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>
55 #include <sys/rwlock.h>
57 #include <sys/sysctl.h>
59 #include <sys/vmmeter.h>
64 #include <vm/vm_param.h>
65 #include <vm/vm_kern.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_page.h>
68 #include <vm/vm_phys.h>
70 _Static_assert(sizeof(long) * NBBY >= VM_PHYSSEG_MAX,
71 "Too many physsegs.");
73 struct mem_affinity *mem_affinity;
77 struct vm_phys_seg vm_phys_segs[VM_PHYSSEG_MAX];
80 struct vm_phys_fictitious_seg;
81 static int vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *,
82 struct vm_phys_fictitious_seg *);
84 RB_HEAD(fict_tree, vm_phys_fictitious_seg) vm_phys_fictitious_tree =
85 RB_INITIALIZER(_vm_phys_fictitious_tree);
87 struct vm_phys_fictitious_seg {
88 RB_ENTRY(vm_phys_fictitious_seg) node;
89 /* Memory region data */
95 RB_GENERATE_STATIC(fict_tree, vm_phys_fictitious_seg, node,
96 vm_phys_fictitious_cmp);
98 static struct rwlock vm_phys_fictitious_reg_lock;
99 MALLOC_DEFINE(M_FICT_PAGES, "vm_fictitious", "Fictitious VM pages");
101 static struct vm_freelist
102 vm_phys_free_queues[MAXMEMDOM][VM_NFREELIST][VM_NFREEPOOL][VM_NFREEORDER];
104 static int vm_nfreelists = VM_FREELIST_DEFAULT + 1;
106 static int cnt_prezero;
107 SYSCTL_INT(_vm_stats_misc, OID_AUTO, cnt_prezero, CTLFLAG_RD,
108 &cnt_prezero, 0, "The number of physical pages prezeroed at idle time");
110 static int sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS);
111 SYSCTL_OID(_vm, OID_AUTO, phys_free, CTLTYPE_STRING | CTLFLAG_RD,
112 NULL, 0, sysctl_vm_phys_free, "A", "Phys Free Info");
114 static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS);
115 SYSCTL_OID(_vm, OID_AUTO, phys_segs, CTLTYPE_STRING | CTLFLAG_RD,
116 NULL, 0, sysctl_vm_phys_segs, "A", "Phys Seg Info");
118 SYSCTL_INT(_vm, OID_AUTO, ndomains, CTLFLAG_RD,
119 &vm_ndomains, 0, "Number of physical memory domains available.");
121 static vm_page_t vm_phys_alloc_domain_pages(int domain, int flind, int pool,
123 static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind,
125 static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind);
126 static int vm_phys_paddr_to_segind(vm_paddr_t pa);
127 static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl,
131 * Red-black tree helpers for vm fictitious range management.
134 vm_phys_fictitious_in_range(struct vm_phys_fictitious_seg *p,
135 struct vm_phys_fictitious_seg *range)
138 KASSERT(range->start != 0 && range->end != 0,
139 ("Invalid range passed on search for vm_fictitious page"));
140 if (p->start >= range->end)
142 if (p->start < range->start)
149 vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *p1,
150 struct vm_phys_fictitious_seg *p2)
153 /* Check if this is a search for a page */
155 return (vm_phys_fictitious_in_range(p1, p2));
157 KASSERT(p2->end != 0,
158 ("Invalid range passed as second parameter to vm fictitious comparison"));
160 /* Searching to add a new range */
161 if (p1->end <= p2->start)
163 if (p1->start >= p2->end)
166 panic("Trying to add overlapping vm fictitious ranges:\n"
167 "[%#jx:%#jx] and [%#jx:%#jx]", (uintmax_t)p1->start,
168 (uintmax_t)p1->end, (uintmax_t)p2->start, (uintmax_t)p2->end);
172 vm_rr_selectdomain(void)
180 td->td_dom_rr_idx %= vm_ndomains;
181 return (td->td_dom_rr_idx);
188 vm_phys_domain_intersects(long mask, vm_paddr_t low, vm_paddr_t high)
190 struct vm_phys_seg *s;
193 while ((idx = ffsl(mask)) != 0) {
194 idx--; /* ffsl counts from 1 */
195 mask &= ~(1UL << idx);
196 s = &vm_phys_segs[idx];
197 if (low < s->end && high > s->start)
204 * Outputs the state of the physical memory allocator, specifically,
205 * the amount of physical memory in each free list.
208 sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS)
211 struct vm_freelist *fl;
212 int dom, error, flind, oind, pind;
214 error = sysctl_wire_old_buffer(req, 0);
217 sbuf_new_for_sysctl(&sbuf, NULL, 128 * vm_ndomains, req);
218 for (dom = 0; dom < vm_ndomains; dom++) {
219 sbuf_printf(&sbuf,"\nDOMAIN %d:\n", dom);
220 for (flind = 0; flind < vm_nfreelists; flind++) {
221 sbuf_printf(&sbuf, "\nFREE LIST %d:\n"
222 "\n ORDER (SIZE) | NUMBER"
224 for (pind = 0; pind < VM_NFREEPOOL; pind++)
225 sbuf_printf(&sbuf, " | POOL %d", pind);
226 sbuf_printf(&sbuf, "\n-- ");
227 for (pind = 0; pind < VM_NFREEPOOL; pind++)
228 sbuf_printf(&sbuf, "-- -- ");
229 sbuf_printf(&sbuf, "--\n");
230 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
231 sbuf_printf(&sbuf, " %2d (%6dK)", oind,
232 1 << (PAGE_SHIFT - 10 + oind));
233 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
234 fl = vm_phys_free_queues[dom][flind][pind];
235 sbuf_printf(&sbuf, " | %6d",
238 sbuf_printf(&sbuf, "\n");
242 error = sbuf_finish(&sbuf);
248 * Outputs the set of physical memory segments.
251 sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS)
254 struct vm_phys_seg *seg;
257 error = sysctl_wire_old_buffer(req, 0);
260 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
261 for (segind = 0; segind < vm_phys_nsegs; segind++) {
262 sbuf_printf(&sbuf, "\nSEGMENT %d:\n\n", segind);
263 seg = &vm_phys_segs[segind];
264 sbuf_printf(&sbuf, "start: %#jx\n",
265 (uintmax_t)seg->start);
266 sbuf_printf(&sbuf, "end: %#jx\n",
267 (uintmax_t)seg->end);
268 sbuf_printf(&sbuf, "domain: %d\n", seg->domain);
269 sbuf_printf(&sbuf, "free list: %p\n", seg->free_queues);
271 error = sbuf_finish(&sbuf);
277 vm_freelist_add(struct vm_freelist *fl, vm_page_t m, int order, int tail)
282 TAILQ_INSERT_TAIL(&fl[order].pl, m, plinks.q);
284 TAILQ_INSERT_HEAD(&fl[order].pl, m, plinks.q);
289 vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order)
292 TAILQ_REMOVE(&fl[order].pl, m, plinks.q);
294 m->order = VM_NFREEORDER;
298 * Create a physical memory segment.
301 _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind, int domain)
303 struct vm_phys_seg *seg;
304 #ifdef VM_PHYSSEG_SPARSE
309 for (segind = 0; segind < vm_phys_nsegs; segind++) {
310 seg = &vm_phys_segs[segind];
311 pages += atop(seg->end - seg->start);
314 KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX,
315 ("vm_phys_create_seg: increase VM_PHYSSEG_MAX"));
316 KASSERT(domain < vm_ndomains,
317 ("vm_phys_create_seg: invalid domain provided"));
318 seg = &vm_phys_segs[vm_phys_nsegs++];
321 seg->domain = domain;
322 #ifdef VM_PHYSSEG_SPARSE
323 seg->first_page = &vm_page_array[pages];
325 seg->first_page = PHYS_TO_VM_PAGE(start);
327 seg->free_queues = &vm_phys_free_queues[domain][flind];
331 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind)
335 if (mem_affinity == NULL) {
336 _vm_phys_create_seg(start, end, flind, 0);
341 if (mem_affinity[i].end == 0)
342 panic("Reached end of affinity info");
343 if (mem_affinity[i].end <= start)
345 if (mem_affinity[i].start > start)
346 panic("No affinity info for start %jx",
348 if (mem_affinity[i].end >= end) {
349 _vm_phys_create_seg(start, end, flind,
350 mem_affinity[i].domain);
353 _vm_phys_create_seg(start, mem_affinity[i].end, flind,
354 mem_affinity[i].domain);
355 start = mem_affinity[i].end;
360 * Initialize the physical memory allocator.
365 struct vm_freelist *fl;
366 int dom, flind, i, oind, pind;
368 for (i = 0; phys_avail[i + 1] != 0; i += 2) {
369 #ifdef VM_FREELIST_ISADMA
370 if (phys_avail[i] < 16777216) {
371 if (phys_avail[i + 1] > 16777216) {
372 vm_phys_create_seg(phys_avail[i], 16777216,
374 vm_phys_create_seg(16777216, phys_avail[i + 1],
375 VM_FREELIST_DEFAULT);
377 vm_phys_create_seg(phys_avail[i],
378 phys_avail[i + 1], VM_FREELIST_ISADMA);
380 if (VM_FREELIST_ISADMA >= vm_nfreelists)
381 vm_nfreelists = VM_FREELIST_ISADMA + 1;
384 #ifdef VM_FREELIST_HIGHMEM
385 if (phys_avail[i + 1] > VM_HIGHMEM_ADDRESS) {
386 if (phys_avail[i] < VM_HIGHMEM_ADDRESS) {
387 vm_phys_create_seg(phys_avail[i],
388 VM_HIGHMEM_ADDRESS, VM_FREELIST_DEFAULT);
389 vm_phys_create_seg(VM_HIGHMEM_ADDRESS,
390 phys_avail[i + 1], VM_FREELIST_HIGHMEM);
392 vm_phys_create_seg(phys_avail[i],
393 phys_avail[i + 1], VM_FREELIST_HIGHMEM);
395 if (VM_FREELIST_HIGHMEM >= vm_nfreelists)
396 vm_nfreelists = VM_FREELIST_HIGHMEM + 1;
399 vm_phys_create_seg(phys_avail[i], phys_avail[i + 1],
400 VM_FREELIST_DEFAULT);
402 for (dom = 0; dom < vm_ndomains; dom++) {
403 for (flind = 0; flind < vm_nfreelists; flind++) {
404 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
405 fl = vm_phys_free_queues[dom][flind][pind];
406 for (oind = 0; oind < VM_NFREEORDER; oind++)
407 TAILQ_INIT(&fl[oind].pl);
411 rw_init(&vm_phys_fictitious_reg_lock, "vmfctr");
415 * Split a contiguous, power of two-sized set of physical pages.
418 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order)
422 while (oind > order) {
424 m_buddy = &m[1 << oind];
425 KASSERT(m_buddy->order == VM_NFREEORDER,
426 ("vm_phys_split_pages: page %p has unexpected order %d",
427 m_buddy, m_buddy->order));
428 vm_freelist_add(fl, m_buddy, oind, 0);
433 * Initialize a physical page and add it to the free lists.
436 vm_phys_add_page(vm_paddr_t pa)
439 struct vm_domain *vmd;
441 vm_cnt.v_page_count++;
442 m = vm_phys_paddr_to_vm_page(pa);
445 m->segind = vm_phys_paddr_to_segind(pa);
446 vmd = vm_phys_domain(m);
447 vmd->vmd_page_count++;
448 vmd->vmd_segs |= 1UL << m->segind;
449 KASSERT(m->order == VM_NFREEORDER,
450 ("vm_phys_add_page: page %p has unexpected order %d",
452 m->pool = VM_FREEPOOL_DEFAULT;
454 mtx_lock(&vm_page_queue_free_mtx);
455 vm_phys_freecnt_adj(m, 1);
456 vm_phys_free_pages(m, 0);
457 mtx_unlock(&vm_page_queue_free_mtx);
461 * Allocate a contiguous, power of two-sized set of physical pages
462 * from the free lists.
464 * The free page queues must be locked.
467 vm_phys_alloc_pages(int pool, int order)
470 int dom, domain, flind;
472 KASSERT(pool < VM_NFREEPOOL,
473 ("vm_phys_alloc_pages: pool %d is out of range", pool));
474 KASSERT(order < VM_NFREEORDER,
475 ("vm_phys_alloc_pages: order %d is out of range", order));
477 for (dom = 0; dom < vm_ndomains; dom++) {
478 domain = vm_rr_selectdomain();
479 for (flind = 0; flind < vm_nfreelists; flind++) {
480 m = vm_phys_alloc_domain_pages(domain, flind, pool,
490 * Find and dequeue a free page on the given free list, with the
491 * specified pool and order
494 vm_phys_alloc_freelist_pages(int flind, int pool, int order)
499 KASSERT(flind < VM_NFREELIST,
500 ("vm_phys_alloc_freelist_pages: freelist %d is out of range", flind));
501 KASSERT(pool < VM_NFREEPOOL,
502 ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
503 KASSERT(order < VM_NFREEORDER,
504 ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
506 for (dom = 0; dom < vm_ndomains; dom++) {
507 domain = vm_rr_selectdomain();
508 m = vm_phys_alloc_domain_pages(domain, flind, pool, order);
516 vm_phys_alloc_domain_pages(int domain, int flind, int pool, int order)
518 struct vm_freelist *fl;
519 struct vm_freelist *alt;
523 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
524 fl = &vm_phys_free_queues[domain][flind][pool][0];
525 for (oind = order; oind < VM_NFREEORDER; oind++) {
526 m = TAILQ_FIRST(&fl[oind].pl);
528 vm_freelist_rem(fl, m, oind);
529 vm_phys_split_pages(m, oind, fl, order);
535 * The given pool was empty. Find the largest
536 * contiguous, power-of-two-sized set of pages in any
537 * pool. Transfer these pages to the given pool, and
538 * use them to satisfy the allocation.
540 for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
541 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
542 alt = &vm_phys_free_queues[domain][flind][pind][0];
543 m = TAILQ_FIRST(&alt[oind].pl);
545 vm_freelist_rem(alt, m, oind);
546 vm_phys_set_pool(pool, m, oind);
547 vm_phys_split_pages(m, oind, fl, order);
556 * Find the vm_page corresponding to the given physical address.
559 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
561 struct vm_phys_seg *seg;
564 for (segind = 0; segind < vm_phys_nsegs; segind++) {
565 seg = &vm_phys_segs[segind];
566 if (pa >= seg->start && pa < seg->end)
567 return (&seg->first_page[atop(pa - seg->start)]);
573 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
575 struct vm_phys_fictitious_seg tmp, *seg;
582 rw_rlock(&vm_phys_fictitious_reg_lock);
583 seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
584 rw_runlock(&vm_phys_fictitious_reg_lock);
588 m = &seg->first_page[atop(pa - seg->start)];
589 KASSERT((m->flags & PG_FICTITIOUS) != 0, ("%p not fictitious", m));
595 vm_phys_fictitious_init_range(vm_page_t range, vm_paddr_t start,
596 long page_count, vm_memattr_t memattr)
600 for (i = 0; i < page_count; i++) {
601 vm_page_initfake(&range[i], start + PAGE_SIZE * i, memattr);
602 range[i].oflags &= ~VPO_UNMANAGED;
603 range[i].busy_lock = VPB_UNBUSIED;
608 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
609 vm_memattr_t memattr)
611 struct vm_phys_fictitious_seg *seg;
614 #ifdef VM_PHYSSEG_DENSE
620 ("Start of segment isn't less than end (start: %jx end: %jx)",
621 (uintmax_t)start, (uintmax_t)end));
623 page_count = (end - start) / PAGE_SIZE;
625 #ifdef VM_PHYSSEG_DENSE
628 if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
629 fp = &vm_page_array[pi - first_page];
630 if ((pe - first_page) > vm_page_array_size) {
632 * We have a segment that starts inside
633 * of vm_page_array, but ends outside of it.
635 * Use vm_page_array pages for those that are
636 * inside of the vm_page_array range, and
637 * allocate the remaining ones.
639 dpage_count = vm_page_array_size - (pi - first_page);
640 vm_phys_fictitious_init_range(fp, start, dpage_count,
642 page_count -= dpage_count;
643 start += ptoa(dpage_count);
647 * We can allocate the full range from vm_page_array,
648 * so there's no need to register the range in the tree.
650 vm_phys_fictitious_init_range(fp, start, page_count, memattr);
652 } else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
654 * We have a segment that ends inside of vm_page_array,
655 * but starts outside of it.
657 fp = &vm_page_array[0];
658 dpage_count = pe - first_page;
659 vm_phys_fictitious_init_range(fp, ptoa(first_page), dpage_count,
661 end -= ptoa(dpage_count);
662 page_count -= dpage_count;
664 } else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
666 * Trying to register a fictitious range that expands before
667 * and after vm_page_array.
673 fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
675 #ifdef VM_PHYSSEG_DENSE
678 vm_phys_fictitious_init_range(fp, start, page_count, memattr);
680 seg = malloc(sizeof(*seg), M_FICT_PAGES, M_WAITOK | M_ZERO);
683 seg->first_page = fp;
685 rw_wlock(&vm_phys_fictitious_reg_lock);
686 RB_INSERT(fict_tree, &vm_phys_fictitious_tree, seg);
687 rw_wunlock(&vm_phys_fictitious_reg_lock);
693 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
695 struct vm_phys_fictitious_seg *seg, tmp;
696 #ifdef VM_PHYSSEG_DENSE
701 ("Start of segment isn't less than end (start: %jx end: %jx)",
702 (uintmax_t)start, (uintmax_t)end));
704 #ifdef VM_PHYSSEG_DENSE
707 if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
708 if ((pe - first_page) <= vm_page_array_size) {
710 * This segment was allocated using vm_page_array
711 * only, there's nothing to do since those pages
712 * were never added to the tree.
717 * We have a segment that starts inside
718 * of vm_page_array, but ends outside of it.
720 * Calculate how many pages were added to the
721 * tree and free them.
723 start = ptoa(first_page + vm_page_array_size);
724 } else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
726 * We have a segment that ends inside of vm_page_array,
727 * but starts outside of it.
729 end = ptoa(first_page);
730 } else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
731 /* Since it's not possible to register such a range, panic. */
733 "Unregistering not registered fictitious range [%#jx:%#jx]",
734 (uintmax_t)start, (uintmax_t)end);
740 rw_wlock(&vm_phys_fictitious_reg_lock);
741 seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
742 if (seg->start != start || seg->end != end) {
743 rw_wunlock(&vm_phys_fictitious_reg_lock);
745 "Unregistering not registered fictitious range [%#jx:%#jx]",
746 (uintmax_t)start, (uintmax_t)end);
748 RB_REMOVE(fict_tree, &vm_phys_fictitious_tree, seg);
749 rw_wunlock(&vm_phys_fictitious_reg_lock);
750 free(seg->first_page, M_FICT_PAGES);
751 free(seg, M_FICT_PAGES);
755 * Find the segment containing the given physical address.
758 vm_phys_paddr_to_segind(vm_paddr_t pa)
760 struct vm_phys_seg *seg;
763 for (segind = 0; segind < vm_phys_nsegs; segind++) {
764 seg = &vm_phys_segs[segind];
765 if (pa >= seg->start && pa < seg->end)
768 panic("vm_phys_paddr_to_segind: paddr %#jx is not in any segment" ,
773 * Free a contiguous, power of two-sized set of physical pages.
775 * The free page queues must be locked.
778 vm_phys_free_pages(vm_page_t m, int order)
780 struct vm_freelist *fl;
781 struct vm_phys_seg *seg;
785 KASSERT(m->order == VM_NFREEORDER,
786 ("vm_phys_free_pages: page %p has unexpected order %d",
788 KASSERT(m->pool < VM_NFREEPOOL,
789 ("vm_phys_free_pages: page %p has unexpected pool %d",
791 KASSERT(order < VM_NFREEORDER,
792 ("vm_phys_free_pages: order %d is out of range", order));
793 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
794 seg = &vm_phys_segs[m->segind];
795 if (order < VM_NFREEORDER - 1) {
796 pa = VM_PAGE_TO_PHYS(m);
798 pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
799 if (pa < seg->start || pa >= seg->end)
801 m_buddy = &seg->first_page[atop(pa - seg->start)];
802 if (m_buddy->order != order)
804 fl = (*seg->free_queues)[m_buddy->pool];
805 vm_freelist_rem(fl, m_buddy, order);
806 if (m_buddy->pool != m->pool)
807 vm_phys_set_pool(m->pool, m_buddy, order);
809 pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
810 m = &seg->first_page[atop(pa - seg->start)];
811 } while (order < VM_NFREEORDER - 1);
813 fl = (*seg->free_queues)[m->pool];
814 vm_freelist_add(fl, m, order, 1);
818 * Free a contiguous, arbitrarily sized set of physical pages.
820 * The free page queues must be locked.
823 vm_phys_free_contig(vm_page_t m, u_long npages)
829 * Avoid unnecessary coalescing by freeing the pages in the largest
830 * possible power-of-two-sized subsets.
832 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
833 for (;; npages -= n) {
835 * Unsigned "min" is used here so that "order" is assigned
836 * "VM_NFREEORDER - 1" when "m"'s physical address is zero
837 * or the low-order bits of its physical address are zero
838 * because the size of a physical address exceeds the size of
841 order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
846 vm_phys_free_pages(m, order);
849 /* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */
850 for (; npages > 0; npages -= n) {
851 order = flsl(npages) - 1;
853 vm_phys_free_pages(m, order);
859 * Set the pool for a contiguous, power of two-sized set of physical pages.
862 vm_phys_set_pool(int pool, vm_page_t m, int order)
866 for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
871 * Search for the given physical page "m" in the free lists. If the search
872 * succeeds, remove "m" from the free lists and return TRUE. Otherwise, return
873 * FALSE, indicating that "m" is not in the free lists.
875 * The free page queues must be locked.
878 vm_phys_unfree_page(vm_page_t m)
880 struct vm_freelist *fl;
881 struct vm_phys_seg *seg;
882 vm_paddr_t pa, pa_half;
883 vm_page_t m_set, m_tmp;
886 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
889 * First, find the contiguous, power of two-sized set of free
890 * physical pages containing the given physical page "m" and
891 * assign it to "m_set".
893 seg = &vm_phys_segs[m->segind];
894 for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
895 order < VM_NFREEORDER - 1; ) {
897 pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
898 if (pa >= seg->start)
899 m_set = &seg->first_page[atop(pa - seg->start)];
903 if (m_set->order < order)
905 if (m_set->order == VM_NFREEORDER)
907 KASSERT(m_set->order < VM_NFREEORDER,
908 ("vm_phys_unfree_page: page %p has unexpected order %d",
909 m_set, m_set->order));
912 * Next, remove "m_set" from the free lists. Finally, extract
913 * "m" from "m_set" using an iterative algorithm: While "m_set"
914 * is larger than a page, shrink "m_set" by returning the half
915 * of "m_set" that does not contain "m" to the free lists.
917 fl = (*seg->free_queues)[m_set->pool];
918 order = m_set->order;
919 vm_freelist_rem(fl, m_set, order);
922 pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
923 if (m->phys_addr < pa_half)
924 m_tmp = &seg->first_page[atop(pa_half - seg->start)];
927 m_set = &seg->first_page[atop(pa_half - seg->start)];
929 vm_freelist_add(fl, m_tmp, order, 0);
931 KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
936 * Try to zero one physical page. Used by an idle priority thread.
939 vm_phys_zero_pages_idle(void)
941 static struct vm_freelist *fl;
942 static int flind, oind, pind;
946 domain = vm_rr_selectdomain();
947 fl = vm_phys_free_queues[domain][0][0];
948 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
950 TAILQ_FOREACH_REVERSE(m, &fl[oind].pl, pglist, plinks.q) {
951 for (m_tmp = m; m_tmp < &m[1 << oind]; m_tmp++) {
952 if ((m_tmp->flags & (PG_CACHED | PG_ZERO)) == 0) {
953 vm_phys_unfree_page(m_tmp);
954 vm_phys_freecnt_adj(m, -1);
955 mtx_unlock(&vm_page_queue_free_mtx);
956 pmap_zero_page_idle(m_tmp);
957 m_tmp->flags |= PG_ZERO;
958 mtx_lock(&vm_page_queue_free_mtx);
959 vm_phys_freecnt_adj(m, 1);
960 vm_phys_free_pages(m_tmp, 0);
961 vm_page_zero_count++;
968 if (oind == VM_NFREEORDER) {
971 if (pind == VM_NFREEPOOL) {
974 if (flind == vm_nfreelists)
977 fl = vm_phys_free_queues[domain][flind][pind];
983 * Allocate a contiguous set of physical pages of the given size
984 * "npages" from the free lists. All of the physical pages must be at
985 * or above the given physical address "low" and below the given
986 * physical address "high". The given value "alignment" determines the
987 * alignment of the first physical page in the set. If the given value
988 * "boundary" is non-zero, then the set of physical pages cannot cross
989 * any physical address boundary that is a multiple of that value. Both
990 * "alignment" and "boundary" must be a power of two.
993 vm_phys_alloc_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
994 u_long alignment, vm_paddr_t boundary)
996 struct vm_freelist *fl;
997 struct vm_phys_seg *seg;
998 vm_paddr_t pa, pa_last, size;
1001 int dom, domain, flind, oind, order, pind;
1003 mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1004 size = npages << PAGE_SHIFT;
1006 ("vm_phys_alloc_contig: size must not be 0"));
1007 KASSERT((alignment & (alignment - 1)) == 0,
1008 ("vm_phys_alloc_contig: alignment must be a power of 2"));
1009 KASSERT((boundary & (boundary - 1)) == 0,
1010 ("vm_phys_alloc_contig: boundary must be a power of 2"));
1011 /* Compute the queue that is the best fit for npages. */
1012 for (order = 0; (1 << order) < npages; order++);
1015 domain = vm_rr_selectdomain();
1016 for (flind = 0; flind < vm_nfreelists; flind++) {
1017 for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER; oind++) {
1018 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1019 fl = &vm_phys_free_queues[domain][flind][pind][0];
1020 TAILQ_FOREACH(m_ret, &fl[oind].pl, plinks.q) {
1022 * A free list may contain physical pages
1023 * from one or more segments.
1025 seg = &vm_phys_segs[m_ret->segind];
1026 if (seg->start > high ||
1031 * Is the size of this allocation request
1032 * larger than the largest block size?
1034 if (order >= VM_NFREEORDER) {
1036 * Determine if a sufficient number
1037 * of subsequent blocks to satisfy
1038 * the allocation request are free.
1040 pa = VM_PAGE_TO_PHYS(m_ret);
1041 pa_last = pa + size;
1043 pa += 1 << (PAGE_SHIFT + VM_NFREEORDER - 1);
1046 if (pa < seg->start ||
1049 m = &seg->first_page[atop(pa - seg->start)];
1050 if (m->order != VM_NFREEORDER - 1)
1053 /* If not, continue to the next block. */
1059 * Determine if the blocks are within the given range,
1060 * satisfy the given alignment, and do not cross the
1063 pa = VM_PAGE_TO_PHYS(m_ret);
1065 pa + size <= high &&
1066 (pa & (alignment - 1)) == 0 &&
1067 ((pa ^ (pa + size - 1)) & ~(boundary - 1)) == 0)
1073 if (++dom < vm_ndomains)
1077 for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
1078 fl = (*seg->free_queues)[m->pool];
1079 vm_freelist_rem(fl, m, m->order);
1081 if (m_ret->pool != VM_FREEPOOL_DEFAULT)
1082 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m_ret, oind);
1083 fl = (*seg->free_queues)[m_ret->pool];
1084 vm_phys_split_pages(m_ret, oind, fl, order);
1085 /* Return excess pages to the free lists. */
1086 npages_end = roundup2(npages, 1 << imin(oind, order));
1087 if (npages < npages_end)
1088 vm_phys_free_contig(&m_ret[npages], npages_end - npages);
1094 * Show the number of physical pages in each of the free lists.
1096 DB_SHOW_COMMAND(freepages, db_show_freepages)
1098 struct vm_freelist *fl;
1099 int flind, oind, pind, dom;
1101 for (dom = 0; dom < vm_ndomains; dom++) {
1102 db_printf("DOMAIN: %d\n", dom);
1103 for (flind = 0; flind < vm_nfreelists; flind++) {
1104 db_printf("FREE LIST %d:\n"
1105 "\n ORDER (SIZE) | NUMBER"
1107 for (pind = 0; pind < VM_NFREEPOOL; pind++)
1108 db_printf(" | POOL %d", pind);
1110 for (pind = 0; pind < VM_NFREEPOOL; pind++)
1111 db_printf("-- -- ");
1113 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
1114 db_printf(" %2.2d (%6.6dK)", oind,
1115 1 << (PAGE_SHIFT - 10 + oind));
1116 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1117 fl = vm_phys_free_queues[dom][flind][pind];
1118 db_printf(" | %6.6d", fl[oind].lcnt);