2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2002-2006 Rice University
5 * Copyright (c) 2007 Alan L. Cox <alc@cs.rice.edu>
8 * This software was developed for the FreeBSD Project by Alan L. Cox,
9 * Olivier Crameri, Peter Druschel, Sitaram Iyer, and Juan Navarro.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
23 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
24 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
27 * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY
30 * WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31 * POSSIBILITY OF SUCH DAMAGE.
35 * Physical memory system implementation
37 * Any external functions defined by this module are only to be used by the
38 * virtual memory system.
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
47 #include <sys/param.h>
48 #include <sys/systm.h>
50 #include <sys/kernel.h>
51 #include <sys/malloc.h>
52 #include <sys/mutex.h>
54 #include <sys/queue.h>
55 #include <sys/rwlock.h>
57 #include <sys/sysctl.h>
59 #include <sys/vmmeter.h>
65 #include <vm/vm_param.h>
66 #include <vm/vm_kern.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_phys.h>
70 #include <vm/vm_pagequeue.h>
72 _Static_assert(sizeof(long) * NBBY >= VM_PHYSSEG_MAX,
73 "Too many physsegs.");
76 struct mem_affinity __read_mostly *mem_affinity;
77 int __read_mostly *mem_locality;
80 int __read_mostly vm_ndomains = 1;
81 domainset_t __read_mostly all_domains = DOMAINSET_T_INITIALIZER(0x1);
83 struct vm_phys_seg __read_mostly vm_phys_segs[VM_PHYSSEG_MAX];
84 int __read_mostly vm_phys_nsegs;
86 struct vm_phys_fictitious_seg;
87 static int vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *,
88 struct vm_phys_fictitious_seg *);
90 RB_HEAD(fict_tree, vm_phys_fictitious_seg) vm_phys_fictitious_tree =
91 RB_INITIALIZER(_vm_phys_fictitious_tree);
93 struct vm_phys_fictitious_seg {
94 RB_ENTRY(vm_phys_fictitious_seg) node;
95 /* Memory region data */
101 RB_GENERATE_STATIC(fict_tree, vm_phys_fictitious_seg, node,
102 vm_phys_fictitious_cmp);
104 static struct rwlock_padalign vm_phys_fictitious_reg_lock;
105 MALLOC_DEFINE(M_FICT_PAGES, "vm_fictitious", "Fictitious VM pages");
107 static struct vm_freelist __aligned(CACHE_LINE_SIZE)
108 vm_phys_free_queues[MAXMEMDOM][VM_NFREELIST][VM_NFREEPOOL][VM_NFREEORDER];
110 static int __read_mostly vm_nfreelists;
113 * Provides the mapping from VM_FREELIST_* to free list indices (flind).
115 static int __read_mostly vm_freelist_to_flind[VM_NFREELIST];
117 CTASSERT(VM_FREELIST_DEFAULT == 0);
119 #ifdef VM_FREELIST_DMA32
120 #define VM_DMA32_BOUNDARY ((vm_paddr_t)1 << 32)
124 * Enforce the assumptions made by vm_phys_add_seg() and vm_phys_init() about
125 * the ordering of the free list boundaries.
127 #if defined(VM_LOWMEM_BOUNDARY) && defined(VM_DMA32_BOUNDARY)
128 CTASSERT(VM_LOWMEM_BOUNDARY < VM_DMA32_BOUNDARY);
131 static int sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS);
132 SYSCTL_OID(_vm, OID_AUTO, phys_free, CTLTYPE_STRING | CTLFLAG_RD,
133 NULL, 0, sysctl_vm_phys_free, "A", "Phys Free Info");
135 static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS);
136 SYSCTL_OID(_vm, OID_AUTO, phys_segs, CTLTYPE_STRING | CTLFLAG_RD,
137 NULL, 0, sysctl_vm_phys_segs, "A", "Phys Seg Info");
140 static int sysctl_vm_phys_locality(SYSCTL_HANDLER_ARGS);
141 SYSCTL_OID(_vm, OID_AUTO, phys_locality, CTLTYPE_STRING | CTLFLAG_RD,
142 NULL, 0, sysctl_vm_phys_locality, "A", "Phys Locality Info");
145 SYSCTL_INT(_vm, OID_AUTO, ndomains, CTLFLAG_RD,
146 &vm_ndomains, 0, "Number of physical memory domains available.");
148 static vm_page_t vm_phys_alloc_seg_contig(struct vm_phys_seg *seg,
149 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
150 vm_paddr_t boundary);
151 static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain);
152 static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end);
153 static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl,
154 int order, int tail);
157 * Red-black tree helpers for vm fictitious range management.
160 vm_phys_fictitious_in_range(struct vm_phys_fictitious_seg *p,
161 struct vm_phys_fictitious_seg *range)
164 KASSERT(range->start != 0 && range->end != 0,
165 ("Invalid range passed on search for vm_fictitious page"));
166 if (p->start >= range->end)
168 if (p->start < range->start)
175 vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *p1,
176 struct vm_phys_fictitious_seg *p2)
179 /* Check if this is a search for a page */
181 return (vm_phys_fictitious_in_range(p1, p2));
183 KASSERT(p2->end != 0,
184 ("Invalid range passed as second parameter to vm fictitious comparison"));
186 /* Searching to add a new range */
187 if (p1->end <= p2->start)
189 if (p1->start >= p2->end)
192 panic("Trying to add overlapping vm fictitious ranges:\n"
193 "[%#jx:%#jx] and [%#jx:%#jx]", (uintmax_t)p1->start,
194 (uintmax_t)p1->end, (uintmax_t)p2->start, (uintmax_t)p2->end);
198 vm_phys_domain_match(int prefer, vm_paddr_t low, vm_paddr_t high)
204 if (vm_ndomains == 1 || mem_affinity == NULL)
207 DOMAINSET_ZERO(&mask);
209 * Check for any memory that overlaps low, high.
211 for (i = 0; mem_affinity[i].end != 0; i++)
212 if (mem_affinity[i].start <= high &&
213 mem_affinity[i].end >= low)
214 DOMAINSET_SET(mem_affinity[i].domain, &mask);
215 if (prefer != -1 && DOMAINSET_ISSET(prefer, &mask))
217 if (DOMAINSET_EMPTY(&mask))
218 panic("vm_phys_domain_match: Impossible constraint");
219 return (DOMAINSET_FFS(&mask) - 1);
226 * Outputs the state of the physical memory allocator, specifically,
227 * the amount of physical memory in each free list.
230 sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS)
233 struct vm_freelist *fl;
234 int dom, error, flind, oind, pind;
236 error = sysctl_wire_old_buffer(req, 0);
239 sbuf_new_for_sysctl(&sbuf, NULL, 128 * vm_ndomains, req);
240 for (dom = 0; dom < vm_ndomains; dom++) {
241 sbuf_printf(&sbuf,"\nDOMAIN %d:\n", dom);
242 for (flind = 0; flind < vm_nfreelists; flind++) {
243 sbuf_printf(&sbuf, "\nFREE LIST %d:\n"
244 "\n ORDER (SIZE) | NUMBER"
246 for (pind = 0; pind < VM_NFREEPOOL; pind++)
247 sbuf_printf(&sbuf, " | POOL %d", pind);
248 sbuf_printf(&sbuf, "\n-- ");
249 for (pind = 0; pind < VM_NFREEPOOL; pind++)
250 sbuf_printf(&sbuf, "-- -- ");
251 sbuf_printf(&sbuf, "--\n");
252 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
253 sbuf_printf(&sbuf, " %2d (%6dK)", oind,
254 1 << (PAGE_SHIFT - 10 + oind));
255 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
256 fl = vm_phys_free_queues[dom][flind][pind];
257 sbuf_printf(&sbuf, " | %6d",
260 sbuf_printf(&sbuf, "\n");
264 error = sbuf_finish(&sbuf);
270 * Outputs the set of physical memory segments.
273 sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS)
276 struct vm_phys_seg *seg;
279 error = sysctl_wire_old_buffer(req, 0);
282 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
283 for (segind = 0; segind < vm_phys_nsegs; segind++) {
284 sbuf_printf(&sbuf, "\nSEGMENT %d:\n\n", segind);
285 seg = &vm_phys_segs[segind];
286 sbuf_printf(&sbuf, "start: %#jx\n",
287 (uintmax_t)seg->start);
288 sbuf_printf(&sbuf, "end: %#jx\n",
289 (uintmax_t)seg->end);
290 sbuf_printf(&sbuf, "domain: %d\n", seg->domain);
291 sbuf_printf(&sbuf, "free list: %p\n", seg->free_queues);
293 error = sbuf_finish(&sbuf);
299 * Return affinity, or -1 if there's no affinity information.
302 vm_phys_mem_affinity(int f, int t)
306 if (mem_locality == NULL)
308 if (f >= vm_ndomains || t >= vm_ndomains)
310 return (mem_locality[f * vm_ndomains + t]);
318 * Outputs the VM locality table.
321 sysctl_vm_phys_locality(SYSCTL_HANDLER_ARGS)
326 error = sysctl_wire_old_buffer(req, 0);
329 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
331 sbuf_printf(&sbuf, "\n");
333 for (i = 0; i < vm_ndomains; i++) {
334 sbuf_printf(&sbuf, "%d: ", i);
335 for (j = 0; j < vm_ndomains; j++) {
336 sbuf_printf(&sbuf, "%d ", vm_phys_mem_affinity(i, j));
338 sbuf_printf(&sbuf, "\n");
340 error = sbuf_finish(&sbuf);
347 vm_freelist_add(struct vm_freelist *fl, vm_page_t m, int order, int tail)
352 TAILQ_INSERT_TAIL(&fl[order].pl, m, listq);
354 TAILQ_INSERT_HEAD(&fl[order].pl, m, listq);
359 vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order)
362 TAILQ_REMOVE(&fl[order].pl, m, listq);
364 m->order = VM_NFREEORDER;
368 * Create a physical memory segment.
371 _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain)
373 struct vm_phys_seg *seg;
375 KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX,
376 ("vm_phys_create_seg: increase VM_PHYSSEG_MAX"));
377 KASSERT(domain >= 0 && domain < vm_ndomains,
378 ("vm_phys_create_seg: invalid domain provided"));
379 seg = &vm_phys_segs[vm_phys_nsegs++];
380 while (seg > vm_phys_segs && (seg - 1)->start >= end) {
386 seg->domain = domain;
390 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end)
395 if (mem_affinity == NULL) {
396 _vm_phys_create_seg(start, end, 0);
401 if (mem_affinity[i].end == 0)
402 panic("Reached end of affinity info");
403 if (mem_affinity[i].end <= start)
405 if (mem_affinity[i].start > start)
406 panic("No affinity info for start %jx",
408 if (mem_affinity[i].end >= end) {
409 _vm_phys_create_seg(start, end,
410 mem_affinity[i].domain);
413 _vm_phys_create_seg(start, mem_affinity[i].end,
414 mem_affinity[i].domain);
415 start = mem_affinity[i].end;
418 _vm_phys_create_seg(start, end, 0);
423 * Add a physical memory segment.
426 vm_phys_add_seg(vm_paddr_t start, vm_paddr_t end)
430 KASSERT((start & PAGE_MASK) == 0,
431 ("vm_phys_define_seg: start is not page aligned"));
432 KASSERT((end & PAGE_MASK) == 0,
433 ("vm_phys_define_seg: end is not page aligned"));
436 * Split the physical memory segment if it spans two or more free
440 #ifdef VM_FREELIST_LOWMEM
441 if (paddr < VM_LOWMEM_BOUNDARY && end > VM_LOWMEM_BOUNDARY) {
442 vm_phys_create_seg(paddr, VM_LOWMEM_BOUNDARY);
443 paddr = VM_LOWMEM_BOUNDARY;
446 #ifdef VM_FREELIST_DMA32
447 if (paddr < VM_DMA32_BOUNDARY && end > VM_DMA32_BOUNDARY) {
448 vm_phys_create_seg(paddr, VM_DMA32_BOUNDARY);
449 paddr = VM_DMA32_BOUNDARY;
452 vm_phys_create_seg(paddr, end);
456 * Initialize the physical memory allocator.
458 * Requires that vm_page_array is initialized!
463 struct vm_freelist *fl;
464 struct vm_phys_seg *end_seg, *prev_seg, *seg, *tmp_seg;
466 int dom, flind, freelist, oind, pind, segind;
469 * Compute the number of free lists, and generate the mapping from the
470 * manifest constants VM_FREELIST_* to the free list indices.
472 * Initially, the entries of vm_freelist_to_flind[] are set to either
473 * 0 or 1 to indicate which free lists should be created.
476 for (segind = vm_phys_nsegs - 1; segind >= 0; segind--) {
477 seg = &vm_phys_segs[segind];
478 #ifdef VM_FREELIST_LOWMEM
479 if (seg->end <= VM_LOWMEM_BOUNDARY)
480 vm_freelist_to_flind[VM_FREELIST_LOWMEM] = 1;
483 #ifdef VM_FREELIST_DMA32
485 #ifdef VM_DMA32_NPAGES_THRESHOLD
487 * Create the DMA32 free list only if the amount of
488 * physical memory above physical address 4G exceeds the
491 npages > VM_DMA32_NPAGES_THRESHOLD &&
493 seg->end <= VM_DMA32_BOUNDARY)
494 vm_freelist_to_flind[VM_FREELIST_DMA32] = 1;
498 npages += atop(seg->end - seg->start);
499 vm_freelist_to_flind[VM_FREELIST_DEFAULT] = 1;
502 /* Change each entry into a running total of the free lists. */
503 for (freelist = 1; freelist < VM_NFREELIST; freelist++) {
504 vm_freelist_to_flind[freelist] +=
505 vm_freelist_to_flind[freelist - 1];
507 vm_nfreelists = vm_freelist_to_flind[VM_NFREELIST - 1];
508 KASSERT(vm_nfreelists > 0, ("vm_phys_init: no free lists"));
509 /* Change each entry into a free list index. */
510 for (freelist = 0; freelist < VM_NFREELIST; freelist++)
511 vm_freelist_to_flind[freelist]--;
514 * Initialize the first_page and free_queues fields of each physical
517 #ifdef VM_PHYSSEG_SPARSE
520 for (segind = 0; segind < vm_phys_nsegs; segind++) {
521 seg = &vm_phys_segs[segind];
522 #ifdef VM_PHYSSEG_SPARSE
523 seg->first_page = &vm_page_array[npages];
524 npages += atop(seg->end - seg->start);
526 seg->first_page = PHYS_TO_VM_PAGE(seg->start);
528 #ifdef VM_FREELIST_LOWMEM
529 if (seg->end <= VM_LOWMEM_BOUNDARY) {
530 flind = vm_freelist_to_flind[VM_FREELIST_LOWMEM];
532 ("vm_phys_init: LOWMEM flind < 0"));
535 #ifdef VM_FREELIST_DMA32
536 if (seg->end <= VM_DMA32_BOUNDARY) {
537 flind = vm_freelist_to_flind[VM_FREELIST_DMA32];
539 ("vm_phys_init: DMA32 flind < 0"));
543 flind = vm_freelist_to_flind[VM_FREELIST_DEFAULT];
545 ("vm_phys_init: DEFAULT flind < 0"));
547 seg->free_queues = &vm_phys_free_queues[seg->domain][flind];
551 * Coalesce physical memory segments that are contiguous and share the
552 * same per-domain free queues.
554 prev_seg = vm_phys_segs;
555 seg = &vm_phys_segs[1];
556 end_seg = &vm_phys_segs[vm_phys_nsegs];
557 while (seg < end_seg) {
558 if (prev_seg->end == seg->start &&
559 prev_seg->free_queues == seg->free_queues) {
560 prev_seg->end = seg->end;
561 KASSERT(prev_seg->domain == seg->domain,
562 ("vm_phys_init: free queues cannot span domains"));
565 for (tmp_seg = seg; tmp_seg < end_seg; tmp_seg++)
566 *tmp_seg = *(tmp_seg + 1);
574 * Initialize the free queues.
576 for (dom = 0; dom < vm_ndomains; dom++) {
577 for (flind = 0; flind < vm_nfreelists; flind++) {
578 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
579 fl = vm_phys_free_queues[dom][flind][pind];
580 for (oind = 0; oind < VM_NFREEORDER; oind++)
581 TAILQ_INIT(&fl[oind].pl);
586 rw_init(&vm_phys_fictitious_reg_lock, "vmfctr");
590 * Split a contiguous, power of two-sized set of physical pages.
592 * When this function is called by a page allocation function, the caller
593 * should request insertion at the head unless the order [order, oind) queues
594 * are known to be empty. The objective being to reduce the likelihood of
595 * long-term fragmentation by promoting contemporaneous allocation and
596 * (hopefully) deallocation.
599 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order,
604 while (oind > order) {
606 m_buddy = &m[1 << oind];
607 KASSERT(m_buddy->order == VM_NFREEORDER,
608 ("vm_phys_split_pages: page %p has unexpected order %d",
609 m_buddy, m_buddy->order));
610 vm_freelist_add(fl, m_buddy, oind, tail);
615 * Add the physical pages [m, m + npages) at the end of a power-of-two aligned
616 * and sized set to the specified free list.
618 * When this function is called by a page allocation function, the caller
619 * should request insertion at the head unless the lower-order queues are
620 * known to be empty. The objective being to reduce the likelihood of long-
621 * term fragmentation by promoting contemporaneous allocation and (hopefully)
624 * The physical page m's buddy must not be free.
627 vm_phys_enq_range(vm_page_t m, u_int npages, struct vm_freelist *fl, int tail)
632 KASSERT(npages > 0, ("vm_phys_enq_range: npages is 0"));
633 KASSERT(((VM_PAGE_TO_PHYS(m) + npages * PAGE_SIZE) &
634 ((PAGE_SIZE << (fls(npages) - 1)) - 1)) == 0,
635 ("vm_phys_enq_range: page %p and npages %u are misaligned",
638 KASSERT(m->order == VM_NFREEORDER,
639 ("vm_phys_enq_range: page %p has unexpected order %d",
641 order = ffs(npages) - 1;
642 KASSERT(order < VM_NFREEORDER,
643 ("vm_phys_enq_range: order %d is out of range", order));
644 vm_freelist_add(fl, m, order, tail);
648 } while (npages > 0);
652 * Tries to allocate the specified number of pages from the specified pool
653 * within the specified domain. Returns the actual number of allocated pages
654 * and a pointer to each page through the array ma[].
656 * The returned pages may not be physically contiguous. However, in contrast
657 * to performing multiple, back-to-back calls to vm_phys_alloc_pages(..., 0),
658 * calling this function once to allocate the desired number of pages will
659 * avoid wasted time in vm_phys_split_pages().
661 * The free page queues for the specified domain must be locked.
664 vm_phys_alloc_npages(int domain, int pool, int npages, vm_page_t ma[])
666 struct vm_freelist *alt, *fl;
668 int avail, end, flind, freelist, i, need, oind, pind;
670 KASSERT(domain >= 0 && domain < vm_ndomains,
671 ("vm_phys_alloc_npages: domain %d is out of range", domain));
672 KASSERT(pool < VM_NFREEPOOL,
673 ("vm_phys_alloc_npages: pool %d is out of range", pool));
674 KASSERT(npages <= 1 << (VM_NFREEORDER - 1),
675 ("vm_phys_alloc_npages: npages %d is out of range", npages));
676 vm_domain_free_assert_locked(VM_DOMAIN(domain));
678 for (freelist = 0; freelist < VM_NFREELIST; freelist++) {
679 flind = vm_freelist_to_flind[freelist];
682 fl = vm_phys_free_queues[domain][flind][pool];
683 for (oind = 0; oind < VM_NFREEORDER; oind++) {
684 while ((m = TAILQ_FIRST(&fl[oind].pl)) != NULL) {
685 vm_freelist_rem(fl, m, oind);
687 need = imin(npages - i, avail);
688 for (end = i + need; i < end;)
692 * Return excess pages to fl. Its
693 * order [0, oind) queues are empty.
695 vm_phys_enq_range(m, avail - need, fl,
698 } else if (i == npages)
702 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
703 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
704 alt = vm_phys_free_queues[domain][flind][pind];
705 while ((m = TAILQ_FIRST(&alt[oind].pl)) !=
707 vm_freelist_rem(alt, m, oind);
708 vm_phys_set_pool(pool, m, oind);
710 need = imin(npages - i, avail);
711 for (end = i + need; i < end;)
715 * Return excess pages to fl.
716 * Its order [0, oind) queues
719 vm_phys_enq_range(m, avail -
722 } else if (i == npages)
732 * Allocate a contiguous, power of two-sized set of physical pages
733 * from the free lists.
735 * The free page queues must be locked.
738 vm_phys_alloc_pages(int domain, int pool, int order)
743 for (freelist = 0; freelist < VM_NFREELIST; freelist++) {
744 m = vm_phys_alloc_freelist_pages(domain, freelist, pool, order);
752 * Allocate a contiguous, power of two-sized set of physical pages from the
753 * specified free list. The free list must be specified using one of the
754 * manifest constants VM_FREELIST_*.
756 * The free page queues must be locked.
759 vm_phys_alloc_freelist_pages(int domain, int freelist, int pool, int order)
761 struct vm_freelist *alt, *fl;
763 int oind, pind, flind;
765 KASSERT(domain >= 0 && domain < vm_ndomains,
766 ("vm_phys_alloc_freelist_pages: domain %d is out of range",
768 KASSERT(freelist < VM_NFREELIST,
769 ("vm_phys_alloc_freelist_pages: freelist %d is out of range",
771 KASSERT(pool < VM_NFREEPOOL,
772 ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
773 KASSERT(order < VM_NFREEORDER,
774 ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
776 flind = vm_freelist_to_flind[freelist];
777 /* Check if freelist is present */
781 vm_domain_free_assert_locked(VM_DOMAIN(domain));
782 fl = &vm_phys_free_queues[domain][flind][pool][0];
783 for (oind = order; oind < VM_NFREEORDER; oind++) {
784 m = TAILQ_FIRST(&fl[oind].pl);
786 vm_freelist_rem(fl, m, oind);
787 /* The order [order, oind) queues are empty. */
788 vm_phys_split_pages(m, oind, fl, order, 1);
794 * The given pool was empty. Find the largest
795 * contiguous, power-of-two-sized set of pages in any
796 * pool. Transfer these pages to the given pool, and
797 * use them to satisfy the allocation.
799 for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
800 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
801 alt = &vm_phys_free_queues[domain][flind][pind][0];
802 m = TAILQ_FIRST(&alt[oind].pl);
804 vm_freelist_rem(alt, m, oind);
805 vm_phys_set_pool(pool, m, oind);
806 /* The order [order, oind) queues are empty. */
807 vm_phys_split_pages(m, oind, fl, order, 1);
816 * Find the vm_page corresponding to the given physical address.
819 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
821 struct vm_phys_seg *seg;
824 for (segind = 0; segind < vm_phys_nsegs; segind++) {
825 seg = &vm_phys_segs[segind];
826 if (pa >= seg->start && pa < seg->end)
827 return (&seg->first_page[atop(pa - seg->start)]);
833 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
835 struct vm_phys_fictitious_seg tmp, *seg;
842 rw_rlock(&vm_phys_fictitious_reg_lock);
843 seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
844 rw_runlock(&vm_phys_fictitious_reg_lock);
848 m = &seg->first_page[atop(pa - seg->start)];
849 KASSERT((m->flags & PG_FICTITIOUS) != 0, ("%p not fictitious", m));
855 vm_phys_fictitious_init_range(vm_page_t range, vm_paddr_t start,
856 long page_count, vm_memattr_t memattr)
860 bzero(range, page_count * sizeof(*range));
861 for (i = 0; i < page_count; i++) {
862 vm_page_initfake(&range[i], start + PAGE_SIZE * i, memattr);
863 range[i].oflags &= ~VPO_UNMANAGED;
864 range[i].busy_lock = VPB_UNBUSIED;
869 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
870 vm_memattr_t memattr)
872 struct vm_phys_fictitious_seg *seg;
875 #ifdef VM_PHYSSEG_DENSE
881 ("Start of segment isn't less than end (start: %jx end: %jx)",
882 (uintmax_t)start, (uintmax_t)end));
884 page_count = (end - start) / PAGE_SIZE;
886 #ifdef VM_PHYSSEG_DENSE
889 if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
890 fp = &vm_page_array[pi - first_page];
891 if ((pe - first_page) > vm_page_array_size) {
893 * We have a segment that starts inside
894 * of vm_page_array, but ends outside of it.
896 * Use vm_page_array pages for those that are
897 * inside of the vm_page_array range, and
898 * allocate the remaining ones.
900 dpage_count = vm_page_array_size - (pi - first_page);
901 vm_phys_fictitious_init_range(fp, start, dpage_count,
903 page_count -= dpage_count;
904 start += ptoa(dpage_count);
908 * We can allocate the full range from vm_page_array,
909 * so there's no need to register the range in the tree.
911 vm_phys_fictitious_init_range(fp, start, page_count, memattr);
913 } else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
915 * We have a segment that ends inside of vm_page_array,
916 * but starts outside of it.
918 fp = &vm_page_array[0];
919 dpage_count = pe - first_page;
920 vm_phys_fictitious_init_range(fp, ptoa(first_page), dpage_count,
922 end -= ptoa(dpage_count);
923 page_count -= dpage_count;
925 } else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
927 * Trying to register a fictitious range that expands before
928 * and after vm_page_array.
934 fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
936 #ifdef VM_PHYSSEG_DENSE
939 vm_phys_fictitious_init_range(fp, start, page_count, memattr);
941 seg = malloc(sizeof(*seg), M_FICT_PAGES, M_WAITOK | M_ZERO);
944 seg->first_page = fp;
946 rw_wlock(&vm_phys_fictitious_reg_lock);
947 RB_INSERT(fict_tree, &vm_phys_fictitious_tree, seg);
948 rw_wunlock(&vm_phys_fictitious_reg_lock);
954 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
956 struct vm_phys_fictitious_seg *seg, tmp;
957 #ifdef VM_PHYSSEG_DENSE
962 ("Start of segment isn't less than end (start: %jx end: %jx)",
963 (uintmax_t)start, (uintmax_t)end));
965 #ifdef VM_PHYSSEG_DENSE
968 if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
969 if ((pe - first_page) <= vm_page_array_size) {
971 * This segment was allocated using vm_page_array
972 * only, there's nothing to do since those pages
973 * were never added to the tree.
978 * We have a segment that starts inside
979 * of vm_page_array, but ends outside of it.
981 * Calculate how many pages were added to the
982 * tree and free them.
984 start = ptoa(first_page + vm_page_array_size);
985 } else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
987 * We have a segment that ends inside of vm_page_array,
988 * but starts outside of it.
990 end = ptoa(first_page);
991 } else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
992 /* Since it's not possible to register such a range, panic. */
994 "Unregistering not registered fictitious range [%#jx:%#jx]",
995 (uintmax_t)start, (uintmax_t)end);
1001 rw_wlock(&vm_phys_fictitious_reg_lock);
1002 seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
1003 if (seg->start != start || seg->end != end) {
1004 rw_wunlock(&vm_phys_fictitious_reg_lock);
1006 "Unregistering not registered fictitious range [%#jx:%#jx]",
1007 (uintmax_t)start, (uintmax_t)end);
1009 RB_REMOVE(fict_tree, &vm_phys_fictitious_tree, seg);
1010 rw_wunlock(&vm_phys_fictitious_reg_lock);
1011 free(seg->first_page, M_FICT_PAGES);
1012 free(seg, M_FICT_PAGES);
1016 * Free a contiguous, power of two-sized set of physical pages.
1018 * The free page queues must be locked.
1021 vm_phys_free_pages(vm_page_t m, int order)
1023 struct vm_freelist *fl;
1024 struct vm_phys_seg *seg;
1028 KASSERT(m->order == VM_NFREEORDER,
1029 ("vm_phys_free_pages: page %p has unexpected order %d",
1031 KASSERT(m->pool < VM_NFREEPOOL,
1032 ("vm_phys_free_pages: page %p has unexpected pool %d",
1034 KASSERT(order < VM_NFREEORDER,
1035 ("vm_phys_free_pages: order %d is out of range", order));
1036 seg = &vm_phys_segs[m->segind];
1037 vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
1038 if (order < VM_NFREEORDER - 1) {
1039 pa = VM_PAGE_TO_PHYS(m);
1041 pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
1042 if (pa < seg->start || pa >= seg->end)
1044 m_buddy = &seg->first_page[atop(pa - seg->start)];
1045 if (m_buddy->order != order)
1047 fl = (*seg->free_queues)[m_buddy->pool];
1048 vm_freelist_rem(fl, m_buddy, order);
1049 if (m_buddy->pool != m->pool)
1050 vm_phys_set_pool(m->pool, m_buddy, order);
1052 pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
1053 m = &seg->first_page[atop(pa - seg->start)];
1054 } while (order < VM_NFREEORDER - 1);
1056 fl = (*seg->free_queues)[m->pool];
1057 vm_freelist_add(fl, m, order, 1);
1061 * Free a contiguous, arbitrarily sized set of physical pages.
1063 * The free page queues must be locked.
1066 vm_phys_free_contig(vm_page_t m, u_long npages)
1072 * Avoid unnecessary coalescing by freeing the pages in the largest
1073 * possible power-of-two-sized subsets.
1075 vm_domain_free_assert_locked(vm_pagequeue_domain(m));
1076 for (;; npages -= n) {
1078 * Unsigned "min" is used here so that "order" is assigned
1079 * "VM_NFREEORDER - 1" when "m"'s physical address is zero
1080 * or the low-order bits of its physical address are zero
1081 * because the size of a physical address exceeds the size of
1084 order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
1089 vm_phys_free_pages(m, order);
1092 /* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */
1093 for (; npages > 0; npages -= n) {
1094 order = flsl(npages) - 1;
1096 vm_phys_free_pages(m, order);
1102 * Scan physical memory between the specified addresses "low" and "high" for a
1103 * run of contiguous physical pages that satisfy the specified conditions, and
1104 * return the lowest page in the run. The specified "alignment" determines
1105 * the alignment of the lowest physical page in the run. If the specified
1106 * "boundary" is non-zero, then the run of physical pages cannot span a
1107 * physical address that is a multiple of "boundary".
1109 * "npages" must be greater than zero. Both "alignment" and "boundary" must
1110 * be a power of two.
1113 vm_phys_scan_contig(int domain, u_long npages, vm_paddr_t low, vm_paddr_t high,
1114 u_long alignment, vm_paddr_t boundary, int options)
1117 vm_page_t m_end, m_run, m_start;
1118 struct vm_phys_seg *seg;
1121 KASSERT(npages > 0, ("npages is 0"));
1122 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1123 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1126 for (segind = 0; segind < vm_phys_nsegs; segind++) {
1127 seg = &vm_phys_segs[segind];
1128 if (seg->domain != domain)
1130 if (seg->start >= high)
1132 if (low >= seg->end)
1134 if (low <= seg->start)
1135 m_start = seg->first_page;
1137 m_start = &seg->first_page[atop(low - seg->start)];
1138 if (high < seg->end)
1142 if (pa_end - VM_PAGE_TO_PHYS(m_start) < ptoa(npages))
1144 m_end = &seg->first_page[atop(pa_end - seg->start)];
1145 m_run = vm_page_scan_contig(npages, m_start, m_end,
1146 alignment, boundary, options);
1154 * Set the pool for a contiguous, power of two-sized set of physical pages.
1157 vm_phys_set_pool(int pool, vm_page_t m, int order)
1161 for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
1166 * Search for the given physical page "m" in the free lists. If the search
1167 * succeeds, remove "m" from the free lists and return TRUE. Otherwise, return
1168 * FALSE, indicating that "m" is not in the free lists.
1170 * The free page queues must be locked.
1173 vm_phys_unfree_page(vm_page_t m)
1175 struct vm_freelist *fl;
1176 struct vm_phys_seg *seg;
1177 vm_paddr_t pa, pa_half;
1178 vm_page_t m_set, m_tmp;
1182 * First, find the contiguous, power of two-sized set of free
1183 * physical pages containing the given physical page "m" and
1184 * assign it to "m_set".
1186 seg = &vm_phys_segs[m->segind];
1187 vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
1188 for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
1189 order < VM_NFREEORDER - 1; ) {
1191 pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
1192 if (pa >= seg->start)
1193 m_set = &seg->first_page[atop(pa - seg->start)];
1197 if (m_set->order < order)
1199 if (m_set->order == VM_NFREEORDER)
1201 KASSERT(m_set->order < VM_NFREEORDER,
1202 ("vm_phys_unfree_page: page %p has unexpected order %d",
1203 m_set, m_set->order));
1206 * Next, remove "m_set" from the free lists. Finally, extract
1207 * "m" from "m_set" using an iterative algorithm: While "m_set"
1208 * is larger than a page, shrink "m_set" by returning the half
1209 * of "m_set" that does not contain "m" to the free lists.
1211 fl = (*seg->free_queues)[m_set->pool];
1212 order = m_set->order;
1213 vm_freelist_rem(fl, m_set, order);
1216 pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
1217 if (m->phys_addr < pa_half)
1218 m_tmp = &seg->first_page[atop(pa_half - seg->start)];
1221 m_set = &seg->first_page[atop(pa_half - seg->start)];
1223 vm_freelist_add(fl, m_tmp, order, 0);
1225 KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
1230 * Allocate a contiguous set of physical pages of the given size
1231 * "npages" from the free lists. All of the physical pages must be at
1232 * or above the given physical address "low" and below the given
1233 * physical address "high". The given value "alignment" determines the
1234 * alignment of the first physical page in the set. If the given value
1235 * "boundary" is non-zero, then the set of physical pages cannot cross
1236 * any physical address boundary that is a multiple of that value. Both
1237 * "alignment" and "boundary" must be a power of two.
1240 vm_phys_alloc_contig(int domain, u_long npages, vm_paddr_t low, vm_paddr_t high,
1241 u_long alignment, vm_paddr_t boundary)
1243 vm_paddr_t pa_end, pa_start;
1245 struct vm_phys_seg *seg;
1248 KASSERT(npages > 0, ("npages is 0"));
1249 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1250 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1251 vm_domain_free_assert_locked(VM_DOMAIN(domain));
1255 for (segind = vm_phys_nsegs - 1; segind >= 0; segind--) {
1256 seg = &vm_phys_segs[segind];
1257 if (seg->start >= high || seg->domain != domain)
1259 if (low >= seg->end)
1261 if (low <= seg->start)
1262 pa_start = seg->start;
1265 if (high < seg->end)
1269 if (pa_end - pa_start < ptoa(npages))
1271 m_run = vm_phys_alloc_seg_contig(seg, npages, low, high,
1272 alignment, boundary);
1280 * Allocate a run of contiguous physical pages from the free list for the
1281 * specified segment.
1284 vm_phys_alloc_seg_contig(struct vm_phys_seg *seg, u_long npages,
1285 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
1287 struct vm_freelist *fl;
1288 vm_paddr_t pa, pa_end, size;
1291 int oind, order, pind;
1293 KASSERT(npages > 0, ("npages is 0"));
1294 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1295 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1296 vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
1297 /* Compute the queue that is the best fit for npages. */
1298 order = flsl(npages - 1);
1299 /* Search for a run satisfying the specified conditions. */
1300 size = npages << PAGE_SHIFT;
1301 for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER;
1303 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1304 fl = (*seg->free_queues)[pind];
1305 TAILQ_FOREACH(m_ret, &fl[oind].pl, listq) {
1307 * Is the size of this allocation request
1308 * larger than the largest block size?
1310 if (order >= VM_NFREEORDER) {
1312 * Determine if a sufficient number of
1313 * subsequent blocks to satisfy the
1314 * allocation request are free.
1316 pa = VM_PAGE_TO_PHYS(m_ret);
1321 pa += 1 << (PAGE_SHIFT +
1327 m = &seg->first_page[atop(pa -
1329 if (m->order != VM_NFREEORDER -
1333 /* If not, go to the next block. */
1339 * Determine if the blocks are within the
1340 * given range, satisfy the given alignment,
1341 * and do not cross the given boundary.
1343 pa = VM_PAGE_TO_PHYS(m_ret);
1345 if (pa >= low && pa_end <= high &&
1346 (pa & (alignment - 1)) == 0 &&
1347 rounddown2(pa ^ (pa_end - 1), boundary) == 0)
1354 for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
1355 fl = (*seg->free_queues)[m->pool];
1356 vm_freelist_rem(fl, m, oind);
1357 if (m->pool != VM_FREEPOOL_DEFAULT)
1358 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, oind);
1360 /* Return excess pages to the free lists. */
1361 npages_end = roundup2(npages, 1 << oind);
1362 if (npages < npages_end) {
1363 fl = (*seg->free_queues)[VM_FREEPOOL_DEFAULT];
1364 vm_phys_enq_range(&m_ret[npages], npages_end - npages, fl, 0);
1371 * Show the number of physical pages in each of the free lists.
1373 DB_SHOW_COMMAND(freepages, db_show_freepages)
1375 struct vm_freelist *fl;
1376 int flind, oind, pind, dom;
1378 for (dom = 0; dom < vm_ndomains; dom++) {
1379 db_printf("DOMAIN: %d\n", dom);
1380 for (flind = 0; flind < vm_nfreelists; flind++) {
1381 db_printf("FREE LIST %d:\n"
1382 "\n ORDER (SIZE) | NUMBER"
1384 for (pind = 0; pind < VM_NFREEPOOL; pind++)
1385 db_printf(" | POOL %d", pind);
1387 for (pind = 0; pind < VM_NFREEPOOL; pind++)
1388 db_printf("-- -- ");
1390 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
1391 db_printf(" %2.2d (%6.6dK)", oind,
1392 1 << (PAGE_SHIFT - 10 + oind));
1393 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1394 fl = vm_phys_free_queues[dom][flind][pind];
1395 db_printf(" | %6.6d", fl[oind].lcnt);