xref: /netbsd-src/sys/uvm/uvm_aobj.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: uvm_aobj.c,v 1.95 2007/12/01 10:40:27 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1998 Chuck Silvers, Charles D. Cranor and
5  *                    Washington University.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *      This product includes software developed by Charles D. Cranor and
19  *      Washington University.
20  * 4. The name of the author may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  *
34  * from: Id: uvm_aobj.c,v 1.1.2.5 1998/02/06 05:14:38 chs Exp
35  */
36 /*
37  * uvm_aobj.c: anonymous memory uvm_object pager
38  *
39  * author: Chuck Silvers <chuq@chuq.com>
40  * started: Jan-1998
41  *
42  * - design mostly from Chuck Cranor
43  */
44 
45 #include <sys/cdefs.h>
46 __KERNEL_RCSID(0, "$NetBSD: uvm_aobj.c,v 1.95 2007/12/01 10:40:27 yamt Exp $");
47 
48 #include "opt_uvmhist.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/proc.h>
53 #include <sys/malloc.h>
54 #include <sys/kernel.h>
55 #include <sys/pool.h>
56 
57 #include <uvm/uvm.h>
58 
59 /*
60  * an aobj manages anonymous-memory backed uvm_objects.   in addition
61  * to keeping the list of resident pages, it also keeps a list of
62  * allocated swap blocks.  depending on the size of the aobj this list
63  * of allocated swap blocks is either stored in an array (small objects)
64  * or in a hash table (large objects).
65  */
66 
67 /*
68  * local structures
69  */
70 
71 /*
72  * for hash tables, we break the address space of the aobj into blocks
73  * of UAO_SWHASH_CLUSTER_SIZE pages.   we require the cluster size to
74  * be a power of two.
75  */
76 
77 #define UAO_SWHASH_CLUSTER_SHIFT 4
78 #define UAO_SWHASH_CLUSTER_SIZE (1 << UAO_SWHASH_CLUSTER_SHIFT)
79 
80 /* get the "tag" for this page index */
81 #define UAO_SWHASH_ELT_TAG(PAGEIDX) \
82 	((PAGEIDX) >> UAO_SWHASH_CLUSTER_SHIFT)
83 
84 #define UAO_SWHASH_ELT_PAGESLOT_IDX(PAGEIDX) \
85 	((PAGEIDX) & (UAO_SWHASH_CLUSTER_SIZE - 1))
86 
87 /* given an ELT and a page index, find the swap slot */
88 #define UAO_SWHASH_ELT_PAGESLOT(ELT, PAGEIDX) \
89 	((ELT)->slots[UAO_SWHASH_ELT_PAGESLOT_IDX(PAGEIDX)])
90 
91 /* given an ELT, return its pageidx base */
92 #define UAO_SWHASH_ELT_PAGEIDX_BASE(ELT) \
93 	((ELT)->tag << UAO_SWHASH_CLUSTER_SHIFT)
94 
95 /*
96  * the swhash hash function
97  */
98 
99 #define UAO_SWHASH_HASH(AOBJ, PAGEIDX) \
100 	(&(AOBJ)->u_swhash[(((PAGEIDX) >> UAO_SWHASH_CLUSTER_SHIFT) \
101 			    & (AOBJ)->u_swhashmask)])
102 
103 /*
104  * the swhash threshhold determines if we will use an array or a
105  * hash table to store the list of allocated swap blocks.
106  */
107 
108 #define UAO_SWHASH_THRESHOLD (UAO_SWHASH_CLUSTER_SIZE * 4)
109 #define UAO_USES_SWHASH(AOBJ) \
110 	((AOBJ)->u_pages > UAO_SWHASH_THRESHOLD)	/* use hash? */
111 
112 /*
113  * the number of buckets in a swhash, with an upper bound
114  */
115 
116 #define UAO_SWHASH_MAXBUCKETS 256
117 #define UAO_SWHASH_BUCKETS(AOBJ) \
118 	(MIN((AOBJ)->u_pages >> UAO_SWHASH_CLUSTER_SHIFT, \
119 	     UAO_SWHASH_MAXBUCKETS))
120 
121 
122 /*
123  * uao_swhash_elt: when a hash table is being used, this structure defines
124  * the format of an entry in the bucket list.
125  */
126 
127 struct uao_swhash_elt {
128 	LIST_ENTRY(uao_swhash_elt) list;	/* the hash list */
129 	voff_t tag;				/* our 'tag' */
130 	int count;				/* our number of active slots */
131 	int slots[UAO_SWHASH_CLUSTER_SIZE];	/* the slots */
132 };
133 
134 /*
135  * uao_swhash: the swap hash table structure
136  */
137 
138 LIST_HEAD(uao_swhash, uao_swhash_elt);
139 
140 /*
141  * uao_swhash_elt_pool: pool of uao_swhash_elt structures
142  * NOTE: Pages for this pool must not come from a pageable kernel map!
143  */
144 POOL_INIT(uao_swhash_elt_pool, sizeof(struct uao_swhash_elt), 0, 0, 0,
145     "uaoeltpl", NULL, IPL_VM);
146 
147 /*
148  * uvm_aobj: the actual anon-backed uvm_object
149  *
150  * => the uvm_object is at the top of the structure, this allows
151  *   (struct uvm_aobj *) == (struct uvm_object *)
152  * => only one of u_swslots and u_swhash is used in any given aobj
153  */
154 
155 struct uvm_aobj {
156 	struct uvm_object u_obj; /* has: lock, pgops, memq, #pages, #refs */
157 	pgoff_t u_pages;	 /* number of pages in entire object */
158 	int u_flags;		 /* the flags (see uvm_aobj.h) */
159 	int *u_swslots;		 /* array of offset->swapslot mappings */
160 				 /*
161 				  * hashtable of offset->swapslot mappings
162 				  * (u_swhash is an array of bucket heads)
163 				  */
164 	struct uao_swhash *u_swhash;
165 	u_long u_swhashmask;		/* mask for hashtable */
166 	LIST_ENTRY(uvm_aobj) u_list;	/* global list of aobjs */
167 };
168 
169 /*
170  * uvm_aobj_pool: pool of uvm_aobj structures
171  */
172 POOL_INIT(uvm_aobj_pool, sizeof(struct uvm_aobj), 0, 0, 0, "aobjpl",
173     &pool_allocator_nointr, IPL_NONE);
174 
175 MALLOC_DEFINE(M_UVMAOBJ, "UVM aobj", "UVM aobj and related structures");
176 
177 /*
178  * local functions
179  */
180 
181 static void	uao_free(struct uvm_aobj *);
182 static int	uao_get(struct uvm_object *, voff_t, struct vm_page **,
183 		    int *, int, vm_prot_t, int, int);
184 static int	uao_put(struct uvm_object *, voff_t, voff_t, int);
185 
186 #if defined(VMSWAP)
187 static struct uao_swhash_elt *uao_find_swhash_elt
188     (struct uvm_aobj *, int, bool);
189 
190 static bool uao_pagein(struct uvm_aobj *, int, int);
191 static bool uao_pagein_page(struct uvm_aobj *, int);
192 static void uao_dropswap_range1(struct uvm_aobj *, voff_t, voff_t);
193 #endif /* defined(VMSWAP) */
194 
195 /*
196  * aobj_pager
197  *
198  * note that some functions (e.g. put) are handled elsewhere
199  */
200 
201 const struct uvm_pagerops aobj_pager = {
202 	.pgo_reference = uao_reference,
203 	.pgo_detach = uao_detach,
204 	.pgo_get = uao_get,
205 	.pgo_put = uao_put,
206 };
207 
208 /*
209  * uao_list: global list of active aobjs, locked by uao_list_lock
210  */
211 
212 static LIST_HEAD(aobjlist, uvm_aobj) uao_list;
213 static kmutex_t uao_list_lock;
214 
215 /*
216  * functions
217  */
218 
219 /*
220  * hash table/array related functions
221  */
222 
223 #if defined(VMSWAP)
224 
225 /*
226  * uao_find_swhash_elt: find (or create) a hash table entry for a page
227  * offset.
228  *
229  * => the object should be locked by the caller
230  */
231 
232 static struct uao_swhash_elt *
233 uao_find_swhash_elt(struct uvm_aobj *aobj, int pageidx, bool create)
234 {
235 	struct uao_swhash *swhash;
236 	struct uao_swhash_elt *elt;
237 	voff_t page_tag;
238 
239 	swhash = UAO_SWHASH_HASH(aobj, pageidx);
240 	page_tag = UAO_SWHASH_ELT_TAG(pageidx);
241 
242 	/*
243 	 * now search the bucket for the requested tag
244 	 */
245 
246 	LIST_FOREACH(elt, swhash, list) {
247 		if (elt->tag == page_tag) {
248 			return elt;
249 		}
250 	}
251 	if (!create) {
252 		return NULL;
253 	}
254 
255 	/*
256 	 * allocate a new entry for the bucket and init/insert it in
257 	 */
258 
259 	elt = pool_get(&uao_swhash_elt_pool, PR_NOWAIT);
260 	if (elt == NULL) {
261 		return NULL;
262 	}
263 	LIST_INSERT_HEAD(swhash, elt, list);
264 	elt->tag = page_tag;
265 	elt->count = 0;
266 	memset(elt->slots, 0, sizeof(elt->slots));
267 	return elt;
268 }
269 
270 /*
271  * uao_find_swslot: find the swap slot number for an aobj/pageidx
272  *
273  * => object must be locked by caller
274  */
275 
276 int
277 uao_find_swslot(struct uvm_object *uobj, int pageidx)
278 {
279 	struct uvm_aobj *aobj = (struct uvm_aobj *)uobj;
280 	struct uao_swhash_elt *elt;
281 
282 	/*
283 	 * if noswap flag is set, then we never return a slot
284 	 */
285 
286 	if (aobj->u_flags & UAO_FLAG_NOSWAP)
287 		return(0);
288 
289 	/*
290 	 * if hashing, look in hash table.
291 	 */
292 
293 	if (UAO_USES_SWHASH(aobj)) {
294 		elt = uao_find_swhash_elt(aobj, pageidx, false);
295 		if (elt)
296 			return(UAO_SWHASH_ELT_PAGESLOT(elt, pageidx));
297 		else
298 			return(0);
299 	}
300 
301 	/*
302 	 * otherwise, look in the array
303 	 */
304 
305 	return(aobj->u_swslots[pageidx]);
306 }
307 
308 /*
309  * uao_set_swslot: set the swap slot for a page in an aobj.
310  *
311  * => setting a slot to zero frees the slot
312  * => object must be locked by caller
313  * => we return the old slot number, or -1 if we failed to allocate
314  *    memory to record the new slot number
315  */
316 
317 int
318 uao_set_swslot(struct uvm_object *uobj, int pageidx, int slot)
319 {
320 	struct uvm_aobj *aobj = (struct uvm_aobj *)uobj;
321 	struct uao_swhash_elt *elt;
322 	int oldslot;
323 	UVMHIST_FUNC("uao_set_swslot"); UVMHIST_CALLED(pdhist);
324 	UVMHIST_LOG(pdhist, "aobj %p pageidx %d slot %d",
325 	    aobj, pageidx, slot, 0);
326 
327 	/*
328 	 * if noswap flag is set, then we can't set a non-zero slot.
329 	 */
330 
331 	if (aobj->u_flags & UAO_FLAG_NOSWAP) {
332 		if (slot == 0)
333 			return(0);
334 
335 		printf("uao_set_swslot: uobj = %p\n", uobj);
336 		panic("uao_set_swslot: NOSWAP object");
337 	}
338 
339 	/*
340 	 * are we using a hash table?  if so, add it in the hash.
341 	 */
342 
343 	if (UAO_USES_SWHASH(aobj)) {
344 
345 		/*
346 		 * Avoid allocating an entry just to free it again if
347 		 * the page had not swap slot in the first place, and
348 		 * we are freeing.
349 		 */
350 
351 		elt = uao_find_swhash_elt(aobj, pageidx, slot != 0);
352 		if (elt == NULL) {
353 			return slot ? -1 : 0;
354 		}
355 
356 		oldslot = UAO_SWHASH_ELT_PAGESLOT(elt, pageidx);
357 		UAO_SWHASH_ELT_PAGESLOT(elt, pageidx) = slot;
358 
359 		/*
360 		 * now adjust the elt's reference counter and free it if we've
361 		 * dropped it to zero.
362 		 */
363 
364 		if (slot) {
365 			if (oldslot == 0)
366 				elt->count++;
367 		} else {
368 			if (oldslot)
369 				elt->count--;
370 
371 			if (elt->count == 0) {
372 				LIST_REMOVE(elt, list);
373 				pool_put(&uao_swhash_elt_pool, elt);
374 			}
375 		}
376 	} else {
377 		/* we are using an array */
378 		oldslot = aobj->u_swslots[pageidx];
379 		aobj->u_swslots[pageidx] = slot;
380 	}
381 	return (oldslot);
382 }
383 
384 #endif /* defined(VMSWAP) */
385 
386 /*
387  * end of hash/array functions
388  */
389 
390 /*
391  * uao_free: free all resources held by an aobj, and then free the aobj
392  *
393  * => the aobj should be dead
394  */
395 
396 static void
397 uao_free(struct uvm_aobj *aobj)
398 {
399 	int swpgonlydelta = 0;
400 
401 #if defined(VMSWAP)
402 	uao_dropswap_range1(aobj, 0, 0);
403 #endif /* defined(VMSWAP) */
404 
405 	simple_unlock(&aobj->u_obj.vmobjlock);
406 
407 #if defined(VMSWAP)
408 	if (UAO_USES_SWHASH(aobj)) {
409 
410 		/*
411 		 * free the hash table itself.
412 		 */
413 
414 		free(aobj->u_swhash, M_UVMAOBJ);
415 	} else {
416 
417 		/*
418 		 * free the array itsself.
419 		 */
420 
421 		free(aobj->u_swslots, M_UVMAOBJ);
422 	}
423 #endif /* defined(VMSWAP) */
424 
425 	/*
426 	 * finally free the aobj itself
427 	 */
428 
429 	pool_put(&uvm_aobj_pool, aobj);
430 
431 	/*
432 	 * adjust the counter of pages only in swap for all
433 	 * the swap slots we've freed.
434 	 */
435 
436 	if (swpgonlydelta > 0) {
437 		mutex_enter(&uvm_swap_data_lock);
438 		KASSERT(uvmexp.swpgonly >= swpgonlydelta);
439 		uvmexp.swpgonly -= swpgonlydelta;
440 		mutex_exit(&uvm_swap_data_lock);
441 	}
442 }
443 
444 /*
445  * pager functions
446  */
447 
448 /*
449  * uao_create: create an aobj of the given size and return its uvm_object.
450  *
451  * => for normal use, flags are always zero
452  * => for the kernel object, the flags are:
453  *	UAO_FLAG_KERNOBJ - allocate the kernel object (can only happen once)
454  *	UAO_FLAG_KERNSWAP - enable swapping of kernel object ("           ")
455  */
456 
457 struct uvm_object *
458 uao_create(vsize_t size, int flags)
459 {
460 	static struct uvm_aobj kernel_object_store;
461 	static int kobj_alloced = 0;
462 	pgoff_t pages = round_page(size) >> PAGE_SHIFT;
463 	struct uvm_aobj *aobj;
464 	int refs;
465 
466 	/*
467 	 * malloc a new aobj unless we are asked for the kernel object
468 	 */
469 
470 	if (flags & UAO_FLAG_KERNOBJ) {
471 		KASSERT(!kobj_alloced);
472 		aobj = &kernel_object_store;
473 		aobj->u_pages = pages;
474 		aobj->u_flags = UAO_FLAG_NOSWAP;
475 		refs = UVM_OBJ_KERN;
476 		kobj_alloced = UAO_FLAG_KERNOBJ;
477 	} else if (flags & UAO_FLAG_KERNSWAP) {
478 		KASSERT(kobj_alloced == UAO_FLAG_KERNOBJ);
479 		aobj = &kernel_object_store;
480 		kobj_alloced = UAO_FLAG_KERNSWAP;
481 		refs = 0xdeadbeaf; /* XXX: gcc */
482 	} else {
483 		aobj = pool_get(&uvm_aobj_pool, PR_WAITOK);
484 		aobj->u_pages = pages;
485 		aobj->u_flags = 0;
486 		refs = 1;
487 	}
488 
489 	/*
490  	 * allocate hash/array if necessary
491  	 *
492  	 * note: in the KERNSWAP case no need to worry about locking since
493  	 * we are still booting we should be the only thread around.
494  	 */
495 
496 	if (flags == 0 || (flags & UAO_FLAG_KERNSWAP) != 0) {
497 #if defined(VMSWAP)
498 		int mflags = (flags & UAO_FLAG_KERNSWAP) != 0 ?
499 		    M_NOWAIT : M_WAITOK;
500 
501 		/* allocate hash table or array depending on object size */
502 		if (UAO_USES_SWHASH(aobj)) {
503 			aobj->u_swhash = hashinit(UAO_SWHASH_BUCKETS(aobj),
504 			    HASH_LIST, M_UVMAOBJ, mflags, &aobj->u_swhashmask);
505 			if (aobj->u_swhash == NULL)
506 				panic("uao_create: hashinit swhash failed");
507 		} else {
508 			aobj->u_swslots = malloc(pages * sizeof(int),
509 			    M_UVMAOBJ, mflags);
510 			if (aobj->u_swslots == NULL)
511 				panic("uao_create: malloc swslots failed");
512 			memset(aobj->u_swslots, 0, pages * sizeof(int));
513 		}
514 #endif /* defined(VMSWAP) */
515 
516 		if (flags) {
517 			aobj->u_flags &= ~UAO_FLAG_NOSWAP; /* clear noswap */
518 			return(&aobj->u_obj);
519 		}
520 	}
521 
522 	/*
523  	 * init aobj fields
524  	 */
525 
526 	UVM_OBJ_INIT(&aobj->u_obj, &aobj_pager, refs);
527 
528 	/*
529  	 * now that aobj is ready, add it to the global list
530  	 */
531 
532 	mutex_enter(&uao_list_lock);
533 	LIST_INSERT_HEAD(&uao_list, aobj, u_list);
534 	mutex_exit(&uao_list_lock);
535 	return(&aobj->u_obj);
536 }
537 
538 
539 
540 /*
541  * uao_init: set up aobj pager subsystem
542  *
543  * => called at boot time from uvm_pager_init()
544  */
545 
546 void
547 uao_init(void)
548 {
549 	static int uao_initialized;
550 
551 	if (uao_initialized)
552 		return;
553 	uao_initialized = true;
554 	LIST_INIT(&uao_list);
555 	/* XXXSMP should be adaptive but vmobjlock needs to be too */
556 	mutex_init(&uao_list_lock, MUTEX_SPIN, IPL_NONE);
557 }
558 
559 /*
560  * uao_reference: add a ref to an aobj
561  *
562  * => aobj must be unlocked
563  * => just lock it and call the locked version
564  */
565 
566 void
567 uao_reference(struct uvm_object *uobj)
568 {
569 	simple_lock(&uobj->vmobjlock);
570 	uao_reference_locked(uobj);
571 	simple_unlock(&uobj->vmobjlock);
572 }
573 
574 /*
575  * uao_reference_locked: add a ref to an aobj that is already locked
576  *
577  * => aobj must be locked
578  * this needs to be separate from the normal routine
579  * since sometimes we need to add a reference to an aobj when
580  * it's already locked.
581  */
582 
583 void
584 uao_reference_locked(struct uvm_object *uobj)
585 {
586 	UVMHIST_FUNC("uao_reference"); UVMHIST_CALLED(maphist);
587 
588 	/*
589  	 * kernel_object already has plenty of references, leave it alone.
590  	 */
591 
592 	if (UVM_OBJ_IS_KERN_OBJECT(uobj))
593 		return;
594 
595 	uobj->uo_refs++;
596 	UVMHIST_LOG(maphist, "<- done (uobj=0x%x, ref = %d)",
597 		    uobj, uobj->uo_refs,0,0);
598 }
599 
600 /*
601  * uao_detach: drop a reference to an aobj
602  *
603  * => aobj must be unlocked
604  * => just lock it and call the locked version
605  */
606 
607 void
608 uao_detach(struct uvm_object *uobj)
609 {
610 	simple_lock(&uobj->vmobjlock);
611 	uao_detach_locked(uobj);
612 }
613 
614 /*
615  * uao_detach_locked: drop a reference to an aobj
616  *
617  * => aobj must be locked, and is unlocked (or freed) upon return.
618  * this needs to be separate from the normal routine
619  * since sometimes we need to detach from an aobj when
620  * it's already locked.
621  */
622 
623 void
624 uao_detach_locked(struct uvm_object *uobj)
625 {
626 	struct uvm_aobj *aobj = (struct uvm_aobj *)uobj;
627 	struct vm_page *pg;
628 	UVMHIST_FUNC("uao_detach"); UVMHIST_CALLED(maphist);
629 
630 	/*
631  	 * detaching from kernel_object is a noop.
632  	 */
633 
634 	if (UVM_OBJ_IS_KERN_OBJECT(uobj)) {
635 		simple_unlock(&uobj->vmobjlock);
636 		return;
637 	}
638 
639 	UVMHIST_LOG(maphist,"  (uobj=0x%x)  ref=%d", uobj,uobj->uo_refs,0,0);
640 	uobj->uo_refs--;
641 	if (uobj->uo_refs) {
642 		simple_unlock(&uobj->vmobjlock);
643 		UVMHIST_LOG(maphist, "<- done (rc>0)", 0,0,0,0);
644 		return;
645 	}
646 
647 	/*
648  	 * remove the aobj from the global list.
649  	 */
650 
651 	mutex_enter(&uao_list_lock);
652 	LIST_REMOVE(aobj, u_list);
653 	mutex_exit(&uao_list_lock);
654 
655 	/*
656  	 * free all the pages left in the aobj.  for each page,
657 	 * when the page is no longer busy (and thus after any disk i/o that
658 	 * it's involved in is complete), release any swap resources and
659 	 * free the page itself.
660  	 */
661 
662 	uvm_lock_pageq();
663 	while ((pg = TAILQ_FIRST(&uobj->memq)) != NULL) {
664 		pmap_page_protect(pg, VM_PROT_NONE);
665 		if (pg->flags & PG_BUSY) {
666 			pg->flags |= PG_WANTED;
667 			uvm_unlock_pageq();
668 			UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, false,
669 			    "uao_det", 0);
670 			simple_lock(&uobj->vmobjlock);
671 			uvm_lock_pageq();
672 			continue;
673 		}
674 		uao_dropswap(&aobj->u_obj, pg->offset >> PAGE_SHIFT);
675 		uvm_pagefree(pg);
676 	}
677 	uvm_unlock_pageq();
678 
679 	/*
680  	 * finally, free the aobj itself.
681  	 */
682 
683 	uao_free(aobj);
684 }
685 
686 /*
687  * uao_put: flush pages out of a uvm object
688  *
689  * => object should be locked by caller.  we may _unlock_ the object
690  *	if (and only if) we need to clean a page (PGO_CLEANIT).
691  *	XXXJRT Currently, however, we don't.  In the case of cleaning
692  *	XXXJRT a page, we simply just deactivate it.  Should probably
693  *	XXXJRT handle this better, in the future (although "flushing"
694  *	XXXJRT anonymous memory isn't terribly important).
695  * => if PGO_CLEANIT is not set, then we will neither unlock the object
696  *	or block.
697  * => if PGO_ALLPAGE is set, then all pages in the object are valid targets
698  *	for flushing.
699  * => NOTE: we rely on the fact that the object's memq is a TAILQ and
700  *	that new pages are inserted on the tail end of the list.  thus,
701  *	we can make a complete pass through the object in one go by starting
702  *	at the head and working towards the tail (new pages are put in
703  *	front of us).
704  * => NOTE: we are allowed to lock the page queues, so the caller
705  *	must not be holding the lock on them [e.g. pagedaemon had
706  *	better not call us with the queues locked]
707  * => we return 0 unless we encountered some sort of I/O error
708  *	XXXJRT currently never happens, as we never directly initiate
709  *	XXXJRT I/O
710  *
711  * note on page traversal:
712  *	we can traverse the pages in an object either by going down the
713  *	linked list in "uobj->memq", or we can go over the address range
714  *	by page doing hash table lookups for each address.  depending
715  *	on how many pages are in the object it may be cheaper to do one
716  *	or the other.  we set "by_list" to true if we are using memq.
717  *	if the cost of a hash lookup was equal to the cost of the list
718  *	traversal we could compare the number of pages in the start->stop
719  *	range to the total number of pages in the object.  however, it
720  *	seems that a hash table lookup is more expensive than the linked
721  *	list traversal, so we multiply the number of pages in the
722  *	start->stop range by a penalty which we define below.
723  */
724 
725 static int
726 uao_put(struct uvm_object *uobj, voff_t start, voff_t stop, int flags)
727 {
728 	struct uvm_aobj *aobj = (struct uvm_aobj *)uobj;
729 	struct vm_page *pg, *nextpg, curmp, endmp;
730 	bool by_list;
731 	voff_t curoff;
732 	UVMHIST_FUNC("uao_put"); UVMHIST_CALLED(maphist);
733 
734 	curoff = 0;
735 	if (flags & PGO_ALLPAGES) {
736 		start = 0;
737 		stop = aobj->u_pages << PAGE_SHIFT;
738 		by_list = true;		/* always go by the list */
739 	} else {
740 		start = trunc_page(start);
741 		if (stop == 0) {
742 			stop = aobj->u_pages << PAGE_SHIFT;
743 		} else {
744 			stop = round_page(stop);
745 		}
746 		if (stop > (aobj->u_pages << PAGE_SHIFT)) {
747 			printf("uao_flush: strange, got an out of range "
748 			    "flush (fixed)\n");
749 			stop = aobj->u_pages << PAGE_SHIFT;
750 		}
751 		by_list = (uobj->uo_npages <=
752 		    ((stop - start) >> PAGE_SHIFT) * UVM_PAGE_HASH_PENALTY);
753 	}
754 	UVMHIST_LOG(maphist,
755 	    " flush start=0x%lx, stop=0x%x, by_list=%d, flags=0x%x",
756 	    start, stop, by_list, flags);
757 
758 	/*
759 	 * Don't need to do any work here if we're not freeing
760 	 * or deactivating pages.
761 	 */
762 
763 	if ((flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) {
764 		simple_unlock(&uobj->vmobjlock);
765 		return 0;
766 	}
767 
768 	/*
769 	 * Initialize the marker pages.  See the comment in
770 	 * genfs_putpages() also.
771 	 */
772 
773 	curmp.uobject = uobj;
774 	curmp.offset = (voff_t)-1;
775 	curmp.flags = PG_BUSY;
776 	endmp.uobject = uobj;
777 	endmp.offset = (voff_t)-1;
778 	endmp.flags = PG_BUSY;
779 
780 	/*
781 	 * now do it.  note: we must update nextpg in the body of loop or we
782 	 * will get stuck.  we need to use nextpg if we'll traverse the list
783 	 * because we may free "pg" before doing the next loop.
784 	 */
785 
786 	if (by_list) {
787 		TAILQ_INSERT_TAIL(&uobj->memq, &endmp, listq);
788 		nextpg = TAILQ_FIRST(&uobj->memq);
789 		uvm_lwp_hold(curlwp);
790 	} else {
791 		curoff = start;
792 		nextpg = NULL;	/* Quell compiler warning */
793 	}
794 
795 	uvm_lock_pageq();
796 
797 	/* locked: both page queues and uobj */
798 	for (;;) {
799 		if (by_list) {
800 			pg = nextpg;
801 			if (pg == &endmp)
802 				break;
803 			nextpg = TAILQ_NEXT(pg, listq);
804 			if (pg->offset < start || pg->offset >= stop)
805 				continue;
806 		} else {
807 			if (curoff < stop) {
808 				pg = uvm_pagelookup(uobj, curoff);
809 				curoff += PAGE_SIZE;
810 			} else
811 				break;
812 			if (pg == NULL)
813 				continue;
814 		}
815 		switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
816 
817 		/*
818 		 * XXX In these first 3 cases, we always just
819 		 * XXX deactivate the page.  We may want to
820 		 * XXX handle the different cases more specifically
821 		 * XXX in the future.
822 		 */
823 
824 		case PGO_CLEANIT|PGO_FREE:
825 		case PGO_CLEANIT|PGO_DEACTIVATE:
826 		case PGO_DEACTIVATE:
827  deactivate_it:
828 			/* skip the page if it's wired */
829 			if (pg->wire_count != 0)
830 				continue;
831 
832 			/* ...and deactivate the page. */
833 			pmap_clear_reference(pg);
834 			uvm_pagedeactivate(pg);
835 			continue;
836 
837 		case PGO_FREE:
838 
839 			/*
840 			 * If there are multiple references to
841 			 * the object, just deactivate the page.
842 			 */
843 
844 			if (uobj->uo_refs > 1)
845 				goto deactivate_it;
846 
847 			/*
848 			 * wait and try again if the page is busy.
849 			 * otherwise free the swap slot and the page.
850 			 */
851 
852 			pmap_page_protect(pg, VM_PROT_NONE);
853 			if (pg->flags & PG_BUSY) {
854 				if (by_list) {
855 					TAILQ_INSERT_BEFORE(pg, &curmp, listq);
856 				}
857 				pg->flags |= PG_WANTED;
858 				uvm_unlock_pageq();
859 				UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
860 				    "uao_put", 0);
861 				simple_lock(&uobj->vmobjlock);
862 				uvm_lock_pageq();
863 				if (by_list) {
864 					nextpg = TAILQ_NEXT(&curmp, listq);
865 					TAILQ_REMOVE(&uobj->memq, &curmp,
866 					    listq);
867 				} else
868 					curoff -= PAGE_SIZE;
869 				continue;
870 			}
871 
872 			/*
873 			 * freeing swapslot here is not strictly necessary.
874 			 * however, leaving it here doesn't save much
875 			 * because we need to update swap accounting anyway.
876 			 */
877 
878 			uao_dropswap(uobj, pg->offset >> PAGE_SHIFT);
879 			uvm_pagefree(pg);
880 			continue;
881 		}
882 	}
883 	uvm_unlock_pageq();
884 	if (by_list) {
885 		TAILQ_REMOVE(&uobj->memq, &endmp, listq);
886 	}
887 	simple_unlock(&uobj->vmobjlock);
888 	if (by_list) {
889 		uvm_lwp_rele(curlwp);
890 	}
891 	return 0;
892 }
893 
894 /*
895  * uao_get: fetch me a page
896  *
897  * we have three cases:
898  * 1: page is resident     -> just return the page.
899  * 2: page is zero-fill    -> allocate a new page and zero it.
900  * 3: page is swapped out  -> fetch the page from swap.
901  *
902  * cases 1 and 2 can be handled with PGO_LOCKED, case 3 cannot.
903  * so, if the "center" page hits case 3 (or any page, with PGO_ALLPAGES),
904  * then we will need to return EBUSY.
905  *
906  * => prefer map unlocked (not required)
907  * => object must be locked!  we will _unlock_ it before starting any I/O.
908  * => flags: PGO_ALLPAGES: get all of the pages
909  *           PGO_LOCKED: fault data structures are locked
910  * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx]
911  * => NOTE: caller must check for released pages!!
912  */
913 
914 static int
915 uao_get(struct uvm_object *uobj, voff_t offset, struct vm_page **pps,
916     int *npagesp, int centeridx, vm_prot_t access_type, int advice, int flags)
917 {
918 #if defined(VMSWAP)
919 	struct uvm_aobj *aobj = (struct uvm_aobj *)uobj;
920 #endif /* defined(VMSWAP) */
921 	voff_t current_offset;
922 	struct vm_page *ptmp = NULL;	/* Quell compiler warning */
923 	int lcv, gotpages, maxpages, swslot, pageidx;
924 	bool done;
925 	UVMHIST_FUNC("uao_get"); UVMHIST_CALLED(pdhist);
926 
927 	UVMHIST_LOG(pdhist, "aobj=%p offset=%d, flags=%d",
928 		    (struct uvm_aobj *)uobj, offset, flags,0);
929 
930 	/*
931  	 * get number of pages
932  	 */
933 
934 	maxpages = *npagesp;
935 
936 	/*
937  	 * step 1: handled the case where fault data structures are locked.
938  	 */
939 
940 	if (flags & PGO_LOCKED) {
941 
942 		/*
943  		 * step 1a: get pages that are already resident.   only do
944 		 * this if the data structures are locked (i.e. the first
945 		 * time through).
946  		 */
947 
948 		done = true;	/* be optimistic */
949 		gotpages = 0;	/* # of pages we got so far */
950 		for (lcv = 0, current_offset = offset ; lcv < maxpages ;
951 		    lcv++, current_offset += PAGE_SIZE) {
952 			/* do we care about this page?  if not, skip it */
953 			if (pps[lcv] == PGO_DONTCARE)
954 				continue;
955 			ptmp = uvm_pagelookup(uobj, current_offset);
956 
957 			/*
958  			 * if page is new, attempt to allocate the page,
959 			 * zero-fill'd.
960  			 */
961 
962 			if (ptmp == NULL && uao_find_swslot(&aobj->u_obj,
963 			    current_offset >> PAGE_SHIFT) == 0) {
964 				ptmp = uvm_pagealloc(uobj, current_offset,
965 				    NULL, UVM_PGA_ZERO);
966 				if (ptmp) {
967 					/* new page */
968 					ptmp->flags &= ~(PG_FAKE);
969 					ptmp->pqflags |= PQ_AOBJ;
970 					goto gotpage;
971 				}
972 			}
973 
974 			/*
975 			 * to be useful must get a non-busy page
976 			 */
977 
978 			if (ptmp == NULL || (ptmp->flags & PG_BUSY) != 0) {
979 				if (lcv == centeridx ||
980 				    (flags & PGO_ALLPAGES) != 0)
981 					/* need to do a wait or I/O! */
982 					done = false;
983 					continue;
984 			}
985 
986 			/*
987 			 * useful page: busy/lock it and plug it in our
988 			 * result array
989 			 */
990 
991 			/* caller must un-busy this page */
992 			ptmp->flags |= PG_BUSY;
993 			UVM_PAGE_OWN(ptmp, "uao_get1");
994 gotpage:
995 			pps[lcv] = ptmp;
996 			gotpages++;
997 		}
998 
999 		/*
1000  		 * step 1b: now we've either done everything needed or we
1001 		 * to unlock and do some waiting or I/O.
1002  		 */
1003 
1004 		UVMHIST_LOG(pdhist, "<- done (done=%d)", done, 0,0,0);
1005 		*npagesp = gotpages;
1006 		if (done)
1007 			return 0;
1008 		else
1009 			return EBUSY;
1010 	}
1011 
1012 	/*
1013  	 * step 2: get non-resident or busy pages.
1014  	 * object is locked.   data structures are unlocked.
1015  	 */
1016 
1017 	if ((flags & PGO_SYNCIO) == 0) {
1018 		goto done;
1019 	}
1020 
1021 	for (lcv = 0, current_offset = offset ; lcv < maxpages ;
1022 	    lcv++, current_offset += PAGE_SIZE) {
1023 
1024 		/*
1025 		 * - skip over pages we've already gotten or don't want
1026 		 * - skip over pages we don't _have_ to get
1027 		 */
1028 
1029 		if (pps[lcv] != NULL ||
1030 		    (lcv != centeridx && (flags & PGO_ALLPAGES) == 0))
1031 			continue;
1032 
1033 		pageidx = current_offset >> PAGE_SHIFT;
1034 
1035 		/*
1036  		 * we have yet to locate the current page (pps[lcv]).   we
1037 		 * first look for a page that is already at the current offset.
1038 		 * if we find a page, we check to see if it is busy or
1039 		 * released.  if that is the case, then we sleep on the page
1040 		 * until it is no longer busy or released and repeat the lookup.
1041 		 * if the page we found is neither busy nor released, then we
1042 		 * busy it (so we own it) and plug it into pps[lcv].   this
1043 		 * 'break's the following while loop and indicates we are
1044 		 * ready to move on to the next page in the "lcv" loop above.
1045  		 *
1046  		 * if we exit the while loop with pps[lcv] still set to NULL,
1047 		 * then it means that we allocated a new busy/fake/clean page
1048 		 * ptmp in the object and we need to do I/O to fill in the data.
1049  		 */
1050 
1051 		/* top of "pps" while loop */
1052 		while (pps[lcv] == NULL) {
1053 			/* look for a resident page */
1054 			ptmp = uvm_pagelookup(uobj, current_offset);
1055 
1056 			/* not resident?   allocate one now (if we can) */
1057 			if (ptmp == NULL) {
1058 
1059 				ptmp = uvm_pagealloc(uobj, current_offset,
1060 				    NULL, 0);
1061 
1062 				/* out of RAM? */
1063 				if (ptmp == NULL) {
1064 					simple_unlock(&uobj->vmobjlock);
1065 					UVMHIST_LOG(pdhist,
1066 					    "sleeping, ptmp == NULL\n",0,0,0,0);
1067 					uvm_wait("uao_getpage");
1068 					simple_lock(&uobj->vmobjlock);
1069 					continue;
1070 				}
1071 
1072 				/*
1073 				 * safe with PQ's unlocked: because we just
1074 				 * alloc'd the page
1075 				 */
1076 
1077 				ptmp->pqflags |= PQ_AOBJ;
1078 
1079 				/*
1080 				 * got new page ready for I/O.  break pps while
1081 				 * loop.  pps[lcv] is still NULL.
1082 				 */
1083 
1084 				break;
1085 			}
1086 
1087 			/* page is there, see if we need to wait on it */
1088 			if ((ptmp->flags & PG_BUSY) != 0) {
1089 				ptmp->flags |= PG_WANTED;
1090 				UVMHIST_LOG(pdhist,
1091 				    "sleeping, ptmp->flags 0x%x\n",
1092 				    ptmp->flags,0,0,0);
1093 				UVM_UNLOCK_AND_WAIT(ptmp, &uobj->vmobjlock,
1094 				    false, "uao_get", 0);
1095 				simple_lock(&uobj->vmobjlock);
1096 				continue;
1097 			}
1098 
1099 			/*
1100  			 * if we get here then the page has become resident and
1101 			 * unbusy between steps 1 and 2.  we busy it now (so we
1102 			 * own it) and set pps[lcv] (so that we exit the while
1103 			 * loop).
1104  			 */
1105 
1106 			/* we own it, caller must un-busy */
1107 			ptmp->flags |= PG_BUSY;
1108 			UVM_PAGE_OWN(ptmp, "uao_get2");
1109 			pps[lcv] = ptmp;
1110 		}
1111 
1112 		/*
1113  		 * if we own the valid page at the correct offset, pps[lcv] will
1114  		 * point to it.   nothing more to do except go to the next page.
1115  		 */
1116 
1117 		if (pps[lcv])
1118 			continue;			/* next lcv */
1119 
1120 		/*
1121  		 * we have a "fake/busy/clean" page that we just allocated.
1122  		 * do the needed "i/o", either reading from swap or zeroing.
1123  		 */
1124 
1125 		swslot = uao_find_swslot(&aobj->u_obj, pageidx);
1126 
1127 		/*
1128  		 * just zero the page if there's nothing in swap.
1129  		 */
1130 
1131 		if (swslot == 0) {
1132 
1133 			/*
1134 			 * page hasn't existed before, just zero it.
1135 			 */
1136 
1137 			uvm_pagezero(ptmp);
1138 		} else {
1139 #if defined(VMSWAP)
1140 			int error;
1141 
1142 			UVMHIST_LOG(pdhist, "pagein from swslot %d",
1143 			     swslot, 0,0,0);
1144 
1145 			/*
1146 			 * page in the swapped-out page.
1147 			 * unlock object for i/o, relock when done.
1148 			 */
1149 
1150 			simple_unlock(&uobj->vmobjlock);
1151 			error = uvm_swap_get(ptmp, swslot, PGO_SYNCIO);
1152 			simple_lock(&uobj->vmobjlock);
1153 
1154 			/*
1155 			 * I/O done.  check for errors.
1156 			 */
1157 
1158 			if (error != 0) {
1159 				UVMHIST_LOG(pdhist, "<- done (error=%d)",
1160 				    error,0,0,0);
1161 				if (ptmp->flags & PG_WANTED)
1162 					wakeup(ptmp);
1163 
1164 				/*
1165 				 * remove the swap slot from the aobj
1166 				 * and mark the aobj as having no real slot.
1167 				 * don't free the swap slot, thus preventing
1168 				 * it from being used again.
1169 				 */
1170 
1171 				swslot = uao_set_swslot(&aobj->u_obj, pageidx,
1172 							SWSLOT_BAD);
1173 				if (swslot > 0) {
1174 					uvm_swap_markbad(swslot, 1);
1175 				}
1176 
1177 				uvm_lock_pageq();
1178 				uvm_pagefree(ptmp);
1179 				uvm_unlock_pageq();
1180 				simple_unlock(&uobj->vmobjlock);
1181 				return error;
1182 			}
1183 #else /* defined(VMSWAP) */
1184 			panic("%s: pagein", __func__);
1185 #endif /* defined(VMSWAP) */
1186 		}
1187 
1188 		if ((access_type & VM_PROT_WRITE) == 0) {
1189 			ptmp->flags |= PG_CLEAN;
1190 			pmap_clear_modify(ptmp);
1191 		}
1192 
1193 		/*
1194  		 * we got the page!   clear the fake flag (indicates valid
1195 		 * data now in page) and plug into our result array.   note
1196 		 * that page is still busy.
1197  		 *
1198  		 * it is the callers job to:
1199  		 * => check if the page is released
1200  		 * => unbusy the page
1201  		 * => activate the page
1202  		 */
1203 
1204 		ptmp->flags &= ~PG_FAKE;
1205 		pps[lcv] = ptmp;
1206 	}
1207 
1208 	/*
1209  	 * finally, unlock object and return.
1210  	 */
1211 
1212 done:
1213 	simple_unlock(&uobj->vmobjlock);
1214 	UVMHIST_LOG(pdhist, "<- done (OK)",0,0,0,0);
1215 	return 0;
1216 }
1217 
1218 #if defined(VMSWAP)
1219 
1220 /*
1221  * uao_dropswap:  release any swap resources from this aobj page.
1222  *
1223  * => aobj must be locked or have a reference count of 0.
1224  */
1225 
1226 void
1227 uao_dropswap(struct uvm_object *uobj, int pageidx)
1228 {
1229 	int slot;
1230 
1231 	slot = uao_set_swslot(uobj, pageidx, 0);
1232 	if (slot) {
1233 		uvm_swap_free(slot, 1);
1234 	}
1235 }
1236 
1237 /*
1238  * page in every page in every aobj that is paged-out to a range of swslots.
1239  *
1240  * => nothing should be locked.
1241  * => returns true if pagein was aborted due to lack of memory.
1242  */
1243 
1244 bool
1245 uao_swap_off(int startslot, int endslot)
1246 {
1247 	struct uvm_aobj *aobj, *nextaobj;
1248 	bool rv;
1249 
1250 	/*
1251 	 * walk the list of all aobjs.
1252 	 */
1253 
1254 restart:
1255 	mutex_enter(&uao_list_lock);
1256 	for (aobj = LIST_FIRST(&uao_list);
1257 	     aobj != NULL;
1258 	     aobj = nextaobj) {
1259 
1260 		/*
1261 		 * try to get the object lock, start all over if we fail.
1262 		 * most of the time we'll get the aobj lock,
1263 		 * so this should be a rare case.
1264 		 */
1265 
1266 		if (!simple_lock_try(&aobj->u_obj.vmobjlock)) {
1267 			mutex_exit(&uao_list_lock);
1268 			goto restart;
1269 		}
1270 
1271 		/*
1272 		 * add a ref to the aobj so it doesn't disappear
1273 		 * while we're working.
1274 		 */
1275 
1276 		uao_reference_locked(&aobj->u_obj);
1277 
1278 		/*
1279 		 * now it's safe to unlock the uao list.
1280 		 */
1281 
1282 		mutex_exit(&uao_list_lock);
1283 
1284 		/*
1285 		 * page in any pages in the swslot range.
1286 		 * if there's an error, abort and return the error.
1287 		 */
1288 
1289 		rv = uao_pagein(aobj, startslot, endslot);
1290 		if (rv) {
1291 			uao_detach_locked(&aobj->u_obj);
1292 			return rv;
1293 		}
1294 
1295 		/*
1296 		 * we're done with this aobj.
1297 		 * relock the list and drop our ref on the aobj.
1298 		 */
1299 
1300 		mutex_enter(&uao_list_lock);
1301 		nextaobj = LIST_NEXT(aobj, u_list);
1302 		uao_detach_locked(&aobj->u_obj);
1303 	}
1304 
1305 	/*
1306 	 * done with traversal, unlock the list
1307 	 */
1308 	mutex_exit(&uao_list_lock);
1309 	return false;
1310 }
1311 
1312 
1313 /*
1314  * page in any pages from aobj in the given range.
1315  *
1316  * => aobj must be locked and is returned locked.
1317  * => returns true if pagein was aborted due to lack of memory.
1318  */
1319 static bool
1320 uao_pagein(struct uvm_aobj *aobj, int startslot, int endslot)
1321 {
1322 	bool rv;
1323 
1324 	if (UAO_USES_SWHASH(aobj)) {
1325 		struct uao_swhash_elt *elt;
1326 		int buck;
1327 
1328 restart:
1329 		for (buck = aobj->u_swhashmask; buck >= 0; buck--) {
1330 			for (elt = LIST_FIRST(&aobj->u_swhash[buck]);
1331 			     elt != NULL;
1332 			     elt = LIST_NEXT(elt, list)) {
1333 				int i;
1334 
1335 				for (i = 0; i < UAO_SWHASH_CLUSTER_SIZE; i++) {
1336 					int slot = elt->slots[i];
1337 
1338 					/*
1339 					 * if the slot isn't in range, skip it.
1340 					 */
1341 
1342 					if (slot < startslot ||
1343 					    slot >= endslot) {
1344 						continue;
1345 					}
1346 
1347 					/*
1348 					 * process the page,
1349 					 * the start over on this object
1350 					 * since the swhash elt
1351 					 * may have been freed.
1352 					 */
1353 
1354 					rv = uao_pagein_page(aobj,
1355 					  UAO_SWHASH_ELT_PAGEIDX_BASE(elt) + i);
1356 					if (rv) {
1357 						return rv;
1358 					}
1359 					goto restart;
1360 				}
1361 			}
1362 		}
1363 	} else {
1364 		int i;
1365 
1366 		for (i = 0; i < aobj->u_pages; i++) {
1367 			int slot = aobj->u_swslots[i];
1368 
1369 			/*
1370 			 * if the slot isn't in range, skip it
1371 			 */
1372 
1373 			if (slot < startslot || slot >= endslot) {
1374 				continue;
1375 			}
1376 
1377 			/*
1378 			 * process the page.
1379 			 */
1380 
1381 			rv = uao_pagein_page(aobj, i);
1382 			if (rv) {
1383 				return rv;
1384 			}
1385 		}
1386 	}
1387 
1388 	return false;
1389 }
1390 
1391 /*
1392  * page in a page from an aobj.  used for swap_off.
1393  * returns true if pagein was aborted due to lack of memory.
1394  *
1395  * => aobj must be locked and is returned locked.
1396  */
1397 
1398 static bool
1399 uao_pagein_page(struct uvm_aobj *aobj, int pageidx)
1400 {
1401 	struct vm_page *pg;
1402 	int rv, npages;
1403 
1404 	pg = NULL;
1405 	npages = 1;
1406 	/* locked: aobj */
1407 	rv = uao_get(&aobj->u_obj, pageidx << PAGE_SHIFT,
1408 	    &pg, &npages, 0, VM_PROT_READ|VM_PROT_WRITE, 0, PGO_SYNCIO);
1409 	/* unlocked: aobj */
1410 
1411 	/*
1412 	 * relock and finish up.
1413 	 */
1414 
1415 	simple_lock(&aobj->u_obj.vmobjlock);
1416 	switch (rv) {
1417 	case 0:
1418 		break;
1419 
1420 	case EIO:
1421 	case ERESTART:
1422 
1423 		/*
1424 		 * nothing more to do on errors.
1425 		 * ERESTART can only mean that the anon was freed,
1426 		 * so again there's nothing to do.
1427 		 */
1428 
1429 		return false;
1430 
1431 	default:
1432 		return true;
1433 	}
1434 
1435 	/*
1436 	 * ok, we've got the page now.
1437 	 * mark it as dirty, clear its swslot and un-busy it.
1438 	 */
1439 	uao_dropswap(&aobj->u_obj, pageidx);
1440 
1441 	/*
1442 	 * make sure it's on a page queue.
1443 	 */
1444 	uvm_lock_pageq();
1445 	if (pg->wire_count == 0)
1446 		uvm_pageenqueue(pg);
1447 	uvm_unlock_pageq();
1448 
1449 	if (pg->flags & PG_WANTED) {
1450 		wakeup(pg);
1451 	}
1452 	pg->flags &= ~(PG_WANTED|PG_BUSY|PG_CLEAN|PG_FAKE);
1453 	UVM_PAGE_OWN(pg, NULL);
1454 
1455 	return false;
1456 }
1457 
1458 /*
1459  * uao_dropswap_range: drop swapslots in the range.
1460  *
1461  * => aobj must be locked and is returned locked.
1462  * => start is inclusive.  end is exclusive.
1463  */
1464 
1465 void
1466 uao_dropswap_range(struct uvm_object *uobj, voff_t start, voff_t end)
1467 {
1468 	struct uvm_aobj *aobj = (struct uvm_aobj *)uobj;
1469 
1470 	LOCK_ASSERT(simple_lock_held(&uobj->vmobjlock));
1471 
1472 	uao_dropswap_range1(aobj, start, end);
1473 }
1474 
1475 static void
1476 uao_dropswap_range1(struct uvm_aobj *aobj, voff_t start, voff_t end)
1477 {
1478 	int swpgonlydelta = 0;
1479 
1480 	if (end == 0) {
1481 		end = INT64_MAX;
1482 	}
1483 
1484 	if (UAO_USES_SWHASH(aobj)) {
1485 		int i, hashbuckets = aobj->u_swhashmask + 1;
1486 		voff_t taghi;
1487 		voff_t taglo;
1488 
1489 		taglo = UAO_SWHASH_ELT_TAG(start);
1490 		taghi = UAO_SWHASH_ELT_TAG(end);
1491 
1492 		for (i = 0; i < hashbuckets; i++) {
1493 			struct uao_swhash_elt *elt, *next;
1494 
1495 			for (elt = LIST_FIRST(&aobj->u_swhash[i]);
1496 			     elt != NULL;
1497 			     elt = next) {
1498 				int startidx, endidx;
1499 				int j;
1500 
1501 				next = LIST_NEXT(elt, list);
1502 
1503 				if (elt->tag < taglo || taghi < elt->tag) {
1504 					continue;
1505 				}
1506 
1507 				if (elt->tag == taglo) {
1508 					startidx =
1509 					    UAO_SWHASH_ELT_PAGESLOT_IDX(start);
1510 				} else {
1511 					startidx = 0;
1512 				}
1513 
1514 				if (elt->tag == taghi) {
1515 					endidx =
1516 					    UAO_SWHASH_ELT_PAGESLOT_IDX(end);
1517 				} else {
1518 					endidx = UAO_SWHASH_CLUSTER_SIZE;
1519 				}
1520 
1521 				for (j = startidx; j < endidx; j++) {
1522 					int slot = elt->slots[j];
1523 
1524 					KASSERT(uvm_pagelookup(&aobj->u_obj,
1525 					    (UAO_SWHASH_ELT_PAGEIDX_BASE(elt)
1526 					    + j) << PAGE_SHIFT) == NULL);
1527 					if (slot > 0) {
1528 						uvm_swap_free(slot, 1);
1529 						swpgonlydelta++;
1530 						KASSERT(elt->count > 0);
1531 						elt->slots[j] = 0;
1532 						elt->count--;
1533 					}
1534 				}
1535 
1536 				if (elt->count == 0) {
1537 					LIST_REMOVE(elt, list);
1538 					pool_put(&uao_swhash_elt_pool, elt);
1539 				}
1540 			}
1541 		}
1542 	} else {
1543 		int i;
1544 
1545 		if (aobj->u_pages < end) {
1546 			end = aobj->u_pages;
1547 		}
1548 		for (i = start; i < end; i++) {
1549 			int slot = aobj->u_swslots[i];
1550 
1551 			if (slot > 0) {
1552 				uvm_swap_free(slot, 1);
1553 				swpgonlydelta++;
1554 			}
1555 		}
1556 	}
1557 
1558 	/*
1559 	 * adjust the counter of pages only in swap for all
1560 	 * the swap slots we've freed.
1561 	 */
1562 
1563 	if (swpgonlydelta > 0) {
1564 		mutex_enter(&uvm_swap_data_lock);
1565 		KASSERT(uvmexp.swpgonly >= swpgonlydelta);
1566 		uvmexp.swpgonly -= swpgonlydelta;
1567 		mutex_exit(&uvm_swap_data_lock);
1568 	}
1569 }
1570 
1571 #endif /* defined(VMSWAP) */
1572