
1598 			vm_map_unlock(dstmap);
1599 
1600 	}
1601 	else
1602 		deadentry = NULL; /* XXX: gcc */
1603 
1604 	/*
1605 	 * step 7: we are done with the source map, unlock.   if copy_ok
1606 	 * is 0 then we have not replaced the dummy mapping in dstmap yet
1607 	 * and we need to do so now.
1608 	 */
1609 
1610 	vm_map_unlock(srcmap);
1611 	if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
1612 		uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
1613 
1614 	/* now do the replacement if we didn't do it in step 5 */
1615 	if (copy_ok == 0) {
1616 		vm_map_lock(dstmap);
1617 		error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1618 		    nchain);
1619 		vm_map_unlock(dstmap);
1620 
1621 		if (error == FALSE) {
1622 			error = EIO;
1623 			goto bad2;
1624 		}
1625 	}
1626 
1627 	/*
1628 	 * done!
1629 	 */
1630 	return(0);
1631 
1632 	/*
1633 	 * bad: failure recovery
1634 	 */
1635 bad:
1636 	vm_map_unlock(srcmap);
1637 bad2:			/* src already unlocked */
1638 	if (chain)
1639 		uvm_unmap_detach(chain,
1640 		    (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1641 	uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
1642 	return(error);
1643 }
1644 
1645 /* end of extraction functions */
1646 
1647 /*
1648  * uvm_map_submap: punch down part of a map into a submap
1649  *
1650  * => only the kernel_map is allowed to be submapped
1651  * => the purpose of submapping is to break up the locking granularity
1652  *	of a larger map
1653  * => the range specified must have been mapped previously with a uvm_map()
1654  *	call [with uobj==NULL] to create a blank map entry in the main map.
1655  *	[And it had better still be blank!]
1656  * => maps which contain submaps should never be copied or forked.
1657  * => to remove a submap, use uvm_unmap() on the main map
1658  *	and then uvm_map_deallocate() the submap.
1659  * => main map must be unlocked.
1660  * => submap must have been init'd and have a zero reference count.
1661  *	[need not be locked as we don't actually reference it]
1662  */
1663 
1664 int
1665 uvm_map_submap(map, start, end, submap)
1666 	vm_map_t map, submap;
1667 	vaddr_t start, end;
1668 {
1669 	vm_map_entry_t entry;
1670 	int result;
1671 	UVMHIST_FUNC("uvm_map_submap"); UVMHIST_CALLED(maphist);
1672 
1673 	vm_map_lock(map);
1674 
1675 	VM_MAP_RANGE_CHECK(map, start, end);
1676 
1677 	if (uvm_map_lookup_entry(map, start, &entry)) {
1678 		UVM_MAP_CLIP_START(map, entry, start);
1679 		UVM_MAP_CLIP_END(map, entry, end);		/* to be safe */
1680 	}
1681 	else {
1682 		entry = NULL;
1683 	}
1684 
1685 	if (entry != NULL &&
1686 	    entry->start == start && entry->end == end &&
1687 	    entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
1688 	    !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1689 
1690 		/*
1691 		 * doit!
1692 		 */
1693 		entry->etype |= UVM_ET_SUBMAP;
1694 		entry->object.sub_map = submap;
1695 		entry->offset = 0;
1696 		uvm_map_reference(submap);
1697 		result = KERN_SUCCESS;
1698 	} else {
1699 		result = KERN_INVALID_ARGUMENT;
1700 	}
1701 	vm_map_unlock(map);
1702 
1703 	return(result);
1704 }
1705 
1706 
1707 /*
1708  * uvm_map_protect: change map protection
1709  *
1710  * => set_max means set max_protection.
1711  * => map must be unlocked.
1712  */
1713 
1714 #define MASK(entry)     (UVM_ET_ISCOPYONWRITE(entry) ? \
1715 			 ~VM_PROT_WRITE : VM_PROT_ALL)
1716 #define max(a,b)        ((a) > (b) ? (a) : (b))
1717 
1718 int
1719 uvm_map_protect(map, start, end, new_prot, set_max)
1720 	vm_map_t map;
1721 	vaddr_t start, end;
1722 	vm_prot_t new_prot;
1723 	boolean_t set_max;
1724 {
1725 	vm_map_entry_t current, entry;
1726 	int rv = KERN_SUCCESS;
1727 	UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
1728 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
1729 	map, start, end, new_prot);
1730 
1731 	vm_map_lock(map);
1732 
1733 	VM_MAP_RANGE_CHECK(map, start, end);
1734 
1735 	if (uvm_map_lookup_entry(map, start, &entry)) {
1736 		UVM_MAP_CLIP_START(map, entry, start);
1737 	} else {
1738 		entry = entry->next;
1739 	}
1740 
1741 	/*
1742 	 * make a first pass to check for protection violations.
1743 	 */
1744 
1745 	current = entry;
1746 	while ((current != &map->header) && (current->start < end)) {
1747 		if (UVM_ET_ISSUBMAP(current)) {
1748 			rv = KERN_INVALID_ARGUMENT;
1749 			goto out;
1750 		}
1751 		if ((new_prot & current->max_protection) != new_prot) {
1752 			rv = KERN_PROTECTION_FAILURE;
1753 			goto out;
1754 		}
1755 		current = current->next;
1756 	}
1757 
1758 	/* go back and fix up protections (no need to clip this time). */
1759 
1760 	current = entry;
1761 
1762 	while ((current != &map->header) && (current->start < end)) {
1763 		vm_prot_t old_prot;
1764 
1765 		UVM_MAP_CLIP_END(map, current, end);
1766 
1767 		old_prot = current->protection;
1768 		if (set_max)
1769 			current->protection =
1770 			    (current->max_protection = new_prot) & old_prot;
1771 		else
1772 			current->protection = new_prot;
1773 
1774 		/*
1775 		 * update physical map if necessary.  worry about copy-on-write
1776 		 * here -- CHECK THIS XXX
1777 		 */
1778 
1779 		if (current->protection != old_prot) {
1780 			/* update pmap! */
1781 			pmap_protect(map->pmap, current->start, current->end,
1782 			    current->protection & MASK(entry));
1783 		}
1784 
1785 		/*
1786 		 * If the map is configured to lock any future mappings,
1787 		 * wire this entry now if the old protection was VM_PROT_NONE
1788 		 * and the new protection is not VM_PROT_NONE.
1789 		 */
1790 
1791 		if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
1792 		    VM_MAPENT_ISWIRED(entry) == 0 &&
1793 		    old_prot == VM_PROT_NONE &&
1794 		    new_prot != VM_PROT_NONE) {
1795 			if (uvm_map_pageable(map, entry->start,
1796 			    entry->end, FALSE,
1797 			    UVM_LK_ENTER|UVM_LK_EXIT) != KERN_SUCCESS) {
1798 				/*
1799 				 * If locking the entry fails, remember the
1800 				 * error if it's the first one.  Note we
1801 				 * still continue setting the protection in
1802 				 * the map, but will return the resource
1803 				 * shortage condition regardless.
1804 				 *
1805 				 * XXX Ignore what the actual error is,
1806 				 * XXX just call it a resource shortage
1807 				 * XXX so that it doesn't get confused
1808 				 * XXX what uvm_map_protect() itself would
1809 				 * XXX normally return.
1810 				 */
1811 				rv = KERN_RESOURCE_SHORTAGE;
1812 			}
1813 		}
1814 
1815 		current = current->next;
1816 	}
1817 
1818  out:
1819 	vm_map_unlock(map);
1820 	UVMHIST_LOG(maphist, "<- done, rv=%d",rv,0,0,0);
1821 	return (rv);
1822 }
1823 
1824 #undef  max
1825 #undef  MASK
1826 
1827 /*
1828  * uvm_map_inherit: set inheritance code for range of addrs in map.
1829  *
1830  * => map must be unlocked
1831  * => note that the inherit code is used during a "fork".  see fork
1832  *	code for details.
1833  */
1834 
1835 int
1836 uvm_map_inherit(map, start, end, new_inheritance)
1837 	vm_map_t map;
1838 	vaddr_t start;
1839 	vaddr_t end;
1840 	vm_inherit_t new_inheritance;
1841 {
1842 	vm_map_entry_t entry, temp_entry;
1843 	UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
1844 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
1845 	    map, start, end, new_inheritance);
1846 
1847 	switch (new_inheritance) {
1848 	case VM_INHERIT_NONE:
1849 	case VM_INHERIT_COPY:
1850 	case VM_INHERIT_SHARE:
1851 		break;
1852 	default:
1853 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1854 		return (KERN_INVALID_ARGUMENT);
1855 	}
1856 
1857 	vm_map_lock(map);
1858 
1859 	VM_MAP_RANGE_CHECK(map, start, end);
1860 
1861 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1862 		entry = temp_entry;
1863 		UVM_MAP_CLIP_START(map, entry, start);
1864 	}  else {
1865 		entry = temp_entry->next;
1866 	}
1867 
1868 	while ((entry != &map->header) && (entry->start < end)) {
1869 		UVM_MAP_CLIP_END(map, entry, end);
1870 
1871 		entry->inheritance = new_inheritance;
1872 
1873 		entry = entry->next;
1874 	}
1875 
1876 	vm_map_unlock(map);
1877 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1878 	return(KERN_SUCCESS);
1879 }
1880 
1881 /*
1882  * uvm_map_advice: set advice code for range of addrs in map.
1883  *
1884  * => map must be unlocked
1885  */
1886 
1887 int
1888 uvm_map_advice(map, start, end, new_advice)
1889 	vm_map_t map;
1890 	vaddr_t start;
1891 	vaddr_t end;
1892 	int new_advice;
1893 {
1894 	vm_map_entry_t entry, temp_entry;
1895 	UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
1896 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
1897 	    map, start, end, new_advice);
1898 
1899 	vm_map_lock(map);
1900 
1901 	VM_MAP_RANGE_CHECK(map, start, end);
1902 
1903 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1904 		entry = temp_entry;
1905 		UVM_MAP_CLIP_START(map, entry, start);
1906 	} else {
1907 		entry = temp_entry->next;
1908 	}
1909 
1910 	/*
1911 	 * XXXJRT: disallow holes?
1912 	 */
1913 
1914 	while ((entry != &map->header) && (entry->start < end)) {
1915 		UVM_MAP_CLIP_END(map, entry, end);
1916 
1917 		switch (new_advice) {
1918 		case MADV_NORMAL:
1919 		case MADV_RANDOM:
1920 		case MADV_SEQUENTIAL:
1921 			/* nothing special here */
1922 			break;
1923 
1924 		default:
1925 			vm_map_unlock(map);
1926 			UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1927 			return (KERN_INVALID_ARGUMENT);
1928 		}
1929 
1930 
1931 		entry->advice = new_advice;
1932 
1933 		entry = entry->next;
1934 	}
1935 
1936 	vm_map_unlock(map);
1937 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1938 	return (KERN_SUCCESS);
1939 }
1940 
1941 /*
1942  * uvm_map_pageable: sets the pageability of a range in a map.
1943  *
1944  * => wires map entries.  should not be used for transient page locking.
1945  *	for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
1946  * => regions sepcified as not pageable require lock-down (wired) memory
1947  *	and page tables.
1948  * => map must never be read-locked
1949  * => if islocked is TRUE, map is already write-locked
1950  * => we always unlock the map, since we must downgrade to a read-lock
1951  *	to call uvm_fault_wire()
1952  * => XXXCDC: check this and try and clean it up.
1953  */
1954 
1955 int
1956 uvm_map_pageable(map, start, end, new_pageable, lockflags)
1957 	vm_map_t map;
1958 	vaddr_t start, end;
1959 	boolean_t new_pageable;
1960 	int lockflags;
1961 {
1962 	vm_map_entry_t entry, start_entry, failed_entry;
1963 	int rv;
1964 #ifdef DIAGNOSTIC
1965 	u_int timestamp_save;
1966 #endif
1967 	UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
1968 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
1969 	map, start, end, new_pageable);
1970 
1971 #ifdef DIAGNOSTIC
1972 	if ((map->flags & VM_MAP_PAGEABLE) == 0)
1973 		panic("uvm_map_pageable: map %p not pageable", map);
1974 #endif
1975 
1976 	if ((lockflags & UVM_LK_ENTER) == 0)
1977 		vm_map_lock(map);
1978 
1979 	VM_MAP_RANGE_CHECK(map, start, end);
1980 
1981 	/*
1982 	 * only one pageability change may take place at one time, since
1983 	 * uvm_fault_wire assumes it will be called only once for each
1984 	 * wiring/unwiring.  therefore, we have to make sure we're actually
1985 	 * changing the pageability for the entire region.  we do so before
1986 	 * making any changes.
1987 	 */
1988 
1989 	if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1990 		if ((lockflags & UVM_LK_EXIT) == 0)
1991 			vm_map_unlock(map);
1992 
1993 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1994 		return (KERN_INVALID_ADDRESS);
1995 	}
1996 	entry = start_entry;
1997 
1998 	/*
1999 	 * handle wiring and unwiring separately.
2000 	 */
2001 
2002 	if (new_pageable) {		/* unwire */
2003 		UVM_MAP_CLIP_START(map, entry, start);
2004 		/*
2005 		 * unwiring.  first ensure that the range to be unwired is
2006 		 * really wired down and that there are no holes.
2007 		 */
2008 		while ((entry != &map->header) && (entry->start < end)) {
2009 			if (entry->wired_count == 0 ||
2010 			    (entry->end < end &&
2011 			     (entry->next == &map->header ||
2012 			      entry->next->start > entry->end))) {
2013 				if ((lockflags & UVM_LK_EXIT) == 0)
2014 					vm_map_unlock(map);
2015 				UVMHIST_LOG(maphist,
2016 				    "<- done (INVALID UNWIRE ARG)",0,0,0,0);
2017 				return (KERN_INVALID_ARGUMENT);
2018 			}
2019 			entry = entry->next;
2020 		}
2021 
2022 		/*
2023 		 * POSIX 1003.1b - a single munlock call unlocks a region,
2024 		 * regardless of the number of mlock calls made on that
2025 		 * region.
2026 		 */
2027 		entry = start_entry;
2028 		while ((entry != &map->header) && (entry->start < end)) {
2029 			UVM_MAP_CLIP_END(map, entry, end);
2030 			if (VM_MAPENT_ISWIRED(entry))
2031 				uvm_map_entry_unwire(map, entry);
2032 			entry = entry->next;
2033 		}
2034 		if ((lockflags & UVM_LK_EXIT) == 0)
2035 			vm_map_unlock(map);
2036 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2037 		return(KERN_SUCCESS);
2038 
2039 		/*
2040 		 * end of unwire case!
2041 		 */
2042 	}
2043 
2044 	/*
2045 	 * wire case: in two passes [XXXCDC: ugly block of code here]
2046 	 *
2047 	 * 1: holding the write lock, we create any anonymous maps that need
2048 	 *    to be created.  then we clip each map entry to the region to
2049 	 *    be wired and increment its wiring count.
2050 	 *
2051 	 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
2052 	 *    in the pages for any newly wired area (wired_count == 1).
2053 	 *
2054 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
2055 	 *    deadlock with another thread that may have faulted on one of
2056 	 *    the pages to be wired (it would mark the page busy, blocking
2057 	 *    us, then in turn block on the map lock that we hold).  because
2058 	 *    of problems in the recursive lock package, we cannot upgrade
2059 	 *    to a write lock in vm_map_lookup.  thus, any actions that
2060 	 *    require the write lock must be done beforehand.  because we
2061 	 *    keep the read lock on the map, the copy-on-write status of the
2062 	 *    entries we modify here cannot change.
2063 	 */
2064 
2065 	while ((entry != &map->header) && (entry->start < end)) {
2066 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2067 			/*
2068 			 * perform actions of vm_map_lookup that need the
2069 			 * write lock on the map: create an anonymous map
2070 			 * for a copy-on-write region, or an anonymous map
2071 			 * for a zero-fill region.  (XXXCDC: submap case
2072 			 * ok?)
2073 			 */
2074 			if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
2075 				if (UVM_ET_ISNEEDSCOPY(entry) &&
2076 				    ((entry->protection & VM_PROT_WRITE) ||
2077 				     (entry->object.uvm_obj == NULL))) {
2078 					amap_copy(map, entry, M_WAITOK, TRUE,
2079 					    start, end);
2080 					/* XXXCDC: wait OK? */
2081 				}
2082 			}
2083 		}
2084 		UVM_MAP_CLIP_START(map, entry, start);
2085 		UVM_MAP_CLIP_END(map, entry, end);
2086 		entry->wired_count++;
2087 
2088 		/*
2089 		 * Check for holes
2090 		 */
2091 		if (entry->protection == VM_PROT_NONE ||
2092 		    (entry->end < end &&
2093 		     (entry->next == &map->header ||
2094 		      entry->next->start > entry->end))) {
2095 			/*
2096 			 * found one.  amap creation actions do not need to
2097 			 * be undone, but the wired counts need to be restored.
2098 			 */
2099 			while (entry != &map->header && entry->end > start) {
2100 				entry->wired_count--;
2101 				entry = entry->prev;
2102 			}
2103 			if ((lockflags & UVM_LK_EXIT) == 0)
2104 				vm_map_unlock(map);
2105 			UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
2106 			return (KERN_INVALID_ARGUMENT);
2107 		}
2108 		entry = entry->next;
2109 	}
2110 
2111 	/*
2112 	 * Pass 2.
2113 	 */
2114 
2115 #ifdef DIAGNOSTIC
2116 	timestamp_save = map->timestamp;
2117 #endif
2118 	vm_map_busy(map);
2119 	vm_map_downgrade(map);
2120 
2121 	rv = 0;
2122 	entry = start_entry;
2123 	while (entry != &map->header && entry->start < end) {
2124 		if (entry->wired_count == 1) {
2125 			rv = uvm_fault_wire(map, entry->start, entry->end,
2126 			    entry->protection);
2127 			if (rv) {
2128 				/*
2129 				 * wiring failed.  break out of the loop.
2130 				 * we'll clean up the map below, once we
2131 				 * have a write lock again.
2132 				 */
2133 				break;
2134 			}
2135 		}
2136 		entry = entry->next;
2137 	}
2138 
2139 	if (rv) {        /* failed? */
2140 		/*
2141 		 * Get back to an exclusive (write) lock.
2142 		 */
2143 		vm_map_upgrade(map);
2144 		vm_map_unbusy(map);
2145 
2146 #ifdef DIAGNOSTIC
2147 		if (timestamp_save != map->timestamp)
2148 			panic("uvm_map_pageable: stale map");
2149 #endif
2150 
2151 		/*
2152 		 * first drop the wiring count on all the entries
2153 		 * which haven't actually been wired yet.
2154 		 */
2155 		failed_entry = entry;
2156 		while (entry != &map->header && entry->start < end) {
2157 			entry->wired_count--;
2158 			entry = entry->next;
2159 		}
2160 
2161 		/*
2162 		 * now, unwire all the entries that were successfully
2163 		 * wired above.
2164 		 */
2165 		entry = start_entry;
2166 		while (entry != failed_entry) {
2167 			entry->wired_count--;
2168 			if (VM_MAPENT_ISWIRED(entry) == 0)
2169 				uvm_map_entry_unwire(map, entry);
2170 			entry = entry->next;
2171 		}
2172 		if ((lockflags & UVM_LK_EXIT) == 0)
2173 			vm_map_unlock(map);
2174 		UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
2175 		return(rv);
2176 	}
2177 
2178 	/* We are holding a read lock here. */
2179 	if ((lockflags & UVM_LK_EXIT) == 0) {
2180 		vm_map_unbusy(map);
2181 		vm_map_unlock_read(map);
2182 	} else {
2183 		/*
2184 		 * Get back to an exclusive (write) lock.
2185 		 */
2186 		vm_map_upgrade(map);
2187 		vm_map_unbusy(map);
2188 	}
2189 
2190 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2191 	return(KERN_SUCCESS);
2192 }
2193 
2194 /*
2195  * uvm_map_pageable_all: special case of uvm_map_pageable - affects
2196  * all mapped regions.
2197  *
2198  * => map must not be locked.
2199  * => if no flags are specified, all regions are unwired.
2200  * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
2201  */
2202 
2203 int
2204 uvm_map_pageable_all(map, flags, limit)
2205 	vm_map_t map;
2206 	int flags;
2207 	vsize_t limit;
2208 {
2209 	vm_map_entry_t entry, failed_entry;
2210 	vsize_t size;
2211 	int rv;
2212 #ifdef DIAGNOSTIC
2213 	u_int timestamp_save;
2214 #endif
2215 	UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
2216 	UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
2217 
2218 #ifdef DIAGNOSTIC
2219 	if ((map->flags & VM_MAP_PAGEABLE) == 0)
2220 		panic("uvm_map_pageable_all: map %p not pageable", map);
2221 #endif
2222 
2223 	vm_map_lock(map);
2224 
2225 	/*
2226 	 * handle wiring and unwiring separately.
2227 	 */
2228 
2229 	if (flags == 0) {			/* unwire */
2230 		/*
2231 		 * POSIX 1003.1b -- munlockall unlocks all regions,
2232 		 * regardless of how many times mlockall has been called.
2233 		 */
2234 		for (entry = map->header.next; entry != &map->header;
2235 		     entry = entry->next) {
2236 			if (VM_MAPENT_ISWIRED(entry))
2237 				uvm_map_entry_unwire(map, entry);
2238 		}
2239 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2240 		vm_map_unlock(map);
2241 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
2242 		return (KERN_SUCCESS);
2243 
2244 		/*
2245 		 * end of unwire case!
2246 		 */
2247 	}
2248 
2249 	if (flags & MCL_FUTURE) {
2250 		/*
2251 		 * must wire all future mappings; remember this.
2252 		 */
2253 		vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
2254 	}
2255 
2256 	if ((flags & MCL_CURRENT) == 0) {
2257 		/*
2258 		 * no more work to do!
2259 		 */
2260 		UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
2261 		vm_map_unlock(map);
2262 		return (KERN_SUCCESS);
2263 	}
2264 
2265 	/*
2266 	 * wire case: in three passes [XXXCDC: ugly block of code here]
2267 	 *
2268 	 * 1: holding the write lock, count all pages mapped by non-wired
2269 	 *    entries.  if this would cause us to go over our limit, we fail.
2270 	 *
2271 	 * 2: still holding the write lock, we create any anonymous maps that
2272 	 *    need to be created.  then we increment its wiring count.
2273 	 *
2274 	 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
2275 	 *    in the pages for any newly wired area (wired_count == 1).
2276 	 *
2277 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
2278 	 *    deadlock with another thread that may have faulted on one of
2279 	 *    the pages to be wired (it would mark the page busy, blocking
2280 	 *    us, then in turn block on the map lock that we hold).  because
2281 	 *    of problems in the recursive lock package, we cannot upgrade
2282 	 *    to a write lock in vm_map_lookup.  thus, any actions that
2283 	 *    require the write lock must be done beforehand.  because we
2284 	 *    keep the read lock on the map, the copy-on-write status of the
2285 	 *    entries we modify here cannot change.
2286 	 */
2287 
2288 	for (size = 0, entry = map->header.next; entry != &map->header;
2289 	     entry = entry->next) {
2290 		if (entry->protection != VM_PROT_NONE &&
2291 		    VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2292 			size += entry->end - entry->start;
2293 		}
2294 	}
2295 
2296 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
2297 		vm_map_unlock(map);
2298 		return (KERN_NO_SPACE);		/* XXX overloaded */
2299 	}
2300 
2301 	/* XXX non-pmap_wired_count case must be handled by caller */
2302 #ifdef pmap_wired_count
2303 	if (limit != 0 &&
2304 	    (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
2305 		vm_map_unlock(map);
2306 		return (KERN_NO_SPACE);		/* XXX overloaded */
2307 	}
2308 #endif
2309 
2310 	/*
2311 	 * Pass 2.
2312 	 */
2313 
2314 	for (entry = map->header.next; entry != &map->header;
2315 	     entry = entry->next) {
2316 		if (entry->protection == VM_PROT_NONE)
2317 			continue;
2318 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
2319 			/*
2320 			 * perform actions of vm_map_lookup that need the
2321 			 * write lock on the map: create an anonymous map
2322 			 * for a copy-on-write region, or an anonymous map
2323 			 * for a zero-fill region.  (XXXCDC: submap case
2324 			 * ok?)
2325 			 */
2326 			if (!UVM_ET_ISSUBMAP(entry)) {	/* not submap */
2327 				if (UVM_ET_ISNEEDSCOPY(entry) &&
2328 				    ((entry->protection & VM_PROT_WRITE) ||
2329 				     (entry->object.uvm_obj == NULL))) {
2330 					amap_copy(map, entry, M_WAITOK, TRUE,
2331 					    entry->start, entry->end);
2332 					/* XXXCDC: wait OK? */
2333 				}
2334 			}
2335 		}
2336 		entry->wired_count++;
2337 	}
2338 
2339 	/*
2340 	 * Pass 3.
2341 	 */
2342 
2343 #ifdef DIAGNOSTIC
2344 	timestamp_save = map->timestamp;
2345 #endif
2346 	vm_map_busy(map);
2347 	vm_map_downgrade(map);
2348 
2349 	rv = KERN_SUCCESS;
2350 	for (entry = map->header.next; entry != &map->header;
2351 	     entry = entry->next) {
2352 		if (entry->wired_count == 1) {
2353 			rv = uvm_fault_wire(map, entry->start, entry->end,
2354 			     entry->protection);
2355 			if (rv) {
2356 				/*
2357 				 * wiring failed.  break out of the loop.
2358 				 * we'll clean up the map below, once we
2359 				 * have a write lock again.
2360 				 */
2361 				break;
2362 			}
2363 		}
2364 	}
2365 
2366 	if (rv) {	/* failed? */
2367 		/*
2368 		 * Get back an exclusive (write) lock.
2369 		 */
2370 		vm_map_upgrade(map);
2371 		vm_map_unbusy(map);
2372 
2373 #ifdef DIAGNOSTIC
2374 		if (timestamp_save != map->timestamp)
2375 			panic("uvm_map_pageable_all: stale map");
2376 #endif
2377 
2378 		/*
2379 		 * first drop the wiring count on all the entries
2380 		 * which haven't actually been wired yet.
2381 		 *
2382 		 * Skip VM_PROT_NONE entries like we did above.
2383 		 */
2384 		failed_entry = entry;
2385 		for (/* nothing */; entry != &map->header;
2386 		     entry = entry->next) {
2387 			if (entry->protection == VM_PROT_NONE)
2388 				continue;
2389 			entry->wired_count--;
2390 		}
2391 
2392 		/*
2393 		 * now, unwire all the entries that were successfully
2394 		 * wired above.
2395 		 *
2396 		 * Skip VM_PROT_NONE entries like we did above.
2397 		 */
2398 		for (entry = map->header.next; entry != failed_entry;
2399 		     entry = entry->next) {
2400 			if (entry->protection == VM_PROT_NONE)
2401 				continue;
2402 			entry->wired_count--;
2403 			if (VM_MAPENT_ISWIRED(entry))
2404 				uvm_map_entry_unwire(map, entry);
2405 		}
2406 		vm_map_unlock(map);
2407 		UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
2408 		return (rv);
2409 	}
2410 
2411 	/* We are holding a read lock here. */
2412 	vm_map_unbusy(map);
2413 	vm_map_unlock_read(map);
2414 
2415 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2416 	return (KERN_SUCCESS);
2417 }
2418 
2419 /*
2420  * uvm_map_clean: clean out a map range
2421  *
2422  * => valid flags:
2423  *   if (flags & PGO_CLEANIT): dirty pages are cleaned first
2424  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
2425  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
2426  *   if (flags & PGO_FREE): any cached pages are freed after clean
2427  * => returns an error if any part of the specified range isn't mapped
2428  * => never a need to flush amap layer since the anonymous memory has
2429  *	no permanent home, but may deactivate pages there
2430  * => called from sys_msync() and sys_madvise()
2431  * => caller must not write-lock map (read OK).
2432  * => we may sleep while cleaning if SYNCIO [with map read-locked]
2433  */
2434 
2435 int	amap_clean_works = 1;	/* XXX for now, just in case... */
2436 
2437 int
2438 uvm_map_clean(map, start, end, flags)
2439 	vm_map_t map;
2440 	vaddr_t start, end;
2441 	int flags;
2442 {
2443 	vm_map_entry_t current, entry;
2444 	struct uvm_object *uobj;
2445 	struct vm_amap *amap;
2446 	struct vm_anon *anon;
2447 	struct vm_page *pg;
2448 	vaddr_t offset;
2449 	vsize_t size;
2450 	int rv, error, refs;
2451 	UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
2452 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
2453 	map, start, end, flags);
2454 
2455 #ifdef DIAGNOSTIC
2456 	if ((flags & (PGO_FREE|PGO_DEACTIVATE)) == (PGO_FREE|PGO_DEACTIVATE))
2457 		panic("uvm_map_clean: FREE and DEACTIVATE");
2458 #endif
2459 
2460 	vm_map_lock_read(map);
2461 	VM_MAP_RANGE_CHECK(map, start, end);
2462 	if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
2463 		vm_map_unlock_read(map);
2464 		return(KERN_INVALID_ADDRESS);
2465 	}
2466 
2467 	/*
2468 	 * Make a first pass to check for holes.
2469 	 */
2470 	for (current = entry; current->start < end; current = current->next) {
2471 		if (UVM_ET_ISSUBMAP(current)) {
2472 			vm_map_unlock_read(map);
2473 			return (KERN_INVALID_ARGUMENT);
2474 		}
2475 		if (end > current->end && (current->next == &map->header ||
2476 		    current->end != current->next->start)) {
2477 			vm_map_unlock_read(map);
2478 			return (KERN_INVALID_ADDRESS);
2479 		}
2480 	}
2481 
2482 	error = KERN_SUCCESS;
2483 
2484 	for (current = entry; current->start < end; current = current->next) {
2485 		amap = current->aref.ar_amap;	/* top layer */
2486 		uobj = current->object.uvm_obj;	/* bottom layer */
2487 
2488 #ifdef DIAGNOSTIC
2489 		if (start < current->start)
2490 			panic("uvm_map_clean: hole");
2491 #endif
2492 
2493 		/*
2494 		 * No amap cleaning necessary if:
2495 		 *
2496 		 *	(1) There's no amap.
2497 		 *
2498 		 *	(2) We're not deactivating or freeing pages.
2499 		 */
2500 		if (amap == NULL ||
2501 		    (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
2502 			goto flush_object;
2503 
2504 		/* XXX for now, just in case... */
2505 		if (amap_clean_works == 0)
2506 			goto flush_object;
2507 
2508 		amap_lock(amap);
2509 
2510 		offset = start - current->start;
2511 		size = (end <= current->end ? end : current->end) -
2512 		    start;
2513 
2514 		for (/* nothing */; size != 0; size -= PAGE_SIZE,
2515 		     offset += PAGE_SIZE) {
2516 			anon = amap_lookup(&current->aref, offset);
2517 			if (anon == NULL)
2518 				continue;
2519 
2520 			simple_lock(&anon->an_lock);
2521 
2522 			pg = anon->u.an_page;
2523 			if (pg == NULL) {
2524 				simple_unlock(&anon->an_lock);
2525 				continue;
2526 			}
2527 
2528 			switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
2529 			/*
2530 			 * XXX In these first 3 cases, we always just
2531 			 * XXX deactivate the page.  We may want to
2532 			 * XXX handle the different cases more
2533 			 * XXX specifically, in the future.
2534 			 */
2535 			case PGO_CLEANIT|PGO_FREE:
2536 			case PGO_CLEANIT|PGO_DEACTIVATE:
2537 			case PGO_DEACTIVATE:
2538  deactivate_it:
2539 				/* skip the page if it's loaned or wired */
2540 				if (pg->loan_count != 0 ||
2541 				    pg->wire_count != 0) {
2542 					simple_unlock(&anon->an_lock);
2543 					continue;
2544 				}
2545 
2546 				uvm_lock_pageq();
2547 
2548 				/*
2549 				 * skip the page if it's not actually owned
2550 				 * by the anon (may simply be loaned to the
2551 				 * anon).
2552 				 */
2553 				if ((pg->pqflags & PQ_ANON) == 0) {
2554 #ifdef DIAGNOSTIC
2555 					if (pg->uobject != NULL)
2556 						panic("uvm_map_clean: "
2557 						    "page anon vs. object "
2558 						    "inconsistency");
2559 #endif
2560 					uvm_unlock_pageq();
2561 					simple_unlock(&anon->an_lock);
2562 					continue;
2563 				}
2564 
2565 #ifdef DIAGNOSTIC
2566 				if (pg->uanon != anon)
2567 					panic("uvm_map_clean: anon "
2568 					    "inconsistency");
2569 #endif
2570 
2571 				/* zap all mappings for the page. */
2572 				pmap_page_protect(pg, VM_PROT_NONE);
2573 
2574 				/* ...and deactivate the page. */
2575 				uvm_pagedeactivate(pg);
2576 
2577 				uvm_unlock_pageq();
2578 				simple_unlock(&anon->an_lock);
2579 				continue;
2580 
2581 			case PGO_FREE:
2582 				/*
2583 				 * If there are multiple references to
2584 				 * the amap, just deactivate the page.
2585 				 */
2586 				if (amap_refs(amap) > 1)
2587 					goto deactivate_it;
2588 
2589 				/* XXX skip the page if it's wired */
2590 				if (pg->wire_count != 0) {
2591 					simple_unlock(&anon->an_lock);
2592 					continue;
2593 				}
2594 				amap_unadd(&current->aref, offset);
2595 				refs = --anon->an_ref;
2596 				simple_unlock(&anon->an_lock);
2597 				if (refs == 0)
2598 					uvm_anfree(anon);
2599 				continue;
2600 
2601 			default:
2602 				panic("uvm_map_clean: wierd flags");
2603 			}
2604 #ifdef DIAGNOSTIC
2605 			panic("uvm_map_clean: unreachable code");
2606 #endif
2607 		}
2608 
2609 		amap_unlock(amap);
2610 
2611  flush_object:
2612 		/*
2613 		 * flush pages if we've got a valid backing object.
2614 		 */
2615 
2616 		offset = current->offset + (start - current->start);
2617 		size = (end <= current->end ? end : current->end) - start;
2618 
2619 		if (uobj != NULL) {
2620 			simple_lock(&uobj->vmobjlock);
2621 			rv = uobj->pgops->pgo_flush(uobj, offset,
2622 			    offset + size, flags);
2623 			simple_unlock(&uobj->vmobjlock);
2624 
2625 			if (rv == FALSE)
2626 				error = KERN_FAILURE;
2627 		}
2628 		start += size;
2629 	}
2630 
2631 	vm_map_unlock_read(map);
2632 	return (error);
2633 }
2634 
2635 
2636 /*
2637  * uvm_map_checkprot: check protection in map
2638  *
2639  * => must allow specified protection in a fully allocated region.
2640  * => map must be read or write locked by caller.
2641  */
2642 
2643 boolean_t
2644 uvm_map_checkprot(map, start, end, protection)
2645 	vm_map_t       map;
2646 	vaddr_t    start, end;
2647 	vm_prot_t      protection;
2648 {
2649 	 vm_map_entry_t entry;
2650 	 vm_map_entry_t tmp_entry;
2651 
2652 	 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
2653 		 return(FALSE);
2654 	 }
2655 
2656 	 entry = tmp_entry;
2657 
2658 	 while (start < end) {
2659 		 if (entry == &map->header) {
2660 			 return(FALSE);
2661 		 }
2662 
2663 		/*
2664 		 * no holes allowed
2665 		 */
2666 
2667 		 if (start < entry->start) {
2668 			 return(FALSE);
2669 		 }
2670 
2671 		/*
2672 		 * check protection associated with entry
2673 		 */
2674 
2675 		 if ((entry->protection & protection) != protection) {
2676 			 return(FALSE);
2677 		 }
2678 
2679 		 /* go to next entry */
2680 
2681 		 start = entry->end;
2682 		 entry = entry->next;
2683 	 }
2684 	 return(TRUE);
2685 }
2686 
2687 /*
2688  * uvmspace_alloc: allocate a vmspace structure.
2689  *
2690  * - structure includes vm_map and pmap
2691  * - XXX: no locking on this structure
2692  * - refcnt set to 1, rest must be init'd by caller
2693  */
2694 struct vmspace *
2695 uvmspace_alloc(min, max, pageable)
2696 	vaddr_t min, max;
2697 	int pageable;
2698 {
2699 	struct vmspace *vm;
2700 	UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
2701 
2702 	vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
2703 	uvmspace_init(vm, NULL, min, max, pageable);
2704 	UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
2705 	return (vm);
2706 }
2707 
2708 /*
2709  * uvmspace_init: initialize a vmspace structure.
2710  *
2711  * - XXX: no locking on this structure
2712  * - refcnt set to 1, rest must me init'd by caller
2713  */
2714 void
2715 uvmspace_init(vm, pmap, min, max, pageable)
2716 	struct vmspace *vm;
2717 	struct pmap *pmap;
2718 	vaddr_t min, max;
2719 	boolean_t pageable;
2720 {
2721 	UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
2722 
2723 	memset(vm, 0, sizeof(*vm));
2724 
2725 	uvm_map_setup(&vm->vm_map, min, max, pageable ? VM_MAP_PAGEABLE : 0);
2726 
2727 	if (pmap)
2728 		pmap_reference(pmap);
2729 	else
2730 		pmap = pmap_create();
2731 	vm->vm_map.pmap = pmap;
2732 
2733 	vm->vm_refcnt = 1;
2734 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
2735 }
2736 
2737 /*
2738  * uvmspace_share: share a vmspace between two proceses
2739  *
2740  * - XXX: no locking on vmspace
2741  * - used for vfork, threads(?)
2742  */
2743 
2744 void
2745 uvmspace_share(p1, p2)
2746 	struct proc *p1, *p2;
2747 {
2748 	p2->p_vmspace = p1->p_vmspace;
2749 	p1->p_vmspace->vm_refcnt++;
2750 }
2751 
2752 /*
2753  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
2754  *
2755  * - XXX: no locking on vmspace
2756  */
2757 
2758 void
2759 uvmspace_unshare(p)
2760 	struct proc *p;
2761 {
2762 	struct vmspace *nvm, *ovm = p->p_vmspace;
2763 
2764 	if (ovm->vm_refcnt == 1)
2765 		/* nothing to do: vmspace isn't shared in the first place */
2766 		return;
2767 
2768 	/* make a new vmspace, still holding old one */
2769 	nvm = uvmspace_fork(ovm);
2770 
2771 	pmap_deactivate(p);		/* unbind old vmspace */
2772 	p->p_vmspace = nvm;
2773 	pmap_activate(p);		/* switch to new vmspace */
2774 
2775 	uvmspace_free(ovm);		/* drop reference to old vmspace */
2776 }
2777 
2778 /*
2779  * uvmspace_exec: the process wants to exec a new program
2780  *
2781  * - XXX: no locking on vmspace
2782  */
2783 
2784 void
2785 uvmspace_exec(p)
2786 	struct proc *p;
2787 {
2788 	struct vmspace *nvm, *ovm = p->p_vmspace;
2789 	vm_map_t map = &ovm->vm_map;
2790 
2791 #ifdef __sparc__
2792 	/* XXX cgd 960926: the sparc #ifdef should be a MD hook */
2793 	kill_user_windows(p);   /* before stack addresses go away */
2794 #endif
2795 
2796 	/*
2797 	 * see if more than one process is using this vmspace...
2798 	 */
2799 
2800 	if (ovm->vm_refcnt == 1) {
2801 
2802 		/*
2803 		 * if p is the only process using its vmspace then we can safely
2804 		 * recycle that vmspace for the program that is being exec'd.
2805 		 */
2806 
2807 #ifdef SYSVSHM
2808 		/*
2809 		 * SYSV SHM semantics require us to kill all segments on an exec
2810 		 */
2811 		if (ovm->vm_shm)
2812 			shmexit(ovm);
2813 #endif
2814 
2815 		/*
2816 		 * POSIX 1003.1b -- "lock future mappings" is revoked
2817 		 * when a process execs another program image.
2818 		 */
2819 		vm_map_lock(map);
2820 		vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
2821 		vm_map_unlock(map);
2822 
2823 		/*
2824 		 * now unmap the old program
2825 		 */
2826 		uvm_unmap(map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
2827 
2828 	} else {
2829 
2830 		/*
2831 		 * p's vmspace is being shared, so we can't reuse it for p since
2832 		 * it is still being used for others.   allocate a new vmspace
2833 		 * for p
2834 		 */
2835 		nvm = uvmspace_alloc(map->min_offset, map->max_offset,
2836 			 (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2837 
2838 		/*
2839 		 * install new vmspace and drop our ref to the old one.
2840 		 */
2841 
2842 		pmap_deactivate(p);
2843 		p->p_vmspace = nvm;
2844 		pmap_activate(p);
2845 
2846 		uvmspace_free(ovm);
2847 	}
2848 }
2849 
2850 /*
2851  * uvmspace_free: free a vmspace data structure
2852  *
2853  * - XXX: no locking on vmspace
2854  */
2855 
2856 void
2857 uvmspace_free(vm)
2858 	struct vmspace *vm;
2859 {
2860 	vm_map_entry_t dead_entries;
2861 	UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
2862 
2863 	UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
2864 	if (--vm->vm_refcnt == 0) {
2865 		/*
2866 		 * lock the map, to wait out all other references to it.  delete
2867 		 * all of the mappings and pages they hold, then call the pmap
2868 		 * module to reclaim anything left.
2869 		 */
2870 		vm_map_lock(&vm->vm_map);
2871 		if (vm->vm_map.nentries) {
2872 			(void)uvm_unmap_remove(&vm->vm_map,
2873 			    vm->vm_map.min_offset, vm->vm_map.max_offset,
2874 			    &dead_entries);
2875 			if (dead_entries != NULL)
2876 				uvm_unmap_detach(dead_entries, 0);
2877 		}
2878 		pmap_destroy(vm->vm_map.pmap);
2879 		vm->vm_map.pmap = NULL;
2880 		pool_put(&uvm_vmspace_pool, vm);
2881 	}
2882 	UVMHIST_LOG(maphist,"<- done", 0,0,0,0);
2883 }
2884 
2885 /*
2886  *   F O R K   -   m a i n   e n t r y   p o i n t
2887  */
2888 /*
2889  * uvmspace_fork: fork a process' main map
2890  *
2891  * => create a new vmspace for child process from parent.
2892  * => parent's map must not be locked.
2893  */
2894 
2895 struct vmspace *
2896 uvmspace_fork(vm1)
2897 	struct vmspace *vm1;
2898 {
2899 	struct vmspace *vm2;
2900 	vm_map_t        old_map = &vm1->vm_map;
2901 	vm_map_t        new_map;
2902 	vm_map_entry_t  old_entry;
2903 	vm_map_entry_t  new_entry;
2904 	pmap_t          new_pmap;
2905 	boolean_t	protect_child;
2906 	UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
2907 
2908 	vm_map_lock(old_map);
2909 
2910 	vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset,
2911 		      (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE);
2912 	memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
2913 	(caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
2914 	new_map = &vm2->vm_map;		  /* XXX */
2915 	new_pmap = new_map->pmap;
2916 
2917 	old_entry = old_map->header.next;
2918 
2919 	/*
2920 	 * go entry-by-entry
2921 	 */
2922 
2923 	while (old_entry != &old_map->header) {
2924 
2925 		/*
2926 		 * first, some sanity checks on the old entry
2927 		 */
2928 		if (UVM_ET_ISSUBMAP(old_entry))
2929 		    panic("fork: encountered a submap during fork (illegal)");
2930 
2931 		if (!UVM_ET_ISCOPYONWRITE(old_entry) &&
2932 			    UVM_ET_ISNEEDSCOPY(old_entry))
2933 	panic("fork: non-copy_on_write map entry marked needs_copy (illegal)");
2934 
2935 
2936 		switch (old_entry->inheritance) {
2937 		case VM_INHERIT_NONE:
2938 			/*
2939 			 * drop the mapping
2940 			 */
2941 			break;
2942 
2943 		case VM_INHERIT_SHARE:
2944 			/*
2945 			 * share the mapping: this means we want the old and
2946 			 * new entries to share amaps and backing objects.
2947 			 */
2948 
2949 			/*
2950 			 * if the old_entry needs a new amap (due to prev fork)
2951 			 * then we need to allocate it now so that we have
2952 			 * something we own to share with the new_entry.   [in
2953 			 * other words, we need to clear needs_copy]
2954 			 */
2955 
2956 			if (UVM_ET_ISNEEDSCOPY(old_entry)) {
2957 				/* get our own amap, clears needs_copy */
2958 				amap_copy(old_map, old_entry, M_WAITOK, FALSE,
2959 				    0, 0);
2960 				/* XXXCDC: WAITOK??? */
2961 			}
2962 
2963 			new_entry = uvm_mapent_alloc(new_map);
2964 			/* old_entry -> new_entry */
2965 			uvm_mapent_copy(old_entry, new_entry);
2966 
2967 			/* new pmap has nothing wired in it */
2968 			new_entry->wired_count = 0;
2969 
2970 			/*
2971 			 * gain reference to object backing the map (can't
2972 			 * be a submap, already checked this case).
2973 			 */
2974 			if (new_entry->aref.ar_amap)
2975 				/* share reference */
2976 				amap_ref(new_entry, AMAP_SHARED);
2977 
2978 			if (new_entry->object.uvm_obj &&
2979 			    new_entry->object.uvm_obj->pgops->pgo_reference)
2980 				new_entry->object.uvm_obj->
2981 				    pgops->pgo_reference(
2982 				        new_entry->object.uvm_obj);
2983 
2984 			/* insert entry at end of new_map's entry list */
2985 			uvm_map_entry_link(new_map, new_map->header.prev,
2986 			    new_entry);
2987 
2988 			/*
2989 			 * pmap_copy the mappings: this routine is optional
2990 			 * but if it is there it will reduce the number of
2991 			 * page faults in the new proc.
2992 			 */
2993 
2994 			pmap_copy(new_pmap, old_map->pmap, new_entry->start,
2995 			    (old_entry->end - old_entry->start),
2996 			    old_entry->start);
2997 
2998 			break;
2999 
3000 		case VM_INHERIT_COPY:
3001 
3002 			/*
3003 			 * copy-on-write the mapping (using mmap's
3004 			 * MAP_PRIVATE semantics)
3005 			 *
3006 			 * allocate new_entry, adjust reference counts.
3007 			 * (note that new references are read-only).
3008 			 */
3009 
3010 			new_entry = uvm_mapent_alloc(new_map);
3011 			/* old_entry -> new_entry */
3012 			uvm_mapent_copy(old_entry, new_entry);
3013 
3014 			if (new_entry->aref.ar_amap)
3015 				amap_ref(new_entry, 0);
3016 
3017 			if (new_entry->object.uvm_obj &&
3018 			    new_entry->object.uvm_obj->pgops->pgo_reference)
3019 				new_entry->object.uvm_obj->pgops->pgo_reference
3020 				    (new_entry->object.uvm_obj);
3021 
3022 			/* new pmap has nothing wired in it */
3023 			new_entry->wired_count = 0;
3024 
3025 			new_entry->etype |=
3026 			    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
3027 			uvm_map_entry_link(new_map, new_map->header.prev,
3028 			    new_entry);
3029 
3030 			/*
3031 			 * the new entry will need an amap.  it will either
3032 			 * need to be copied from the old entry or created
3033 			 * from scratch (if the old entry does not have an
3034 			 * amap).  can we defer this process until later
3035 			 * (by setting "needs_copy") or do we need to copy
3036 			 * the amap now?
3037 			 *
3038 			 * we must copy the amap now if any of the following
3039 			 * conditions hold:
3040 			 * 1. the old entry has an amap and that amap is
3041 			 *    being shared.  this means that the old (parent)
3042 			 *    process is sharing the amap with another
3043 			 *    process.  if we do not clear needs_copy here
3044 			 *    we will end up in a situation where both the
3045 			 *    parent and child process are refering to the
3046 			 *    same amap with "needs_copy" set.  if the
3047 			 *    parent write-faults, the fault routine will
3048 			 *    clear "needs_copy" in the parent by allocating
3049 			 *    a new amap.   this is wrong because the
3050 			 *    parent is supposed to be sharing the old amap
3051 			 *    and the new amap will break that.
3052 			 *
3053 			 * 2. if the old entry has an amap and a non-zero
3054 			 *    wire count then we are going to have to call
3055 			 *    amap_cow_now to avoid page faults in the
3056 			 *    parent process.   since amap_cow_now requires
3057 			 *    "needs_copy" to be clear we might as well
3058 			 *    clear it here as well.
3059 			 *
3060 			 */
3061 
3062 			if (old_entry->aref.ar_amap != NULL) {
3063 
3064 			  if ((amap_flags(old_entry->aref.ar_amap) &
3065 			       AMAP_SHARED) != 0 ||
3066 			      VM_MAPENT_ISWIRED(old_entry)) {
3067 
3068 			    amap_copy(new_map, new_entry, M_WAITOK, FALSE,
3069 				      0, 0);
3070 			    /* XXXCDC: M_WAITOK ... ok? */
3071 			  }
3072 			}
3073 
3074 			/*
3075 			 * if the parent's entry is wired down, then the
3076 			 * parent process does not want page faults on
3077 			 * access to that memory.  this means that we
3078 			 * cannot do copy-on-write because we can't write
3079 			 * protect the old entry.   in this case we
3080 			 * resolve all copy-on-write faults now, using
3081 			 * amap_cow_now.   note that we have already
3082 			 * allocated any needed amap (above).
3083 			 */
3084 
3085 			if (VM_MAPENT_ISWIRED(old_entry)) {
3086 
3087 			  /*
3088 			   * resolve all copy-on-write faults now
3089 			   * (note that there is nothing to do if
3090 			   * the old mapping does not have an amap).
3091 			   * XXX: is it worthwhile to bother with pmap_copy
3092 			   * in this case?
3093 			   */
3094 			  if (old_entry->aref.ar_amap)
3095 			    amap_cow_now(new_map, new_entry);
3096 
3097 			} else {
3098 
3099 			  /*
3100 			   * setup mappings to trigger copy-on-write faults
3101 			   * we must write-protect the parent if it has
3102 			   * an amap and it is not already "needs_copy"...
3103 			   * if it is already "needs_copy" then the parent
3104 			   * has already been write-protected by a previous
3105 			   * fork operation.
3106 			   *
3107 			   * if we do not write-protect the parent, then
3108 			   * we must be sure to write-protect the child
3109 			   * after the pmap_copy() operation.
3110 			   *
3111 			   * XXX: pmap_copy should have some way of telling
3112 			   * us that it didn't do anything so we can avoid
3113 			   * calling pmap_protect needlessly.
3114 			   */
3115 
3116 			  if (old_entry->aref.ar_amap) {
3117 
3118 			    if (!UVM_ET_ISNEEDSCOPY(old_entry)) {
3119 			      if (old_entry->max_protection & VM_PROT_WRITE) {
3120 				pmap_protect(old_map->pmap,
3121 					     old_entry->start,
3122 					     old_entry->end,
3123 					     old_entry->protection &
3124 					     ~VM_PROT_WRITE);
3125 			      }
3126 			      old_entry->etype |= UVM_ET_NEEDSCOPY;
3127 			    }
3128 
3129 			    /*
3130 			     * parent must now be write-protected
3131 			     */
3132 			    protect_child = FALSE;
3133 			  } else {
3134 
3135 			    /*
3136 			     * we only need to protect the child if the
3137 			     * parent has write access.
3138 			     */
3139 			    if (old_entry->max_protection & VM_PROT_WRITE)
3140 			      protect_child = TRUE;
3141 			    else
3142 			      protect_child = FALSE;
3143 
3144 			  }
3145 
3146 			  /*
3147 			   * copy the mappings
3148 			   * XXX: need a way to tell if this does anything
3149 			   */
3150 
3151 			  pmap_copy(new_pmap, old_map->pmap,
3152 				    new_entry->start,
3153 				    (old_entry->end - old_entry->start),
3154 				    old_entry->start);
3155 
3156 			  /*
3157 			   * protect the child's mappings if necessary
3158 			   */
3159 			  if (protect_child) {
3160 			    pmap_protect(new_pmap, new_entry->start,
3161 					 new_entry->end,
3162 					 new_entry->protection &
3163 					          ~VM_PROT_WRITE);
3164 			  }
3165 
3166 			}
3167 			break;
3168 		}  /* end of switch statement */
3169 		old_entry = old_entry->next;
3170 	}
3171 
3172 	new_map->size = old_map->size;
3173 	vm_map_unlock(old_map);
3174 
3175 #ifdef SYSVSHM
3176 	if (vm1->vm_shm)
3177 		shmfork(vm1, vm2);
3178 #endif
3179 
3180 #ifdef PMAP_FORK
3181 	pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
3182 #endif
3183 
3184 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3185 	return(vm2);
3186 }
3187 
3188 
3189 #if defined(DDB)
3190 
3191 /*
3192  * DDB hooks
3193  */
3194 
3195 /*
3196  * uvm_map_print: print out a map
3197  */
3198 
3199 void
3200 uvm_map_print(map, full)
3201 	vm_map_t map;
3202 	boolean_t full;
3203 {
3204 
3205 	uvm_map_printit(map, full, printf);
3206 }
3207 
3208 /*
3209  * uvm_map_printit: actually prints the map
3210  */
3211 
3212 void
3213 uvm_map_printit(map, full, pr)
3214 	vm_map_t map;
3215 	boolean_t full;
3216 	int (*pr) __P((const char *, ...));
3217 {
3218 	vm_map_entry_t entry;
3219 
3220 	(*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
3221 	(*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
3222 	    map->nentries, map->size, map->ref_count, map->timestamp,
3223 	    map->flags);
3224 #ifdef pmap_resident_count
3225 	(*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
3226 	    pmap_resident_count(map->pmap));
3227 #else
3228 	/* XXXCDC: this should be required ... */
3229 	(*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap);
3230 #endif
3231 	if (!full)
3232 		return;
3233 	for (entry = map->header.next; entry != &map->header;
3234 	    entry = entry->next) {
3235 		(*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
3236 		    entry, entry->start, entry->end, entry->object.uvm_obj,
3237 		    (long long)entry->offset, entry->aref.ar_amap, entry->aref.ar_pageoff);
3238 		(*pr)(
3239 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, wc=%d, adv=%d\n",
3240 		    (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
3241 		    (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
3242 		    (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
3243 		    entry->protection, entry->max_protection,
3244 		    entry->inheritance, entry->wired_count, entry->advice);
3245 	}
3246 }
3247 
3248 /*
3249  * uvm_object_print: print out an object
3250  */
3251 
3252 void
3253 uvm_object_print(uobj, full)
3254 	struct uvm_object *uobj;
3255 	boolean_t full;
3256 {
3257 
3258 	uvm_object_printit(uobj, full, printf);
3259 }
3260 
3261 /*
3262  * uvm_object_printit: actually prints the object
3263  */
3264 
3265 void
3266 uvm_object_printit(uobj, full, pr)
3267 	struct uvm_object *uobj;
3268 	boolean_t full;
3269 	int (*pr) __P((const char *, ...));
3270 {
3271 	struct vm_page *pg;
3272 	int cnt = 0;
3273 
3274 	(*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
3275 	    uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
3276 	if (UVM_OBJ_IS_KERN_OBJECT(uobj))
3277 		(*pr)("refs=<SYSTEM>\n");
3278 	else
3279 		(*pr)("refs=%d\n", uobj->uo_refs);
3280 
3281 	if (!full) {
3282 		return;
3283 	}
3284 	(*pr)("  PAGES <pg,offset>:\n  ");
3285 	for (pg = TAILQ_FIRST(&uobj->memq);
3286 	     pg != NULL;
3287 	     pg = TAILQ_NEXT(pg, listq), cnt++) {
3288 		(*pr)("<%p,0x%lx> ", pg, pg->offset);
3289 		if ((cnt % 3) == 2) {
3290 			(*pr)("\n  ");
3291 		}
3292 	}
3293 	if ((cnt % 3) != 2) {
3294 		(*pr)("\n");
3295 	}
3296 }
3297 
3298 const char page_flagbits[] =
3299 	"\20\4CLEAN\5BUSY\6WANTED\7TABLED\12FAKE\13FILLED\14DIRTY\15RELEASED"
3300 	"\16FAULTING\17CLEANCHK";
3301 const char page_pqflagbits[] =
3302 	"\20\1FREE\2INACTIVE\3ACTIVE\4LAUNDRY\5ANON\6AOBJ";
3303 
3304 /*
3305  * uvm_page_print: print out a page
3306  */
3307 
3308 void
3309 uvm_page_print(pg, full)
3310 	struct vm_page *pg;
3311 	boolean_t full;
3312 {
3313 
3314 	uvm_page_printit(pg, full, printf);
3315 }
3316 
3317 /*
3318  * uvm_page_printit: actually print the page
3319  */
3320 
3321 void
3322 uvm_page_printit(pg, full, pr)
3323 	struct vm_page *pg;
3324 	boolean_t full;
3325 	int (*pr) __P((const char *, ...));
3326 {
3327 	struct vm_page *lcv;
3328 	struct uvm_object *uobj;
3329 	struct pglist *pgl;
3330 	char pgbuf[128];
3331 	char pqbuf[128];
3332 
3333 	(*pr)("PAGE %p:\n", pg);
3334 	snprintf(pgbuf, sizeof(pgbuf), "%b", pg->flags, page_flagbits);
3335 	snprintf(pqbuf, sizeof(pqbuf), "%b", pg->pqflags, page_pqflagbits);
3336 	(*pr)("  flags=%s, pqflags=%s, vers=%d, wire_count=%d, pa=0x%lx\n",
3337 	    pgbuf, pqbuf, pg->version, pg->wire_count, (long)pg->phys_addr);
3338 	(*pr)("  uobject=%p, uanon=%p, offset=0x%lx loan_count=%d\n",
3339 	    pg->uobject, pg->uanon, pg->offset, pg->loan_count);
3340 #if defined(UVM_PAGE_TRKOWN)
3341 	if (pg->flags & PG_BUSY)
3342 		(*pr)("  owning process = %d, tag=%s\n",
3343 		    pg->owner, pg->owner_tag);
3344 	else
3345 		(*pr)("  page not busy, no owner\n");
3346 #else
3347 	(*pr)("  [page ownership tracking disabled]\n");
3348 #endif
3349 
3350 	if (!full)
3351 		return;
3352 
3353 	/* cross-verify object/anon */
3354 	if ((pg->pqflags & PQ_FREE) == 0) {
3355 		if (pg->pqflags & PQ_ANON) {
3356 			if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
3357 			    (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
3358 				(pg->uanon) ? pg->uanon->u.an_page : NULL);
3359 			else
3360 				(*pr)("  anon backpointer is OK\n");
3361 		} else {
3362 			uobj = pg->uobject;
3363 			if (uobj) {
3364 				(*pr)("  checking object list\n");
3365 				for (lcv = uobj->memq.tqh_first ; lcv ;
3366 				    lcv = lcv->listq.tqe_next) {
3367 					if (lcv == pg) break;
3368 				}
3369 				if (lcv)
3370 					(*pr)("  page found on object list\n");
3371 				else
3372 			(*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
3373 			}
3374 		}
3375 	}
3376 
3377 	/* cross-verify page queue */
3378 	if (pg->pqflags & PQ_FREE) {
3379 		int fl = uvm_page_lookup_freelist(pg);
3380 		pgl = &uvm.page_free[fl].pgfl_queues[((pg)->flags & PG_ZERO) ?
3381 		    PGFL_ZEROS : PGFL_UNKNOWN];
3382 	}
3383 	else if (pg->pqflags & PQ_INACTIVE)
3384 		pgl = (pg->pqflags & PQ_SWAPBACKED) ?
3385 		    &uvm.page_inactive_swp : &uvm.page_inactive_obj;
3386 	else if (pg->pqflags & PQ_ACTIVE)
3387 		pgl = &uvm.page_active;
3388 	else
3389 		pgl = NULL;
3390 
3391 	if (pgl) {
3392 		(*pr)("  checking pageq list\n");
3393 		for (lcv = pgl->tqh_first ; lcv ; lcv = lcv->pageq.tqe_next) {
3394 			if (lcv == pg) break;
3395 		}
3396 		if (lcv)
3397 			(*pr)("  page found on pageq list\n");
3398 		else
3399 			(*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
3400 	}
3401 }
3402 #endif
3403