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>
51 #include <sys/sysctl.h>
52 #include <sys/systm.h>
55 #include <vm/vm_param.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_phys.h>
59 #include <vm/vm_radix.h>
60 #include <vm/vm_reserv.h>
63 * The reservation system supports the speculative allocation of large physical
64 * pages ("superpages"). Speculative allocation enables the fully-automatic
65 * utilization of superpages by the virtual memory system. In other words, no
66 * programmatic directives are required to use superpages.
69 #if VM_NRESERVLEVEL > 0
72 * The number of small pages that are contained in a level 0 reservation
74 #define VM_LEVEL_0_NPAGES (1 << VM_LEVEL_0_ORDER)
77 * The number of bits by which a physical address is shifted to obtain the
80 #define VM_LEVEL_0_SHIFT (VM_LEVEL_0_ORDER + PAGE_SHIFT)
83 * The size of a level 0 reservation in bytes
85 #define VM_LEVEL_0_SIZE (1 << VM_LEVEL_0_SHIFT)
88 * Computes the index of the small page underlying the given (object, pindex)
89 * within the reservation's array of small pages.
91 #define VM_RESERV_INDEX(object, pindex) \
92 (((object)->pg_color + (pindex)) & (VM_LEVEL_0_NPAGES - 1))
95 * The reservation structure
97 * A reservation structure is constructed whenever a large physical page is
98 * speculatively allocated to an object. The reservation provides the small
99 * physical pages for the range [pindex, pindex + VM_LEVEL_0_NPAGES) of offsets
100 * within that object. The reservation's "popcnt" tracks the number of these
101 * small physical pages that are in use at any given time. When and if the
102 * reservation is not fully utilized, it appears in the queue of partially-
103 * populated reservations. The reservation always appears on the containing
104 * object's list of reservations.
106 * A partially-populated reservation can be broken and reclaimed at any time.
109 TAILQ_ENTRY(vm_reserv) partpopq;
110 LIST_ENTRY(vm_reserv) objq;
111 vm_object_t object; /* containing object */
112 vm_pindex_t pindex; /* offset within object */
113 vm_page_t pages; /* first page of a superpage */
114 int popcnt; /* # of pages in use */
119 * The reservation array
121 * This array is analoguous in function to vm_page_array. It differs in the
122 * respect that it may contain a greater number of useful reservation
123 * structures than there are (physical) superpages. These "invalid"
124 * reservation structures exist to trade-off space for time in the
125 * implementation of vm_reserv_from_page(). Invalid reservation structures are
126 * distinguishable from "valid" reservation structures by inspecting the
127 * reservation's "pages" field. Invalid reservation structures have a NULL
130 * vm_reserv_from_page() maps a small (physical) page to an element of this
131 * array by computing a physical reservation number from the page's physical
132 * address. The physical reservation number is used as the array index.
134 * An "active" reservation is a valid reservation structure that has a non-NULL
135 * "object" field and a non-zero "popcnt" field. In other words, every active
136 * reservation belongs to a particular object. Moreover, every active
137 * reservation has an entry in the containing object's list of reservations.
139 static vm_reserv_t vm_reserv_array;
142 * The partially-populated reservation queue
144 * This queue enables the fast recovery of an unused cached or free small page
145 * from a partially-populated reservation. The reservation at the head of
146 * this queue is the least-recently-changed, partially-populated reservation.
148 * Access to this queue is synchronized by the free page queue lock.
150 static TAILQ_HEAD(, vm_reserv) vm_rvq_partpop =
151 TAILQ_HEAD_INITIALIZER(vm_rvq_partpop);
153 static SYSCTL_NODE(_vm, OID_AUTO, reserv, CTLFLAG_RD, 0, "Reservation Info");
155 static long vm_reserv_broken;
156 SYSCTL_LONG(_vm_reserv, OID_AUTO, broken, CTLFLAG_RD,
157 &vm_reserv_broken, 0, "Cumulative number of broken reservations");
159 static long vm_reserv_freed;
160 SYSCTL_LONG(_vm_reserv, OID_AUTO, freed, CTLFLAG_RD,
161 &vm_reserv_freed, 0, "Cumulative number of freed reservations");
163 static int sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS);
165 SYSCTL_OID(_vm_reserv, OID_AUTO, partpopq, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
166 sysctl_vm_reserv_partpopq, "A", "Partially-populated reservation queues");
168 static long vm_reserv_reclaimed;
169 SYSCTL_LONG(_vm_reserv, OID_AUTO, reclaimed, CTLFLAG_RD,
170 &vm_reserv_reclaimed, 0, "Cumulative number of reclaimed reservations");
172 static void vm_reserv_depopulate(vm_reserv_t rv);
173 static vm_reserv_t vm_reserv_from_page(vm_page_t m);
174 static boolean_t vm_reserv_has_pindex(vm_reserv_t rv,
176 static void vm_reserv_populate(vm_reserv_t rv);
177 static void vm_reserv_reclaim(vm_reserv_t rv);
180 * Describes the current state of the partially-populated reservation queue.
183 sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS)
187 int counter, error, level, unused_pages;
189 error = sysctl_wire_old_buffer(req, 0);
192 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
193 sbuf_printf(&sbuf, "\nLEVEL SIZE NUMBER\n\n");
194 for (level = -1; level <= VM_NRESERVLEVEL - 2; level++) {
197 mtx_lock(&vm_page_queue_free_mtx);
198 TAILQ_FOREACH(rv, &vm_rvq_partpop/*[level]*/, partpopq) {
200 unused_pages += VM_LEVEL_0_NPAGES - rv->popcnt;
202 mtx_unlock(&vm_page_queue_free_mtx);
203 sbuf_printf(&sbuf, "%5d: %6dK, %6d\n", level,
204 unused_pages * ((int)PAGE_SIZE / 1024), counter);
206 error = sbuf_finish(&sbuf);
212 * Reduces the given reservation's population count. If the population count
213 * becomes zero, the reservation is destroyed. Additionally, moves the
214 * reservation to the tail of the partially-populated reservations queue if the
215 * population count is non-zero.
217 * The free page queue lock must be held.
220 vm_reserv_depopulate(vm_reserv_t rv)
223 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
224 KASSERT(rv->object != NULL,
225 ("vm_reserv_depopulate: reserv %p is free", rv));
226 KASSERT(rv->popcnt > 0,
227 ("vm_reserv_depopulate: reserv %p's popcnt is corrupted", rv));
228 if (rv->inpartpopq) {
229 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
230 rv->inpartpopq = FALSE;
233 if (rv->popcnt == 0) {
234 LIST_REMOVE(rv, objq);
236 vm_phys_free_pages(rv->pages, VM_LEVEL_0_ORDER);
239 rv->inpartpopq = TRUE;
240 TAILQ_INSERT_TAIL(&vm_rvq_partpop, rv, partpopq);
245 * Returns the reservation to which the given page might belong.
247 static __inline vm_reserv_t
248 vm_reserv_from_page(vm_page_t m)
251 return (&vm_reserv_array[VM_PAGE_TO_PHYS(m) >> VM_LEVEL_0_SHIFT]);
255 * Returns TRUE if the given reservation contains the given page index and
258 static __inline boolean_t
259 vm_reserv_has_pindex(vm_reserv_t rv, vm_pindex_t pindex)
262 return (((pindex - rv->pindex) & ~(VM_LEVEL_0_NPAGES - 1)) == 0);
266 * Increases the given reservation's population count. Moves the reservation
267 * to the tail of the partially-populated reservation queue.
269 * The free page queue must be locked.
272 vm_reserv_populate(vm_reserv_t rv)
275 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
276 KASSERT(rv->object != NULL,
277 ("vm_reserv_populate: reserv %p is free", rv));
278 KASSERT(rv->popcnt < VM_LEVEL_0_NPAGES,
279 ("vm_reserv_populate: reserv %p is already full", rv));
280 if (rv->inpartpopq) {
281 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
282 rv->inpartpopq = FALSE;
285 if (rv->popcnt < VM_LEVEL_0_NPAGES) {
286 rv->inpartpopq = TRUE;
287 TAILQ_INSERT_TAIL(&vm_rvq_partpop, rv, partpopq);
292 * Allocates a contiguous set of physical pages of the given size "npages"
293 * from an existing or newly-created reservation. All of the physical pages
294 * must be at or above the given physical address "low" and below the given
295 * physical address "high". The given value "alignment" determines the
296 * alignment of the first physical page in the set. If the given value
297 * "boundary" is non-zero, then the set of physical pages cannot cross any
298 * physical address boundary that is a multiple of that value. Both
299 * "alignment" and "boundary" must be a power of two.
301 * The object and free page queue must be locked.
304 vm_reserv_alloc_contig(vm_object_t object, vm_pindex_t pindex, u_long npages,
305 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
308 vm_page_t m, m_ret, mpred, msucc;
309 vm_pindex_t first, leftcap, rightcap;
311 u_long allocpages, maxpages, minpages;
314 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
315 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
316 KASSERT(npages != 0, ("vm_reserv_alloc_contig: npages is 0"));
319 * Is a reservation fundamentally impossible?
321 if (pindex < VM_RESERV_INDEX(object, pindex) ||
322 pindex + npages > object->size)
326 * All reservations of a particular size have the same alignment.
327 * Assuming that the first page is allocated from a reservation, the
328 * least significant bits of its physical address can be determined
329 * from its offset from the beginning of the reservation and the size
330 * of the reservation.
332 * Could the specified index within a reservation of the smallest
333 * possible size satisfy the alignment and boundary requirements?
335 pa = VM_RESERV_INDEX(object, pindex) << PAGE_SHIFT;
336 if ((pa & (alignment - 1)) != 0)
338 size = npages << PAGE_SHIFT;
339 if (((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0)
343 * Look for an existing reservation.
345 mpred = vm_radix_lookup_le(&object->rtree, pindex);
347 KASSERT(mpred->pindex < pindex,
348 ("vm_reserv_alloc_contig: pindex already allocated"));
349 rv = vm_reserv_from_page(mpred);
350 if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
352 msucc = TAILQ_NEXT(mpred, listq);
354 msucc = vm_radix_lookup_ge(&object->rtree, pindex);
356 KASSERT(msucc->pindex > pindex,
357 ("vm_reserv_alloc_page: pindex already allocated"));
358 rv = vm_reserv_from_page(msucc);
359 if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
364 * Could at least one reservation fit between the first index to the
365 * left that can be used and the first index to the right that cannot
368 first = pindex - VM_RESERV_INDEX(object, pindex);
370 if ((rv = vm_reserv_from_page(mpred))->object != object)
371 leftcap = mpred->pindex + 1;
373 leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
377 minpages = VM_RESERV_INDEX(object, pindex) + npages;
378 maxpages = roundup2(minpages, VM_LEVEL_0_NPAGES);
379 allocpages = maxpages;
381 if ((rv = vm_reserv_from_page(msucc))->object != object)
382 rightcap = msucc->pindex;
384 rightcap = rv->pindex;
385 if (first + maxpages > rightcap) {
386 if (maxpages == VM_LEVEL_0_NPAGES)
388 allocpages = minpages;
393 * Would the last new reservation extend past the end of the object?
395 if (first + maxpages > object->size) {
397 * Don't allocate the last new reservation if the object is a
398 * vnode or backed by another object that is a vnode.
400 if (object->type == OBJT_VNODE ||
401 (object->backing_object != NULL &&
402 object->backing_object->type == OBJT_VNODE)) {
403 if (maxpages == VM_LEVEL_0_NPAGES)
405 allocpages = minpages;
407 /* Speculate that the object may grow. */
411 * Allocate and populate the new reservations. The alignment and
412 * boundary specified for this allocation may be different from the
413 * alignment and boundary specified for the requested pages. For
414 * instance, the specified index may not be the first page within the
415 * first new reservation.
417 m = vm_phys_alloc_contig(allocpages, low, high, ulmax(alignment,
418 VM_LEVEL_0_SIZE), boundary > VM_LEVEL_0_SIZE ? boundary : 0);
422 index = VM_RESERV_INDEX(object, pindex);
424 rv = vm_reserv_from_page(m);
425 KASSERT(rv->pages == m,
426 ("vm_reserv_alloc_contig: reserv %p's pages is corrupted",
428 KASSERT(rv->object == NULL,
429 ("vm_reserv_alloc_contig: reserv %p isn't free", rv));
430 LIST_INSERT_HEAD(&object->rvq, rv, objq);
433 KASSERT(rv->popcnt == 0,
434 ("vm_reserv_alloc_contig: reserv %p's popcnt is corrupted",
436 KASSERT(!rv->inpartpopq,
437 ("vm_reserv_alloc_contig: reserv %p's inpartpopq is TRUE",
439 n = ulmin(VM_LEVEL_0_NPAGES - index, npages);
440 for (i = 0; i < n; i++)
441 vm_reserv_populate(rv);
444 m_ret = &rv->pages[index];
447 m += VM_LEVEL_0_NPAGES;
448 first += VM_LEVEL_0_NPAGES;
449 allocpages -= VM_LEVEL_0_NPAGES;
450 } while (allocpages > 0);
454 * Found a matching reservation.
457 index = VM_RESERV_INDEX(object, pindex);
458 /* Does the allocation fit within the reservation? */
459 if (index + npages > VM_LEVEL_0_NPAGES)
461 m = &rv->pages[index];
462 pa = VM_PAGE_TO_PHYS(m);
463 if (pa < low || pa + size > high || (pa & (alignment - 1)) != 0 ||
464 ((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0)
466 /* Handle vm_page_rename(m, new_object, ...). */
467 for (i = 0; i < npages; i++)
468 if ((rv->pages[index + i].flags & (PG_CACHED | PG_FREE)) == 0)
470 for (i = 0; i < npages; i++)
471 vm_reserv_populate(rv);
476 * Allocates a page from an existing or newly-created reservation.
478 * The object and free page queue must be locked.
481 vm_reserv_alloc_page(vm_object_t object, vm_pindex_t pindex)
483 vm_page_t m, mpred, msucc;
484 vm_pindex_t first, leftcap, rightcap;
487 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
488 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
491 * Is a reservation fundamentally impossible?
493 if (pindex < VM_RESERV_INDEX(object, pindex) ||
494 pindex >= object->size)
498 * Look for an existing reservation.
500 mpred = vm_radix_lookup_le(&object->rtree, pindex);
502 KASSERT(mpred->pindex < pindex,
503 ("vm_reserv_alloc_page: pindex already allocated"));
504 rv = vm_reserv_from_page(mpred);
505 if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
507 msucc = TAILQ_NEXT(mpred, listq);
509 msucc = vm_radix_lookup_ge(&object->rtree, pindex);
511 KASSERT(msucc->pindex > pindex,
512 ("vm_reserv_alloc_page: pindex already allocated"));
513 rv = vm_reserv_from_page(msucc);
514 if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
519 * Could a reservation fit between the first index to the left that
520 * can be used and the first index to the right that cannot be used?
522 first = pindex - VM_RESERV_INDEX(object, pindex);
524 if ((rv = vm_reserv_from_page(mpred))->object != object)
525 leftcap = mpred->pindex + 1;
527 leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
532 if ((rv = vm_reserv_from_page(msucc))->object != object)
533 rightcap = msucc->pindex;
535 rightcap = rv->pindex;
536 if (first + VM_LEVEL_0_NPAGES > rightcap)
541 * Would a new reservation extend past the end of the object?
543 if (first + VM_LEVEL_0_NPAGES > object->size) {
545 * Don't allocate a new reservation if the object is a vnode or
546 * backed by another object that is a vnode.
548 if (object->type == OBJT_VNODE ||
549 (object->backing_object != NULL &&
550 object->backing_object->type == OBJT_VNODE))
552 /* Speculate that the object may grow. */
556 * Allocate and populate the new reservation.
558 m = vm_phys_alloc_pages(VM_FREEPOOL_DEFAULT, VM_LEVEL_0_ORDER);
561 rv = vm_reserv_from_page(m);
562 KASSERT(rv->pages == m,
563 ("vm_reserv_alloc_page: reserv %p's pages is corrupted", rv));
564 KASSERT(rv->object == NULL,
565 ("vm_reserv_alloc_page: reserv %p isn't free", rv));
566 LIST_INSERT_HEAD(&object->rvq, rv, objq);
569 KASSERT(rv->popcnt == 0,
570 ("vm_reserv_alloc_page: reserv %p's popcnt is corrupted", rv));
571 KASSERT(!rv->inpartpopq,
572 ("vm_reserv_alloc_page: reserv %p's inpartpopq is TRUE", rv));
573 vm_reserv_populate(rv);
574 return (&rv->pages[VM_RESERV_INDEX(object, pindex)]);
577 * Found a matching reservation.
580 m = &rv->pages[VM_RESERV_INDEX(object, pindex)];
581 /* Handle vm_page_rename(m, new_object, ...). */
582 if ((m->flags & (PG_CACHED | PG_FREE)) == 0)
584 vm_reserv_populate(rv);
589 * Breaks all reservations belonging to the given object.
592 vm_reserv_break_all(vm_object_t object)
597 mtx_lock(&vm_page_queue_free_mtx);
598 while ((rv = LIST_FIRST(&object->rvq)) != NULL) {
599 KASSERT(rv->object == object,
600 ("vm_reserv_break_all: reserv %p is corrupted", rv));
601 if (rv->inpartpopq) {
602 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
603 rv->inpartpopq = FALSE;
605 LIST_REMOVE(rv, objq);
607 for (i = 0; i < VM_LEVEL_0_NPAGES; i++) {
608 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
609 vm_phys_free_pages(&rv->pages[i], 0);
613 KASSERT(rv->popcnt == 0,
614 ("vm_reserv_break_all: reserv %p's popcnt is corrupted",
618 mtx_unlock(&vm_page_queue_free_mtx);
622 * Frees the given page if it belongs to a reservation. Returns TRUE if the
623 * page is freed and FALSE otherwise.
625 * The free page queue lock must be held.
628 vm_reserv_free_page(vm_page_t m)
632 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
633 rv = vm_reserv_from_page(m);
634 if (rv->object == NULL)
636 if ((m->flags & PG_CACHED) != 0 && m->pool != VM_FREEPOOL_CACHE)
637 vm_phys_set_pool(VM_FREEPOOL_CACHE, rv->pages,
639 vm_reserv_depopulate(rv);
644 * Initializes the reservation management system. Specifically, initializes
645 * the reservation array.
647 * Requires that vm_page_array and first_page are initialized!
656 * Initialize the reservation array. Specifically, initialize the
657 * "pages" field for every element that has an underlying superpage.
659 for (i = 0; phys_avail[i + 1] != 0; i += 2) {
660 paddr = roundup2(phys_avail[i], VM_LEVEL_0_SIZE);
661 while (paddr + VM_LEVEL_0_SIZE <= phys_avail[i + 1]) {
662 vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT].pages =
663 PHYS_TO_VM_PAGE(paddr);
664 paddr += VM_LEVEL_0_SIZE;
670 * Returns a reservation level if the given page belongs to a fully-populated
671 * reservation and -1 otherwise.
674 vm_reserv_level_iffullpop(vm_page_t m)
678 rv = vm_reserv_from_page(m);
679 return (rv->popcnt == VM_LEVEL_0_NPAGES ? 0 : -1);
683 * Prepare for the reactivation of a cached page.
685 * First, suppose that the given page "m" was allocated individually, i.e., not
686 * as part of a reservation, and cached. Then, suppose a reservation
687 * containing "m" is allocated by the same object. Although "m" and the
688 * reservation belong to the same object, "m"'s pindex may not match the
691 * The free page queue must be locked.
694 vm_reserv_reactivate_page(vm_page_t m)
699 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
700 rv = vm_reserv_from_page(m);
701 if (rv->object == NULL)
703 KASSERT((m->flags & PG_CACHED) != 0,
704 ("vm_reserv_uncache_page: page %p is not cached", m));
705 if (m->object == rv->object &&
706 m->pindex - rv->pindex == VM_RESERV_INDEX(m->object, m->pindex))
707 vm_reserv_populate(rv);
709 KASSERT(rv->inpartpopq,
710 ("vm_reserv_uncache_page: reserv %p's inpartpopq is FALSE",
712 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
713 rv->inpartpopq = FALSE;
714 LIST_REMOVE(rv, objq);
716 /* Don't vm_phys_free_pages(m, 0). */
717 m_index = m - rv->pages;
718 for (i = 0; i < m_index; i++) {
719 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
720 vm_phys_free_pages(&rv->pages[i], 0);
724 for (i++; i < VM_LEVEL_0_NPAGES; i++) {
725 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
726 vm_phys_free_pages(&rv->pages[i], 0);
730 KASSERT(rv->popcnt == 0,
731 ("vm_reserv_uncache_page: reserv %p's popcnt is corrupted",
739 * Breaks the given partially-populated reservation, releasing its cached and
740 * free pages to the physical memory allocator.
742 * The free page queue lock must be held.
745 vm_reserv_reclaim(vm_reserv_t rv)
749 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
750 KASSERT(rv->inpartpopq,
751 ("vm_reserv_reclaim: reserv %p's inpartpopq is corrupted", rv));
752 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
753 rv->inpartpopq = FALSE;
754 KASSERT(rv->object != NULL,
755 ("vm_reserv_reclaim: reserv %p is free", rv));
756 LIST_REMOVE(rv, objq);
758 for (i = 0; i < VM_LEVEL_0_NPAGES; i++) {
759 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
760 vm_phys_free_pages(&rv->pages[i], 0);
764 KASSERT(rv->popcnt == 0,
765 ("vm_reserv_reclaim: reserv %p's popcnt is corrupted", rv));
766 vm_reserv_reclaimed++;
770 * Breaks the reservation at the head of the partially-populated reservation
771 * queue, releasing its cached and free pages to the physical memory
772 * allocator. Returns TRUE if a reservation is broken and FALSE otherwise.
774 * The free page queue lock must be held.
777 vm_reserv_reclaim_inactive(void)
781 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
782 if ((rv = TAILQ_FIRST(&vm_rvq_partpop)) != NULL) {
783 vm_reserv_reclaim(rv);
790 * Searches the partially-populated reservation queue for the least recently
791 * active reservation with unused pages, i.e., cached or free, that satisfy the
792 * given request for contiguous physical memory. If a satisfactory reservation
793 * is found, it is broken. Returns TRUE if a reservation is broken and FALSE
796 * The free page queue lock must be held.
799 vm_reserv_reclaim_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
800 u_long alignment, vm_paddr_t boundary)
802 vm_paddr_t pa, pa_length, size;
806 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
807 if (npages > VM_LEVEL_0_NPAGES - 1)
809 size = npages << PAGE_SHIFT;
810 TAILQ_FOREACH(rv, &vm_rvq_partpop, partpopq) {
811 pa = VM_PAGE_TO_PHYS(&rv->pages[VM_LEVEL_0_NPAGES - 1]);
812 if (pa + PAGE_SIZE - size < low) {
813 /* this entire reservation is too low; go to next */
817 for (i = 0; i < VM_LEVEL_0_NPAGES; i++)
818 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0) {
819 pa_length += PAGE_SIZE;
820 if (pa_length == PAGE_SIZE) {
821 pa = VM_PAGE_TO_PHYS(&rv->pages[i]);
822 if (pa + size > high) {
823 /* skip to next reservation */
825 } else if (pa < low ||
826 (pa & (alignment - 1)) != 0 ||
827 ((pa ^ (pa + size - 1)) &
828 ~(boundary - 1)) != 0)
831 if (pa_length >= size) {
832 vm_reserv_reclaim(rv);
842 * Transfers the reservation underlying the given page to a new object.
844 * The object must be locked.
847 vm_reserv_rename(vm_page_t m, vm_object_t new_object, vm_object_t old_object,
848 vm_pindex_t old_object_offset)
852 VM_OBJECT_LOCK_ASSERT(new_object, MA_OWNED);
853 rv = vm_reserv_from_page(m);
854 if (rv->object == old_object) {
855 mtx_lock(&vm_page_queue_free_mtx);
856 if (rv->object == old_object) {
857 LIST_REMOVE(rv, objq);
858 LIST_INSERT_HEAD(&new_object->rvq, rv, objq);
859 rv->object = new_object;
860 rv->pindex -= old_object_offset;
862 mtx_unlock(&vm_page_queue_free_mtx);
867 * Allocates the virtual and physical memory required by the reservation
868 * management system's data structures, in particular, the reservation array.
871 vm_reserv_startup(vm_offset_t *vaddr, vm_paddr_t end, vm_paddr_t high_water)
877 * Calculate the size (in bytes) of the reservation array. Round up
878 * from "high_water" because every small page is mapped to an element
879 * in the reservation array based on its physical address. Thus, the
880 * number of elements in the reservation array can be greater than the
881 * number of superpages.
883 size = howmany(high_water, VM_LEVEL_0_SIZE) * sizeof(struct vm_reserv);
886 * Allocate and map the physical memory for the reservation array. The
887 * next available virtual address is returned by reference.
889 new_end = end - round_page(size);
890 vm_reserv_array = (void *)(uintptr_t)pmap_map(vaddr, new_end, end,
891 VM_PROT_READ | VM_PROT_WRITE);
892 bzero(vm_reserv_array, size);
895 * Return the next available physical address.
900 #endif /* VM_NRESERVLEVEL > 0 */