FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_object.c
1 /*-
2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
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 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94
33 *
34 *
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
37 *
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39 *
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
45 *
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49 *
50 * Carnegie Mellon requests users of this software to return to
51 *
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
56 *
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
59 */
60
61 /*
62 * Virtual memory object module.
63 */
64
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD: releng/6.1/sys/vm/vm_object.c 158179 2006-04-30 16:44:43Z cvs2svn $");
67
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/lock.h>
71 #include <sys/mman.h>
72 #include <sys/mount.h>
73 #include <sys/kernel.h>
74 #include <sys/sysctl.h>
75 #include <sys/mutex.h>
76 #include <sys/proc.h> /* for curproc, pageproc */
77 #include <sys/socket.h>
78 #include <sys/vnode.h>
79 #include <sys/vmmeter.h>
80 #include <sys/sx.h>
81
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pageout.h>
89 #include <vm/vm_pager.h>
90 #include <vm/swap_pager.h>
91 #include <vm/vm_kern.h>
92 #include <vm/vm_extern.h>
93 #include <vm/uma.h>
94
95 #define EASY_SCAN_FACTOR 8
96
97 #define MSYNC_FLUSH_HARDSEQ 0x01
98 #define MSYNC_FLUSH_SOFTSEQ 0x02
99
100 /*
101 * msync / VM object flushing optimizations
102 */
103 static int msync_flush_flags = MSYNC_FLUSH_HARDSEQ | MSYNC_FLUSH_SOFTSEQ;
104 SYSCTL_INT(_vm, OID_AUTO, msync_flush_flags,
105 CTLFLAG_RW, &msync_flush_flags, 0, "");
106
107 static int old_msync;
108 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
109 "Use old (insecure) msync behavior");
110
111 static void vm_object_qcollapse(vm_object_t object);
112 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int curgeneration, int pagerflags);
113
114 /*
115 * Virtual memory objects maintain the actual data
116 * associated with allocated virtual memory. A given
117 * page of memory exists within exactly one object.
118 *
119 * An object is only deallocated when all "references"
120 * are given up. Only one "reference" to a given
121 * region of an object should be writeable.
122 *
123 * Associated with each object is a list of all resident
124 * memory pages belonging to that object; this list is
125 * maintained by the "vm_page" module, and locked by the object's
126 * lock.
127 *
128 * Each object also records a "pager" routine which is
129 * used to retrieve (and store) pages to the proper backing
130 * storage. In addition, objects may be backed by other
131 * objects from which they were virtual-copied.
132 *
133 * The only items within the object structure which are
134 * modified after time of creation are:
135 * reference count locked by object's lock
136 * pager routine locked by object's lock
137 *
138 */
139
140 struct object_q vm_object_list;
141 struct mtx vm_object_list_mtx; /* lock for object list and count */
142
143 struct vm_object kernel_object_store;
144 struct vm_object kmem_object_store;
145
146 static long object_collapses;
147 static long object_bypasses;
148
149 /*
150 * next_index determines the page color that is assigned to the next
151 * allocated object. Accesses to next_index are not synchronized
152 * because the effects of two or more object allocations using
153 * next_index simultaneously are inconsequential. At any given time,
154 * numerous objects have the same page color.
155 */
156 static int next_index;
157
158 static uma_zone_t obj_zone;
159 #define VM_OBJECTS_INIT 256
160
161 static int vm_object_zinit(void *mem, int size, int flags);
162
163 #ifdef INVARIANTS
164 static void vm_object_zdtor(void *mem, int size, void *arg);
165
166 static void
167 vm_object_zdtor(void *mem, int size, void *arg)
168 {
169 vm_object_t object;
170
171 object = (vm_object_t)mem;
172 KASSERT(TAILQ_EMPTY(&object->memq),
173 ("object %p has resident pages",
174 object));
175 KASSERT(object->paging_in_progress == 0,
176 ("object %p paging_in_progress = %d",
177 object, object->paging_in_progress));
178 KASSERT(object->resident_page_count == 0,
179 ("object %p resident_page_count = %d",
180 object, object->resident_page_count));
181 KASSERT(object->shadow_count == 0,
182 ("object %p shadow_count = %d",
183 object, object->shadow_count));
184 }
185 #endif
186
187 static int
188 vm_object_zinit(void *mem, int size, int flags)
189 {
190 vm_object_t object;
191
192 object = (vm_object_t)mem;
193 bzero(&object->mtx, sizeof(object->mtx));
194 VM_OBJECT_LOCK_INIT(object, "standard object");
195
196 /* These are true for any object that has been freed */
197 object->paging_in_progress = 0;
198 object->resident_page_count = 0;
199 object->shadow_count = 0;
200 return (0);
201 }
202
203 void
204 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
205 {
206 int incr;
207
208 TAILQ_INIT(&object->memq);
209 LIST_INIT(&object->shadow_head);
210
211 object->root = NULL;
212 object->type = type;
213 object->size = size;
214 object->generation = 1;
215 object->ref_count = 1;
216 object->flags = 0;
217 if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
218 object->flags = OBJ_ONEMAPPING;
219 if (size > (PQ_L2_SIZE / 3 + PQ_PRIME1))
220 incr = PQ_L2_SIZE / 3 + PQ_PRIME1;
221 else
222 incr = size;
223 object->pg_color = next_index;
224 next_index = (object->pg_color + incr) & PQ_L2_MASK;
225 object->handle = NULL;
226 object->backing_object = NULL;
227 object->backing_object_offset = (vm_ooffset_t) 0;
228
229 mtx_lock(&vm_object_list_mtx);
230 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
231 mtx_unlock(&vm_object_list_mtx);
232 }
233
234 /*
235 * vm_object_init:
236 *
237 * Initialize the VM objects module.
238 */
239 void
240 vm_object_init(void)
241 {
242 TAILQ_INIT(&vm_object_list);
243 mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
244
245 VM_OBJECT_LOCK_INIT(&kernel_object_store, "kernel object");
246 _vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
247 kernel_object);
248
249 VM_OBJECT_LOCK_INIT(&kmem_object_store, "kmem object");
250 _vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
251 kmem_object);
252
253 /*
254 * The lock portion of struct vm_object must be type stable due
255 * to vm_pageout_fallback_object_lock locking a vm object
256 * without holding any references to it.
257 */
258 obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
259 #ifdef INVARIANTS
260 vm_object_zdtor,
261 #else
262 NULL,
263 #endif
264 vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM|UMA_ZONE_NOFREE);
265 uma_prealloc(obj_zone, VM_OBJECTS_INIT);
266 }
267
268 void
269 vm_object_clear_flag(vm_object_t object, u_short bits)
270 {
271
272 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
273 object->flags &= ~bits;
274 }
275
276 void
277 vm_object_pip_add(vm_object_t object, short i)
278 {
279
280 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
281 object->paging_in_progress += i;
282 }
283
284 void
285 vm_object_pip_subtract(vm_object_t object, short i)
286 {
287
288 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
289 object->paging_in_progress -= i;
290 }
291
292 void
293 vm_object_pip_wakeup(vm_object_t object)
294 {
295
296 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
297 object->paging_in_progress--;
298 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
299 vm_object_clear_flag(object, OBJ_PIPWNT);
300 wakeup(object);
301 }
302 }
303
304 void
305 vm_object_pip_wakeupn(vm_object_t object, short i)
306 {
307
308 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
309 if (i)
310 object->paging_in_progress -= i;
311 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
312 vm_object_clear_flag(object, OBJ_PIPWNT);
313 wakeup(object);
314 }
315 }
316
317 void
318 vm_object_pip_wait(vm_object_t object, char *waitid)
319 {
320
321 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
322 while (object->paging_in_progress) {
323 object->flags |= OBJ_PIPWNT;
324 msleep(object, VM_OBJECT_MTX(object), PVM, waitid, 0);
325 }
326 }
327
328 /*
329 * vm_object_allocate:
330 *
331 * Returns a new object with the given size.
332 */
333 vm_object_t
334 vm_object_allocate(objtype_t type, vm_pindex_t size)
335 {
336 vm_object_t object;
337
338 object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
339 _vm_object_allocate(type, size, object);
340 return (object);
341 }
342
343
344 /*
345 * vm_object_reference:
346 *
347 * Gets another reference to the given object. Note: OBJ_DEAD
348 * objects can be referenced during final cleaning.
349 */
350 void
351 vm_object_reference(vm_object_t object)
352 {
353 struct vnode *vp;
354
355 if (object == NULL)
356 return;
357 VM_OBJECT_LOCK(object);
358 object->ref_count++;
359 if (object->type == OBJT_VNODE) {
360 int vfslocked;
361
362 vp = object->handle;
363 VM_OBJECT_UNLOCK(object);
364 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
365 vget(vp, LK_RETRY, curthread);
366 VFS_UNLOCK_GIANT(vfslocked);
367 } else
368 VM_OBJECT_UNLOCK(object);
369 }
370
371 /*
372 * vm_object_reference_locked:
373 *
374 * Gets another reference to the given object.
375 *
376 * The object must be locked.
377 */
378 void
379 vm_object_reference_locked(vm_object_t object)
380 {
381 struct vnode *vp;
382
383 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
384 KASSERT((object->flags & OBJ_DEAD) == 0,
385 ("vm_object_reference_locked: dead object referenced"));
386 object->ref_count++;
387 if (object->type == OBJT_VNODE) {
388 vp = object->handle;
389 vref(vp);
390 }
391 }
392
393 /*
394 * Handle deallocating an object of type OBJT_VNODE.
395 */
396 void
397 vm_object_vndeallocate(vm_object_t object)
398 {
399 struct vnode *vp = (struct vnode *) object->handle;
400
401 VFS_ASSERT_GIANT(vp->v_mount);
402 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
403 KASSERT(object->type == OBJT_VNODE,
404 ("vm_object_vndeallocate: not a vnode object"));
405 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
406 #ifdef INVARIANTS
407 if (object->ref_count == 0) {
408 vprint("vm_object_vndeallocate", vp);
409 panic("vm_object_vndeallocate: bad object reference count");
410 }
411 #endif
412
413 object->ref_count--;
414 if (object->ref_count == 0) {
415 mp_fixme("Unlocked vflag access.");
416 vp->v_vflag &= ~VV_TEXT;
417 }
418 VM_OBJECT_UNLOCK(object);
419 /*
420 * vrele may need a vop lock
421 */
422 vrele(vp);
423 }
424
425 /*
426 * vm_object_deallocate:
427 *
428 * Release a reference to the specified object,
429 * gained either through a vm_object_allocate
430 * or a vm_object_reference call. When all references
431 * are gone, storage associated with this object
432 * may be relinquished.
433 *
434 * No object may be locked.
435 */
436 void
437 vm_object_deallocate(vm_object_t object)
438 {
439 vm_object_t temp;
440
441 while (object != NULL) {
442 int vfslocked;
443 /*
444 * In general, the object should be locked when working with
445 * its type. In this case, in order to maintain proper lock
446 * ordering, an exception is possible because a vnode-backed
447 * object never changes its type.
448 */
449 vfslocked = 0;
450 if (object->type == OBJT_VNODE) {
451 struct vnode *vp = (struct vnode *) object->handle;
452 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
453 }
454 VM_OBJECT_LOCK(object);
455 if (object->type == OBJT_VNODE) {
456 vm_object_vndeallocate(object);
457 VFS_UNLOCK_GIANT(vfslocked);
458 return;
459 }
460
461 KASSERT(object->ref_count != 0,
462 ("vm_object_deallocate: object deallocated too many times: %d", object->type));
463
464 /*
465 * If the reference count goes to 0 we start calling
466 * vm_object_terminate() on the object chain.
467 * A ref count of 1 may be a special case depending on the
468 * shadow count being 0 or 1.
469 */
470 object->ref_count--;
471 if (object->ref_count > 1) {
472 VM_OBJECT_UNLOCK(object);
473 return;
474 } else if (object->ref_count == 1) {
475 if (object->shadow_count == 0) {
476 vm_object_set_flag(object, OBJ_ONEMAPPING);
477 } else if ((object->shadow_count == 1) &&
478 (object->handle == NULL) &&
479 (object->type == OBJT_DEFAULT ||
480 object->type == OBJT_SWAP)) {
481 vm_object_t robject;
482
483 robject = LIST_FIRST(&object->shadow_head);
484 KASSERT(robject != NULL,
485 ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
486 object->ref_count,
487 object->shadow_count));
488 if (!VM_OBJECT_TRYLOCK(robject)) {
489 /*
490 * Avoid a potential deadlock.
491 */
492 object->ref_count++;
493 VM_OBJECT_UNLOCK(object);
494 /*
495 * More likely than not the thread
496 * holding robject's lock has lower
497 * priority than the current thread.
498 * Let the lower priority thread run.
499 */
500 tsleep(&proc0, PVM, "vmo_de", 1);
501 continue;
502 }
503 /*
504 * Collapse object into its shadow unless its
505 * shadow is dead. In that case, object will
506 * be deallocated by the thread that is
507 * deallocating its shadow.
508 */
509 if ((robject->flags & OBJ_DEAD) == 0 &&
510 (robject->handle == NULL) &&
511 (robject->type == OBJT_DEFAULT ||
512 robject->type == OBJT_SWAP)) {
513
514 robject->ref_count++;
515 retry:
516 if (robject->paging_in_progress) {
517 VM_OBJECT_UNLOCK(object);
518 vm_object_pip_wait(robject,
519 "objde1");
520 VM_OBJECT_LOCK(object);
521 goto retry;
522 } else if (object->paging_in_progress) {
523 VM_OBJECT_UNLOCK(robject);
524 object->flags |= OBJ_PIPWNT;
525 msleep(object,
526 VM_OBJECT_MTX(object),
527 PDROP | PVM, "objde2", 0);
528 VM_OBJECT_LOCK(robject);
529 VM_OBJECT_LOCK(object);
530 goto retry;
531 }
532 VM_OBJECT_UNLOCK(object);
533 if (robject->ref_count == 1) {
534 robject->ref_count--;
535 object = robject;
536 goto doterm;
537 }
538 object = robject;
539 vm_object_collapse(object);
540 VM_OBJECT_UNLOCK(object);
541 continue;
542 }
543 VM_OBJECT_UNLOCK(robject);
544 }
545 VM_OBJECT_UNLOCK(object);
546 return;
547 }
548 doterm:
549 temp = object->backing_object;
550 if (temp != NULL) {
551 VM_OBJECT_LOCK(temp);
552 LIST_REMOVE(object, shadow_list);
553 temp->shadow_count--;
554 temp->generation++;
555 VM_OBJECT_UNLOCK(temp);
556 object->backing_object = NULL;
557 }
558 /*
559 * Don't double-terminate, we could be in a termination
560 * recursion due to the terminate having to sync data
561 * to disk.
562 */
563 if ((object->flags & OBJ_DEAD) == 0)
564 vm_object_terminate(object);
565 else
566 VM_OBJECT_UNLOCK(object);
567 object = temp;
568 }
569 }
570
571 /*
572 * vm_object_terminate actually destroys the specified object, freeing
573 * up all previously used resources.
574 *
575 * The object must be locked.
576 * This routine may block.
577 */
578 void
579 vm_object_terminate(vm_object_t object)
580 {
581 vm_page_t p;
582
583 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
584
585 /*
586 * Make sure no one uses us.
587 */
588 vm_object_set_flag(object, OBJ_DEAD);
589
590 /*
591 * wait for the pageout daemon to be done with the object
592 */
593 vm_object_pip_wait(object, "objtrm");
594
595 KASSERT(!object->paging_in_progress,
596 ("vm_object_terminate: pageout in progress"));
597
598 /*
599 * Clean and free the pages, as appropriate. All references to the
600 * object are gone, so we don't need to lock it.
601 */
602 if (object->type == OBJT_VNODE) {
603 struct vnode *vp = (struct vnode *)object->handle;
604
605 /*
606 * Clean pages and flush buffers.
607 */
608 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
609 VM_OBJECT_UNLOCK(object);
610
611 vinvalbuf(vp, V_SAVE, NULL, 0, 0);
612
613 VM_OBJECT_LOCK(object);
614 }
615
616 KASSERT(object->ref_count == 0,
617 ("vm_object_terminate: object with references, ref_count=%d",
618 object->ref_count));
619
620 /*
621 * Now free any remaining pages. For internal objects, this also
622 * removes them from paging queues. Don't free wired pages, just
623 * remove them from the object.
624 */
625 vm_page_lock_queues();
626 while ((p = TAILQ_FIRST(&object->memq)) != NULL) {
627 KASSERT(!p->busy && (p->flags & PG_BUSY) == 0,
628 ("vm_object_terminate: freeing busy page %p "
629 "p->busy = %d, p->flags %x\n", p, p->busy, p->flags));
630 if (p->wire_count == 0) {
631 vm_page_free(p);
632 cnt.v_pfree++;
633 } else {
634 vm_page_remove(p);
635 }
636 }
637 vm_page_unlock_queues();
638
639 /*
640 * Let the pager know object is dead.
641 */
642 vm_pager_deallocate(object);
643 VM_OBJECT_UNLOCK(object);
644
645 /*
646 * Remove the object from the global object list.
647 */
648 mtx_lock(&vm_object_list_mtx);
649 TAILQ_REMOVE(&vm_object_list, object, object_list);
650 mtx_unlock(&vm_object_list_mtx);
651
652 /*
653 * Free the space for the object.
654 */
655 uma_zfree(obj_zone, object);
656 }
657
658 /*
659 * vm_object_page_clean
660 *
661 * Clean all dirty pages in the specified range of object. Leaves page
662 * on whatever queue it is currently on. If NOSYNC is set then do not
663 * write out pages with PG_NOSYNC set (originally comes from MAP_NOSYNC),
664 * leaving the object dirty.
665 *
666 * When stuffing pages asynchronously, allow clustering. XXX we need a
667 * synchronous clustering mode implementation.
668 *
669 * Odd semantics: if start == end, we clean everything.
670 *
671 * The object must be locked.
672 */
673 void
674 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end, int flags)
675 {
676 vm_page_t p, np;
677 vm_pindex_t tstart, tend;
678 vm_pindex_t pi;
679 int clearobjflags;
680 int pagerflags;
681 int curgeneration;
682
683 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
684 if (object->type != OBJT_VNODE ||
685 (object->flags & OBJ_MIGHTBEDIRTY) == 0)
686 return;
687
688 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ? VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
689 pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
690
691 vm_object_set_flag(object, OBJ_CLEANING);
692
693 tstart = start;
694 if (end == 0) {
695 tend = object->size;
696 } else {
697 tend = end;
698 }
699
700 vm_page_lock_queues();
701 /*
702 * If the caller is smart and only msync()s a range he knows is
703 * dirty, we may be able to avoid an object scan. This results in
704 * a phenominal improvement in performance. We cannot do this
705 * as a matter of course because the object may be huge - e.g.
706 * the size might be in the gigabytes or terrabytes.
707 */
708 if (msync_flush_flags & MSYNC_FLUSH_HARDSEQ) {
709 vm_pindex_t tscan;
710 int scanlimit;
711 int scanreset;
712
713 scanreset = object->resident_page_count / EASY_SCAN_FACTOR;
714 if (scanreset < 16)
715 scanreset = 16;
716 pagerflags |= VM_PAGER_IGNORE_CLEANCHK;
717
718 scanlimit = scanreset;
719 tscan = tstart;
720 while (tscan < tend) {
721 curgeneration = object->generation;
722 p = vm_page_lookup(object, tscan);
723 if (p == NULL || p->valid == 0 ||
724 (p->queue - p->pc) == PQ_CACHE) {
725 if (--scanlimit == 0)
726 break;
727 ++tscan;
728 continue;
729 }
730 vm_page_test_dirty(p);
731 if ((p->dirty & p->valid) == 0) {
732 if (--scanlimit == 0)
733 break;
734 ++tscan;
735 continue;
736 }
737 /*
738 * If we have been asked to skip nosync pages and
739 * this is a nosync page, we can't continue.
740 */
741 if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
742 if (--scanlimit == 0)
743 break;
744 ++tscan;
745 continue;
746 }
747 scanlimit = scanreset;
748
749 /*
750 * This returns 0 if it was unable to busy the first
751 * page (i.e. had to sleep).
752 */
753 tscan += vm_object_page_collect_flush(object, p, curgeneration, pagerflags);
754 }
755
756 /*
757 * If everything was dirty and we flushed it successfully,
758 * and the requested range is not the entire object, we
759 * don't have to mess with CLEANCHK or MIGHTBEDIRTY and can
760 * return immediately.
761 */
762 if (tscan >= tend && (tstart || tend < object->size)) {
763 vm_page_unlock_queues();
764 vm_object_clear_flag(object, OBJ_CLEANING);
765 return;
766 }
767 pagerflags &= ~VM_PAGER_IGNORE_CLEANCHK;
768 }
769
770 /*
771 * Generally set CLEANCHK interlock and make the page read-only so
772 * we can then clear the object flags.
773 *
774 * However, if this is a nosync mmap then the object is likely to
775 * stay dirty so do not mess with the page and do not clear the
776 * object flags.
777 */
778 clearobjflags = 1;
779 TAILQ_FOREACH(p, &object->memq, listq) {
780 vm_page_flag_set(p, PG_CLEANCHK);
781 if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC))
782 clearobjflags = 0;
783 else
784 pmap_page_protect(p, VM_PROT_READ);
785 }
786
787 if (clearobjflags && (tstart == 0) && (tend == object->size)) {
788 struct vnode *vp;
789
790 vm_object_clear_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
791 if (object->type == OBJT_VNODE &&
792 (vp = (struct vnode *)object->handle) != NULL) {
793 VI_LOCK(vp);
794 if (vp->v_iflag & VI_OBJDIRTY)
795 vp->v_iflag &= ~VI_OBJDIRTY;
796 VI_UNLOCK(vp);
797 }
798 }
799
800 rescan:
801 curgeneration = object->generation;
802
803 for (p = TAILQ_FIRST(&object->memq); p; p = np) {
804 int n;
805
806 np = TAILQ_NEXT(p, listq);
807
808 again:
809 pi = p->pindex;
810 if (((p->flags & PG_CLEANCHK) == 0) ||
811 (pi < tstart) || (pi >= tend) ||
812 (p->valid == 0) ||
813 ((p->queue - p->pc) == PQ_CACHE)) {
814 vm_page_flag_clear(p, PG_CLEANCHK);
815 continue;
816 }
817
818 vm_page_test_dirty(p);
819 if ((p->dirty & p->valid) == 0) {
820 vm_page_flag_clear(p, PG_CLEANCHK);
821 continue;
822 }
823
824 /*
825 * If we have been asked to skip nosync pages and this is a
826 * nosync page, skip it. Note that the object flags were
827 * not cleared in this case so we do not have to set them.
828 */
829 if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
830 vm_page_flag_clear(p, PG_CLEANCHK);
831 continue;
832 }
833
834 n = vm_object_page_collect_flush(object, p,
835 curgeneration, pagerflags);
836 if (n == 0)
837 goto rescan;
838
839 if (object->generation != curgeneration)
840 goto rescan;
841
842 /*
843 * Try to optimize the next page. If we can't we pick up
844 * our (random) scan where we left off.
845 */
846 if (msync_flush_flags & MSYNC_FLUSH_SOFTSEQ) {
847 if ((p = vm_page_lookup(object, pi + n)) != NULL)
848 goto again;
849 }
850 }
851 vm_page_unlock_queues();
852 #if 0
853 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC)?MNT_WAIT:0, curproc);
854 #endif
855
856 vm_object_clear_flag(object, OBJ_CLEANING);
857 return;
858 }
859
860 static int
861 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int curgeneration, int pagerflags)
862 {
863 int runlen;
864 int maxf;
865 int chkb;
866 int maxb;
867 int i;
868 vm_pindex_t pi;
869 vm_page_t maf[vm_pageout_page_count];
870 vm_page_t mab[vm_pageout_page_count];
871 vm_page_t ma[vm_pageout_page_count];
872
873 mtx_assert(&vm_page_queue_mtx, MA_OWNED);
874 pi = p->pindex;
875 while (vm_page_sleep_if_busy(p, TRUE, "vpcwai")) {
876 vm_page_lock_queues();
877 if (object->generation != curgeneration) {
878 return(0);
879 }
880 }
881 maxf = 0;
882 for(i = 1; i < vm_pageout_page_count; i++) {
883 vm_page_t tp;
884
885 if ((tp = vm_page_lookup(object, pi + i)) != NULL) {
886 if ((tp->flags & PG_BUSY) ||
887 ((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
888 (tp->flags & PG_CLEANCHK) == 0) ||
889 (tp->busy != 0))
890 break;
891 if((tp->queue - tp->pc) == PQ_CACHE) {
892 vm_page_flag_clear(tp, PG_CLEANCHK);
893 break;
894 }
895 vm_page_test_dirty(tp);
896 if ((tp->dirty & tp->valid) == 0) {
897 vm_page_flag_clear(tp, PG_CLEANCHK);
898 break;
899 }
900 maf[ i - 1 ] = tp;
901 maxf++;
902 continue;
903 }
904 break;
905 }
906
907 maxb = 0;
908 chkb = vm_pageout_page_count - maxf;
909 if (chkb) {
910 for(i = 1; i < chkb;i++) {
911 vm_page_t tp;
912
913 if ((tp = vm_page_lookup(object, pi - i)) != NULL) {
914 if ((tp->flags & PG_BUSY) ||
915 ((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
916 (tp->flags & PG_CLEANCHK) == 0) ||
917 (tp->busy != 0))
918 break;
919 if ((tp->queue - tp->pc) == PQ_CACHE) {
920 vm_page_flag_clear(tp, PG_CLEANCHK);
921 break;
922 }
923 vm_page_test_dirty(tp);
924 if ((tp->dirty & tp->valid) == 0) {
925 vm_page_flag_clear(tp, PG_CLEANCHK);
926 break;
927 }
928 mab[ i - 1 ] = tp;
929 maxb++;
930 continue;
931 }
932 break;
933 }
934 }
935
936 for(i = 0; i < maxb; i++) {
937 int index = (maxb - i) - 1;
938 ma[index] = mab[i];
939 vm_page_flag_clear(ma[index], PG_CLEANCHK);
940 }
941 vm_page_flag_clear(p, PG_CLEANCHK);
942 ma[maxb] = p;
943 for(i = 0; i < maxf; i++) {
944 int index = (maxb + i) + 1;
945 ma[index] = maf[i];
946 vm_page_flag_clear(ma[index], PG_CLEANCHK);
947 }
948 runlen = maxb + maxf + 1;
949
950 vm_pageout_flush(ma, runlen, pagerflags);
951 for (i = 0; i < runlen; i++) {
952 if (ma[i]->valid & ma[i]->dirty) {
953 pmap_page_protect(ma[i], VM_PROT_READ);
954 vm_page_flag_set(ma[i], PG_CLEANCHK);
955
956 /*
957 * maxf will end up being the actual number of pages
958 * we wrote out contiguously, non-inclusive of the
959 * first page. We do not count look-behind pages.
960 */
961 if (i >= maxb + 1 && (maxf > i - maxb - 1))
962 maxf = i - maxb - 1;
963 }
964 }
965 return(maxf + 1);
966 }
967
968 /*
969 * Note that there is absolutely no sense in writing out
970 * anonymous objects, so we track down the vnode object
971 * to write out.
972 * We invalidate (remove) all pages from the address space
973 * for semantic correctness.
974 *
975 * Note: certain anonymous maps, such as MAP_NOSYNC maps,
976 * may start out with a NULL object.
977 */
978 void
979 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
980 boolean_t syncio, boolean_t invalidate)
981 {
982 vm_object_t backing_object;
983 struct vnode *vp;
984 struct mount *mp;
985 int flags;
986
987 if (object == NULL)
988 return;
989 VM_OBJECT_LOCK(object);
990 while ((backing_object = object->backing_object) != NULL) {
991 VM_OBJECT_LOCK(backing_object);
992 offset += object->backing_object_offset;
993 VM_OBJECT_UNLOCK(object);
994 object = backing_object;
995 if (object->size < OFF_TO_IDX(offset + size))
996 size = IDX_TO_OFF(object->size) - offset;
997 }
998 /*
999 * Flush pages if writing is allowed, invalidate them
1000 * if invalidation requested. Pages undergoing I/O
1001 * will be ignored by vm_object_page_remove().
1002 *
1003 * We cannot lock the vnode and then wait for paging
1004 * to complete without deadlocking against vm_fault.
1005 * Instead we simply call vm_object_page_remove() and
1006 * allow it to block internally on a page-by-page
1007 * basis when it encounters pages undergoing async
1008 * I/O.
1009 */
1010 if (object->type == OBJT_VNODE &&
1011 (object->flags & OBJ_MIGHTBEDIRTY) != 0) {
1012 int vfslocked;
1013 vp = object->handle;
1014 VM_OBJECT_UNLOCK(object);
1015 (void) vn_start_write(vp, &mp, V_WAIT);
1016 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1017 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
1018 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1019 flags |= invalidate ? OBJPC_INVAL : 0;
1020 VM_OBJECT_LOCK(object);
1021 vm_object_page_clean(object,
1022 OFF_TO_IDX(offset),
1023 OFF_TO_IDX(offset + size + PAGE_MASK),
1024 flags);
1025 VM_OBJECT_UNLOCK(object);
1026 VOP_UNLOCK(vp, 0, curthread);
1027 VFS_UNLOCK_GIANT(vfslocked);
1028 vn_finished_write(mp);
1029 VM_OBJECT_LOCK(object);
1030 }
1031 if ((object->type == OBJT_VNODE ||
1032 object->type == OBJT_DEVICE) && invalidate) {
1033 boolean_t purge;
1034 purge = old_msync || (object->type == OBJT_DEVICE);
1035 vm_object_page_remove(object,
1036 OFF_TO_IDX(offset),
1037 OFF_TO_IDX(offset + size + PAGE_MASK),
1038 purge ? FALSE : TRUE);
1039 }
1040 VM_OBJECT_UNLOCK(object);
1041 }
1042
1043 /*
1044 * vm_object_madvise:
1045 *
1046 * Implements the madvise function at the object/page level.
1047 *
1048 * MADV_WILLNEED (any object)
1049 *
1050 * Activate the specified pages if they are resident.
1051 *
1052 * MADV_DONTNEED (any object)
1053 *
1054 * Deactivate the specified pages if they are resident.
1055 *
1056 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects,
1057 * OBJ_ONEMAPPING only)
1058 *
1059 * Deactivate and clean the specified pages if they are
1060 * resident. This permits the process to reuse the pages
1061 * without faulting or the kernel to reclaim the pages
1062 * without I/O.
1063 */
1064 void
1065 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
1066 {
1067 vm_pindex_t end, tpindex;
1068 vm_object_t backing_object, tobject;
1069 vm_page_t m;
1070
1071 if (object == NULL)
1072 return;
1073 VM_OBJECT_LOCK(object);
1074 end = pindex + count;
1075 /*
1076 * Locate and adjust resident pages
1077 */
1078 for (; pindex < end; pindex += 1) {
1079 relookup:
1080 tobject = object;
1081 tpindex = pindex;
1082 shadowlookup:
1083 /*
1084 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1085 * and those pages must be OBJ_ONEMAPPING.
1086 */
1087 if (advise == MADV_FREE) {
1088 if ((tobject->type != OBJT_DEFAULT &&
1089 tobject->type != OBJT_SWAP) ||
1090 (tobject->flags & OBJ_ONEMAPPING) == 0) {
1091 goto unlock_tobject;
1092 }
1093 }
1094 m = vm_page_lookup(tobject, tpindex);
1095 if (m == NULL) {
1096 /*
1097 * There may be swap even if there is no backing page
1098 */
1099 if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1100 swap_pager_freespace(tobject, tpindex, 1);
1101 /*
1102 * next object
1103 */
1104 backing_object = tobject->backing_object;
1105 if (backing_object == NULL)
1106 goto unlock_tobject;
1107 VM_OBJECT_LOCK(backing_object);
1108 tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1109 if (tobject != object)
1110 VM_OBJECT_UNLOCK(tobject);
1111 tobject = backing_object;
1112 goto shadowlookup;
1113 }
1114 /*
1115 * If the page is busy or not in a normal active state,
1116 * we skip it. If the page is not managed there are no
1117 * page queues to mess with. Things can break if we mess
1118 * with pages in any of the below states.
1119 */
1120 vm_page_lock_queues();
1121 if (m->hold_count ||
1122 m->wire_count ||
1123 (m->flags & PG_UNMANAGED) ||
1124 m->valid != VM_PAGE_BITS_ALL) {
1125 vm_page_unlock_queues();
1126 goto unlock_tobject;
1127 }
1128 if ((m->flags & PG_BUSY) || m->busy) {
1129 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
1130 if (object != tobject)
1131 VM_OBJECT_UNLOCK(object);
1132 VM_OBJECT_UNLOCK(tobject);
1133 msleep(m, &vm_page_queue_mtx, PDROP | PVM, "madvpo", 0);
1134 VM_OBJECT_LOCK(object);
1135 goto relookup;
1136 }
1137 if (advise == MADV_WILLNEED) {
1138 vm_page_activate(m);
1139 } else if (advise == MADV_DONTNEED) {
1140 vm_page_dontneed(m);
1141 } else if (advise == MADV_FREE) {
1142 /*
1143 * Mark the page clean. This will allow the page
1144 * to be freed up by the system. However, such pages
1145 * are often reused quickly by malloc()/free()
1146 * so we do not do anything that would cause
1147 * a page fault if we can help it.
1148 *
1149 * Specifically, we do not try to actually free
1150 * the page now nor do we try to put it in the
1151 * cache (which would cause a page fault on reuse).
1152 *
1153 * But we do make the page is freeable as we
1154 * can without actually taking the step of unmapping
1155 * it.
1156 */
1157 pmap_clear_modify(m);
1158 m->dirty = 0;
1159 m->act_count = 0;
1160 vm_page_dontneed(m);
1161 }
1162 vm_page_unlock_queues();
1163 if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1164 swap_pager_freespace(tobject, tpindex, 1);
1165 unlock_tobject:
1166 if (tobject != object)
1167 VM_OBJECT_UNLOCK(tobject);
1168 }
1169 VM_OBJECT_UNLOCK(object);
1170 }
1171
1172 /*
1173 * vm_object_shadow:
1174 *
1175 * Create a new object which is backed by the
1176 * specified existing object range. The source
1177 * object reference is deallocated.
1178 *
1179 * The new object and offset into that object
1180 * are returned in the source parameters.
1181 */
1182 void
1183 vm_object_shadow(
1184 vm_object_t *object, /* IN/OUT */
1185 vm_ooffset_t *offset, /* IN/OUT */
1186 vm_size_t length)
1187 {
1188 vm_object_t source;
1189 vm_object_t result;
1190
1191 source = *object;
1192
1193 /*
1194 * Don't create the new object if the old object isn't shared.
1195 */
1196 if (source != NULL) {
1197 VM_OBJECT_LOCK(source);
1198 if (source->ref_count == 1 &&
1199 source->handle == NULL &&
1200 (source->type == OBJT_DEFAULT ||
1201 source->type == OBJT_SWAP)) {
1202 VM_OBJECT_UNLOCK(source);
1203 return;
1204 }
1205 VM_OBJECT_UNLOCK(source);
1206 }
1207
1208 /*
1209 * Allocate a new object with the given length.
1210 */
1211 result = vm_object_allocate(OBJT_DEFAULT, length);
1212
1213 /*
1214 * The new object shadows the source object, adding a reference to it.
1215 * Our caller changes his reference to point to the new object,
1216 * removing a reference to the source object. Net result: no change
1217 * of reference count.
1218 *
1219 * Try to optimize the result object's page color when shadowing
1220 * in order to maintain page coloring consistency in the combined
1221 * shadowed object.
1222 */
1223 result->backing_object = source;
1224 /*
1225 * Store the offset into the source object, and fix up the offset into
1226 * the new object.
1227 */
1228 result->backing_object_offset = *offset;
1229 if (source != NULL) {
1230 VM_OBJECT_LOCK(source);
1231 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1232 source->shadow_count++;
1233 source->generation++;
1234 if (length < source->size)
1235 length = source->size;
1236 if (length > PQ_L2_SIZE / 3 + PQ_PRIME1 ||
1237 source->generation > 1)
1238 length = PQ_L2_SIZE / 3 + PQ_PRIME1;
1239 result->pg_color = (source->pg_color +
1240 length * source->generation) & PQ_L2_MASK;
1241 result->flags |= source->flags & OBJ_NEEDGIANT;
1242 VM_OBJECT_UNLOCK(source);
1243 next_index = (result->pg_color + PQ_L2_SIZE / 3 + PQ_PRIME1) &
1244 PQ_L2_MASK;
1245 }
1246
1247
1248 /*
1249 * Return the new things
1250 */
1251 *offset = 0;
1252 *object = result;
1253 }
1254
1255 /*
1256 * vm_object_split:
1257 *
1258 * Split the pages in a map entry into a new object. This affords
1259 * easier removal of unused pages, and keeps object inheritance from
1260 * being a negative impact on memory usage.
1261 */
1262 void
1263 vm_object_split(vm_map_entry_t entry)
1264 {
1265 vm_page_t m;
1266 vm_object_t orig_object, new_object, source;
1267 vm_pindex_t offidxstart, offidxend;
1268 vm_size_t idx, size;
1269
1270 orig_object = entry->object.vm_object;
1271 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1272 return;
1273 if (orig_object->ref_count <= 1)
1274 return;
1275 VM_OBJECT_UNLOCK(orig_object);
1276
1277 offidxstart = OFF_TO_IDX(entry->offset);
1278 offidxend = offidxstart + OFF_TO_IDX(entry->end - entry->start);
1279 size = offidxend - offidxstart;
1280
1281 /*
1282 * If swap_pager_copy() is later called, it will convert new_object
1283 * into a swap object.
1284 */
1285 new_object = vm_object_allocate(OBJT_DEFAULT, size);
1286
1287 VM_OBJECT_LOCK(new_object);
1288 VM_OBJECT_LOCK(orig_object);
1289 source = orig_object->backing_object;
1290 if (source != NULL) {
1291 VM_OBJECT_LOCK(source);
1292 LIST_INSERT_HEAD(&source->shadow_head,
1293 new_object, shadow_list);
1294 source->shadow_count++;
1295 source->generation++;
1296 vm_object_reference_locked(source); /* for new_object */
1297 vm_object_clear_flag(source, OBJ_ONEMAPPING);
1298 VM_OBJECT_UNLOCK(source);
1299 new_object->backing_object_offset =
1300 orig_object->backing_object_offset + entry->offset;
1301 new_object->backing_object = source;
1302 }
1303 new_object->flags |= orig_object->flags & OBJ_NEEDGIANT;
1304 vm_page_lock_queues();
1305 for (idx = 0; idx < size; idx++) {
1306 retry:
1307 m = vm_page_lookup(orig_object, offidxstart + idx);
1308 if (m == NULL)
1309 continue;
1310
1311 /*
1312 * We must wait for pending I/O to complete before we can
1313 * rename the page.
1314 *
1315 * We do not have to VM_PROT_NONE the page as mappings should
1316 * not be changed by this operation.
1317 */
1318 if ((m->flags & PG_BUSY) || m->busy) {
1319 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
1320 VM_OBJECT_UNLOCK(orig_object);
1321 VM_OBJECT_UNLOCK(new_object);
1322 msleep(m, &vm_page_queue_mtx, PDROP | PVM, "spltwt", 0);
1323 VM_OBJECT_LOCK(new_object);
1324 VM_OBJECT_LOCK(orig_object);
1325 vm_page_lock_queues();
1326 goto retry;
1327 }
1328 vm_page_rename(m, new_object, idx);
1329 /* page automatically made dirty by rename and cache handled */
1330 vm_page_busy(m);
1331 }
1332 vm_page_unlock_queues();
1333 if (orig_object->type == OBJT_SWAP) {
1334 /*
1335 * swap_pager_copy() can sleep, in which case the orig_object's
1336 * and new_object's locks are released and reacquired.
1337 */
1338 swap_pager_copy(orig_object, new_object, offidxstart, 0);
1339 }
1340 VM_OBJECT_UNLOCK(orig_object);
1341 vm_page_lock_queues();
1342 TAILQ_FOREACH(m, &new_object->memq, listq)
1343 vm_page_wakeup(m);
1344 vm_page_unlock_queues();
1345 VM_OBJECT_UNLOCK(new_object);
1346 entry->object.vm_object = new_object;
1347 entry->offset = 0LL;
1348 vm_object_deallocate(orig_object);
1349 VM_OBJECT_LOCK(new_object);
1350 }
1351
1352 #define OBSC_TEST_ALL_SHADOWED 0x0001
1353 #define OBSC_COLLAPSE_NOWAIT 0x0002
1354 #define OBSC_COLLAPSE_WAIT 0x0004
1355
1356 static int
1357 vm_object_backing_scan(vm_object_t object, int op)
1358 {
1359 int r = 1;
1360 vm_page_t p;
1361 vm_object_t backing_object;
1362 vm_pindex_t backing_offset_index;
1363
1364 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1365 VM_OBJECT_LOCK_ASSERT(object->backing_object, MA_OWNED);
1366
1367 backing_object = object->backing_object;
1368 backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1369
1370 /*
1371 * Initial conditions
1372 */
1373 if (op & OBSC_TEST_ALL_SHADOWED) {
1374 /*
1375 * We do not want to have to test for the existence of
1376 * swap pages in the backing object. XXX but with the
1377 * new swapper this would be pretty easy to do.
1378 *
1379 * XXX what about anonymous MAP_SHARED memory that hasn't
1380 * been ZFOD faulted yet? If we do not test for this, the
1381 * shadow test may succeed! XXX
1382 */
1383 if (backing_object->type != OBJT_DEFAULT) {
1384 return (0);
1385 }
1386 }
1387 if (op & OBSC_COLLAPSE_WAIT) {
1388 vm_object_set_flag(backing_object, OBJ_DEAD);
1389 }
1390
1391 /*
1392 * Our scan
1393 */
1394 p = TAILQ_FIRST(&backing_object->memq);
1395 while (p) {
1396 vm_page_t next = TAILQ_NEXT(p, listq);
1397 vm_pindex_t new_pindex = p->pindex - backing_offset_index;
1398
1399 if (op & OBSC_TEST_ALL_SHADOWED) {
1400 vm_page_t pp;
1401
1402 /*
1403 * Ignore pages outside the parent object's range
1404 * and outside the parent object's mapping of the
1405 * backing object.
1406 *
1407 * note that we do not busy the backing object's
1408 * page.
1409 */
1410 if (
1411 p->pindex < backing_offset_index ||
1412 new_pindex >= object->size
1413 ) {
1414 p = next;
1415 continue;
1416 }
1417
1418 /*
1419 * See if the parent has the page or if the parent's
1420 * object pager has the page. If the parent has the
1421 * page but the page is not valid, the parent's
1422 * object pager must have the page.
1423 *
1424 * If this fails, the parent does not completely shadow
1425 * the object and we might as well give up now.
1426 */
1427
1428 pp = vm_page_lookup(object, new_pindex);
1429 if (
1430 (pp == NULL || pp->valid == 0) &&
1431 !vm_pager_has_page(object, new_pindex, NULL, NULL)
1432 ) {
1433 r = 0;
1434 break;
1435 }
1436 }
1437
1438 /*
1439 * Check for busy page
1440 */
1441 if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1442 vm_page_t pp;
1443
1444 if (op & OBSC_COLLAPSE_NOWAIT) {
1445 if ((p->flags & PG_BUSY) ||
1446 !p->valid ||
1447 p->busy) {
1448 p = next;
1449 continue;
1450 }
1451 } else if (op & OBSC_COLLAPSE_WAIT) {
1452 if ((p->flags & PG_BUSY) || p->busy) {
1453 vm_page_lock_queues();
1454 vm_page_flag_set(p,
1455 PG_WANTED | PG_REFERENCED);
1456 VM_OBJECT_UNLOCK(backing_object);
1457 VM_OBJECT_UNLOCK(object);
1458 msleep(p, &vm_page_queue_mtx,
1459 PDROP | PVM, "vmocol", 0);
1460 VM_OBJECT_LOCK(object);
1461 VM_OBJECT_LOCK(backing_object);
1462 /*
1463 * If we slept, anything could have
1464 * happened. Since the object is
1465 * marked dead, the backing offset
1466 * should not have changed so we
1467 * just restart our scan.
1468 */
1469 p = TAILQ_FIRST(&backing_object->memq);
1470 continue;
1471 }
1472 }
1473
1474 KASSERT(
1475 p->object == backing_object,
1476 ("vm_object_backing_scan: object mismatch")
1477 );
1478
1479 /*
1480 * Destroy any associated swap
1481 */
1482 if (backing_object->type == OBJT_SWAP) {
1483 swap_pager_freespace(
1484 backing_object,
1485 p->pindex,
1486 1
1487 );
1488 }
1489
1490 if (
1491 p->pindex < backing_offset_index ||
1492 new_pindex >= object->size
1493 ) {
1494 /*
1495 * Page is out of the parent object's range, we
1496 * can simply destroy it.
1497 */
1498 vm_page_lock_queues();
1499 KASSERT(!pmap_page_is_mapped(p),
1500 ("freeing mapped page %p", p));
1501 if (p->wire_count == 0)
1502 vm_page_free(p);
1503 else
1504 vm_page_remove(p);
1505 vm_page_unlock_queues();
1506 p = next;
1507 continue;
1508 }
1509
1510 pp = vm_page_lookup(object, new_pindex);
1511 if (
1512 pp != NULL ||
1513 vm_pager_has_page(object, new_pindex, NULL, NULL)
1514 ) {
1515 /*
1516 * page already exists in parent OR swap exists
1517 * for this location in the parent. Destroy
1518 * the original page from the backing object.
1519 *
1520 * Leave the parent's page alone
1521 */
1522 vm_page_lock_queues();
1523 KASSERT(!pmap_page_is_mapped(p),
1524 ("freeing mapped page %p", p));
1525 if (p->wire_count == 0)
1526 vm_page_free(p);
1527 else
1528 vm_page_remove(p);
1529 vm_page_unlock_queues();
1530 p = next;
1531 continue;
1532 }
1533
1534 /*
1535 * Page does not exist in parent, rename the
1536 * page from the backing object to the main object.
1537 *
1538 * If the page was mapped to a process, it can remain
1539 * mapped through the rename.
1540 */
1541 vm_page_lock_queues();
1542 vm_page_rename(p, object, new_pindex);
1543 vm_page_unlock_queues();
1544 /* page automatically made dirty by rename */
1545 }
1546 p = next;
1547 }
1548 return (r);
1549 }
1550
1551
1552 /*
1553 * this version of collapse allows the operation to occur earlier and
1554 * when paging_in_progress is true for an object... This is not a complete
1555 * operation, but should plug 99.9% of the rest of the leaks.
1556 */
1557 static void
1558 vm_object_qcollapse(vm_object_t object)
1559 {
1560 vm_object_t backing_object = object->backing_object;
1561
1562 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1563 VM_OBJECT_LOCK_ASSERT(backing_object, MA_OWNED);
1564
1565 if (backing_object->ref_count != 1)
1566 return;
1567
1568 vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1569 }
1570
1571 /*
1572 * vm_object_collapse:
1573 *
1574 * Collapse an object with the object backing it.
1575 * Pages in the backing object are moved into the
1576 * parent, and the backing object is deallocated.
1577 */
1578 void
1579 vm_object_collapse(vm_object_t object)
1580 {
1581 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1582
1583 while (TRUE) {
1584 vm_object_t backing_object;
1585
1586 /*
1587 * Verify that the conditions are right for collapse:
1588 *
1589 * The object exists and the backing object exists.
1590 */
1591 if ((backing_object = object->backing_object) == NULL)
1592 break;
1593
1594 /*
1595 * we check the backing object first, because it is most likely
1596 * not collapsable.
1597 */
1598 VM_OBJECT_LOCK(backing_object);
1599 if (backing_object->handle != NULL ||
1600 (backing_object->type != OBJT_DEFAULT &&
1601 backing_object->type != OBJT_SWAP) ||
1602 (backing_object->flags & OBJ_DEAD) ||
1603 object->handle != NULL ||
1604 (object->type != OBJT_DEFAULT &&
1605 object->type != OBJT_SWAP) ||
1606 (object->flags & OBJ_DEAD)) {
1607 VM_OBJECT_UNLOCK(backing_object);
1608 break;
1609 }
1610
1611 if (
1612 object->paging_in_progress != 0 ||
1613 backing_object->paging_in_progress != 0
1614 ) {
1615 vm_object_qcollapse(object);
1616 VM_OBJECT_UNLOCK(backing_object);
1617 break;
1618 }
1619 /*
1620 * We know that we can either collapse the backing object (if
1621 * the parent is the only reference to it) or (perhaps) have
1622 * the parent bypass the object if the parent happens to shadow
1623 * all the resident pages in the entire backing object.
1624 *
1625 * This is ignoring pager-backed pages such as swap pages.
1626 * vm_object_backing_scan fails the shadowing test in this
1627 * case.
1628 */
1629 if (backing_object->ref_count == 1) {
1630 /*
1631 * If there is exactly one reference to the backing
1632 * object, we can collapse it into the parent.
1633 */
1634 vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1635
1636 /*
1637 * Move the pager from backing_object to object.
1638 */
1639 if (backing_object->type == OBJT_SWAP) {
1640 /*
1641 * swap_pager_copy() can sleep, in which case
1642 * the backing_object's and object's locks are
1643 * released and reacquired.
1644 */
1645 swap_pager_copy(
1646 backing_object,
1647 object,
1648 OFF_TO_IDX(object->backing_object_offset), TRUE);
1649 }
1650 /*
1651 * Object now shadows whatever backing_object did.
1652 * Note that the reference to
1653 * backing_object->backing_object moves from within
1654 * backing_object to within object.
1655 */
1656 LIST_REMOVE(object, shadow_list);
1657 backing_object->shadow_count--;
1658 backing_object->generation++;
1659 if (backing_object->backing_object) {
1660 VM_OBJECT_LOCK(backing_object->backing_object);
1661 LIST_REMOVE(backing_object, shadow_list);
1662 LIST_INSERT_HEAD(
1663 &backing_object->backing_object->shadow_head,
1664 object, shadow_list);
1665 /*
1666 * The shadow_count has not changed.
1667 */
1668 backing_object->backing_object->generation++;
1669 VM_OBJECT_UNLOCK(backing_object->backing_object);
1670 }
1671 object->backing_object = backing_object->backing_object;
1672 object->backing_object_offset +=
1673 backing_object->backing_object_offset;
1674
1675 /*
1676 * Discard backing_object.
1677 *
1678 * Since the backing object has no pages, no pager left,
1679 * and no object references within it, all that is
1680 * necessary is to dispose of it.
1681 */
1682 KASSERT(backing_object->ref_count == 1, ("backing_object %p was somehow re-referenced during collapse!", backing_object));
1683 VM_OBJECT_UNLOCK(backing_object);
1684
1685 mtx_lock(&vm_object_list_mtx);
1686 TAILQ_REMOVE(
1687 &vm_object_list,
1688 backing_object,
1689 object_list
1690 );
1691 mtx_unlock(&vm_object_list_mtx);
1692
1693 uma_zfree(obj_zone, backing_object);
1694
1695 object_collapses++;
1696 } else {
1697 vm_object_t new_backing_object;
1698
1699 /*
1700 * If we do not entirely shadow the backing object,
1701 * there is nothing we can do so we give up.
1702 */
1703 if (object->resident_page_count != object->size &&
1704 vm_object_backing_scan(object,
1705 OBSC_TEST_ALL_SHADOWED) == 0) {
1706 VM_OBJECT_UNLOCK(backing_object);
1707 break;
1708 }
1709
1710 /*
1711 * Make the parent shadow the next object in the
1712 * chain. Deallocating backing_object will not remove
1713 * it, since its reference count is at least 2.
1714 */
1715 LIST_REMOVE(object, shadow_list);
1716 backing_object->shadow_count--;
1717 backing_object->generation++;
1718
1719 new_backing_object = backing_object->backing_object;
1720 if ((object->backing_object = new_backing_object) != NULL) {
1721 VM_OBJECT_LOCK(new_backing_object);
1722 LIST_INSERT_HEAD(
1723 &new_backing_object->shadow_head,
1724 object,
1725 shadow_list
1726 );
1727 new_backing_object->shadow_count++;
1728 new_backing_object->generation++;
1729 vm_object_reference_locked(new_backing_object);
1730 VM_OBJECT_UNLOCK(new_backing_object);
1731 object->backing_object_offset +=
1732 backing_object->backing_object_offset;
1733 }
1734
1735 /*
1736 * Drop the reference count on backing_object. Since
1737 * its ref_count was at least 2, it will not vanish.
1738 */
1739 backing_object->ref_count--;
1740 VM_OBJECT_UNLOCK(backing_object);
1741 object_bypasses++;
1742 }
1743
1744 /*
1745 * Try again with this object's new backing object.
1746 */
1747 }
1748 }
1749
1750 /*
1751 * vm_object_page_remove:
1752 *
1753 * Removes all physical pages in the given range from the
1754 * object's list of pages. If the range's end is zero, all
1755 * physical pages from the range's start to the end of the object
1756 * are deleted.
1757 *
1758 * The object must be locked.
1759 */
1760 void
1761 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1762 boolean_t clean_only)
1763 {
1764 vm_page_t p, next;
1765
1766 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1767 if (object->resident_page_count == 0)
1768 return;
1769
1770 /*
1771 * Since physically-backed objects do not use managed pages, we can't
1772 * remove pages from the object (we must instead remove the page
1773 * references, and then destroy the object).
1774 */
1775 KASSERT(object->type != OBJT_PHYS,
1776 ("attempt to remove pages from a physical object"));
1777
1778 vm_object_pip_add(object, 1);
1779 again:
1780 vm_page_lock_queues();
1781 if ((p = TAILQ_FIRST(&object->memq)) != NULL) {
1782 if (p->pindex < start) {
1783 p = vm_page_splay(start, object->root);
1784 if ((object->root = p)->pindex < start)
1785 p = TAILQ_NEXT(p, listq);
1786 }
1787 }
1788 /*
1789 * Assert: the variable p is either (1) the page with the
1790 * least pindex greater than or equal to the parameter pindex
1791 * or (2) NULL.
1792 */
1793 for (;
1794 p != NULL && (p->pindex < end || end == 0);
1795 p = next) {
1796 next = TAILQ_NEXT(p, listq);
1797
1798 if (p->wire_count != 0) {
1799 pmap_remove_all(p);
1800 if (!clean_only)
1801 p->valid = 0;
1802 continue;
1803 }
1804 if (vm_page_sleep_if_busy(p, TRUE, "vmopar"))
1805 goto again;
1806 if (clean_only && p->valid) {
1807 pmap_page_protect(p, VM_PROT_READ | VM_PROT_EXECUTE);
1808 if (p->valid & p->dirty)
1809 continue;
1810 }
1811 pmap_remove_all(p);
1812 vm_page_free(p);
1813 }
1814 vm_page_unlock_queues();
1815 vm_object_pip_wakeup(object);
1816 }
1817
1818 /*
1819 * Routine: vm_object_coalesce
1820 * Function: Coalesces two objects backing up adjoining
1821 * regions of memory into a single object.
1822 *
1823 * returns TRUE if objects were combined.
1824 *
1825 * NOTE: Only works at the moment if the second object is NULL -
1826 * if it's not, which object do we lock first?
1827 *
1828 * Parameters:
1829 * prev_object First object to coalesce
1830 * prev_offset Offset into prev_object
1831 * prev_size Size of reference to prev_object
1832 * next_size Size of reference to the second object
1833 *
1834 * Conditions:
1835 * The object must *not* be locked.
1836 */
1837 boolean_t
1838 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
1839 vm_size_t prev_size, vm_size_t next_size)
1840 {
1841 vm_pindex_t next_pindex;
1842
1843 if (prev_object == NULL)
1844 return (TRUE);
1845 VM_OBJECT_LOCK(prev_object);
1846 if (prev_object->type != OBJT_DEFAULT &&
1847 prev_object->type != OBJT_SWAP) {
1848 VM_OBJECT_UNLOCK(prev_object);
1849 return (FALSE);
1850 }
1851
1852 /*
1853 * Try to collapse the object first
1854 */
1855 vm_object_collapse(prev_object);
1856
1857 /*
1858 * Can't coalesce if: . more than one reference . paged out . shadows
1859 * another object . has a copy elsewhere (any of which mean that the
1860 * pages not mapped to prev_entry may be in use anyway)
1861 */
1862 if (prev_object->backing_object != NULL) {
1863 VM_OBJECT_UNLOCK(prev_object);
1864 return (FALSE);
1865 }
1866
1867 prev_size >>= PAGE_SHIFT;
1868 next_size >>= PAGE_SHIFT;
1869 next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
1870
1871 if ((prev_object->ref_count > 1) &&
1872 (prev_object->size != next_pindex)) {
1873 VM_OBJECT_UNLOCK(prev_object);
1874 return (FALSE);
1875 }
1876
1877 /*
1878 * Remove any pages that may still be in the object from a previous
1879 * deallocation.
1880 */
1881 if (next_pindex < prev_object->size) {
1882 vm_object_page_remove(prev_object,
1883 next_pindex,
1884 next_pindex + next_size, FALSE);
1885 if (prev_object->type == OBJT_SWAP)
1886 swap_pager_freespace(prev_object,
1887 next_pindex, next_size);
1888 }
1889
1890 /*
1891 * Extend the object if necessary.
1892 */
1893 if (next_pindex + next_size > prev_object->size)
1894 prev_object->size = next_pindex + next_size;
1895
1896 VM_OBJECT_UNLOCK(prev_object);
1897 return (TRUE);
1898 }
1899
1900 void
1901 vm_object_set_writeable_dirty(vm_object_t object)
1902 {
1903 struct vnode *vp;
1904
1905 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1906 if ((object->flags & (OBJ_MIGHTBEDIRTY|OBJ_WRITEABLE)) ==
1907 (OBJ_MIGHTBEDIRTY|OBJ_WRITEABLE))
1908 return;
1909 vm_object_set_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
1910 if (object->type == OBJT_VNODE &&
1911 (vp = (struct vnode *)object->handle) != NULL) {
1912 VI_LOCK(vp);
1913 vp->v_iflag |= VI_OBJDIRTY;
1914 VI_UNLOCK(vp);
1915 }
1916 }
1917
1918 #include "opt_ddb.h"
1919 #ifdef DDB
1920 #include <sys/kernel.h>
1921
1922 #include <sys/cons.h>
1923
1924 #include <ddb/ddb.h>
1925
1926 static int
1927 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
1928 {
1929 vm_map_t tmpm;
1930 vm_map_entry_t tmpe;
1931 vm_object_t obj;
1932 int entcount;
1933
1934 if (map == 0)
1935 return 0;
1936
1937 if (entry == 0) {
1938 tmpe = map->header.next;
1939 entcount = map->nentries;
1940 while (entcount-- && (tmpe != &map->header)) {
1941 if (_vm_object_in_map(map, object, tmpe)) {
1942 return 1;
1943 }
1944 tmpe = tmpe->next;
1945 }
1946 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
1947 tmpm = entry->object.sub_map;
1948 tmpe = tmpm->header.next;
1949 entcount = tmpm->nentries;
1950 while (entcount-- && tmpe != &tmpm->header) {
1951 if (_vm_object_in_map(tmpm, object, tmpe)) {
1952 return 1;
1953 }
1954 tmpe = tmpe->next;
1955 }
1956 } else if ((obj = entry->object.vm_object) != NULL) {
1957 for (; obj; obj = obj->backing_object)
1958 if (obj == object) {
1959 return 1;
1960 }
1961 }
1962 return 0;
1963 }
1964
1965 static int
1966 vm_object_in_map(vm_object_t object)
1967 {
1968 struct proc *p;
1969
1970 /* sx_slock(&allproc_lock); */
1971 LIST_FOREACH(p, &allproc, p_list) {
1972 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
1973 continue;
1974 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
1975 /* sx_sunlock(&allproc_lock); */
1976 return 1;
1977 }
1978 }
1979 /* sx_sunlock(&allproc_lock); */
1980 if (_vm_object_in_map(kernel_map, object, 0))
1981 return 1;
1982 if (_vm_object_in_map(kmem_map, object, 0))
1983 return 1;
1984 if (_vm_object_in_map(pager_map, object, 0))
1985 return 1;
1986 if (_vm_object_in_map(buffer_map, object, 0))
1987 return 1;
1988 return 0;
1989 }
1990
1991 DB_SHOW_COMMAND(vmochk, vm_object_check)
1992 {
1993 vm_object_t object;
1994
1995 /*
1996 * make sure that internal objs are in a map somewhere
1997 * and none have zero ref counts.
1998 */
1999 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2000 if (object->handle == NULL &&
2001 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2002 if (object->ref_count == 0) {
2003 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2004 (long)object->size);
2005 }
2006 if (!vm_object_in_map(object)) {
2007 db_printf(
2008 "vmochk: internal obj is not in a map: "
2009 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2010 object->ref_count, (u_long)object->size,
2011 (u_long)object->size,
2012 (void *)object->backing_object);
2013 }
2014 }
2015 }
2016 }
2017
2018 /*
2019 * vm_object_print: [ debug ]
2020 */
2021 DB_SHOW_COMMAND(object, vm_object_print_static)
2022 {
2023 /* XXX convert args. */
2024 vm_object_t object = (vm_object_t)addr;
2025 boolean_t full = have_addr;
2026
2027 vm_page_t p;
2028
2029 /* XXX count is an (unused) arg. Avoid shadowing it. */
2030 #define count was_count
2031
2032 int count;
2033
2034 if (object == NULL)
2035 return;
2036
2037 db_iprintf(
2038 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x\n",
2039 object, (int)object->type, (uintmax_t)object->size,
2040 object->resident_page_count, object->ref_count, object->flags);
2041 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2042 object->shadow_count,
2043 object->backing_object ? object->backing_object->ref_count : 0,
2044 object->backing_object, (uintmax_t)object->backing_object_offset);
2045
2046 if (!full)
2047 return;
2048
2049 db_indent += 2;
2050 count = 0;
2051 TAILQ_FOREACH(p, &object->memq, listq) {
2052 if (count == 0)
2053 db_iprintf("memory:=");
2054 else if (count == 6) {
2055 db_printf("\n");
2056 db_iprintf(" ...");
2057 count = 0;
2058 } else
2059 db_printf(",");
2060 count++;
2061
2062 db_printf("(off=0x%jx,page=0x%jx)",
2063 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2064 }
2065 if (count != 0)
2066 db_printf("\n");
2067 db_indent -= 2;
2068 }
2069
2070 /* XXX. */
2071 #undef count
2072
2073 /* XXX need this non-static entry for calling from vm_map_print. */
2074 void
2075 vm_object_print(
2076 /* db_expr_t */ long addr,
2077 boolean_t have_addr,
2078 /* db_expr_t */ long count,
2079 char *modif)
2080 {
2081 vm_object_print_static(addr, have_addr, count, modif);
2082 }
2083
2084 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2085 {
2086 vm_object_t object;
2087 int nl = 0;
2088 int c;
2089
2090 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2091 vm_pindex_t idx, fidx;
2092 vm_pindex_t osize;
2093 vm_paddr_t pa = -1, padiff;
2094 int rcount;
2095 vm_page_t m;
2096
2097 db_printf("new object: %p\n", (void *)object);
2098 if (nl > 18) {
2099 c = cngetc();
2100 if (c != ' ')
2101 return;
2102 nl = 0;
2103 }
2104 nl++;
2105 rcount = 0;
2106 fidx = 0;
2107 osize = object->size;
2108 if (osize > 128)
2109 osize = 128;
2110 for (idx = 0; idx < osize; idx++) {
2111 m = vm_page_lookup(object, idx);
2112 if (m == NULL) {
2113 if (rcount) {
2114 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2115 (long)fidx, rcount, (long)pa);
2116 if (nl > 18) {
2117 c = cngetc();
2118 if (c != ' ')
2119 return;
2120 nl = 0;
2121 }
2122 nl++;
2123 rcount = 0;
2124 }
2125 continue;
2126 }
2127
2128
2129 if (rcount &&
2130 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2131 ++rcount;
2132 continue;
2133 }
2134 if (rcount) {
2135 padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
2136 padiff >>= PAGE_SHIFT;
2137 padiff &= PQ_L2_MASK;
2138 if (padiff == 0) {
2139 pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
2140 ++rcount;
2141 continue;
2142 }
2143 db_printf(" index(%ld)run(%d)pa(0x%lx)",
2144 (long)fidx, rcount, (long)pa);
2145 db_printf("pd(%ld)\n", (long)padiff);
2146 if (nl > 18) {
2147 c = cngetc();
2148 if (c != ' ')
2149 return;
2150 nl = 0;
2151 }
2152 nl++;
2153 }
2154 fidx = idx;
2155 pa = VM_PAGE_TO_PHYS(m);
2156 rcount = 1;
2157 }
2158 if (rcount) {
2159 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2160 (long)fidx, rcount, (long)pa);
2161 if (nl > 18) {
2162 c = cngetc();
2163 if (c != ' ')
2164 return;
2165 nl = 0;
2166 }
2167 nl++;
2168 }
2169 }
2170 }
2171 #endif /* DDB */
Cache object: b291f35ce93751f0acf0b05f7ad37bba
|