xref: /netbsd-src/sys/uvm/uvm_map.c (revision bada23909e740596d0a3785a73bd3583a9807fb8)
1 /*	$NetBSD: uvm_map.c,v 1.34 1999/01/24 23:53:15 chuck Exp $	*/
2 
3 /*
4  * XXXCDC: "ROUGH DRAFT" QUALITY UVM PRE-RELEASE FILE!
5  *         >>>USE AT YOUR OWN RISK, WORK IS NOT FINISHED<<<
6  */
7 /*
8  * Copyright (c) 1997 Charles D. Cranor and Washington University.
9  * Copyright (c) 1991, 1993, The Regents of the University of California.
10  *
11  * All rights reserved.
12  *
13  * This code is derived from software contributed to Berkeley by
14  * The Mach Operating System project at Carnegie-Mellon University.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  * 3. All advertising materials mentioning features or use of this software
25  *    must display the following acknowledgement:
26  *	This product includes software developed by Charles D. Cranor,
27  *      Washington University, the University of California, Berkeley and
28  *      its contributors.
29  * 4. Neither the name of the University nor the names of its contributors
30  *    may be used to endorse or promote products derived from this software
31  *    without specific prior written permission.
32  *
33  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
34  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
35  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
36  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
37  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
38  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
39  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
40  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
41  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
42  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43  * SUCH DAMAGE.
44  *
45  *	@(#)vm_map.c    8.3 (Berkeley) 1/12/94
46  * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp
47  *
48  *
49  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
50  * All rights reserved.
51  *
52  * Permission to use, copy, modify and distribute this software and
53  * its documentation is hereby granted, provided that both the copyright
54  * notice and this permission notice appear in all copies of the
55  * software, derivative works or modified versions, and any portions
56  * thereof, and that both notices appear in supporting documentation.
57  *
58  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
59  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
60  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
61  *
62  * Carnegie Mellon requests users of this software to return to
63  *
64  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
65  *  School of Computer Science
66  *  Carnegie Mellon University
67  *  Pittsburgh PA 15213-3890
68  *
69  * any improvements or extensions that they make and grant Carnegie the
70  * rights to redistribute these changes.
71  */
72 
73 #include "opt_ddb.h"
74 #include "opt_uvmhist.h"
75 #include "opt_pmap_new.h"
76 #include "opt_sysv.h"
77 
78 /*
79  * uvm_map.c: uvm map operations
80  */
81 
82 #include <sys/param.h>
83 #include <sys/systm.h>
84 #include <sys/mman.h>
85 #include <sys/proc.h>
86 #include <sys/malloc.h>
87 #include <sys/pool.h>
88 
89 #ifdef SYSVSHM
90 #include <sys/shm.h>
91 #endif
92 
93 #include <vm/vm.h>
94 #include <vm/vm_page.h>
95 #include <vm/vm_kern.h>
96 
97 #define UVM_MAP
98 #include <uvm/uvm.h>
99 
100 #ifdef DDB
101 #include <uvm/uvm_ddb.h>
102 #endif
103 
104 
105 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge;
106 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint;
107 
108 /*
109  * pool for vmspace structures.
110  */
111 
112 struct pool uvm_vmspace_pool;
113 
114 /*
115  * pool for dynamically-allocated map entries.
116  */
117 
118 struct pool uvm_map_entry_pool;
119 
120 /*
121  * macros
122  */
123 
124 /*
125  * uvm_map_entry_link: insert entry into a map
126  *
127  * => map must be locked
128  */
129 #define uvm_map_entry_link(map, after_where, entry) do { \
130 	(map)->nentries++; \
131 	(entry)->prev = (after_where); \
132 	(entry)->next = (after_where)->next; \
133 	(entry)->prev->next = (entry); \
134 	(entry)->next->prev = (entry); \
135 } while (0)
136 
137 /*
138  * uvm_map_entry_unlink: remove entry from a map
139  *
140  * => map must be locked
141  */
142 #define uvm_map_entry_unlink(map, entry) do { \
143 	(map)->nentries--; \
144 	(entry)->next->prev = (entry)->prev; \
145 	(entry)->prev->next = (entry)->next; \
146 } while (0)
147 
148 /*
149  * SAVE_HINT: saves the specified entry as the hint for future lookups.
150  *
151  * => map need not be locked (protected by hint_lock).
152  */
153 #define SAVE_HINT(map,value) do { \
154 	simple_lock(&(map)->hint_lock); \
155 	(map)->hint = (value); \
156 	simple_unlock(&(map)->hint_lock); \
157 } while (0)
158 
159 /*
160  * VM_MAP_RANGE_CHECK: check and correct range
161  *
162  * => map must at least be read locked
163  */
164 
165 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
166 	if (start < vm_map_min(map)) 		\
167 		start = vm_map_min(map);        \
168 	if (end > vm_map_max(map))              \
169 		end = vm_map_max(map);          \
170 	if (start > end)                        \
171 		start = end;                    \
172 } while (0)
173 
174 /*
175  * local prototypes
176  */
177 
178 static vm_map_entry_t	uvm_mapent_alloc __P((vm_map_t));
179 static void		uvm_mapent_copy __P((vm_map_entry_t,vm_map_entry_t));
180 static void		uvm_mapent_free __P((vm_map_entry_t));
181 static void		uvm_map_entry_unwire __P((vm_map_t, vm_map_entry_t));
182 
183 /*
184  * local inlines
185  */
186 
187 /*
188  * uvm_mapent_alloc: allocate a map entry
189  *
190  * => XXX: static pool for kernel map?
191  */
192 
193 static __inline vm_map_entry_t
194 uvm_mapent_alloc(map)
195 	vm_map_t map;
196 {
197 	vm_map_entry_t me;
198 	int s;
199 	UVMHIST_FUNC("uvm_mapent_alloc");
200 	UVMHIST_CALLED(maphist);
201 
202 	if (map->entries_pageable) {
203 		me = pool_get(&uvm_map_entry_pool, PR_WAITOK);
204 		me->flags = 0;
205 		/* me can't be null, wait ok */
206 
207 	} else {
208 		s = splimp();	/* protect kentry_free list with splimp */
209 		simple_lock(&uvm.kentry_lock);
210 		me = uvm.kentry_free;
211 		if (me) uvm.kentry_free = me->next;
212 		simple_unlock(&uvm.kentry_lock);
213 		splx(s);
214 		if (!me)
215 	panic("mapent_alloc: out of kernel map entries, check MAX_KMAPENT");
216 		me->flags = UVM_MAP_STATIC;
217 	}
218 
219 	UVMHIST_LOG(maphist, "<- new entry=0x%x [pageable=%d]",
220 		me, map->entries_pageable, 0, 0);
221 	return(me);
222 
223 }
224 
225 /*
226  * uvm_mapent_free: free map entry
227  *
228  * => XXX: static pool for kernel map?
229  */
230 
231 static __inline void
232 uvm_mapent_free(me)
233 	vm_map_entry_t me;
234 {
235 	int s;
236 	UVMHIST_FUNC("uvm_mapent_free");
237 	UVMHIST_CALLED(maphist);
238 	UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
239 		me, me->flags, 0, 0);
240 	if ((me->flags & UVM_MAP_STATIC) == 0) {
241 		pool_put(&uvm_map_entry_pool, me);
242 	} else {
243 		s = splimp();	/* protect kentry_free list with splimp */
244 		simple_lock(&uvm.kentry_lock);
245 		me->next = uvm.kentry_free;
246 		uvm.kentry_free = me;
247 		simple_unlock(&uvm.kentry_lock);
248 		splx(s);
249 	}
250 }
251 
252 /*
253  * uvm_mapent_copy: copy a map entry, preserving flags
254  */
255 
256 static __inline void
257 uvm_mapent_copy(src, dst)
258 	vm_map_entry_t src;
259 	vm_map_entry_t dst;
260 {
261 
262 	memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - ((char*)src));
263 }
264 
265 /*
266  * uvm_map_entry_unwire: unwire a map entry
267  *
268  * => map should be locked by caller
269  */
270 
271 static __inline void
272 uvm_map_entry_unwire(map, entry)
273 	vm_map_t map;
274 	vm_map_entry_t entry;
275 {
276 
277 	uvm_fault_unwire(map->pmap, entry->start, entry->end);
278 	entry->wired_count = 0;
279 }
280 
281 /*
282  * uvm_map_init: init mapping system at boot time.   note that we allocate
283  * and init the static pool of vm_map_entry_t's for the kernel here.
284  */
285 
286 void
287 uvm_map_init()
288 {
289 	static struct vm_map_entry kernel_map_entry[MAX_KMAPENT];
290 #if defined(UVMHIST)
291 	static struct uvm_history_ent maphistbuf[100];
292 	static struct uvm_history_ent pdhistbuf[100];
293 #endif
294 	int lcv;
295 
296 	/*
297 	 * first, init logging system.
298 	 */
299 
300 	UVMHIST_FUNC("uvm_map_init");
301 	UVMHIST_INIT_STATIC(maphist, maphistbuf);
302 	UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
303 	UVMHIST_CALLED(maphist);
304 	UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
305 	UVMCNT_INIT(uvm_map_call,  UVMCNT_CNT, 0,
306 	    "# uvm_map() successful calls", 0);
307 	UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0);
308 	UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward",
309 	    0);
310 	UVMCNT_INIT(uvm_mlk_call,  UVMCNT_CNT, 0, "# map lookup calls", 0);
311 	UVMCNT_INIT(uvm_mlk_hint,  UVMCNT_CNT, 0, "# map lookup hint hits", 0);
312 
313 	/*
314 	 * now set up static pool of kernel map entrys ...
315 	 */
316 
317 	simple_lock_init(&uvm.kentry_lock);
318 	uvm.kentry_free = NULL;
319 	for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) {
320 		kernel_map_entry[lcv].next = uvm.kentry_free;
321 		uvm.kentry_free = &kernel_map_entry[lcv];
322 	}
323 
324 	/*
325 	 * initialize the map-related pools.
326 	 */
327 	pool_init(&uvm_vmspace_pool, sizeof(struct vmspace),
328 	    0, 0, 0, "vmsppl", 0,
329 	    pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
330 	pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry),
331 	    0, 0, 0, "vmmpepl", 0,
332 	    pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP);
333 }
334 
335 /*
336  * clippers
337  */
338 
339 /*
340  * uvm_map_clip_start: ensure that the entry begins at or after
341  *	the starting address, if it doesn't we split the entry.
342  *
343  * => caller should use UVM_MAP_CLIP_START macro rather than calling
344  *    this directly
345  * => map must be locked by caller
346  */
347 
348 void uvm_map_clip_start(map, entry, start)
349 	vm_map_t       map;
350 	vm_map_entry_t entry;
351 	vaddr_t    start;
352 {
353 	vm_map_entry_t new_entry;
354 	vaddr_t new_adj;
355 
356 	/* uvm_map_simplify_entry(map, entry); */ /* XXX */
357 
358 	/*
359 	 * Split off the front portion.  note that we must insert the new
360 	 * entry BEFORE this one, so that this entry has the specified
361 	 * starting address.
362 	 */
363 
364 	new_entry = uvm_mapent_alloc(map);
365 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
366 
367 	new_entry->end = start;
368 	new_adj = start - new_entry->start;
369 	if (entry->object.uvm_obj)
370 		entry->offset += new_adj;	/* shift start over */
371 	entry->start = start;
372 
373 	if (new_entry->aref.ar_amap) {
374 		amap_splitref(&new_entry->aref, &entry->aref, new_adj);
375 	}
376 
377 	uvm_map_entry_link(map, entry->prev, new_entry);
378 
379 	if (UVM_ET_ISSUBMAP(entry)) {
380 		/* ... unlikely to happen, but play it safe */
381 		 uvm_map_reference(new_entry->object.sub_map);
382 	} else {
383 		if (UVM_ET_ISOBJ(entry) &&
384 		    entry->object.uvm_obj->pgops &&
385 		    entry->object.uvm_obj->pgops->pgo_reference)
386 			entry->object.uvm_obj->pgops->pgo_reference(
387 			    entry->object.uvm_obj);
388 	}
389 }
390 
391 /*
392  * uvm_map_clip_end: ensure that the entry ends at or before
393  *	the ending address, if it does't we split the reference
394  *
395  * => caller should use UVM_MAP_CLIP_END macro rather than calling
396  *    this directly
397  * => map must be locked by caller
398  */
399 
400 void
401 uvm_map_clip_end(map, entry, end)
402 	vm_map_t	map;
403 	vm_map_entry_t	entry;
404 	vaddr_t	end;
405 {
406 	vm_map_entry_t	new_entry;
407 	vaddr_t new_adj; /* #bytes we move start forward */
408 
409 	/*
410 	 *	Create a new entry and insert it
411 	 *	AFTER the specified entry
412 	 */
413 
414 	new_entry = uvm_mapent_alloc(map);
415 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
416 
417 	new_entry->start = entry->end = end;
418 	new_adj = end - entry->start;
419 	if (new_entry->object.uvm_obj)
420 		new_entry->offset += new_adj;
421 
422 	if (entry->aref.ar_amap)
423 		amap_splitref(&entry->aref, &new_entry->aref, new_adj);
424 
425 	uvm_map_entry_link(map, entry, new_entry);
426 
427 	if (UVM_ET_ISSUBMAP(entry)) {
428 		/* ... unlikely to happen, but play it safe */
429 	 	uvm_map_reference(new_entry->object.sub_map);
430 	} else {
431 		if (UVM_ET_ISOBJ(entry) &&
432 		    entry->object.uvm_obj->pgops &&
433 		    entry->object.uvm_obj->pgops->pgo_reference)
434 			entry->object.uvm_obj->pgops->pgo_reference(
435 			    entry->object.uvm_obj);
436 	}
437 }
438 
439 
440 /*
441  *   M A P   -   m a i n   e n t r y   p o i n t
442  */
443 /*
444  * uvm_map: establish a valid mapping in a map
445  *
446  * => assume startp is page aligned.
447  * => assume size is a multiple of PAGE_SIZE.
448  * => assume sys_mmap provides enough of a "hint" to have us skip
449  *	over text/data/bss area.
450  * => map must be unlocked (we will lock it)
451  * => <uobj,uoffset> value meanings (4 cases):
452  *	 [1] <NULL,uoffset> 		== uoffset is a hint for PMAP_PREFER
453  *	 [2] <NULL,UVM_UNKNOWN_OFFSET>	== don't PMAP_PREFER
454  *	 [3] <uobj,uoffset>		== normal mapping
455  *	 [4] <uobj,UVM_UNKNOWN_OFFSET>	== uvm_map finds offset based on VA
456  *
457  *    case [4] is for kernel mappings where we don't know the offset until
458  *    we've found a virtual address.   note that kernel object offsets are
459  *    always relative to vm_map_min(kernel_map).
460  * => XXXCDC: need way to map in external amap?
461  */
462 
463 int
464 uvm_map(map, startp, size, uobj, uoffset, flags)
465 	vm_map_t map;
466 	vaddr_t *startp;	/* IN/OUT */
467 	vsize_t size;
468 	struct uvm_object *uobj;
469 	vaddr_t uoffset;
470 	uvm_flag_t flags;
471 {
472 	vm_map_entry_t prev_entry, new_entry;
473 	vm_prot_t prot = UVM_PROTECTION(flags), maxprot =
474 	    UVM_MAXPROTECTION(flags);
475 	vm_inherit_t inherit = UVM_INHERIT(flags);
476 	int advice = UVM_ADVICE(flags);
477 	UVMHIST_FUNC("uvm_map");
478 	UVMHIST_CALLED(maphist);
479 
480 	UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)",
481 	    map, *startp, size, flags);
482 	UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
483 
484 	/*
485 	 * step 0: sanity check of protection code
486 	 */
487 
488 	if ((prot & maxprot) != prot) {
489 		UVMHIST_LOG(maphist, "<- prot. failure:  prot=0x%x, max=0x%x",
490 		prot, maxprot,0,0);
491 		return(KERN_PROTECTION_FAILURE);
492 	}
493 
494 	/*
495 	 * step 1: figure out where to put new VM range
496 	 */
497 
498 	if (vm_map_lock_try(map) == FALSE) {
499 		if (flags & UVM_FLAG_TRYLOCK)
500 			return(KERN_FAILURE);
501 		vm_map_lock(map); /* could sleep here */
502 	}
503 	if ((prev_entry = uvm_map_findspace(map, *startp, size, startp,
504 	    uobj, uoffset, flags & UVM_FLAG_FIXED)) == NULL) {
505 		UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0);
506 		vm_map_unlock(map);
507 		return (KERN_NO_SPACE);
508 	}
509 
510 #if defined(PMAP_GROWKERNEL)	/* hack */
511 	{
512 		/* locked by kernel_map lock */
513 		static vaddr_t maxkaddr = 0;
514 
515 		/*
516 		 * hack: grow kernel PTPs in advance.
517 		 */
518 		if (map == kernel_map && maxkaddr < (*startp + size)) {
519 			pmap_growkernel(*startp + size);
520 			maxkaddr = *startp + size;
521 		}
522 	}
523 #endif
524 
525 	UVMCNT_INCR(uvm_map_call);
526 
527 	/*
528 	 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
529 	 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET.   in
530 	 * either case we want to zero it  before storing it in the map entry
531 	 * (because it looks strange and confusing when debugging...)
532 	 *
533 	 * if uobj is not null
534 	 *   if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
535 	 *      and we do not need to change uoffset.
536 	 *   if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
537 	 *      now (based on the starting address of the map).   this case is
538 	 *      for kernel object mappings where we don't know the offset until
539 	 *      the virtual address is found (with uvm_map_findspace).   the
540 	 *      offset is the distance we are from the start of the map.
541 	 */
542 
543 	if (uobj == NULL) {
544 		uoffset = 0;
545 	} else {
546 		if (uoffset == UVM_UNKNOWN_OFFSET) {
547 #ifdef DIAGNOSTIC
548 			if (uobj->uo_refs != UVM_OBJ_KERN)
549 	panic("uvm_map: unknown offset with 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 		/* wired_count's must match (new area is unwired) */
582 		if (prev_entry->wired_count)
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_amap = NULL;
671 	}
672 
673 	uvm_map_entry_link(map, prev_entry, new_entry);
674 
675 	map->size += size;
676 
677 	/*
678 	 *      Update the free space hint
679 	 */
680 
681 	if ((map->first_free == prev_entry) &&
682 	    (prev_entry->end >= new_entry->start))
683 		map->first_free = new_entry;
684 
685 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
686 	vm_map_unlock(map);
687 	return(KERN_SUCCESS);
688 }
689 
690 /*
691  * uvm_map_lookup_entry: find map entry at or before an address
692  *
693  * => map must at least be read-locked by caller
694  * => entry is returned in "entry"
695  * => return value is true if address is in the returned entry
696  */
697 
698 boolean_t
699 uvm_map_lookup_entry(map, address, entry)
700 	vm_map_t	map;
701 	vaddr_t	address;
702 	vm_map_entry_t		*entry;		/* OUT */
703 {
704 	vm_map_entry_t		cur;
705 	vm_map_entry_t		last;
706 	UVMHIST_FUNC("uvm_map_lookup_entry");
707 	UVMHIST_CALLED(maphist);
708 
709 	UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
710 	    map, address, entry, 0);
711 
712 	/*
713 	 * start looking either from the head of the
714 	 * list, or from the hint.
715 	 */
716 
717 	simple_lock(&map->hint_lock);
718 	cur = map->hint;
719 	simple_unlock(&map->hint_lock);
720 
721 	if (cur == &map->header)
722 		cur = cur->next;
723 
724 	UVMCNT_INCR(uvm_mlk_call);
725 	if (address >= cur->start) {
726 	    	/*
727 		 * go from hint to end of list.
728 		 *
729 		 * but first, make a quick check to see if
730 		 * we are already looking at the entry we
731 		 * want (which is usually the case).
732 		 * note also that we don't need to save the hint
733 		 * here... it is the same hint (unless we are
734 		 * at the header, in which case the hint didn't
735 		 * buy us anything anyway).
736 		 */
737 		last = &map->header;
738 		if ((cur != last) && (cur->end > address)) {
739 			UVMCNT_INCR(uvm_mlk_hint);
740 			*entry = cur;
741 			UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
742 			    cur, 0, 0, 0);
743 			return (TRUE);
744 		}
745 	} else {
746 	    	/*
747 		 * go from start to hint, *inclusively*
748 		 */
749 		last = cur->next;
750 		cur = map->header.next;
751 	}
752 
753 	/*
754 	 * search linearly
755 	 */
756 
757 	while (cur != last) {
758 		if (cur->end > address) {
759 			if (address >= cur->start) {
760 			    	/*
761 				 * save this lookup for future
762 				 * hints, and return
763 				 */
764 
765 				*entry = cur;
766 				SAVE_HINT(map, cur);
767 				UVMHIST_LOG(maphist,"<- search got it (0x%x)",
768 					cur, 0, 0, 0);
769 				return (TRUE);
770 			}
771 			break;
772 		}
773 		cur = cur->next;
774 	}
775 	*entry = cur->prev;
776 	SAVE_HINT(map, *entry);
777 	UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
778 	return (FALSE);
779 }
780 
781 
782 /*
783  * uvm_map_findspace: find "length" sized space in "map".
784  *
785  * => "hint" is a hint about where we want it, unless fixed is true
786  *	(in which case we insist on using "hint").
787  * => "result" is VA returned
788  * => uobj/uoffset are to be used to handle VAC alignment, if required
789  * => caller must at least have read-locked map
790  * => returns NULL on failure, or pointer to prev. map entry if success
791  * => note this is a cross between the old vm_map_findspace and vm_map_find
792  */
793 
794 vm_map_entry_t
795 uvm_map_findspace(map, hint, length, result, uobj, uoffset, fixed)
796 	vm_map_t map;
797 	vaddr_t hint;
798 	vsize_t length;
799 	vaddr_t *result; /* OUT */
800 	struct uvm_object *uobj;
801 	vaddr_t uoffset;
802 	boolean_t fixed;
803 {
804 	vm_map_entry_t entry, next, tmp;
805 	vaddr_t end;
806 	UVMHIST_FUNC("uvm_map_findspace");
807 	UVMHIST_CALLED(maphist);
808 
809 	UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, fixed=%d)",
810 		map, hint, length, fixed);
811 
812 	if (hint < map->min_offset) {	/* check ranges ... */
813 		if (fixed) {
814 			UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
815 			return(NULL);
816 		}
817 		hint = map->min_offset;
818 	}
819 	if (hint > map->max_offset) {
820 		UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
821 				hint, map->min_offset, map->max_offset, 0);
822 		return(NULL);
823 	}
824 
825 	/*
826 	 * Look for the first possible address; if there's already
827 	 * something at this address, we have to start after it.
828 	 */
829 
830 	if (!fixed && hint == map->min_offset) {
831 		if ((entry = map->first_free) != &map->header)
832 			hint = entry->end;
833 	} else {
834 		if (uvm_map_lookup_entry(map, hint, &tmp)) {
835 			/* "hint" address already in use ... */
836 			if (fixed) {
837 				UVMHIST_LOG(maphist,"<- fixed & VA in use",
838 				    0, 0, 0, 0);
839 				return(NULL);
840 			}
841 			hint = tmp->end;
842 		}
843 		entry = tmp;
844 	}
845 
846 	/*
847 	 * Look through the rest of the map, trying to fit a new region in
848 	 * the gap between existing regions, or after the very last region.
849 	 * note: entry->end   = base VA of current gap,
850 	 *	 next->start  = VA of end of current gap
851 	 */
852 	for (;; hint = (entry = next)->end) {
853 		/*
854 		 * Find the end of the proposed new region.  Be sure we didn't
855 		 * go beyond the end of the map, or wrap around the address;
856 		 * if so, we lose.  Otherwise, if this is the last entry, or
857 		 * if the proposed new region fits before the next entry, we
858 		 * win.
859 		 */
860 
861 #ifdef PMAP_PREFER
862 		/*
863 		 * push hint forward as needed to avoid VAC alias problems.
864 		 * we only do this if a valid offset is specified.
865 		 */
866 		if (!fixed && uoffset != UVM_UNKNOWN_OFFSET)
867 		  PMAP_PREFER(uoffset, &hint);
868 #endif
869 		end = hint + length;
870 		if (end > map->max_offset || end < hint) {
871 			UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0);
872 			return (NULL);
873 		}
874 		next = entry->next;
875 		if (next == &map->header || next->start >= end)
876 			break;
877 		if (fixed) {
878 			UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0);
879 			return(NULL); /* only one shot at it ... */
880 		}
881 	}
882 	SAVE_HINT(map, entry);
883 	*result = hint;
884 	UVMHIST_LOG(maphist,"<- got it!  (result=0x%x)", hint, 0,0,0);
885 	return (entry);
886 }
887 
888 /*
889  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
890  */
891 
892 /*
893  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
894  *
895  * => caller must check alignment and size
896  * => map must be locked by caller
897  * => we return a list of map entries that we've remove from the map
898  *    in "entry_list"
899  */
900 
901 int
902 uvm_unmap_remove(map, start, end, entry_list)
903 	vm_map_t map;
904 	vaddr_t start,end;
905 	vm_map_entry_t *entry_list;	/* OUT */
906 {
907 	vm_map_entry_t entry, first_entry, next;
908 	vaddr_t len;
909 	UVMHIST_FUNC("uvm_unmap_remove");
910 	UVMHIST_CALLED(maphist);
911 
912 	UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
913 	    map, start, end, 0);
914 
915 	VM_MAP_RANGE_CHECK(map, start, end);
916 
917 	/*
918 	 * find first entry
919 	 */
920 	if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
921 		/* clip and go... */
922 		entry = first_entry;
923 		UVM_MAP_CLIP_START(map, entry, start);
924 		/* critical!  prevents stale hint */
925 		SAVE_HINT(map, entry->prev);
926 
927 	} else {
928 		entry = first_entry->next;
929 	}
930 
931 	/*
932 	 * Save the free space hint
933 	 */
934 
935 	if (map->first_free->start >= start)
936 		map->first_free = entry->prev;
937 
938 	/*
939 	 * note: we now re-use first_entry for a different task.  we remove
940 	 * a number of map entries from the map and save them in a linked
941 	 * list headed by "first_entry".  once we remove them from the map
942 	 * the caller should unlock the map and drop the references to the
943 	 * backing objects [c.f. uvm_unmap_detach].  the object is to
944 	 * seperate unmapping from reference dropping.  why?
945 	 *   [1] the map has to be locked for unmapping
946 	 *   [2] the map need not be locked for reference dropping
947 	 *   [3] dropping references may trigger pager I/O, and if we hit
948 	 *       a pager that does synchronous I/O we may have to wait for it.
949 	 *   [4] we would like all waiting for I/O to occur with maps unlocked
950 	 *       so that we don't block other threads.
951 	 */
952 	first_entry = NULL;
953 	*entry_list = NULL;		/* to be safe */
954 
955 	/*
956 	 * break up the area into map entry sized regions and unmap.  note
957 	 * that all mappings have to be removed before we can even consider
958 	 * dropping references to amaps or VM objects (otherwise we could end
959 	 * up with a mapping to a page on the free list which would be very bad)
960 	 */
961 
962 	while ((entry != &map->header) && (entry->start < end)) {
963 
964 		UVM_MAP_CLIP_END(map, entry, end);
965 		next = entry->next;
966 		len = entry->end - entry->start;
967 
968 		/*
969 		 * unwire before removing addresses from the pmap; otherwise
970 		 * unwiring will put the entries back into the pmap (XXX).
971 		 */
972 
973 		if (entry->wired_count)
974 			uvm_map_entry_unwire(map, entry);
975 
976 		/*
977 		 * special case: handle mappings to anonymous kernel objects.
978 		 * we want to free these pages right away...
979 		 */
980 		if (UVM_ET_ISOBJ(entry) &&
981 		    entry->object.uvm_obj->uo_refs == UVM_OBJ_KERN) {
982 
983 #ifdef DIAGNOSTIC
984 			if (vm_map_pmap(map) != pmap_kernel())
985 	panic("uvm_unmap_remove: kernel object mapped by non-kernel map");
986 #endif
987 
988 			/*
989 			 * note: kernel object mappings are currently used in
990 			 * two ways:
991 			 *  [1] "normal" mappings of pages in the kernel object
992 			 *  [2] uvm_km_valloc'd allocations in which we
993 			 *      pmap_enter in some non-kernel-object page
994 			 *      (e.g. vmapbuf).
995 			 *
996 			 * for case [1], we need to remove the mapping from
997 			 * the pmap and then remove the page from the kernel
998 			 * object (because, once pages in a kernel object are
999 			 * unmapped they are no longer needed, unlike, say,
1000 			 * a vnode where you might want the data to persist
1001 			 * until flushed out of a queue).
1002 			 *
1003 			 * for case [2], we need to remove the mapping from
1004 			 * the pmap.  there shouldn't be any pages at the
1005 			 * specified offset in the kernel object [but it
1006 			 * doesn't hurt to call uvm_km_pgremove just to be
1007 			 * safe?]
1008 			 *
1009 			 * uvm_km_pgremove currently does the following:
1010 			 *   for pages in the kernel object in range:
1011 			 *     - pmap_page_protect them out of all pmaps
1012 			 *     - uvm_pagefree the page
1013 			 *
1014 			 * note that in case [1] the pmap_page_protect call
1015 			 * in uvm_km_pgremove may very well be redundant
1016 			 * because we have already removed the mappings
1017 			 * beforehand with pmap_remove (or pmap_kremove).
1018 			 * in the PMAP_NEW case, the pmap_page_protect call
1019 			 * may not do anything, since PMAP_NEW allows the
1020 			 * kernel to enter/remove kernel mappings without
1021 			 * bothing to keep track of the mappings (e.g. via
1022 			 * pv_entry lists).    XXX: because of this, in the
1023 			 * future we should consider removing the
1024 			 * pmap_page_protect from uvm_km_pgremove some time
1025 			 * in the future.
1026 			 */
1027 
1028 			/*
1029 			 * remove mappings from pmap
1030 			 */
1031 #if defined(PMAP_NEW)
1032 			pmap_kremove(entry->start, len);
1033 #else
1034 			pmap_remove(pmap_kernel(), entry->start,
1035 			    entry->start+len);
1036 #endif
1037 
1038 			/*
1039 			 * remove pages from a kernel object (offsets are
1040 			 * always relative to vm_map_min(kernel_map)).
1041 			 */
1042 			uvm_km_pgremove(entry->object.uvm_obj,
1043 			entry->start - vm_map_min(kernel_map),
1044 			entry->end - vm_map_min(kernel_map));
1045 
1046 			/*
1047 			 * null out kernel_object reference, we've just
1048 			 * dropped it
1049 			 */
1050 			entry->etype &= ~UVM_ET_OBJ;
1051 			entry->object.uvm_obj = NULL;	/* to be safe */
1052 
1053 		} else {
1054 			/*
1055 		 	 * remove mappings the standard way.
1056 		 	 */
1057 			pmap_remove(map->pmap, entry->start, entry->end);
1058 		}
1059 
1060 		/*
1061 		 * remove entry from map and put it on our list of entries
1062 		 * that we've nuked.  then go do next entry.
1063 		 */
1064 		UVMHIST_LOG(maphist, "  removed map entry 0x%x", entry, 0, 0,0);
1065 		uvm_map_entry_unlink(map, entry);
1066 		map->size -= len;
1067 		entry->next = first_entry;
1068 		first_entry = entry;
1069 		entry = next;		/* next entry, please */
1070 	}
1071 
1072 	/*
1073 	 * now we've cleaned up the map and are ready for the caller to drop
1074 	 * references to the mapped objects.
1075 	 */
1076 
1077 	*entry_list = first_entry;
1078 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1079 	return(KERN_SUCCESS);
1080 }
1081 
1082 /*
1083  * uvm_unmap_detach: drop references in a chain of map entries
1084  *
1085  * => we will free the map entries as we traverse the list.
1086  */
1087 
1088 void
1089 uvm_unmap_detach(first_entry, amap_unref_flags)
1090 	vm_map_entry_t first_entry;
1091 	int amap_unref_flags;
1092 {
1093 	vm_map_entry_t next_entry;
1094 	UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1095 
1096 	while (first_entry) {
1097 
1098 #ifdef DIAGNOSTIC
1099 		/*
1100 		 * sanity check
1101 		 */
1102 		/* was part of vm_map_entry_delete() */
1103 		if (first_entry->wired_count)
1104 			panic("unmap: still wired!");
1105 #endif
1106 
1107 		UVMHIST_LOG(maphist,
1108 		    "  detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
1109 		    first_entry, first_entry->aref.ar_amap,
1110 		    first_entry->object.uvm_obj,
1111 		    UVM_ET_ISSUBMAP(first_entry));
1112 
1113 		/*
1114 		 * drop reference to amap, if we've got one
1115 		 */
1116 
1117 		if (first_entry->aref.ar_amap)
1118 			amap_unref(first_entry, amap_unref_flags);
1119 
1120 		/*
1121 		 * drop reference to our backing object, if we've got one
1122 		 */
1123 
1124 		if (UVM_ET_ISSUBMAP(first_entry)) {
1125 			/* ... unlikely to happen, but play it safe */
1126 			uvm_map_deallocate(first_entry->object.sub_map);
1127 		} else {
1128 			if (UVM_ET_ISOBJ(first_entry) &&
1129 			    first_entry->object.uvm_obj->pgops->pgo_detach)
1130 				first_entry->object.uvm_obj->pgops->
1131 				    pgo_detach(first_entry->object.uvm_obj);
1132 		}
1133 
1134 		/*
1135 		 * next entry
1136 		 */
1137 		next_entry = first_entry->next;
1138 		uvm_mapent_free(first_entry);
1139 		first_entry = next_entry;
1140 	}
1141 
1142 	/*
1143 	 * done!
1144 	 */
1145 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1146 	return;
1147 }
1148 
1149 /*
1150  *   E X T R A C T I O N   F U N C T I O N S
1151  */
1152 
1153 /*
1154  * uvm_map_reserve: reserve space in a vm_map for future use.
1155  *
1156  * => we reserve space in a map by putting a dummy map entry in the
1157  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
1158  * => map should be unlocked (we will write lock it)
1159  * => we return true if we were able to reserve space
1160  * => XXXCDC: should be inline?
1161  */
1162 
1163 int
1164 uvm_map_reserve(map, size, offset, raddr)
1165 	vm_map_t map;
1166 	vsize_t size;
1167 	vaddr_t offset;    /* hint for pmap_prefer */
1168 	vaddr_t *raddr;	/* OUT: reserved VA */
1169 {
1170 	UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1171 
1172 	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1173 	      map,size,offset,raddr);
1174 
1175 	size = round_page(size);
1176 	if (*raddr < vm_map_min(map))
1177 		*raddr = vm_map_min(map);                /* hint */
1178 
1179 	/*
1180 	 * reserve some virtual space.
1181 	 */
1182 
1183 	if (uvm_map(map, raddr, size, NULL, offset,
1184 	    UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1185 	    UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != KERN_SUCCESS) {
1186 	    UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1187 		return (FALSE);
1188 	}
1189 
1190 	UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1191 	return (TRUE);
1192 }
1193 
1194 /*
1195  * uvm_map_replace: replace a reserved (blank) area of memory with
1196  * real mappings.
1197  *
1198  * => caller must WRITE-LOCK the map
1199  * => we return TRUE if replacement was a success
1200  * => we expect the newents chain to have nnewents entrys on it and
1201  *    we expect newents->prev to point to the last entry on the list
1202  * => note newents is allowed to be NULL
1203  */
1204 
1205 int
1206 uvm_map_replace(map, start, end, newents, nnewents)
1207 	struct vm_map *map;
1208 	vaddr_t start, end;
1209 	vm_map_entry_t newents;
1210 	int nnewents;
1211 {
1212 	vm_map_entry_t oldent, last;
1213 	UVMHIST_FUNC("uvm_map_replace");
1214 	UVMHIST_CALLED(maphist);
1215 
1216 	/*
1217 	 * first find the blank map entry at the specified address
1218 	 */
1219 
1220 	if (!uvm_map_lookup_entry(map, start, &oldent)) {
1221 		return(FALSE);
1222 	}
1223 
1224 	/*
1225 	 * check to make sure we have a proper blank entry
1226 	 */
1227 
1228 	if (oldent->start != start || oldent->end != end ||
1229 	    oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1230 		return (FALSE);
1231 	}
1232 
1233 #ifdef DIAGNOSTIC
1234 	/*
1235 	 * sanity check the newents chain
1236 	 */
1237 	{
1238 		vm_map_entry_t tmpent = newents;
1239 		int nent = 0;
1240 		vaddr_t cur = start;
1241 
1242 		while (tmpent) {
1243 			nent++;
1244 			if (tmpent->start < cur)
1245 				panic("uvm_map_replace1");
1246 			if (tmpent->start > tmpent->end || tmpent->end > end) {
1247 		printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
1248 			    tmpent->start, tmpent->end, end);
1249 				panic("uvm_map_replace2");
1250 			}
1251 			cur = tmpent->end;
1252 			if (tmpent->next) {
1253 				if (tmpent->next->prev != tmpent)
1254 					panic("uvm_map_replace3");
1255 			} else {
1256 				if (newents->prev != tmpent)
1257 					panic("uvm_map_replace4");
1258 			}
1259 			tmpent = tmpent->next;
1260 		}
1261 		if (nent != nnewents)
1262 			panic("uvm_map_replace5");
1263 	}
1264 #endif
1265 
1266 	/*
1267 	 * map entry is a valid blank!   replace it.   (this does all the
1268 	 * work of map entry link/unlink...).
1269 	 */
1270 
1271 	if (newents) {
1272 
1273 		last = newents->prev;		/* we expect this */
1274 
1275 		/* critical: flush stale hints out of map */
1276 		SAVE_HINT(map, newents);
1277 		if (map->first_free == oldent)
1278 			map->first_free = last;
1279 
1280 		last->next = oldent->next;
1281 		last->next->prev = last;
1282 		newents->prev = oldent->prev;
1283 		newents->prev->next = newents;
1284 		map->nentries = map->nentries + (nnewents - 1);
1285 
1286 	} else {
1287 
1288 		/* critical: flush stale hints out of map */
1289 		SAVE_HINT(map, oldent->prev);
1290 		if (map->first_free == oldent)
1291 			map->first_free = oldent->prev;
1292 
1293 		/* NULL list of new entries: just remove the old one */
1294 		uvm_map_entry_unlink(map, oldent);
1295 	}
1296 
1297 
1298 	/*
1299 	 * now we can free the old blank entry, unlock the map and return.
1300 	 */
1301 
1302 	uvm_mapent_free(oldent);
1303 	return(TRUE);
1304 }
1305 
1306 /*
1307  * uvm_map_extract: extract a mapping from a map and put it somewhere
1308  *	(maybe removing the old mapping)
1309  *
1310  * => maps should be unlocked (we will write lock them)
1311  * => returns 0 on success, error code otherwise
1312  * => start must be page aligned
1313  * => len must be page sized
1314  * => flags:
1315  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
1316  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
1317  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
1318  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
1319  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
1320  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
1321  *             be used from within the kernel in a kernel level map <<<
1322  */
1323 
1324 int
1325 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags)
1326 	vm_map_t srcmap, dstmap;
1327 	vaddr_t start, *dstaddrp;
1328 	vsize_t len;
1329 	int flags;
1330 {
1331 	vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge,
1332 	    oldstart;
1333 	vm_map_entry_t chain, endchain, entry, orig_entry, newentry, deadentry;
1334 	vm_map_entry_t oldentry;
1335 	vsize_t elen;
1336 	int nchain, error, copy_ok;
1337 	UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1338 	UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
1339 	    len,0);
1340 	UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
1341 
1342 #ifdef DIAGNOSTIC
1343 	/*
1344 	 * step 0: sanity check: start must be on a page boundary, length
1345 	 * must be page sized.  can't ask for CONTIG/QREF if you asked for
1346 	 * REMOVE.
1347 	 */
1348 	if ((start & PAGE_MASK) || (len & PAGE_MASK))
1349 		panic("uvm_map_extract1");
1350 	if (flags & UVM_EXTRACT_REMOVE)
1351 		if (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF))
1352 			panic("uvm_map_extract2");
1353 #endif
1354 
1355 
1356 	/*
1357 	 * step 1: reserve space in the target map for the extracted area
1358 	 */
1359 
1360 	dstaddr = *dstaddrp;
1361 	if (uvm_map_reserve(dstmap, len, start, &dstaddr) == FALSE)
1362 		return(ENOMEM);
1363 	*dstaddrp = dstaddr;	/* pass address back to caller */
1364 	UVMHIST_LOG(maphist, "  dstaddr=0x%x", dstaddr,0,0,0);
1365 
1366 
1367 	/*
1368 	 * step 2: setup for the extraction process loop by init'ing the
1369 	 * map entry chain, locking src map, and looking up the first useful
1370 	 * entry in the map.
1371 	 */
1372 
1373 	end = start + len;
1374 	newend = dstaddr + len;
1375 	chain = endchain = NULL;
1376 	nchain = 0;
1377 	vm_map_lock(srcmap);
1378 
1379 	if (uvm_map_lookup_entry(srcmap, start, &entry)) {
1380 
1381 		/* "start" is within an entry */
1382 		if (flags & UVM_EXTRACT_QREF) {
1383 			/*
1384 			 * for quick references we don't clip the entry, so
1385 			 * the entry may map space "before" the starting
1386 			 * virtual address... this is the "fudge" factor
1387 			 * (which can be non-zero only the first time
1388 			 * through the "while" loop in step 3).
1389 			 */
1390 			fudge = start - entry->start;
1391 		} else {
1392 			/*
1393 			 * normal reference: we clip the map to fit (thus
1394 			 * fudge is zero)
1395 			 */
1396 			UVM_MAP_CLIP_START(srcmap, entry, start);
1397 			SAVE_HINT(srcmap, entry->prev);
1398 			fudge = 0;
1399 		}
1400 
1401 	} else {
1402 
1403 		/* "start" is not within an entry ... skip to next entry */
1404 		if (flags & UVM_EXTRACT_CONTIG) {
1405 			error = EINVAL;
1406 			goto bad;    /* definite hole here ... */
1407 		}
1408 
1409 		entry = entry->next;
1410 		fudge = 0;
1411 	}
1412 	/* save values from srcmap for step 6 */
1413 	orig_entry = entry;
1414 	orig_fudge = fudge;
1415 
1416 
1417 	/*
1418 	 * step 3: now start looping through the map entries, extracting
1419 	 * as we go.
1420 	 */
1421 
1422 	while (entry->start < end && entry != &srcmap->header) {
1423 
1424 		/* if we are not doing a quick reference, clip it */
1425 		if ((flags & UVM_EXTRACT_QREF) == 0)
1426 			UVM_MAP_CLIP_END(srcmap, entry, end);
1427 
1428 		/* clear needs_copy (allow chunking) */
1429 		if (UVM_ET_ISNEEDSCOPY(entry)) {
1430 			if (fudge)
1431 				oldstart = entry->start;
1432 			else
1433 				oldstart = 0;	/* XXX: gcc */
1434 			amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
1435 			if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
1436 				error = ENOMEM;
1437 				goto bad;
1438 			}
1439 			/* amap_copy could clip (during chunk)!  update fudge */
1440 			if (fudge) {
1441 				fudge = fudge - (entry->start - oldstart);
1442 				orig_fudge = fudge;
1443 			}
1444 		}
1445 
1446 		/* calculate the offset of this from "start" */
1447 		oldoffset = (entry->start + fudge) - start;
1448 
1449 		/* allocate a new map entry */
1450 		newentry = uvm_mapent_alloc(dstmap);
1451 		if (newentry == NULL) {
1452 			error = ENOMEM;
1453 			goto bad;
1454 		}
1455 
1456 		/* set up new map entry */
1457 		newentry->next = NULL;
1458 		newentry->prev = endchain;
1459 		newentry->start = dstaddr + oldoffset;
1460 		newentry->end =
1461 		    newentry->start + (entry->end - (entry->start + fudge));
1462 		if (newentry->end > newend)
1463 			newentry->end = newend;
1464 		newentry->object.uvm_obj = entry->object.uvm_obj;
1465 		if (newentry->object.uvm_obj) {
1466 			if (newentry->object.uvm_obj->pgops->pgo_reference)
1467 				newentry->object.uvm_obj->pgops->
1468 				    pgo_reference(newentry->object.uvm_obj);
1469 				newentry->offset = entry->offset + fudge;
1470 		} else {
1471 			newentry->offset = 0;
1472 		}
1473 		newentry->etype = entry->etype;
1474 		newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
1475 			entry->max_protection : entry->protection;
1476 		newentry->max_protection = entry->max_protection;
1477 		newentry->inheritance = entry->inheritance;
1478 		newentry->wired_count = 0;
1479 		newentry->aref.ar_amap = entry->aref.ar_amap;
1480 		if (newentry->aref.ar_amap) {
1481 			newentry->aref.ar_pageoff =
1482 			    entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1483 			amap_ref(newentry, AMAP_SHARED |
1484 			    ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
1485 		} else {
1486 			newentry->aref.ar_pageoff = 0;
1487 		}
1488 		newentry->advice = entry->advice;
1489 
1490 		/* now link it on the chain */
1491 		nchain++;
1492 		if (endchain == NULL) {
1493 			chain = endchain = newentry;
1494 		} else {
1495 			endchain->next = newentry;
1496 			endchain = newentry;
1497 		}
1498 
1499 		/* end of 'while' loop! */
1500 		if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1501 		    (entry->next == &srcmap->header ||
1502 		    entry->next->start != entry->end)) {
1503 			error = EINVAL;
1504 			goto bad;
1505 		}
1506 		entry = entry->next;
1507 		fudge = 0;
1508 	}
1509 
1510 
1511 	/*
1512 	 * step 4: close off chain (in format expected by uvm_map_replace)
1513 	 */
1514 
1515 	if (chain)
1516 		chain->prev = endchain;
1517 
1518 
1519 	/*
1520 	 * step 5: attempt to lock the dest map so we can pmap_copy.
1521 	 * note usage of copy_ok:
1522 	 *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
1523 	 *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
1524 	 */
1525 
1526 	if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
1527 
1528 		copy_ok = 1;
1529 		if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1530 		    nchain)) {
1531 			if (srcmap != dstmap)
1532 				vm_map_unlock(dstmap);
1533 			error = EIO;
1534 			goto bad;
1535 		}
1536 
1537 	} else {
1538 
1539 		copy_ok = 0;
1540 		/* replace defered until step 7 */
1541 
1542 	}
1543 
1544 
1545 	/*
1546 	 * step 6: traverse the srcmap a second time to do the following:
1547 	 *  - if we got a lock on the dstmap do pmap_copy
1548 	 *  - if UVM_EXTRACT_REMOVE remove the entries
1549 	 * we make use of orig_entry and orig_fudge (saved in step 2)
1550 	 */
1551 
1552 	if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
1553 
1554 		/* purge possible stale hints from srcmap */
1555 		if (flags & UVM_EXTRACT_REMOVE) {
1556 			SAVE_HINT(srcmap, orig_entry->prev);
1557 			if (srcmap->first_free->start >= start)
1558 				srcmap->first_free = orig_entry->prev;
1559 		}
1560 
1561 		entry = orig_entry;
1562 		fudge = orig_fudge;
1563 		deadentry = NULL;	/* for UVM_EXTRACT_REMOVE */
1564 
1565 		while (entry->start < end && entry != &srcmap->header) {
1566 
1567 			if (copy_ok) {
1568 	oldoffset = (entry->start + fudge) - start;
1569 	elen = min(end, entry->end) - (entry->start + fudge);
1570 	pmap_copy(dstmap->pmap, srcmap->pmap, dstaddr + oldoffset,
1571 		  elen, entry->start + fudge);
1572 			}
1573 
1574       /* we advance "entry" in the following if statement */
1575 			if (flags & UVM_EXTRACT_REMOVE) {
1576 				pmap_remove(srcmap->pmap, entry->start,
1577 						entry->end);
1578         			oldentry = entry;	/* save entry */
1579         			entry = entry->next;	/* advance */
1580 				uvm_map_entry_unlink(srcmap, oldentry);
1581 							/* add to dead list */
1582 				oldentry->next = deadentry;
1583 				deadentry = oldentry;
1584       			} else {
1585         			entry = entry->next;		/* advance */
1586 			}
1587 
1588 			/* end of 'while' loop */
1589 			fudge = 0;
1590 		}
1591 
1592 		/*
1593 		 * unlock dstmap.  we will dispose of deadentry in
1594 		 * step 7 if needed
1595 		 */
1596 		if (copy_ok && srcmap != dstmap)
1597 			vm_map_unlock(dstmap);
1598 
1599 	}
1600 	else
1601 		deadentry = NULL; /* XXX: gcc */
1602 
1603 	/*
1604 	 * step 7: we are done with the source map, unlock.   if copy_ok
1605 	 * is 0 then we have not replaced the dummy mapping in dstmap yet
1606 	 * and we need to do so now.
1607 	 */
1608 
1609 	vm_map_unlock(srcmap);
1610 	if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
1611 		uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
1612 
1613 	/* now do the replacement if we didn't do it in step 5 */
1614 	if (copy_ok == 0) {
1615 		vm_map_lock(dstmap);
1616 		error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
1617 		    nchain);
1618 		vm_map_unlock(dstmap);
1619 
1620 		if (error == FALSE) {
1621 			error = EIO;
1622 			goto bad2;
1623 		}
1624 	}
1625 
1626 	/*
1627 	 * done!
1628 	 */
1629 	return(0);
1630 
1631 	/*
1632 	 * bad: failure recovery
1633 	 */
1634 bad:
1635 	vm_map_unlock(srcmap);
1636 bad2:			/* src already unlocked */
1637 	if (chain)
1638 		uvm_unmap_detach(chain,
1639 		    (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1640 	uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
1641 	return(error);
1642 }
1643 
1644 /* end of extraction functions */
1645 
1646 /*
1647  * uvm_map_submap: punch down part of a map into a submap
1648  *
1649  * => only the kernel_map is allowed to be submapped
1650  * => the purpose of submapping is to break up the locking granularity
1651  *	of a larger map
1652  * => the range specified must have been mapped previously with a uvm_map()
1653  *	call [with uobj==NULL] to create a blank map entry in the main map.
1654  *	[And it had better still be blank!]
1655  * => maps which contain submaps should never be copied or forked.
1656  * => to remove a submap, use uvm_unmap() on the main map
1657  *	and then uvm_map_deallocate() the submap.
1658  * => main map must be unlocked.
1659  * => submap must have been init'd and have a zero reference count.
1660  *	[need not be locked as we don't actually reference it]
1661  */
1662 
1663 int
1664 uvm_map_submap(map, start, end, submap)
1665 	vm_map_t map, submap;
1666 	vaddr_t start, end;
1667 {
1668 	vm_map_entry_t entry;
1669 	int result;
1670 	UVMHIST_FUNC("uvm_map_submap"); UVMHIST_CALLED(maphist);
1671 
1672 	vm_map_lock(map);
1673 
1674 	VM_MAP_RANGE_CHECK(map, start, end);
1675 
1676 	if (uvm_map_lookup_entry(map, start, &entry)) {
1677 		UVM_MAP_CLIP_START(map, entry, start);
1678 		UVM_MAP_CLIP_END(map, entry, end);		/* to be safe */
1679 	}
1680 	else {
1681 		entry = NULL;
1682 	}
1683 
1684 	if (entry != NULL &&
1685 	    entry->start == start && entry->end == end &&
1686 	    entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
1687 	    !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1688 
1689 		/*
1690 		 * doit!
1691 		 */
1692 		entry->etype |= UVM_ET_SUBMAP;
1693 		entry->object.sub_map = submap;
1694 		entry->offset = 0;
1695 		uvm_map_reference(submap);
1696 		result = KERN_SUCCESS;
1697 	} else {
1698 		result = KERN_INVALID_ARGUMENT;
1699 	}
1700 	vm_map_unlock(map);
1701 
1702 	return(result);
1703 }
1704 
1705 
1706 /*
1707  * uvm_map_protect: change map protection
1708  *
1709  * => set_max means set max_protection.
1710  * => map must be unlocked.
1711  * => XXXCDC: does not work properly with share maps.  rethink.
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 	UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
1727 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
1728 	map, start, end, new_prot);
1729 
1730 	vm_map_lock(map);
1731 
1732 	VM_MAP_RANGE_CHECK(map, start, end);
1733 
1734 	if (uvm_map_lookup_entry(map, start, &entry)) {
1735 		UVM_MAP_CLIP_START(map, entry, start);
1736 	} else {
1737 		entry = entry->next;
1738 	}
1739 
1740 	/*
1741 	 * make a first pass to check for protection violations.
1742 	 */
1743 
1744 	current = entry;
1745 	while ((current != &map->header) && (current->start < end)) {
1746 		if (UVM_ET_ISSUBMAP(current))
1747 			return(KERN_INVALID_ARGUMENT);
1748 		if ((new_prot & current->max_protection) != new_prot) {
1749 			vm_map_unlock(map);
1750 			return(KERN_PROTECTION_FAILURE);
1751 		}
1752 			current = current->next;
1753 	}
1754 
1755 	/* go back and fix up protections (no need to clip this time). */
1756 
1757 	current = entry;
1758 
1759 	while ((current != &map->header) && (current->start < end)) {
1760 		vm_prot_t old_prot;
1761 
1762 		UVM_MAP_CLIP_END(map, current, end);
1763 
1764 		old_prot = current->protection;
1765 		if (set_max)
1766 			current->protection =
1767 			    (current->max_protection = new_prot) & old_prot;
1768 		else
1769 			current->protection = new_prot;
1770 
1771 		/*
1772 		 * update physical map if necessary.  worry about copy-on-write
1773 		 * here -- CHECK THIS XXX
1774 		 */
1775 
1776 		if (current->protection != old_prot) {
1777 
1778 			/* update pmap! */
1779 			pmap_protect(map->pmap, current->start, current->end,
1780 			    current->protection & MASK(entry));
1781 
1782 		}
1783 		current = current->next;
1784 	}
1785 
1786 	vm_map_unlock(map);
1787 	UVMHIST_LOG(maphist, "<- done",0,0,0,0);
1788 	return(KERN_SUCCESS);
1789 }
1790 
1791 #undef  max
1792 #undef  MASK
1793 
1794 /*
1795  * uvm_map_inherit: set inheritance code for range of addrs in map.
1796  *
1797  * => map must be unlocked
1798  * => note that the inherit code is used during a "fork".  see fork
1799  *	code for details.
1800  * => XXXCDC: currently only works in main map.  what about share map?
1801  */
1802 
1803 int
1804 uvm_map_inherit(map, start, end, new_inheritance)
1805 	vm_map_t map;
1806 	vaddr_t start;
1807 	vaddr_t end;
1808 	vm_inherit_t new_inheritance;
1809 {
1810 	vm_map_entry_t entry, temp_entry;
1811 	UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
1812 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
1813 	    map, start, end, new_inheritance);
1814 
1815 	switch (new_inheritance) {
1816 	case VM_INHERIT_NONE:
1817 	case VM_INHERIT_COPY:
1818 	case VM_INHERIT_SHARE:
1819 		break;
1820 	default:
1821 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1822 		return(KERN_INVALID_ARGUMENT);
1823 	}
1824 
1825 	vm_map_lock(map);
1826 
1827 	VM_MAP_RANGE_CHECK(map, start, end);
1828 
1829 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
1830 		entry = temp_entry;
1831 		UVM_MAP_CLIP_START(map, entry, start);
1832 	}  else {
1833 		entry = temp_entry->next;
1834 	}
1835 
1836 	while ((entry != &map->header) && (entry->start < end)) {
1837 		UVM_MAP_CLIP_END(map, entry, end);
1838 
1839 		entry->inheritance = new_inheritance;
1840 
1841 		entry = entry->next;
1842 	}
1843 
1844 	vm_map_unlock(map);
1845 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1846 	return(KERN_SUCCESS);
1847 }
1848 
1849 /*
1850  * uvm_map_pageable: sets the pageability of a range in a map.
1851  *
1852  * => regions sepcified as not pageable require lock-down (wired) memory
1853  *	and page tables.
1854  * => map must not be locked.
1855  * => XXXCDC: check this and try and clean it up.
1856  */
1857 
1858 int
1859 uvm_map_pageable(map, start, end, new_pageable)
1860 	vm_map_t map;
1861 	vaddr_t start, end;
1862 	boolean_t new_pageable;
1863 {
1864 	vm_map_entry_t entry, start_entry;
1865 	vaddr_t failed = 0;
1866 	int rv;
1867 	UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
1868 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
1869 	map, start, end, new_pageable);
1870 
1871 	vm_map_lock(map);
1872 	VM_MAP_RANGE_CHECK(map, start, end);
1873 
1874 	/*
1875 	 * only one pageability change may take place at one time, since
1876 	 * uvm_fault_wire assumes it will be called only once for each
1877 	 * wiring/unwiring.  therefore, we have to make sure we're actually
1878 	 * changing the pageability for the entire region.  we do so before
1879 	 * making any changes.
1880 	 */
1881 
1882 	if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1883 		vm_map_unlock(map);
1884 
1885 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1886 		return (KERN_INVALID_ADDRESS);
1887 	}
1888 	entry = start_entry;
1889 
1890 	/*
1891 	 * handle wiring and unwiring seperately.
1892 	 */
1893 
1894 	if (new_pageable) {               /* unwire */
1895 
1896 		UVM_MAP_CLIP_START(map, entry, start);
1897 
1898 		/*
1899 		 * unwiring.  first ensure that the range to be unwired is
1900 		 * really wired down and that there are no holes.
1901 		 */
1902 		while ((entry != &map->header) && (entry->start < end)) {
1903 
1904 			if (entry->wired_count == 0 ||
1905 			    (entry->end < end &&
1906 			    (entry->next == &map->header ||
1907 			    entry->next->start > entry->end))) {
1908 				vm_map_unlock(map);
1909 				UVMHIST_LOG(maphist,
1910 				    "<- done (INVALID UNWIRE ARG)",0,0,0,0);
1911 				return (KERN_INVALID_ARGUMENT);
1912 			}
1913 			entry = entry->next;
1914 		}
1915 
1916 		/*
1917 		 * now decrement the wiring count for each region.  if a region
1918 		 * becomes completely unwired, unwire its physical pages and
1919 		 * mappings.
1920 		 */
1921 #if 0		/* not necessary: uvm_fault_unwire does not lock */
1922 		lock_set_recursive(&map->lock);
1923 #endif  /* XXXCDC */
1924 
1925 		entry = start_entry;
1926 		while ((entry != &map->header) && (entry->start < end)) {
1927 			UVM_MAP_CLIP_END(map, entry, end);
1928 
1929 			entry->wired_count--;
1930 			if (entry->wired_count == 0)
1931 				uvm_map_entry_unwire(map, entry);
1932 
1933 			entry = entry->next;
1934 		}
1935 #if 0 /* XXXCDC: not necessary, see above */
1936 		lock_clear_recursive(&map->lock);
1937 #endif
1938 		vm_map_unlock(map);
1939 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
1940 		return(KERN_SUCCESS);
1941 
1942 		/*
1943 		 * end of unwire case!
1944 		 */
1945 	}
1946 
1947 	/*
1948 	 * wire case: in two passes [XXXCDC: ugly block of code here]
1949 	 *
1950 	 * 1: holding the write lock, we create any anonymous maps that need
1951 	 *    to be created.  then we clip each map entry to the region to
1952 	 *    be wired and increment its wiring count.
1953 	 *
1954 	 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
1955 	 *    in the pages for any newly wired area (wired_count is 1).
1956 	 *
1957 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
1958 	 *    deadlock with another thread that may have faulted on one of
1959 	 *    the pages to be wired (it would mark the page busy, blocking
1960 	 *    us, then in turn block on the map lock that we hold).  because
1961 	 *    of problems in the recursive lock package, we cannot upgrade
1962 	 *    to a write lock in vm_map_lookup.  thus, any actions that
1963 	 *    require the write lock must be done beforehand.  because we
1964 	 *    keep the read lock on the map, the copy-on-write status of the
1965 	 *    entries we modify here cannot change.
1966 	 */
1967 
1968 	while ((entry != &map->header) && (entry->start < end)) {
1969 
1970 		if (entry->wired_count == 0) {  /* not already wired? */
1971 
1972 			/*
1973 			 * perform actions of vm_map_lookup that need the
1974 			 * write lock on the map: create an anonymous map
1975 			 * for a copy-on-write region, or an anonymous map
1976 			 * for a zero-fill region.  (XXXCDC: submap case
1977 			 * ok?)
1978 			 */
1979 
1980 			if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
1981 				/*
1982 				 * XXXCDC: protection vs. max_protection??
1983 				 * (wirefault uses max?)
1984 				 * XXXCDC: used to do it always if
1985 				 * uvm_obj == NULL (wrong?)
1986 				 */
1987 				if ( UVM_ET_ISNEEDSCOPY(entry) &&
1988 				    (entry->protection & VM_PROT_WRITE) != 0) {
1989 					amap_copy(map, entry, M_WAITOK, TRUE,
1990 					    start, end);
1991 					/* XXXCDC: wait OK? */
1992 				}
1993 			}
1994 		}     /* wired_count == 0 */
1995 		UVM_MAP_CLIP_START(map, entry, start);
1996 		UVM_MAP_CLIP_END(map, entry, end);
1997 		entry->wired_count++;
1998 
1999 		/*
2000 		 * Check for holes
2001 		 */
2002 		if (entry->end < end && (entry->next == &map->header ||
2003 			     entry->next->start > entry->end)) {
2004 			/*
2005 			 * found one.  amap creation actions do not need to
2006 			 * be undone, but the wired counts need to be restored.
2007 			 */
2008 			while (entry != &map->header && entry->end > start) {
2009 				entry->wired_count--;
2010 				entry = entry->prev;
2011 			}
2012 			vm_map_unlock(map);
2013 			UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
2014 			return(KERN_INVALID_ARGUMENT);
2015 		}
2016 		entry = entry->next;
2017 	}
2018 
2019 	/*
2020 	 * Pass 2.
2021 	 */
2022 	/*
2023 	 * HACK HACK HACK HACK
2024 	 *
2025 	 * if we are wiring in the kernel map or a submap of it, unlock the
2026 	 * map to avoid deadlocks.  we trust that the kernel threads are
2027 	 * well-behaved, and therefore will not do anything destructive to
2028 	 * this region of the map while we have it unlocked.  we cannot
2029 	 * trust user threads to do the same.
2030 	 *
2031 	 * HACK HACK HACK HACK
2032 	 */
2033 	if (vm_map_pmap(map) == pmap_kernel()) {
2034 		vm_map_unlock(map);         /* trust me ... */
2035 	} else {
2036 		vm_map_set_recursive(&map->lock);
2037 		lockmgr(&map->lock, LK_DOWNGRADE, (void *)0);
2038 	}
2039 
2040 	rv = 0;
2041 	entry = start_entry;
2042 	while (entry != &map->header && entry->start < end) {
2043 		/*
2044 		 * if uvm_fault_wire fails for any page we need to undo what has
2045 		 * been done.  we decrement the wiring count for those pages
2046 		 * which have not yet been wired (now) and unwire those that
2047 		 * have * (later).
2048 		 *
2049 		 * XXX this violates the locking protocol on the map, needs to
2050 		 * be fixed.  [because we only have a read lock on map we
2051 		 * shouldn't be changing wired_count?]
2052 		 */
2053 		if (rv) {
2054 			entry->wired_count--;
2055 		} else if (entry->wired_count == 1) {
2056 			rv = uvm_fault_wire(map, entry->start, entry->end);
2057 			if (rv) {
2058 				failed = entry->start;
2059 				entry->wired_count--;
2060 			}
2061 		}
2062 		entry = entry->next;
2063 	}
2064 
2065 	if (vm_map_pmap(map) == pmap_kernel()) {
2066 		vm_map_lock(map);     /* relock */
2067 	} else {
2068 		vm_map_clear_recursive(&map->lock);
2069 	}
2070 
2071 	if (rv) {        /* failed? */
2072 		vm_map_unlock(map);
2073 		(void) uvm_map_pageable(map, start, failed, TRUE);
2074 		UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
2075 		return(rv);
2076 	}
2077 	vm_map_unlock(map);
2078 
2079 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
2080 	return(KERN_SUCCESS);
2081 }
2082 
2083 /*
2084  * uvm_map_clean: push dirty pages off to backing store.
2085  *
2086  * => valid flags:
2087  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
2088  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
2089  *   if (flags & PGO_FREE): any cached pages are freed after clean
2090  * => returns an error if any part of the specified range isn't mapped
2091  * => never a need to flush amap layer since the anonymous memory has
2092  *	no permanent home...
2093  * => called from sys_msync()
2094  * => caller must not write-lock map (read OK).
2095  * => we may sleep while cleaning if SYNCIO [with map read-locked]
2096  * => XXX: does this handle share maps properly?
2097  */
2098 
2099 int
2100 uvm_map_clean(map, start, end, flags)
2101 	vm_map_t map;
2102 	vaddr_t start, end;
2103 	int flags;
2104 {
2105 	vm_map_entry_t current;
2106 	vm_map_entry_t entry;
2107 	vsize_t size;
2108 	struct uvm_object *object;
2109 	vaddr_t offset;
2110 	UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
2111 	UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
2112 	map, start, end, flags);
2113 
2114 	vm_map_lock_read(map);
2115 	VM_MAP_RANGE_CHECK(map, start, end);
2116 	if (!uvm_map_lookup_entry(map, start, &entry)) {
2117 		vm_map_unlock_read(map);
2118 		return(KERN_INVALID_ADDRESS);
2119 	}
2120 
2121 	/*
2122 	 * Make a first pass to check for holes.
2123 	 */
2124 	for (current = entry; current->start < end; current = current->next) {
2125 		if (UVM_ET_ISSUBMAP(current)) {
2126 			vm_map_unlock_read(map);
2127 			return(KERN_INVALID_ARGUMENT);
2128 		}
2129 		if (end > current->end && (current->next == &map->header ||
2130 		    current->end != current->next->start)) {
2131 			vm_map_unlock_read(map);
2132 			return(KERN_INVALID_ADDRESS);
2133 		}
2134 	}
2135 
2136 	/*
2137 	 * add "cleanit" flag to flags (for generic flush routine).
2138 	 * then make a second pass, cleaning/uncaching pages from
2139 	 * the indicated objects as we go.
2140 	 */
2141 	flags = flags | PGO_CLEANIT;
2142 	for (current = entry; current->start < end; current = current->next) {
2143 		offset = current->offset + (start - current->start);
2144 		size = (end <= current->end ? end : current->end) - start;
2145 
2146 		/*
2147 		 * get object/offset.  can't be submap (checked above).
2148 		 */
2149 		object = current->object.uvm_obj;
2150 		simple_lock(&object->vmobjlock);
2151 
2152 		/*
2153 		 * flush pages if we've got a valid backing object.
2154 		 * note that object is locked.
2155 		 * XXX should we continue on an error?
2156 		 */
2157 
2158 		if (object && object->pgops) {
2159 			if (!object->pgops->pgo_flush(object, offset,
2160 			    offset+size, flags)) {
2161 				simple_unlock(&object->vmobjlock);
2162 				vm_map_unlock_read(map);
2163 				return (KERN_FAILURE);
2164 			}
2165 		}
2166 		simple_unlock(&object->vmobjlock);
2167 		start += size;
2168 	}
2169 	vm_map_unlock_read(map);
2170 	return(KERN_SUCCESS);
2171 }
2172 
2173 
2174 /*
2175  * uvm_map_checkprot: check protection in map
2176  *
2177  * => must allow specified protection in a fully allocated region.
2178  * => map must be read or write locked by caller.
2179  */
2180 
2181 boolean_t
2182 uvm_map_checkprot(map, start, end, protection)
2183 	vm_map_t       map;
2184 	vaddr_t    start, end;
2185 	vm_prot_t      protection;
2186 {
2187 	 vm_map_entry_t entry;
2188 	 vm_map_entry_t tmp_entry;
2189 
2190 	 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
2191 		 return(FALSE);
2192 	 }
2193 
2194 	 entry = tmp_entry;
2195 
2196 	 while (start < end) {
2197 		 if (entry == &map->header) {
2198 			 return(FALSE);
2199 		 }
2200 
2201 		/*
2202 		 * no holes allowed
2203 		 */
2204 
2205 		 if (start < entry->start) {
2206 			 return(FALSE);
2207 		 }
2208 
2209 		/*
2210 		 * check protection associated with entry
2211 		 */
2212 
2213 		 if ((entry->protection & protection) != protection) {
2214 			 return(FALSE);
2215 		 }
2216 
2217 		 /* go to next entry */
2218 
2219 		 start = entry->end;
2220 		 entry = entry->next;
2221 	 }
2222 	 return(TRUE);
2223 }
2224 
2225 /*
2226  * uvmspace_alloc: allocate a vmspace structure.
2227  *
2228  * - structure includes vm_map and pmap
2229  * - XXX: no locking on this structure
2230  * - refcnt set to 1, rest must be init'd by caller
2231  */
2232 struct vmspace *
2233 uvmspace_alloc(min, max, pageable)
2234 	vaddr_t min, max;
2235 	int pageable;
2236 {
2237 	struct vmspace *vm;
2238 	UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
2239 
2240 	vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
2241 	uvmspace_init(vm, NULL, min, max, pageable);
2242 	UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
2243 	return (vm);
2244 }
2245 
2246 /*
2247  * uvmspace_init: initialize a vmspace structure.
2248  *
2249  * - XXX: no locking on this structure
2250  * - refcnt set to 1, rest must me init'd by caller
2251  */
2252 void
2253 uvmspace_init(vm, pmap, min, max, pageable)
2254 	struct vmspace *vm;
2255 	struct pmap *pmap;
2256 	vaddr_t min, max;
2257 	boolean_t pageable;
2258 {
2259 	UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
2260 
2261 	memset(vm, 0, sizeof(*vm));
2262 
2263 	uvm_map_setup(&vm->vm_map, min, max, pageable);
2264 
2265 	if (pmap)
2266 		pmap_reference(pmap);
2267 	else
2268 #if defined(PMAP_NEW)
2269 		pmap = pmap_create();
2270 #else
2271 		pmap = pmap_create(0);
2272 #endif
2273 	vm->vm_map.pmap = pmap;
2274 
2275 	vm->vm_refcnt = 1;
2276 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
2277 }
2278 
2279 /*
2280  * uvmspace_share: share a vmspace between two proceses
2281  *
2282  * - XXX: no locking on vmspace
2283  * - used for vfork, threads(?)
2284  */
2285 
2286 void
2287 uvmspace_share(p1, p2)
2288 	struct proc *p1, *p2;
2289 {
2290 	p2->p_vmspace = p1->p_vmspace;
2291 	p1->p_vmspace->vm_refcnt++;
2292 }
2293 
2294 /*
2295  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
2296  *
2297  * - XXX: no locking on vmspace
2298  */
2299 
2300 void
2301 uvmspace_unshare(p)
2302 	struct proc *p;
2303 {
2304 	struct vmspace *nvm, *ovm = p->p_vmspace;
2305 	int s;
2306 
2307 	if (ovm->vm_refcnt == 1)
2308 		/* nothing to do: vmspace isn't shared in the first place */
2309 		return;
2310 
2311 	/* make a new vmspace, still holding old one */
2312 	nvm = uvmspace_fork(ovm);
2313 
2314 	s = splhigh();			/* make this `atomic' */
2315 	pmap_deactivate(p);		/* unbind old vmspace */
2316 	p->p_vmspace = nvm;
2317 	pmap_activate(p);		/* switch to new vmspace */
2318 	splx(s);			/* end of critical section */
2319 
2320 	uvmspace_free(ovm);		/* drop reference to old vmspace */
2321 }
2322 
2323 /*
2324  * uvmspace_exec: the process wants to exec a new program
2325  *
2326  * - XXX: no locking on vmspace
2327  */
2328 
2329 void
2330 uvmspace_exec(p)
2331 	struct proc *p;
2332 {
2333 	struct vmspace *nvm, *ovm = p->p_vmspace;
2334 	vm_map_t map = &ovm->vm_map;
2335 	int s;
2336 
2337 #ifdef sparc
2338 	/* XXX cgd 960926: the sparc #ifdef should be a MD hook */
2339 	kill_user_windows(p);   /* before stack addresses go away */
2340 #endif
2341 
2342 	/*
2343 	 * see if more than one process is using this vmspace...
2344 	 */
2345 
2346 	if (ovm->vm_refcnt == 1) {
2347 
2348 		/*
2349 		 * if p is the only process using its vmspace then we can safely
2350 		 * recycle that vmspace for the program that is being exec'd.
2351 		 */
2352 
2353 #ifdef SYSVSHM
2354 		/*
2355 		 * SYSV SHM semantics require us to kill all segments on an exec
2356 		 */
2357 		if (ovm->vm_shm)
2358 			shmexit(ovm);
2359 #endif
2360 
2361 		/*
2362 		 * now unmap the old program
2363 		 */
2364 		uvm_unmap(map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
2365 
2366 	} else {
2367 
2368 		/*
2369 		 * p's vmspace is being shared, so we can't reuse it for p since
2370 		 * it is still being used for others.   allocate a new vmspace
2371 		 * for p
2372 		 */
2373 		nvm = uvmspace_alloc(map->min_offset, map->max_offset,
2374 			 map->entries_pageable);
2375 
2376 #if (defined(i386) || defined(pc532)) && !defined(PMAP_NEW)
2377 		/*
2378 		 * allocate zero fill area in the new vmspace's map for user
2379 		 * page tables for ports that have old style pmaps that keep
2380 		 * user page tables in the top part of the process' address
2381 		 * space.
2382 		 *
2383 		 * XXXCDC: this should go away once all pmaps are fixed
2384 		 */
2385 		{
2386 			vaddr_t addr = VM_MAXUSER_ADDRESS;
2387 			if (uvm_map(&nvm->vm_map, &addr, VM_MAX_ADDRESS - addr,
2388 			    NULL, UVM_UNKNOWN_OFFSET, UVM_MAPFLAG(UVM_PROT_ALL,
2389 			    UVM_PROT_ALL, UVM_INH_NONE, UVM_ADV_NORMAL,
2390 			    UVM_FLAG_FIXED|UVM_FLAG_COPYONW)) != KERN_SUCCESS)
2391 				panic("vm_allocate of PT page area failed");
2392 		}
2393 #endif
2394 
2395 		/*
2396 		 * install new vmspace and drop our ref to the old one.
2397 		 */
2398 
2399 		s = splhigh();
2400 		pmap_deactivate(p);
2401 		p->p_vmspace = nvm;
2402 		pmap_activate(p);
2403 		splx(s);
2404 
2405 		uvmspace_free(ovm);
2406 	}
2407 }
2408 
2409 /*
2410  * uvmspace_free: free a vmspace data structure
2411  *
2412  * - XXX: no locking on vmspace
2413  */
2414 
2415 void
2416 uvmspace_free(vm)
2417 	struct vmspace *vm;
2418 {
2419 	vm_map_entry_t dead_entries;
2420 	UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
2421 
2422 	UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
2423 	if (--vm->vm_refcnt == 0) {
2424 		/*
2425 		 * lock the map, to wait out all other references to it.  delete
2426 		 * all of the mappings and pages they hold, then call the pmap
2427 		 * module to reclaim anything left.
2428 		 */
2429 		vm_map_lock(&vm->vm_map);
2430 		if (vm->vm_map.nentries) {
2431 			(void)uvm_unmap_remove(&vm->vm_map,
2432 			    vm->vm_map.min_offset, vm->vm_map.max_offset,
2433 			    &dead_entries);
2434 			if (dead_entries != NULL)
2435 				uvm_unmap_detach(dead_entries, 0);
2436 		}
2437 		pmap_destroy(vm->vm_map.pmap);
2438 		vm->vm_map.pmap = NULL;
2439 		pool_put(&uvm_vmspace_pool, vm);
2440 	}
2441 	UVMHIST_LOG(maphist,"<- done", 0,0,0,0);
2442 }
2443 
2444 /*
2445  *   F O R K   -   m a i n   e n t r y   p o i n t
2446  */
2447 /*
2448  * uvmspace_fork: fork a process' main map
2449  *
2450  * => create a new vmspace for child process from parent.
2451  * => parent's map must not be locked.
2452  */
2453 
2454 struct vmspace *
2455 uvmspace_fork(vm1)
2456 	struct vmspace *vm1;
2457 {
2458 	struct vmspace *vm2;
2459 	vm_map_t        old_map = &vm1->vm_map;
2460 	vm_map_t        new_map;
2461 	vm_map_entry_t  old_entry;
2462 	vm_map_entry_t  new_entry;
2463 	pmap_t          new_pmap;
2464 	boolean_t	protect_child;
2465 	UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
2466 
2467 #if (defined(i386) || defined(pc532)) && !defined(PMAP_NEW)
2468 	/*
2469 	 * avoid copying any of the parent's pagetables or other per-process
2470 	 * objects that reside in the map by marking all of them non-inheritable
2471 	 * XXXCDC: should go away
2472 	 */
2473 	(void) uvm_map_inherit(old_map, VM_MAXUSER_ADDRESS, VM_MAX_ADDRESS,
2474 			 VM_INHERIT_NONE);
2475 #endif
2476 
2477 	vm_map_lock(old_map);
2478 
2479 	vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset,
2480 		      old_map->entries_pageable);
2481 	memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
2482 	(caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
2483 	new_map = &vm2->vm_map;		  /* XXX */
2484 	new_pmap = new_map->pmap;
2485 
2486 	old_entry = old_map->header.next;
2487 
2488 	/*
2489 	 * go entry-by-entry
2490 	 */
2491 
2492 	while (old_entry != &old_map->header) {
2493 
2494 		/*
2495 		 * first, some sanity checks on the old entry
2496 		 */
2497 		if (UVM_ET_ISSUBMAP(old_entry))
2498 		    panic("fork: encountered a submap during fork (illegal)");
2499 
2500 		if (!UVM_ET_ISCOPYONWRITE(old_entry) &&
2501 			    UVM_ET_ISNEEDSCOPY(old_entry))
2502 	panic("fork: non-copy_on_write map entry marked needs_copy (illegal)");
2503 
2504 
2505 		switch (old_entry->inheritance) {
2506 		case VM_INHERIT_NONE:
2507 			/*
2508 			 * drop the mapping
2509 			 */
2510 			break;
2511 
2512 		case VM_INHERIT_SHARE:
2513 			/*
2514 			 * share the mapping: this means we want the old and
2515 			 * new entries to share amaps and backing objects.
2516 			 */
2517 
2518 			/*
2519 			 * if the old_entry needs a new amap (due to prev fork)
2520 			 * then we need to allocate it now so that we have
2521 			 * something we own to share with the new_entry.   [in
2522 			 * other words, we need to clear needs_copy]
2523 			 */
2524 
2525 			if (UVM_ET_ISNEEDSCOPY(old_entry)) {
2526 				/* get our own amap, clears needs_copy */
2527 				amap_copy(old_map, old_entry, M_WAITOK, FALSE,
2528 				    0, 0);
2529 				/* XXXCDC: WAITOK??? */
2530 			}
2531 
2532 			new_entry = uvm_mapent_alloc(new_map);
2533 			/* old_entry -> new_entry */
2534 			uvm_mapent_copy(old_entry, new_entry);
2535 
2536 			/* new pmap has nothing wired in it */
2537 			new_entry->wired_count = 0;
2538 
2539 			/*
2540 			 * gain reference to object backing the map (can't
2541 			 * be a submap, already checked this case).
2542 			 */
2543 			if (new_entry->aref.ar_amap)
2544 				/* share reference */
2545 				amap_ref(new_entry, AMAP_SHARED);
2546 
2547 			if (new_entry->object.uvm_obj &&
2548 			    new_entry->object.uvm_obj->pgops->pgo_reference)
2549 				new_entry->object.uvm_obj->
2550 				    pgops->pgo_reference(
2551 				        new_entry->object.uvm_obj);
2552 
2553 			/* insert entry at end of new_map's entry list */
2554 			uvm_map_entry_link(new_map, new_map->header.prev,
2555 			    new_entry);
2556 
2557 			/*
2558 			 * pmap_copy the mappings: this routine is optional
2559 			 * but if it is there it will reduce the number of
2560 			 * page faults in the new proc.
2561 			 */
2562 
2563 			pmap_copy(new_pmap, old_map->pmap, new_entry->start,
2564 			    (old_entry->end - old_entry->start),
2565 			    old_entry->start);
2566 
2567 			break;
2568 
2569 		case VM_INHERIT_COPY:
2570 
2571 			/*
2572 			 * copy-on-write the mapping (using mmap's
2573 			 * MAP_PRIVATE semantics)
2574 			 *
2575 			 * allocate new_entry, adjust reference counts.
2576 			 * (note that new references are read-only).
2577 			 */
2578 
2579 			new_entry = uvm_mapent_alloc(new_map);
2580 			/* old_entry -> new_entry */
2581 			uvm_mapent_copy(old_entry, new_entry);
2582 
2583 			if (new_entry->aref.ar_amap)
2584 				amap_ref(new_entry, 0);
2585 
2586 			if (new_entry->object.uvm_obj &&
2587 			    new_entry->object.uvm_obj->pgops->pgo_reference)
2588 				new_entry->object.uvm_obj->pgops->pgo_reference
2589 				    (new_entry->object.uvm_obj);
2590 
2591 			/* new pmap has nothing wired in it */
2592 			new_entry->wired_count = 0;
2593 
2594 			new_entry->etype |=
2595 			    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
2596 			uvm_map_entry_link(new_map, new_map->header.prev,
2597 			    new_entry);
2598 
2599 			/*
2600 			 * the new entry will need an amap.  it will either
2601 			 * need to be copied from the old entry or created
2602 			 * from scratch (if the old entry does not have an
2603 			 * amap).  can we defer this process until later
2604 			 * (by setting "needs_copy") or do we need to copy
2605 			 * the amap now?
2606 			 *
2607 			 * we must copy the amap now if any of the following
2608 			 * conditions hold:
2609 			 * 1. the old entry has an amap and that amap is
2610 			 *    being shared.  this means that the old (parent)
2611 			 *    process is sharing the amap with another
2612 			 *    process.  if we do not clear needs_copy here
2613 			 *    we will end up in a situation where both the
2614 			 *    parent and child process are refering to the
2615 			 *    same amap with "needs_copy" set.  if the
2616 			 *    parent write-faults, the fault routine will
2617 			 *    clear "needs_copy" in the parent by allocating
2618 			 *    a new amap.   this is wrong because the
2619 			 *    parent is supposed to be sharing the old amap
2620 			 *    and the new amap will break that.
2621 			 *
2622 			 * 2. if the old entry has an amap and a non-zero
2623 			 *    wire count then we are going to have to call
2624 			 *    amap_cow_now to avoid page faults in the
2625 			 *    parent process.   since amap_cow_now requires
2626 			 *    "needs_copy" to be clear we might as well
2627 			 *    clear it here as well.
2628 			 *
2629 			 */
2630 
2631 			if (old_entry->aref.ar_amap != NULL) {
2632 
2633 			  if ((amap_flags(old_entry->aref.ar_amap) &
2634 			       AMAP_SHARED) != 0 ||
2635 			      old_entry->wired_count != 0) {
2636 
2637 			    amap_copy(new_map, new_entry, M_WAITOK, FALSE,
2638 				      0, 0);
2639 			    /* XXXCDC: M_WAITOK ... ok? */
2640 			  }
2641 			}
2642 
2643 			/*
2644 			 * if the parent's entry is wired down, then the
2645 			 * parent process does not want page faults on
2646 			 * access to that memory.  this means that we
2647 			 * cannot do copy-on-write because we can't write
2648 			 * protect the old entry.   in this case we
2649 			 * resolve all copy-on-write faults now, using
2650 			 * amap_cow_now.   note that we have already
2651 			 * allocated any needed amap (above).
2652 			 */
2653 
2654 			if (old_entry->wired_count != 0) {
2655 
2656 			  /*
2657 			   * resolve all copy-on-write faults now
2658 			   * (note that there is nothing to do if
2659 			   * the old mapping does not have an amap).
2660 			   * XXX: is it worthwhile to bother with pmap_copy
2661 			   * in this case?
2662 			   */
2663 			  if (old_entry->aref.ar_amap)
2664 			    amap_cow_now(new_map, new_entry);
2665 
2666 			} else {
2667 
2668 			  /*
2669 			   * setup mappings to trigger copy-on-write faults
2670 			   * we must write-protect the parent if it has
2671 			   * an amap and it is not already "needs_copy"...
2672 			   * if it is already "needs_copy" then the parent
2673 			   * has already been write-protected by a previous
2674 			   * fork operation.
2675 			   *
2676 			   * if we do not write-protect the parent, then
2677 			   * we must be sure to write-protect the child
2678 			   * after the pmap_copy() operation.
2679 			   *
2680 			   * XXX: pmap_copy should have some way of telling
2681 			   * us that it didn't do anything so we can avoid
2682 			   * calling pmap_protect needlessly.
2683 			   */
2684 
2685 			  if (old_entry->aref.ar_amap) {
2686 
2687 			    if (!UVM_ET_ISNEEDSCOPY(old_entry)) {
2688 			      if (old_entry->max_protection & VM_PROT_WRITE) {
2689 				pmap_protect(old_map->pmap,
2690 					     old_entry->start,
2691 					     old_entry->end,
2692 					     old_entry->protection &
2693 					     ~VM_PROT_WRITE);
2694 			      }
2695 			      old_entry->etype |= UVM_ET_NEEDSCOPY;
2696 			    }
2697 
2698 			    /*
2699 			     * parent must now be write-protected
2700 			     */
2701 			    protect_child = FALSE;
2702 			  } else {
2703 
2704 			    /*
2705 			     * we only need to protect the child if the
2706 			     * parent has write access.
2707 			     */
2708 			    if (old_entry->max_protection & VM_PROT_WRITE)
2709 			      protect_child = TRUE;
2710 			    else
2711 			      protect_child = FALSE;
2712 
2713 			  }
2714 
2715 			  /*
2716 			   * copy the mappings
2717 			   * XXX: need a way to tell if this does anything
2718 			   */
2719 
2720 			  pmap_copy(new_pmap, old_map->pmap,
2721 				    new_entry->start,
2722 				    (old_entry->end - old_entry->start),
2723 				    old_entry->start);
2724 
2725 			  /*
2726 			   * protect the child's mappings if necessary
2727 			   */
2728 			  if (protect_child) {
2729 			    pmap_protect(new_pmap, new_entry->start,
2730 					 new_entry->end,
2731 					 new_entry->protection &
2732 					          ~VM_PROT_WRITE);
2733 			  }
2734 
2735 			}
2736 			break;
2737 		}  /* end of switch statement */
2738 		old_entry = old_entry->next;
2739 	}
2740 
2741 	new_map->size = old_map->size;
2742 	vm_map_unlock(old_map);
2743 
2744 #if (defined(i386) || defined(pc532)) && !defined(PMAP_NEW)
2745 	/*
2746 	 * allocate zero fill area in the new vmspace's map for user
2747 	 * page tables for ports that have old style pmaps that keep
2748 	 * user page tables in the top part of the process' address
2749 	 * space.
2750 	 *
2751 	 * XXXCDC: this should go away once all pmaps are fixed
2752 	 */
2753 	{
2754 		vaddr_t addr = VM_MAXUSER_ADDRESS;
2755 		if (uvm_map(new_map, &addr, VM_MAX_ADDRESS - addr, NULL,
2756 		    UVM_UNKNOWN_OFFSET, UVM_MAPFLAG(UVM_PROT_ALL,
2757 		    UVM_PROT_ALL, UVM_INH_NONE, UVM_ADV_NORMAL,
2758 		    UVM_FLAG_FIXED|UVM_FLAG_COPYONW)) != KERN_SUCCESS)
2759 			panic("vm_allocate of PT page area failed");
2760 	}
2761 #endif
2762 
2763 #ifdef SYSVSHM
2764 	if (vm1->vm_shm)
2765 		shmfork(vm1, vm2);
2766 #endif
2767 
2768 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
2769 	return(vm2);
2770 }
2771 
2772 
2773 #if defined(DDB)
2774 
2775 /*
2776  * DDB hooks
2777  */
2778 
2779 /*
2780  * uvm_map_print: print out a map
2781  */
2782 
2783 void
2784 uvm_map_print(map, full)
2785 	vm_map_t map;
2786 	boolean_t full;
2787 {
2788 
2789 	uvm_map_printit(map, full, printf);
2790 }
2791 
2792 /*
2793  * uvm_map_printit: actually prints the map
2794  */
2795 
2796 void
2797 uvm_map_printit(map, full, pr)
2798 	vm_map_t map;
2799 	boolean_t full;
2800 	void (*pr) __P((const char *, ...));
2801 {
2802 	vm_map_entry_t entry;
2803 
2804 	(*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
2805 	(*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d\n",
2806 	    map->nentries, map->size, map->ref_count, map->timestamp);
2807 #ifdef pmap_resident_count
2808 	(*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
2809 	    pmap_resident_count(map->pmap));
2810 #else
2811 	/* XXXCDC: this should be required ... */
2812 	(*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap);
2813 #endif
2814 	if (!full)
2815 		return;
2816 	for (entry = map->header.next; entry != &map->header;
2817 	    entry = entry->next) {
2818 		(*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%x, amap=%p/%d\n",
2819 		    entry, entry->start, entry->end, entry->object.uvm_obj,
2820 		    entry->offset, entry->aref.ar_amap, entry->aref.ar_pageoff);
2821 		(*pr)(
2822 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, wc=%d, adv=%d\n",
2823 		    (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
2824 		    (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
2825 		    (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
2826 		    entry->protection, entry->max_protection,
2827 		    entry->inheritance, entry->wired_count, entry->advice);
2828 	}
2829 }
2830 
2831 /*
2832  * uvm_object_print: print out an object
2833  */
2834 
2835 void
2836 uvm_object_print(uobj, full)
2837 	struct uvm_object *uobj;
2838 	boolean_t full;
2839 {
2840 
2841 	uvm_object_printit(uobj, full, printf);
2842 }
2843 
2844 /*
2845  * uvm_object_printit: actually prints the object
2846  */
2847 
2848 void
2849 uvm_object_printit(uobj, full, pr)
2850 	struct uvm_object *uobj;
2851 	boolean_t full;
2852 	void (*pr) __P((const char *, ...));
2853 {
2854 	struct vm_page *pg;
2855 	int cnt = 0;
2856 
2857 	(*pr)("OBJECT %p: pgops=%p, npages=%d, ", uobj, uobj->pgops,
2858 	    uobj->uo_npages);
2859 	if (uobj->uo_refs == UVM_OBJ_KERN)
2860 		(*pr)("refs=<SYSTEM>\n");
2861 	else
2862 		(*pr)("refs=%d\n", uobj->uo_refs);
2863 
2864 	if (!full) return;
2865 	(*pr)("  PAGES <pg,offset>:\n  ");
2866 	for (pg = uobj->memq.tqh_first ; pg ; pg = pg->listq.tqe_next, cnt++) {
2867 		(*pr)("<%p,0x%lx> ", pg, pg->offset);
2868 		if ((cnt % 3) == 2) (*pr)("\n  ");
2869 	}
2870 	if ((cnt % 3) != 2) (*pr)("\n");
2871 }
2872 
2873 /*
2874  * uvm_page_print: print out a page
2875  */
2876 
2877 void
2878 uvm_page_print(pg, full)
2879 	struct vm_page *pg;
2880 	boolean_t full;
2881 {
2882 
2883 	uvm_page_printit(pg, full, printf);
2884 }
2885 
2886 /*
2887  * uvm_page_printit: actually print the page
2888  */
2889 
2890 void
2891 uvm_page_printit(pg, full, pr)
2892 	struct vm_page *pg;
2893 	boolean_t full;
2894 	void (*pr) __P((const char *, ...));
2895 {
2896 	struct vm_page *lcv;
2897 	struct uvm_object *uobj;
2898 	struct pglist *pgl;
2899 
2900 	(*pr)("PAGE %p:\n", pg);
2901 	(*pr)("  flags=0x%x, pqflags=0x%x, vers=%d, wire_count=%d, pa=0x%lx\n",
2902 	pg->flags, pg->pqflags, pg->version, pg->wire_count, (long)pg->phys_addr);
2903 	(*pr)("  uobject=%p, uanon=%p, offset=0x%lx loan_count=%d\n",
2904 	pg->uobject, pg->uanon, pg->offset, pg->loan_count);
2905 #if defined(UVM_PAGE_TRKOWN)
2906 	if (pg->flags & PG_BUSY)
2907 		(*pr)("  owning process = %d, tag=%s\n",
2908 		    pg->owner, pg->owner_tag);
2909 	else
2910 		(*pr)("  page not busy, no owner\n");
2911 #else
2912 	(*pr)("  [page ownership tracking disabled]\n");
2913 #endif
2914 
2915 	if (!full)
2916 		return;
2917 
2918 	/* cross-verify object/anon */
2919 	if ((pg->pqflags & PQ_FREE) == 0) {
2920 		if (pg->pqflags & PQ_ANON) {
2921 			if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
2922 			    (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
2923 				(pg->uanon) ? pg->uanon->u.an_page : NULL);
2924 			else
2925 				(*pr)("  anon backpointer is OK\n");
2926 		} else {
2927 			uobj = pg->uobject;
2928 			if (uobj) {
2929 				(*pr)("  checking object list\n");
2930 				for (lcv = uobj->memq.tqh_first ; lcv ;
2931 				    lcv = lcv->listq.tqe_next) {
2932 					if (lcv == pg) break;
2933 				}
2934 				if (lcv)
2935 					(*pr)("  page found on object list\n");
2936 				else
2937 			(*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
2938 			}
2939 		}
2940 	}
2941 
2942 	/* cross-verify page queue */
2943 	if (pg->pqflags & PQ_FREE)
2944 		pgl = &uvm.page_free[uvm_page_lookup_freelist(pg)];
2945 	else if (pg->pqflags & PQ_INACTIVE)
2946 		pgl = (pg->pqflags & PQ_SWAPBACKED) ?
2947 		    &uvm.page_inactive_swp : &uvm.page_inactive_obj;
2948 	else if (pg->pqflags & PQ_ACTIVE)
2949 		pgl = &uvm.page_active;
2950 	else
2951 		pgl = NULL;
2952 
2953 	if (pgl) {
2954 		(*pr)("  checking pageq list\n");
2955 		for (lcv = pgl->tqh_first ; lcv ; lcv = lcv->pageq.tqe_next) {
2956 			if (lcv == pg) break;
2957 		}
2958 		if (lcv)
2959 			(*pr)("  page found on pageq list\n");
2960 		else
2961 			(*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
2962 	}
2963 }
2964 #endif
2965