2 * Copyright (c) 2002-2006 Rice University
3 * Copyright (c) 2007-2008 Alan L. Cox <alc@cs.rice.edu>
6 * This software was developed for the FreeBSD Project by Alan L. Cox,
7 * Olivier Crameri, Peter Druschel, Sitaram Iyer, and Juan Navarro.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
21 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
24 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
25 * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
26 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY
28 * WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
33 * Superpage reservation management module
35 * Any external functions defined by this module are only to be used by the
36 * virtual memory system.
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
44 #include <sys/param.h>
45 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mutex.h>
49 #include <sys/queue.h>
50 #include <sys/rwlock.h>
52 #include <sys/sysctl.h>
53 #include <sys/systm.h>
56 #include <vm/vm_param.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_phys.h>
60 #include <vm/vm_radix.h>
61 #include <vm/vm_reserv.h>
64 * The reservation system supports the speculative allocation of large physical
65 * pages ("superpages"). Speculative allocation enables the fully-automatic
66 * utilization of superpages by the virtual memory system. In other words, no
67 * programmatic directives are required to use superpages.
70 #if VM_NRESERVLEVEL > 0
73 * The number of small pages that are contained in a level 0 reservation
75 #define VM_LEVEL_0_NPAGES (1 << VM_LEVEL_0_ORDER)
78 * The number of bits by which a physical address is shifted to obtain the
81 #define VM_LEVEL_0_SHIFT (VM_LEVEL_0_ORDER + PAGE_SHIFT)
84 * The size of a level 0 reservation in bytes
86 #define VM_LEVEL_0_SIZE (1 << VM_LEVEL_0_SHIFT)
89 * Computes the index of the small page underlying the given (object, pindex)
90 * within the reservation's array of small pages.
92 #define VM_RESERV_INDEX(object, pindex) \
93 (((object)->pg_color + (pindex)) & (VM_LEVEL_0_NPAGES - 1))
96 * The reservation structure
98 * A reservation structure is constructed whenever a large physical page is
99 * speculatively allocated to an object. The reservation provides the small
100 * physical pages for the range [pindex, pindex + VM_LEVEL_0_NPAGES) of offsets
101 * within that object. The reservation's "popcnt" tracks the number of these
102 * small physical pages that are in use at any given time. When and if the
103 * reservation is not fully utilized, it appears in the queue of partially-
104 * populated reservations. The reservation always appears on the containing
105 * object's list of reservations.
107 * A partially-populated reservation can be broken and reclaimed at any time.
110 TAILQ_ENTRY(vm_reserv) partpopq;
111 LIST_ENTRY(vm_reserv) objq;
112 vm_object_t object; /* containing object */
113 vm_pindex_t pindex; /* offset within object */
114 vm_page_t pages; /* first page of a superpage */
115 int popcnt; /* # of pages in use */
120 * The reservation array
122 * This array is analoguous in function to vm_page_array. It differs in the
123 * respect that it may contain a greater number of useful reservation
124 * structures than there are (physical) superpages. These "invalid"
125 * reservation structures exist to trade-off space for time in the
126 * implementation of vm_reserv_from_page(). Invalid reservation structures are
127 * distinguishable from "valid" reservation structures by inspecting the
128 * reservation's "pages" field. Invalid reservation structures have a NULL
131 * vm_reserv_from_page() maps a small (physical) page to an element of this
132 * array by computing a physical reservation number from the page's physical
133 * address. The physical reservation number is used as the array index.
135 * An "active" reservation is a valid reservation structure that has a non-NULL
136 * "object" field and a non-zero "popcnt" field. In other words, every active
137 * reservation belongs to a particular object. Moreover, every active
138 * reservation has an entry in the containing object's list of reservations.
140 static vm_reserv_t vm_reserv_array;
143 * The partially-populated reservation queue
145 * This queue enables the fast recovery of an unused cached or free small page
146 * from a partially-populated reservation. The reservation at the head of
147 * this queue is the least-recently-changed, partially-populated reservation.
149 * Access to this queue is synchronized by the free page queue lock.
151 static TAILQ_HEAD(, vm_reserv) vm_rvq_partpop =
152 TAILQ_HEAD_INITIALIZER(vm_rvq_partpop);
154 static SYSCTL_NODE(_vm, OID_AUTO, reserv, CTLFLAG_RD, 0, "Reservation Info");
156 static long vm_reserv_broken;
157 SYSCTL_LONG(_vm_reserv, OID_AUTO, broken, CTLFLAG_RD,
158 &vm_reserv_broken, 0, "Cumulative number of broken reservations");
160 static long vm_reserv_freed;
161 SYSCTL_LONG(_vm_reserv, OID_AUTO, freed, CTLFLAG_RD,
162 &vm_reserv_freed, 0, "Cumulative number of freed reservations");
164 static int sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS);
166 SYSCTL_OID(_vm_reserv, OID_AUTO, partpopq, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
167 sysctl_vm_reserv_partpopq, "A", "Partially-populated reservation queues");
169 static long vm_reserv_reclaimed;
170 SYSCTL_LONG(_vm_reserv, OID_AUTO, reclaimed, CTLFLAG_RD,
171 &vm_reserv_reclaimed, 0, "Cumulative number of reclaimed reservations");
173 static void vm_reserv_depopulate(vm_reserv_t rv);
174 static vm_reserv_t vm_reserv_from_page(vm_page_t m);
175 static boolean_t vm_reserv_has_pindex(vm_reserv_t rv,
177 static void vm_reserv_populate(vm_reserv_t rv);
178 static void vm_reserv_reclaim(vm_reserv_t rv);
181 * Describes the current state of the partially-populated reservation queue.
184 sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS)
188 int counter, error, level, unused_pages;
190 error = sysctl_wire_old_buffer(req, 0);
193 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
194 sbuf_printf(&sbuf, "\nLEVEL SIZE NUMBER\n\n");
195 for (level = -1; level <= VM_NRESERVLEVEL - 2; level++) {
198 mtx_lock(&vm_page_queue_free_mtx);
199 TAILQ_FOREACH(rv, &vm_rvq_partpop/*[level]*/, partpopq) {
201 unused_pages += VM_LEVEL_0_NPAGES - rv->popcnt;
203 mtx_unlock(&vm_page_queue_free_mtx);
204 sbuf_printf(&sbuf, "%5d: %6dK, %6d\n", level,
205 unused_pages * ((int)PAGE_SIZE / 1024), counter);
207 error = sbuf_finish(&sbuf);
213 * Reduces the given reservation's population count. If the population count
214 * becomes zero, the reservation is destroyed. Additionally, moves the
215 * reservation to the tail of the partially-populated reservations queue if the
216 * population count is non-zero.
218 * The free page queue lock must be held.
221 vm_reserv_depopulate(vm_reserv_t rv)
224 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
225 KASSERT(rv->object != NULL,
226 ("vm_reserv_depopulate: reserv %p is free", rv));
227 KASSERT(rv->popcnt > 0,
228 ("vm_reserv_depopulate: reserv %p's popcnt is corrupted", rv));
229 if (rv->inpartpopq) {
230 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
231 rv->inpartpopq = FALSE;
234 if (rv->popcnt == 0) {
235 LIST_REMOVE(rv, objq);
237 vm_phys_free_pages(rv->pages, VM_LEVEL_0_ORDER);
240 rv->inpartpopq = TRUE;
241 TAILQ_INSERT_TAIL(&vm_rvq_partpop, rv, partpopq);
246 * Returns the reservation to which the given page might belong.
248 static __inline vm_reserv_t
249 vm_reserv_from_page(vm_page_t m)
252 return (&vm_reserv_array[VM_PAGE_TO_PHYS(m) >> VM_LEVEL_0_SHIFT]);
256 * Returns TRUE if the given reservation contains the given page index and
259 static __inline boolean_t
260 vm_reserv_has_pindex(vm_reserv_t rv, vm_pindex_t pindex)
263 return (((pindex - rv->pindex) & ~(VM_LEVEL_0_NPAGES - 1)) == 0);
267 * Increases the given reservation's population count. Moves the reservation
268 * to the tail of the partially-populated reservation queue.
270 * The free page queue must be locked.
273 vm_reserv_populate(vm_reserv_t rv)
276 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
277 KASSERT(rv->object != NULL,
278 ("vm_reserv_populate: reserv %p is free", rv));
279 KASSERT(rv->popcnt < VM_LEVEL_0_NPAGES,
280 ("vm_reserv_populate: reserv %p is already full", rv));
281 if (rv->inpartpopq) {
282 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
283 rv->inpartpopq = FALSE;
286 if (rv->popcnt < VM_LEVEL_0_NPAGES) {
287 rv->inpartpopq = TRUE;
288 TAILQ_INSERT_TAIL(&vm_rvq_partpop, rv, partpopq);
293 * Allocates a contiguous set of physical pages of the given size "npages"
294 * from an existing or newly-created reservation. All of the physical pages
295 * must be at or above the given physical address "low" and below the given
296 * physical address "high". The given value "alignment" determines the
297 * alignment of the first physical page in the set. If the given value
298 * "boundary" is non-zero, then the set of physical pages cannot cross any
299 * physical address boundary that is a multiple of that value. Both
300 * "alignment" and "boundary" must be a power of two.
302 * The object and free page queue must be locked.
305 vm_reserv_alloc_contig(vm_object_t object, vm_pindex_t pindex, u_long npages,
306 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
309 vm_page_t m, m_ret, mpred, msucc;
310 vm_pindex_t first, leftcap, rightcap;
312 u_long allocpages, maxpages, minpages;
315 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
316 VM_OBJECT_ASSERT_WLOCKED(object);
317 KASSERT(npages != 0, ("vm_reserv_alloc_contig: npages is 0"));
320 * Is a reservation fundamentally impossible?
322 if (pindex < VM_RESERV_INDEX(object, pindex) ||
323 pindex + npages > object->size)
327 * All reservations of a particular size have the same alignment.
328 * Assuming that the first page is allocated from a reservation, the
329 * least significant bits of its physical address can be determined
330 * from its offset from the beginning of the reservation and the size
331 * of the reservation.
333 * Could the specified index within a reservation of the smallest
334 * possible size satisfy the alignment and boundary requirements?
336 pa = VM_RESERV_INDEX(object, pindex) << PAGE_SHIFT;
337 if ((pa & (alignment - 1)) != 0)
339 size = npages << PAGE_SHIFT;
340 if (((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0)
344 * Look for an existing reservation.
346 mpred = vm_radix_lookup_le(&object->rtree, pindex);
348 KASSERT(mpred->pindex < pindex,
349 ("vm_reserv_alloc_contig: pindex already allocated"));
350 rv = vm_reserv_from_page(mpred);
351 if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
353 msucc = TAILQ_NEXT(mpred, listq);
355 msucc = TAILQ_FIRST(&object->memq);
357 KASSERT(msucc->pindex > pindex,
358 ("vm_reserv_alloc_page: pindex already allocated"));
359 rv = vm_reserv_from_page(msucc);
360 if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
365 * Could at least one reservation fit between the first index to the
366 * left that can be used and the first index to the right that cannot
369 first = pindex - VM_RESERV_INDEX(object, pindex);
371 if ((rv = vm_reserv_from_page(mpred))->object != object)
372 leftcap = mpred->pindex + 1;
374 leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
378 minpages = VM_RESERV_INDEX(object, pindex) + npages;
379 maxpages = roundup2(minpages, VM_LEVEL_0_NPAGES);
380 allocpages = maxpages;
382 if ((rv = vm_reserv_from_page(msucc))->object != object)
383 rightcap = msucc->pindex;
385 rightcap = rv->pindex;
386 if (first + maxpages > rightcap) {
387 if (maxpages == VM_LEVEL_0_NPAGES)
389 allocpages = minpages;
394 * Would the last new reservation extend past the end of the object?
396 if (first + maxpages > object->size) {
398 * Don't allocate the last new reservation if the object is a
399 * vnode or backed by another object that is a vnode.
401 if (object->type == OBJT_VNODE ||
402 (object->backing_object != NULL &&
403 object->backing_object->type == OBJT_VNODE)) {
404 if (maxpages == VM_LEVEL_0_NPAGES)
406 allocpages = minpages;
408 /* Speculate that the object may grow. */
412 * Allocate and populate the new reservations. The alignment and
413 * boundary specified for this allocation may be different from the
414 * alignment and boundary specified for the requested pages. For
415 * instance, the specified index may not be the first page within the
416 * first new reservation.
418 m = vm_phys_alloc_contig(allocpages, low, high, ulmax(alignment,
419 VM_LEVEL_0_SIZE), boundary > VM_LEVEL_0_SIZE ? boundary : 0);
423 index = VM_RESERV_INDEX(object, pindex);
425 rv = vm_reserv_from_page(m);
426 KASSERT(rv->pages == m,
427 ("vm_reserv_alloc_contig: reserv %p's pages is corrupted",
429 KASSERT(rv->object == NULL,
430 ("vm_reserv_alloc_contig: reserv %p isn't free", rv));
431 LIST_INSERT_HEAD(&object->rvq, rv, objq);
434 KASSERT(rv->popcnt == 0,
435 ("vm_reserv_alloc_contig: reserv %p's popcnt is corrupted",
437 KASSERT(!rv->inpartpopq,
438 ("vm_reserv_alloc_contig: reserv %p's inpartpopq is TRUE",
440 n = ulmin(VM_LEVEL_0_NPAGES - index, npages);
441 for (i = 0; i < n; i++)
442 vm_reserv_populate(rv);
445 m_ret = &rv->pages[index];
448 m += VM_LEVEL_0_NPAGES;
449 first += VM_LEVEL_0_NPAGES;
450 allocpages -= VM_LEVEL_0_NPAGES;
451 } while (allocpages > 0);
455 * Found a matching reservation.
458 index = VM_RESERV_INDEX(object, pindex);
459 /* Does the allocation fit within the reservation? */
460 if (index + npages > VM_LEVEL_0_NPAGES)
462 m = &rv->pages[index];
463 pa = VM_PAGE_TO_PHYS(m);
464 if (pa < low || pa + size > high || (pa & (alignment - 1)) != 0 ||
465 ((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0)
467 /* Handle vm_page_rename(m, new_object, ...). */
468 for (i = 0; i < npages; i++)
469 if ((rv->pages[index + i].flags & (PG_CACHED | PG_FREE)) == 0)
471 for (i = 0; i < npages; i++)
472 vm_reserv_populate(rv);
477 * Allocates a page from an existing or newly-created reservation.
479 * The page "mpred" must immediately precede the offset "pindex" within the
482 * The object and free page queue must be locked.
485 vm_reserv_alloc_page(vm_object_t object, vm_pindex_t pindex, vm_page_t mpred)
488 vm_pindex_t first, leftcap, rightcap;
491 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
492 VM_OBJECT_ASSERT_WLOCKED(object);
495 * Is a reservation fundamentally impossible?
497 if (pindex < VM_RESERV_INDEX(object, pindex) ||
498 pindex >= object->size)
502 * Look for an existing reservation.
505 KASSERT(mpred->object == object,
506 ("vm_reserv_alloc_page: object doesn't contain mpred"));
507 KASSERT(mpred->pindex < pindex,
508 ("vm_reserv_alloc_page: mpred doesn't precede pindex"));
509 rv = vm_reserv_from_page(mpred);
510 if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
512 msucc = TAILQ_NEXT(mpred, listq);
514 msucc = TAILQ_FIRST(&object->memq);
516 KASSERT(msucc->pindex > pindex,
517 ("vm_reserv_alloc_page: msucc doesn't succeed pindex"));
518 rv = vm_reserv_from_page(msucc);
519 if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
524 * Could a reservation fit between the first index to the left that
525 * can be used and the first index to the right that cannot be used?
527 first = pindex - VM_RESERV_INDEX(object, pindex);
529 if ((rv = vm_reserv_from_page(mpred))->object != object)
530 leftcap = mpred->pindex + 1;
532 leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
537 if ((rv = vm_reserv_from_page(msucc))->object != object)
538 rightcap = msucc->pindex;
540 rightcap = rv->pindex;
541 if (first + VM_LEVEL_0_NPAGES > rightcap)
546 * Would a new reservation extend past the end of the object?
548 if (first + VM_LEVEL_0_NPAGES > object->size) {
550 * Don't allocate a new reservation if the object is a vnode or
551 * backed by another object that is a vnode.
553 if (object->type == OBJT_VNODE ||
554 (object->backing_object != NULL &&
555 object->backing_object->type == OBJT_VNODE))
557 /* Speculate that the object may grow. */
561 * Allocate and populate the new reservation.
563 m = vm_phys_alloc_pages(VM_FREEPOOL_DEFAULT, VM_LEVEL_0_ORDER);
566 rv = vm_reserv_from_page(m);
567 KASSERT(rv->pages == m,
568 ("vm_reserv_alloc_page: reserv %p's pages is corrupted", rv));
569 KASSERT(rv->object == NULL,
570 ("vm_reserv_alloc_page: reserv %p isn't free", rv));
571 LIST_INSERT_HEAD(&object->rvq, rv, objq);
574 KASSERT(rv->popcnt == 0,
575 ("vm_reserv_alloc_page: reserv %p's popcnt is corrupted", rv));
576 KASSERT(!rv->inpartpopq,
577 ("vm_reserv_alloc_page: reserv %p's inpartpopq is TRUE", rv));
578 vm_reserv_populate(rv);
579 return (&rv->pages[VM_RESERV_INDEX(object, pindex)]);
582 * Found a matching reservation.
585 m = &rv->pages[VM_RESERV_INDEX(object, pindex)];
586 /* Handle vm_page_rename(m, new_object, ...). */
587 if ((m->flags & (PG_CACHED | PG_FREE)) == 0)
589 vm_reserv_populate(rv);
594 * Breaks all reservations belonging to the given object.
597 vm_reserv_break_all(vm_object_t object)
602 mtx_lock(&vm_page_queue_free_mtx);
603 while ((rv = LIST_FIRST(&object->rvq)) != NULL) {
604 KASSERT(rv->object == object,
605 ("vm_reserv_break_all: reserv %p is corrupted", rv));
606 if (rv->inpartpopq) {
607 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
608 rv->inpartpopq = FALSE;
610 LIST_REMOVE(rv, objq);
612 for (i = 0; i < VM_LEVEL_0_NPAGES; i++) {
613 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
614 vm_phys_free_pages(&rv->pages[i], 0);
618 KASSERT(rv->popcnt == 0,
619 ("vm_reserv_break_all: reserv %p's popcnt is corrupted",
623 mtx_unlock(&vm_page_queue_free_mtx);
627 * Frees the given page if it belongs to a reservation. Returns TRUE if the
628 * page is freed and FALSE otherwise.
630 * The free page queue lock must be held.
633 vm_reserv_free_page(vm_page_t m)
637 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
638 rv = vm_reserv_from_page(m);
639 if (rv->object == NULL)
641 if ((m->flags & PG_CACHED) != 0 && m->pool != VM_FREEPOOL_CACHE)
642 vm_phys_set_pool(VM_FREEPOOL_CACHE, rv->pages,
644 vm_reserv_depopulate(rv);
649 * Initializes the reservation management system. Specifically, initializes
650 * the reservation array.
652 * Requires that vm_page_array and first_page are initialized!
661 * Initialize the reservation array. Specifically, initialize the
662 * "pages" field for every element that has an underlying superpage.
664 for (i = 0; phys_avail[i + 1] != 0; i += 2) {
665 paddr = roundup2(phys_avail[i], VM_LEVEL_0_SIZE);
666 while (paddr + VM_LEVEL_0_SIZE <= phys_avail[i + 1]) {
667 vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT].pages =
668 PHYS_TO_VM_PAGE(paddr);
669 paddr += VM_LEVEL_0_SIZE;
675 * Returns a reservation level if the given page belongs to a fully-populated
676 * reservation and -1 otherwise.
679 vm_reserv_level_iffullpop(vm_page_t m)
683 rv = vm_reserv_from_page(m);
684 return (rv->popcnt == VM_LEVEL_0_NPAGES ? 0 : -1);
688 * Prepare for the reactivation of a cached page.
690 * First, suppose that the given page "m" was allocated individually, i.e., not
691 * as part of a reservation, and cached. Then, suppose a reservation
692 * containing "m" is allocated by the same object. Although "m" and the
693 * reservation belong to the same object, "m"'s pindex may not match the
696 * The free page queue must be locked.
699 vm_reserv_reactivate_page(vm_page_t m)
704 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
705 rv = vm_reserv_from_page(m);
706 if (rv->object == NULL)
708 KASSERT((m->flags & PG_CACHED) != 0,
709 ("vm_reserv_uncache_page: page %p is not cached", m));
710 if (m->object == rv->object &&
711 m->pindex - rv->pindex == VM_RESERV_INDEX(m->object, m->pindex))
712 vm_reserv_populate(rv);
714 KASSERT(rv->inpartpopq,
715 ("vm_reserv_uncache_page: reserv %p's inpartpopq is FALSE",
717 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
718 rv->inpartpopq = FALSE;
719 LIST_REMOVE(rv, objq);
721 /* Don't vm_phys_free_pages(m, 0). */
722 m_index = m - rv->pages;
723 for (i = 0; i < m_index; i++) {
724 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
725 vm_phys_free_pages(&rv->pages[i], 0);
729 for (i++; i < VM_LEVEL_0_NPAGES; i++) {
730 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
731 vm_phys_free_pages(&rv->pages[i], 0);
735 KASSERT(rv->popcnt == 0,
736 ("vm_reserv_uncache_page: reserv %p's popcnt is corrupted",
744 * Breaks the given partially-populated reservation, releasing its cached and
745 * free pages to the physical memory allocator.
747 * The free page queue lock must be held.
750 vm_reserv_reclaim(vm_reserv_t rv)
754 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
755 KASSERT(rv->inpartpopq,
756 ("vm_reserv_reclaim: reserv %p's inpartpopq is corrupted", rv));
757 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
758 rv->inpartpopq = FALSE;
759 KASSERT(rv->object != NULL,
760 ("vm_reserv_reclaim: reserv %p is free", rv));
761 LIST_REMOVE(rv, objq);
763 for (i = 0; i < VM_LEVEL_0_NPAGES; i++) {
764 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
765 vm_phys_free_pages(&rv->pages[i], 0);
769 KASSERT(rv->popcnt == 0,
770 ("vm_reserv_reclaim: reserv %p's popcnt is corrupted", rv));
771 vm_reserv_reclaimed++;
775 * Breaks the reservation at the head of the partially-populated reservation
776 * queue, releasing its cached and free pages to the physical memory
777 * allocator. Returns TRUE if a reservation is broken and FALSE otherwise.
779 * The free page queue lock must be held.
782 vm_reserv_reclaim_inactive(void)
786 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
787 if ((rv = TAILQ_FIRST(&vm_rvq_partpop)) != NULL) {
788 vm_reserv_reclaim(rv);
795 * Searches the partially-populated reservation queue for the least recently
796 * active reservation with unused pages, i.e., cached or free, that satisfy the
797 * given request for contiguous physical memory. If a satisfactory reservation
798 * is found, it is broken. Returns TRUE if a reservation is broken and FALSE
801 * The free page queue lock must be held.
804 vm_reserv_reclaim_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
805 u_long alignment, vm_paddr_t boundary)
807 vm_paddr_t pa, pa_length, size;
811 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
812 if (npages > VM_LEVEL_0_NPAGES - 1)
814 size = npages << PAGE_SHIFT;
815 TAILQ_FOREACH(rv, &vm_rvq_partpop, partpopq) {
816 pa = VM_PAGE_TO_PHYS(&rv->pages[VM_LEVEL_0_NPAGES - 1]);
817 if (pa + PAGE_SIZE - size < low) {
818 /* this entire reservation is too low; go to next */
822 for (i = 0; i < VM_LEVEL_0_NPAGES; i++)
823 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0) {
824 pa_length += PAGE_SIZE;
825 if (pa_length == PAGE_SIZE) {
826 pa = VM_PAGE_TO_PHYS(&rv->pages[i]);
827 if (pa + size > high) {
828 /* skip to next reservation */
830 } else if (pa < low ||
831 (pa & (alignment - 1)) != 0 ||
832 ((pa ^ (pa + size - 1)) &
833 ~(boundary - 1)) != 0)
836 if (pa_length >= size) {
837 vm_reserv_reclaim(rv);
847 * Transfers the reservation underlying the given page to a new object.
849 * The object must be locked.
852 vm_reserv_rename(vm_page_t m, vm_object_t new_object, vm_object_t old_object,
853 vm_pindex_t old_object_offset)
857 VM_OBJECT_ASSERT_WLOCKED(new_object);
858 rv = vm_reserv_from_page(m);
859 if (rv->object == old_object) {
860 mtx_lock(&vm_page_queue_free_mtx);
861 if (rv->object == old_object) {
862 LIST_REMOVE(rv, objq);
863 LIST_INSERT_HEAD(&new_object->rvq, rv, objq);
864 rv->object = new_object;
865 rv->pindex -= old_object_offset;
867 mtx_unlock(&vm_page_queue_free_mtx);
872 * Allocates the virtual and physical memory required by the reservation
873 * management system's data structures, in particular, the reservation array.
876 vm_reserv_startup(vm_offset_t *vaddr, vm_paddr_t end, vm_paddr_t high_water)
882 * Calculate the size (in bytes) of the reservation array. Round up
883 * from "high_water" because every small page is mapped to an element
884 * in the reservation array based on its physical address. Thus, the
885 * number of elements in the reservation array can be greater than the
886 * number of superpages.
888 size = howmany(high_water, VM_LEVEL_0_SIZE) * sizeof(struct vm_reserv);
891 * Allocate and map the physical memory for the reservation array. The
892 * next available virtual address is returned by reference.
894 new_end = end - round_page(size);
895 vm_reserv_array = (void *)(uintptr_t)pmap_map(vaddr, new_end, end,
896 VM_PROT_READ | VM_PROT_WRITE);
897 bzero(vm_reserv_array, size);
900 * Return the next available physical address.
905 #endif /* VM_NRESERVLEVEL > 0 */