2 * Copyright (c) 2002-2006 Rice University
3 * Copyright (c) 2007 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 * Physical memory system implementation
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("$FreeBSD$");
45 #include <sys/param.h>
46 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mutex.h>
54 #include <sys/queue.h>
56 #include <sys/sysctl.h>
57 #include <sys/vmmeter.h>
62 #include <vm/vm_param.h>
63 #include <vm/vm_kern.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_page.h>
66 #include <vm/vm_phys.h>
68 _Static_assert(sizeof(long) * NBBY >= VM_PHYSSEG_MAX,
69 "Too many physsegs.");
71 struct mem_affinity *mem_affinity;
75 struct vm_phys_seg vm_phys_segs[VM_PHYSSEG_MAX];
78 #define VM_PHYS_FICTITIOUS_NSEGS 8
79 static struct vm_phys_fictitious_seg {
83 } vm_phys_fictitious_segs[VM_PHYS_FICTITIOUS_NSEGS];
84 static struct mtx vm_phys_fictitious_reg_mtx;
85 MALLOC_DEFINE(M_FICT_PAGES, "vm_fictitious", "Fictitious VM pages");
87 static struct vm_freelist
88 vm_phys_free_queues[MAXMEMDOM][VM_NFREELIST][VM_NFREEPOOL][VM_NFREEORDER];
90 static int vm_nfreelists = VM_FREELIST_DEFAULT + 1;
92 static int cnt_prezero;
93 SYSCTL_INT(_vm_stats_misc, OID_AUTO, cnt_prezero, CTLFLAG_RD,
94 &cnt_prezero, 0, "The number of physical pages prezeroed at idle time");
96 static int sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS);
97 SYSCTL_OID(_vm, OID_AUTO, phys_free, CTLTYPE_STRING | CTLFLAG_RD,
98 NULL, 0, sysctl_vm_phys_free, "A", "Phys Free Info");
100 static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS);
101 SYSCTL_OID(_vm, OID_AUTO, phys_segs, CTLTYPE_STRING | CTLFLAG_RD,
102 NULL, 0, sysctl_vm_phys_segs, "A", "Phys Seg Info");
104 SYSCTL_INT(_vm, OID_AUTO, ndomains, CTLFLAG_RD,
105 &vm_ndomains, 0, "Number of physical memory domains available.");
107 static vm_page_t vm_phys_alloc_domain_pages(int domain, int flind, int pool,
109 static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind,
111 static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind);
112 static int vm_phys_paddr_to_segind(vm_paddr_t pa);
113 static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl,
117 vm_rr_selectdomain(void)
125 td->td_dom_rr_idx %= vm_ndomains;
126 return (td->td_dom_rr_idx);
133 vm_phys_domain_intersects(long mask, vm_paddr_t low, vm_paddr_t high)
135 struct vm_phys_seg *s;
138 while ((idx = ffsl(mask)) != 0) {
139 idx--; /* ffsl counts from 1 */
140 mask &= ~(1UL << idx);
141 s = &vm_phys_segs[idx];
142 if (low < s->end && high > s->start)
149 * Outputs the state of the physical memory allocator, specifically,
150 * the amount of physical memory in each free list.
153 sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS)
156 struct vm_freelist *fl;
157 int dom, error, flind, oind, pind;
159 error = sysctl_wire_old_buffer(req, 0);
162 sbuf_new_for_sysctl(&sbuf, NULL, 128 * vm_ndomains, req);
163 for (dom = 0; dom < vm_ndomains; dom++) {
164 sbuf_printf(&sbuf,"\nDOMAIN %d:\n", dom);
165 for (flind = 0; flind < vm_nfreelists; flind++) {
166 sbuf_printf(&sbuf, "\nFREE LIST %d:\n"
167 "\n ORDER (SIZE) | NUMBER"
169 for (pind = 0; pind < VM_NFREEPOOL; pind++)
170 sbuf_printf(&sbuf, " | POOL %d", pind);
171 sbuf_printf(&sbuf, "\n-- ");
172 for (pind = 0; pind < VM_NFREEPOOL; pind++)
173 sbuf_printf(&sbuf, "-- -- ");
174 sbuf_printf(&sbuf, "--\n");
175 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
176 sbuf_printf(&sbuf, " %2d (%6dK)", oind,
177 1 << (PAGE_SHIFT - 10 + oind));
178 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
179 fl = vm_phys_free_queues[dom][flind][pind];
180 sbuf_printf(&sbuf, " | %6d",
183 sbuf_printf(&sbuf, "\n");
187 error = sbuf_finish(&sbuf);
193 * Outputs the set of physical memory segments.
196 sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS)
199 struct vm_phys_seg *seg;
202 error = sysctl_wire_old_buffer(req, 0);
205 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
206 for (segind = 0; segind < vm_phys_nsegs; segind++) {
207 sbuf_printf(&sbuf, "\nSEGMENT %d:\n\n", segind);
208 seg = &vm_phys_segs[segind];
209 sbuf_printf(&sbuf, "start: %#jx\n",
210 (uintmax_t)seg->start);
211 sbuf_printf(&sbuf, "end: %#jx\n",
212 (uintmax_t)seg->end);
213 sbuf_printf(&sbuf, "domain: %d\n", seg->domain);
214 sbuf_printf(&sbuf, "free list: %p\n", seg->free_queues);
216 error = sbuf_finish(&sbuf);
222 vm_freelist_add(struct vm_freelist *fl, vm_page_t m, int order, int tail)
227 TAILQ_INSERT_TAIL(&fl[order].pl, m, plinks.q);
229 TAILQ_INSERT_HEAD(&fl[order].pl, m, plinks.q);
234 vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order)
237 TAILQ_REMOVE(&fl[order].pl, m, plinks.q);
239 m->order = VM_NFREEORDER;
243 * Create a physical memory segment.
246 _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind, int domain)
248 struct vm_phys_seg *seg;
249 #ifdef VM_PHYSSEG_SPARSE
254 for (segind = 0; segind < vm_phys_nsegs; segind++) {
255 seg = &vm_phys_segs[segind];
256 pages += atop(seg->end - seg->start);
259 KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX,
260 ("vm_phys_create_seg: increase VM_PHYSSEG_MAX"));
261 KASSERT(domain < vm_ndomains,
262 ("vm_phys_create_seg: invalid domain provided"));
263 seg = &vm_phys_segs[vm_phys_nsegs++];
266 seg->domain = domain;
267 #ifdef VM_PHYSSEG_SPARSE
268 seg->first_page = &vm_page_array[pages];
270 seg->first_page = PHYS_TO_VM_PAGE(start);
272 seg->free_queues = &vm_phys_free_queues[domain][flind];
276 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind)
280 if (mem_affinity == NULL) {
281 _vm_phys_create_seg(start, end, flind, 0);
286 if (mem_affinity[i].end == 0)
287 panic("Reached end of affinity info");
288 if (mem_affinity[i].end <= start)
290 if (mem_affinity[i].start > start)
291 panic("No affinity info for start %jx",
293 if (mem_affinity[i].end >= end) {
294 _vm_phys_create_seg(start, end, flind,
295 mem_affinity[i].domain);
298 _vm_phys_create_seg(start, mem_affinity[i].end, flind,
299 mem_affinity[i].domain);
300 start = mem_affinity[i].end;
305 * Initialize the physical memory allocator.
310 struct vm_freelist *fl;
311 int dom, flind, i, oind, pind;
313 for (i = 0; phys_avail[i + 1] != 0; i += 2) {
314 #ifdef VM_FREELIST_ISADMA
315 if (phys_avail[i] < 16777216) {
316 if (phys_avail[i + 1] > 16777216) {
317 vm_phys_create_seg(phys_avail[i], 16777216,
319 vm_phys_create_seg(16777216, phys_avail[i + 1],
320 VM_FREELIST_DEFAULT);
322 vm_phys_create_seg(phys_avail[i],
323 phys_avail[i + 1], VM_FREELIST_ISADMA);
325 if (VM_FREELIST_ISADMA >= vm_nfreelists)
326 vm_nfreelists = VM_FREELIST_ISADMA + 1;
329 #ifdef VM_FREELIST_HIGHMEM
330 if (phys_avail[i + 1] > VM_HIGHMEM_ADDRESS) {
331 if (phys_avail[i] < VM_HIGHMEM_ADDRESS) {
332 vm_phys_create_seg(phys_avail[i],
333 VM_HIGHMEM_ADDRESS, VM_FREELIST_DEFAULT);
334 vm_phys_create_seg(VM_HIGHMEM_ADDRESS,
335 phys_avail[i + 1], VM_FREELIST_HIGHMEM);
337 vm_phys_create_seg(phys_avail[i],
338 phys_avail[i + 1], VM_FREELIST_HIGHMEM);
340 if (VM_FREELIST_HIGHMEM >= vm_nfreelists)
341 vm_nfreelists = VM_FREELIST_HIGHMEM + 1;
344 vm_phys_create_seg(phys_avail[i], phys_avail[i + 1],
345 VM_FREELIST_DEFAULT);
347 for (dom = 0; dom < vm_ndomains; dom++) {
348 for (flind = 0; flind < vm_nfreelists; flind++) {
349 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
350 fl = vm_phys_free_queues[dom][flind][pind];
351 for (oind = 0; oind < VM_NFREEORDER; oind++)
352 TAILQ_INIT(&fl[oind].pl);
356 mtx_init(&vm_phys_fictitious_reg_mtx, "vmfctr", NULL, MTX_DEF);
360 * Split a contiguous, power of two-sized set of physical pages.
363 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order)
367 while (oind > order) {
369 m_buddy = &m[1 << oind];
370 KASSERT(m_buddy->order == VM_NFREEORDER,
371 ("vm_phys_split_pages: page %p has unexpected order %d",
372 m_buddy, m_buddy->order));
373 vm_freelist_add(fl, m_buddy, oind, 0);
378 * Initialize a physical page and add it to the free lists.
381 vm_phys_add_page(vm_paddr_t pa)
384 struct vm_domain *vmd;
386 vm_cnt.v_page_count++;
387 m = vm_phys_paddr_to_vm_page(pa);
390 m->segind = vm_phys_paddr_to_segind(pa);
391 vmd = vm_phys_domain(m);
392 vmd->vmd_page_count++;
393 vmd->vmd_segs |= 1UL << m->segind;
394 KASSERT(m->order == VM_NFREEORDER,
395 ("vm_phys_add_page: page %p has unexpected order %d",
397 m->pool = VM_FREEPOOL_DEFAULT;
399 mtx_lock(&vm_page_queue_free_mtx);
400 vm_phys_freecnt_adj(m, 1);
401 vm_phys_free_pages(m, 0);
402 mtx_unlock(&vm_page_queue_free_mtx);
406 * Allocate a contiguous, power of two-sized set of physical pages
407 * from the free lists.
409 * The free page queues must be locked.
412 vm_phys_alloc_pages(int pool, int order)
415 int dom, domain, flind;
417 KASSERT(pool < VM_NFREEPOOL,
418 ("vm_phys_alloc_pages: pool %d is out of range", pool));
419 KASSERT(order < VM_NFREEORDER,
420 ("vm_phys_alloc_pages: order %d is out of range", order));
422 for (dom = 0; dom < vm_ndomains; dom++) {
423 domain = vm_rr_selectdomain();
424 for (flind = 0; flind < vm_nfreelists; flind++) {
425 m = vm_phys_alloc_domain_pages(domain, flind, pool,
435 * Find and dequeue a free page on the given free list, with the
436 * specified pool and order
439 vm_phys_alloc_freelist_pages(int flind, int pool, int order)
444 KASSERT(flind < VM_NFREELIST,
445 ("vm_phys_alloc_freelist_pages: freelist %d is out of range", flind));
446 KASSERT(pool < VM_NFREEPOOL,
447 ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
448 KASSERT(order < VM_NFREEORDER,
449 ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
451 for (dom = 0; dom < vm_ndomains; dom++) {
452 domain = vm_rr_selectdomain();
453 m = vm_phys_alloc_domain_pages(domain, flind, pool, order);
461 vm_phys_alloc_domain_pages(int domain, int flind, int pool, int order)
463 struct vm_freelist *fl;
464 struct vm_freelist *alt;
468 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
469 fl = &vm_phys_free_queues[domain][flind][pool][0];
470 for (oind = order; oind < VM_NFREEORDER; oind++) {
471 m = TAILQ_FIRST(&fl[oind].pl);
473 vm_freelist_rem(fl, m, oind);
474 vm_phys_split_pages(m, oind, fl, order);
480 * The given pool was empty. Find the largest
481 * contiguous, power-of-two-sized set of pages in any
482 * pool. Transfer these pages to the given pool, and
483 * use them to satisfy the allocation.
485 for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
486 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
487 alt = &vm_phys_free_queues[domain][flind][pind][0];
488 m = TAILQ_FIRST(&alt[oind].pl);
490 vm_freelist_rem(alt, m, oind);
491 vm_phys_set_pool(pool, m, oind);
492 vm_phys_split_pages(m, oind, fl, order);
501 * Find the vm_page corresponding to the given physical address.
504 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
506 struct vm_phys_seg *seg;
509 for (segind = 0; segind < vm_phys_nsegs; segind++) {
510 seg = &vm_phys_segs[segind];
511 if (pa >= seg->start && pa < seg->end)
512 return (&seg->first_page[atop(pa - seg->start)]);
518 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
520 struct vm_phys_fictitious_seg *seg;
525 for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
526 seg = &vm_phys_fictitious_segs[segind];
527 if (pa >= seg->start && pa < seg->end) {
528 m = &seg->first_page[atop(pa - seg->start)];
529 KASSERT((m->flags & PG_FICTITIOUS) != 0,
530 ("%p not fictitious", m));
538 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
539 vm_memattr_t memattr)
541 struct vm_phys_fictitious_seg *seg;
545 #ifdef VM_PHYSSEG_DENSE
550 page_count = (end - start) / PAGE_SIZE;
552 #ifdef VM_PHYSSEG_DENSE
554 if (pi >= first_page && pi < vm_page_array_size + first_page) {
555 if (atop(end) >= vm_page_array_size + first_page)
557 fp = &vm_page_array[pi - first_page];
562 fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
564 #ifdef VM_PHYSSEG_DENSE
568 for (i = 0; i < page_count; i++) {
569 vm_page_initfake(&fp[i], start + PAGE_SIZE * i, memattr);
570 fp[i].oflags &= ~VPO_UNMANAGED;
571 fp[i].busy_lock = VPB_UNBUSIED;
573 mtx_lock(&vm_phys_fictitious_reg_mtx);
574 for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
575 seg = &vm_phys_fictitious_segs[segind];
576 if (seg->start == 0 && seg->end == 0) {
579 seg->first_page = fp;
580 mtx_unlock(&vm_phys_fictitious_reg_mtx);
584 mtx_unlock(&vm_phys_fictitious_reg_mtx);
585 #ifdef VM_PHYSSEG_DENSE
588 free(fp, M_FICT_PAGES);
593 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
595 struct vm_phys_fictitious_seg *seg;
598 #ifdef VM_PHYSSEG_DENSE
602 #ifdef VM_PHYSSEG_DENSE
606 mtx_lock(&vm_phys_fictitious_reg_mtx);
607 for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
608 seg = &vm_phys_fictitious_segs[segind];
609 if (seg->start == start && seg->end == end) {
610 seg->start = seg->end = 0;
611 fp = seg->first_page;
612 seg->first_page = NULL;
613 mtx_unlock(&vm_phys_fictitious_reg_mtx);
614 #ifdef VM_PHYSSEG_DENSE
615 if (pi < first_page || atop(end) >= vm_page_array_size)
617 free(fp, M_FICT_PAGES);
621 mtx_unlock(&vm_phys_fictitious_reg_mtx);
622 KASSERT(0, ("Unregistering not registered fictitious range"));
626 * Find the segment containing the given physical address.
629 vm_phys_paddr_to_segind(vm_paddr_t pa)
631 struct vm_phys_seg *seg;
634 for (segind = 0; segind < vm_phys_nsegs; segind++) {
635 seg = &vm_phys_segs[segind];
636 if (pa >= seg->start && pa < seg->end)
639 panic("vm_phys_paddr_to_segind: paddr %#jx is not in any segment" ,
644 * Free a contiguous, power of two-sized set of physical pages.
646 * The free page queues must be locked.
649 vm_phys_free_pages(vm_page_t m, int order)
651 struct vm_freelist *fl;
652 struct vm_phys_seg *seg;
656 KASSERT(m->order == VM_NFREEORDER,
657 ("vm_phys_free_pages: page %p has unexpected order %d",
659 KASSERT(m->pool < VM_NFREEPOOL,
660 ("vm_phys_free_pages: page %p has unexpected pool %d",
662 KASSERT(order < VM_NFREEORDER,
663 ("vm_phys_free_pages: order %d is out of range", order));
664 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
665 seg = &vm_phys_segs[m->segind];
666 if (order < VM_NFREEORDER - 1) {
667 pa = VM_PAGE_TO_PHYS(m);
669 pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
670 if (pa < seg->start || pa >= seg->end)
672 m_buddy = &seg->first_page[atop(pa - seg->start)];
673 if (m_buddy->order != order)
675 fl = (*seg->free_queues)[m_buddy->pool];
676 vm_freelist_rem(fl, m_buddy, order);
677 if (m_buddy->pool != m->pool)
678 vm_phys_set_pool(m->pool, m_buddy, order);
680 pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
681 m = &seg->first_page[atop(pa - seg->start)];
682 } while (order < VM_NFREEORDER - 1);
684 fl = (*seg->free_queues)[m->pool];
685 vm_freelist_add(fl, m, order, 1);
689 * Free a contiguous, arbitrarily sized set of physical pages.
691 * The free page queues must be locked.
694 vm_phys_free_contig(vm_page_t m, u_long npages)
700 * Avoid unnecessary coalescing by freeing the pages in the largest
701 * possible power-of-two-sized subsets.
703 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
704 for (;; npages -= n) {
706 * Unsigned "min" is used here so that "order" is assigned
707 * "VM_NFREEORDER - 1" when "m"'s physical address is zero
708 * or the low-order bits of its physical address are zero
709 * because the size of a physical address exceeds the size of
712 order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
717 vm_phys_free_pages(m, order);
720 /* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */
721 for (; npages > 0; npages -= n) {
722 order = flsl(npages) - 1;
724 vm_phys_free_pages(m, order);
730 * Set the pool for a contiguous, power of two-sized set of physical pages.
733 vm_phys_set_pool(int pool, vm_page_t m, int order)
737 for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
742 * Search for the given physical page "m" in the free lists. If the search
743 * succeeds, remove "m" from the free lists and return TRUE. Otherwise, return
744 * FALSE, indicating that "m" is not in the free lists.
746 * The free page queues must be locked.
749 vm_phys_unfree_page(vm_page_t m)
751 struct vm_freelist *fl;
752 struct vm_phys_seg *seg;
753 vm_paddr_t pa, pa_half;
754 vm_page_t m_set, m_tmp;
757 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
760 * First, find the contiguous, power of two-sized set of free
761 * physical pages containing the given physical page "m" and
762 * assign it to "m_set".
764 seg = &vm_phys_segs[m->segind];
765 for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
766 order < VM_NFREEORDER - 1; ) {
768 pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
769 if (pa >= seg->start)
770 m_set = &seg->first_page[atop(pa - seg->start)];
774 if (m_set->order < order)
776 if (m_set->order == VM_NFREEORDER)
778 KASSERT(m_set->order < VM_NFREEORDER,
779 ("vm_phys_unfree_page: page %p has unexpected order %d",
780 m_set, m_set->order));
783 * Next, remove "m_set" from the free lists. Finally, extract
784 * "m" from "m_set" using an iterative algorithm: While "m_set"
785 * is larger than a page, shrink "m_set" by returning the half
786 * of "m_set" that does not contain "m" to the free lists.
788 fl = (*seg->free_queues)[m_set->pool];
789 order = m_set->order;
790 vm_freelist_rem(fl, m_set, order);
793 pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
794 if (m->phys_addr < pa_half)
795 m_tmp = &seg->first_page[atop(pa_half - seg->start)];
798 m_set = &seg->first_page[atop(pa_half - seg->start)];
800 vm_freelist_add(fl, m_tmp, order, 0);
802 KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
807 * Try to zero one physical page. Used by an idle priority thread.
810 vm_phys_zero_pages_idle(void)
812 static struct vm_freelist *fl;
813 static int flind, oind, pind;
817 domain = vm_rr_selectdomain();
818 fl = vm_phys_free_queues[domain][0][0];
819 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
821 TAILQ_FOREACH_REVERSE(m, &fl[oind].pl, pglist, plinks.q) {
822 for (m_tmp = m; m_tmp < &m[1 << oind]; m_tmp++) {
823 if ((m_tmp->flags & (PG_CACHED | PG_ZERO)) == 0) {
824 vm_phys_unfree_page(m_tmp);
825 vm_phys_freecnt_adj(m, -1);
826 mtx_unlock(&vm_page_queue_free_mtx);
827 pmap_zero_page_idle(m_tmp);
828 m_tmp->flags |= PG_ZERO;
829 mtx_lock(&vm_page_queue_free_mtx);
830 vm_phys_freecnt_adj(m, 1);
831 vm_phys_free_pages(m_tmp, 0);
832 vm_page_zero_count++;
839 if (oind == VM_NFREEORDER) {
842 if (pind == VM_NFREEPOOL) {
845 if (flind == vm_nfreelists)
848 fl = vm_phys_free_queues[domain][flind][pind];
854 * Allocate a contiguous set of physical pages of the given size
855 * "npages" from the free lists. All of the physical pages must be at
856 * or above the given physical address "low" and below the given
857 * physical address "high". The given value "alignment" determines the
858 * alignment of the first physical page in the set. If the given value
859 * "boundary" is non-zero, then the set of physical pages cannot cross
860 * any physical address boundary that is a multiple of that value. Both
861 * "alignment" and "boundary" must be a power of two.
864 vm_phys_alloc_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
865 u_long alignment, vm_paddr_t boundary)
867 struct vm_freelist *fl;
868 struct vm_phys_seg *seg;
869 vm_paddr_t pa, pa_last, size;
872 int dom, domain, flind, oind, order, pind;
874 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
875 size = npages << PAGE_SHIFT;
877 ("vm_phys_alloc_contig: size must not be 0"));
878 KASSERT((alignment & (alignment - 1)) == 0,
879 ("vm_phys_alloc_contig: alignment must be a power of 2"));
880 KASSERT((boundary & (boundary - 1)) == 0,
881 ("vm_phys_alloc_contig: boundary must be a power of 2"));
882 /* Compute the queue that is the best fit for npages. */
883 for (order = 0; (1 << order) < npages; order++);
886 domain = vm_rr_selectdomain();
887 for (flind = 0; flind < vm_nfreelists; flind++) {
888 for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER; oind++) {
889 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
890 fl = &vm_phys_free_queues[domain][flind][pind][0];
891 TAILQ_FOREACH(m_ret, &fl[oind].pl, plinks.q) {
893 * A free list may contain physical pages
894 * from one or more segments.
896 seg = &vm_phys_segs[m_ret->segind];
897 if (seg->start > high ||
902 * Is the size of this allocation request
903 * larger than the largest block size?
905 if (order >= VM_NFREEORDER) {
907 * Determine if a sufficient number
908 * of subsequent blocks to satisfy
909 * the allocation request are free.
911 pa = VM_PAGE_TO_PHYS(m_ret);
914 pa += 1 << (PAGE_SHIFT + VM_NFREEORDER - 1);
917 if (pa < seg->start ||
920 m = &seg->first_page[atop(pa - seg->start)];
921 if (m->order != VM_NFREEORDER - 1)
924 /* If not, continue to the next block. */
930 * Determine if the blocks are within the given range,
931 * satisfy the given alignment, and do not cross the
934 pa = VM_PAGE_TO_PHYS(m_ret);
937 (pa & (alignment - 1)) == 0 &&
938 ((pa ^ (pa + size - 1)) & ~(boundary - 1)) == 0)
944 if (++dom < vm_ndomains)
948 for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
949 fl = (*seg->free_queues)[m->pool];
950 vm_freelist_rem(fl, m, m->order);
952 if (m_ret->pool != VM_FREEPOOL_DEFAULT)
953 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m_ret, oind);
954 fl = (*seg->free_queues)[m_ret->pool];
955 vm_phys_split_pages(m_ret, oind, fl, order);
956 /* Return excess pages to the free lists. */
957 npages_end = roundup2(npages, 1 << imin(oind, order));
958 if (npages < npages_end)
959 vm_phys_free_contig(&m_ret[npages], npages_end - npages);
965 * Show the number of physical pages in each of the free lists.
967 DB_SHOW_COMMAND(freepages, db_show_freepages)
969 struct vm_freelist *fl;
970 int flind, oind, pind, dom;
972 for (dom = 0; dom < vm_ndomains; dom++) {
973 db_printf("DOMAIN: %d\n", dom);
974 for (flind = 0; flind < vm_nfreelists; flind++) {
975 db_printf("FREE LIST %d:\n"
976 "\n ORDER (SIZE) | NUMBER"
978 for (pind = 0; pind < VM_NFREEPOOL; pind++)
979 db_printf(" | POOL %d", pind);
981 for (pind = 0; pind < VM_NFREEPOOL; pind++)
984 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
985 db_printf(" %2.2d (%6.6dK)", oind,
986 1 << (PAGE_SHIFT - 10 + oind));
987 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
988 fl = vm_phys_free_queues[dom][flind][pind];
989 db_printf(" | %6.6d", fl[oind].lcnt);