2 * SPDX-License-Identifier: BSD-2-Clause
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>
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/domainset.h>
49 #include <sys/kernel.h>
50 #include <sys/malloc.h>
51 #include <sys/mutex.h>
53 #include <sys/queue.h>
54 #include <sys/rwlock.h>
56 #include <sys/sysctl.h>
58 #include <sys/vmmeter.h>
63 #include <vm/vm_extern.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>
69 #include <vm/vm_pagequeue.h>
71 _Static_assert(sizeof(long) * NBBY >= VM_PHYSSEG_MAX,
72 "Too many physsegs.");
75 struct mem_affinity __read_mostly *mem_affinity;
76 int __read_mostly *mem_locality;
78 static int numa_disabled;
79 static SYSCTL_NODE(_vm, OID_AUTO, numa, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
81 SYSCTL_INT(_vm_numa, OID_AUTO, disabled, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
82 &numa_disabled, 0, "NUMA-awareness in the allocators is disabled");
85 int __read_mostly vm_ndomains = 1;
86 domainset_t __read_mostly all_domains = DOMAINSET_T_INITIALIZER(0x1);
88 struct vm_phys_seg __read_mostly vm_phys_segs[VM_PHYSSEG_MAX];
89 int __read_mostly vm_phys_nsegs;
90 static struct vm_phys_seg vm_phys_early_segs[8];
91 static int vm_phys_early_nsegs;
93 struct vm_phys_fictitious_seg;
94 static int vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *,
95 struct vm_phys_fictitious_seg *);
97 RB_HEAD(fict_tree, vm_phys_fictitious_seg) vm_phys_fictitious_tree =
98 RB_INITIALIZER(&vm_phys_fictitious_tree);
100 struct vm_phys_fictitious_seg {
101 RB_ENTRY(vm_phys_fictitious_seg) node;
102 /* Memory region data */
105 vm_page_t first_page;
108 RB_GENERATE_STATIC(fict_tree, vm_phys_fictitious_seg, node,
109 vm_phys_fictitious_cmp);
111 static struct rwlock_padalign vm_phys_fictitious_reg_lock;
112 MALLOC_DEFINE(M_FICT_PAGES, "vm_fictitious", "Fictitious VM pages");
114 static struct vm_freelist __aligned(CACHE_LINE_SIZE)
115 vm_phys_free_queues[MAXMEMDOM][VM_NFREELIST][VM_NFREEPOOL]
118 static int __read_mostly vm_nfreelists;
121 * These "avail lists" are globals used to communicate boot-time physical
122 * memory layout to other parts of the kernel. Each physically contiguous
123 * region of memory is defined by a start address at an even index and an
124 * end address at the following odd index. Each list is terminated by a
125 * pair of zero entries.
127 * dump_avail tells the dump code what regions to include in a crash dump, and
128 * phys_avail is all of the remaining physical memory that is available for
131 * Initially dump_avail and phys_avail are identical. Boot time memory
132 * allocations remove extents from phys_avail that may still be included
135 vm_paddr_t phys_avail[PHYS_AVAIL_COUNT];
136 vm_paddr_t dump_avail[PHYS_AVAIL_COUNT];
139 * Provides the mapping from VM_FREELIST_* to free list indices (flind).
141 static int __read_mostly vm_freelist_to_flind[VM_NFREELIST];
143 CTASSERT(VM_FREELIST_DEFAULT == 0);
145 #ifdef VM_FREELIST_DMA32
146 #define VM_DMA32_BOUNDARY ((vm_paddr_t)1 << 32)
150 * Enforce the assumptions made by vm_phys_add_seg() and vm_phys_init() about
151 * the ordering of the free list boundaries.
153 #if defined(VM_LOWMEM_BOUNDARY) && defined(VM_DMA32_BOUNDARY)
154 CTASSERT(VM_LOWMEM_BOUNDARY < VM_DMA32_BOUNDARY);
157 static int sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS);
158 SYSCTL_OID(_vm, OID_AUTO, phys_free,
159 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
160 sysctl_vm_phys_free, "A",
163 static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS);
164 SYSCTL_OID(_vm, OID_AUTO, phys_segs,
165 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
166 sysctl_vm_phys_segs, "A",
170 static int sysctl_vm_phys_locality(SYSCTL_HANDLER_ARGS);
171 SYSCTL_OID(_vm, OID_AUTO, phys_locality,
172 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
173 sysctl_vm_phys_locality, "A",
174 "Phys Locality Info");
177 SYSCTL_INT(_vm, OID_AUTO, ndomains, CTLFLAG_RD,
178 &vm_ndomains, 0, "Number of physical memory domains available.");
180 static vm_page_t vm_phys_alloc_seg_contig(struct vm_phys_seg *seg,
181 u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
182 vm_paddr_t boundary);
183 static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain);
184 static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end);
185 static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl,
186 int order, int tail);
189 * Red-black tree helpers for vm fictitious range management.
192 vm_phys_fictitious_in_range(struct vm_phys_fictitious_seg *p,
193 struct vm_phys_fictitious_seg *range)
196 KASSERT(range->start != 0 && range->end != 0,
197 ("Invalid range passed on search for vm_fictitious page"));
198 if (p->start >= range->end)
200 if (p->start < range->start)
207 vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *p1,
208 struct vm_phys_fictitious_seg *p2)
211 /* Check if this is a search for a page */
213 return (vm_phys_fictitious_in_range(p1, p2));
215 KASSERT(p2->end != 0,
216 ("Invalid range passed as second parameter to vm fictitious comparison"));
218 /* Searching to add a new range */
219 if (p1->end <= p2->start)
221 if (p1->start >= p2->end)
224 panic("Trying to add overlapping vm fictitious ranges:\n"
225 "[%#jx:%#jx] and [%#jx:%#jx]", (uintmax_t)p1->start,
226 (uintmax_t)p1->end, (uintmax_t)p2->start, (uintmax_t)p2->end);
230 vm_phys_domain_match(int prefer, vm_paddr_t low, vm_paddr_t high)
236 if (vm_ndomains == 1 || mem_affinity == NULL)
239 DOMAINSET_ZERO(&mask);
241 * Check for any memory that overlaps low, high.
243 for (i = 0; mem_affinity[i].end != 0; i++)
244 if (mem_affinity[i].start <= high &&
245 mem_affinity[i].end >= low)
246 DOMAINSET_SET(mem_affinity[i].domain, &mask);
247 if (prefer != -1 && DOMAINSET_ISSET(prefer, &mask))
249 if (DOMAINSET_EMPTY(&mask))
250 panic("vm_phys_domain_match: Impossible constraint");
251 return (DOMAINSET_FFS(&mask) - 1);
258 * Outputs the state of the physical memory allocator, specifically,
259 * the amount of physical memory in each free list.
262 sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS)
265 struct vm_freelist *fl;
266 int dom, error, flind, oind, pind;
268 error = sysctl_wire_old_buffer(req, 0);
271 sbuf_new_for_sysctl(&sbuf, NULL, 128 * vm_ndomains, req);
272 for (dom = 0; dom < vm_ndomains; dom++) {
273 sbuf_printf(&sbuf,"\nDOMAIN %d:\n", dom);
274 for (flind = 0; flind < vm_nfreelists; flind++) {
275 sbuf_printf(&sbuf, "\nFREE LIST %d:\n"
276 "\n ORDER (SIZE) | NUMBER"
278 for (pind = 0; pind < VM_NFREEPOOL; pind++)
279 sbuf_printf(&sbuf, " | POOL %d", pind);
280 sbuf_printf(&sbuf, "\n-- ");
281 for (pind = 0; pind < VM_NFREEPOOL; pind++)
282 sbuf_printf(&sbuf, "-- -- ");
283 sbuf_printf(&sbuf, "--\n");
284 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
285 sbuf_printf(&sbuf, " %2d (%6dK)", oind,
286 1 << (PAGE_SHIFT - 10 + oind));
287 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
288 fl = vm_phys_free_queues[dom][flind][pind];
289 sbuf_printf(&sbuf, " | %6d",
292 sbuf_printf(&sbuf, "\n");
296 error = sbuf_finish(&sbuf);
302 * Outputs the set of physical memory segments.
305 sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS)
308 struct vm_phys_seg *seg;
311 error = sysctl_wire_old_buffer(req, 0);
314 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
315 for (segind = 0; segind < vm_phys_nsegs; segind++) {
316 sbuf_printf(&sbuf, "\nSEGMENT %d:\n\n", segind);
317 seg = &vm_phys_segs[segind];
318 sbuf_printf(&sbuf, "start: %#jx\n",
319 (uintmax_t)seg->start);
320 sbuf_printf(&sbuf, "end: %#jx\n",
321 (uintmax_t)seg->end);
322 sbuf_printf(&sbuf, "domain: %d\n", seg->domain);
323 sbuf_printf(&sbuf, "free list: %p\n", seg->free_queues);
325 error = sbuf_finish(&sbuf);
331 * Return affinity, or -1 if there's no affinity information.
334 vm_phys_mem_affinity(int f, int t)
338 if (mem_locality == NULL)
340 if (f >= vm_ndomains || t >= vm_ndomains)
342 return (mem_locality[f * vm_ndomains + t]);
350 * Outputs the VM locality table.
353 sysctl_vm_phys_locality(SYSCTL_HANDLER_ARGS)
358 error = sysctl_wire_old_buffer(req, 0);
361 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
363 sbuf_printf(&sbuf, "\n");
365 for (i = 0; i < vm_ndomains; i++) {
366 sbuf_printf(&sbuf, "%d: ", i);
367 for (j = 0; j < vm_ndomains; j++) {
368 sbuf_printf(&sbuf, "%d ", vm_phys_mem_affinity(i, j));
370 sbuf_printf(&sbuf, "\n");
372 error = sbuf_finish(&sbuf);
379 vm_freelist_add(struct vm_freelist *fl, vm_page_t m, int order, int tail)
384 TAILQ_INSERT_TAIL(&fl[order].pl, m, listq);
386 TAILQ_INSERT_HEAD(&fl[order].pl, m, listq);
391 vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order)
394 TAILQ_REMOVE(&fl[order].pl, m, listq);
396 m->order = VM_NFREEORDER;
400 * Create a physical memory segment.
403 _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain)
405 struct vm_phys_seg *seg;
407 KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX,
408 ("vm_phys_create_seg: increase VM_PHYSSEG_MAX"));
409 KASSERT(domain >= 0 && domain < vm_ndomains,
410 ("vm_phys_create_seg: invalid domain provided"));
411 seg = &vm_phys_segs[vm_phys_nsegs++];
412 while (seg > vm_phys_segs && (seg - 1)->start >= end) {
418 seg->domain = domain;
422 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end)
427 if (mem_affinity == NULL) {
428 _vm_phys_create_seg(start, end, 0);
433 if (mem_affinity[i].end == 0)
434 panic("Reached end of affinity info");
435 if (mem_affinity[i].end <= start)
437 if (mem_affinity[i].start > start)
438 panic("No affinity info for start %jx",
440 if (mem_affinity[i].end >= end) {
441 _vm_phys_create_seg(start, end,
442 mem_affinity[i].domain);
445 _vm_phys_create_seg(start, mem_affinity[i].end,
446 mem_affinity[i].domain);
447 start = mem_affinity[i].end;
450 _vm_phys_create_seg(start, end, 0);
455 * Add a physical memory segment.
458 vm_phys_add_seg(vm_paddr_t start, vm_paddr_t end)
462 KASSERT((start & PAGE_MASK) == 0,
463 ("vm_phys_define_seg: start is not page aligned"));
464 KASSERT((end & PAGE_MASK) == 0,
465 ("vm_phys_define_seg: end is not page aligned"));
468 * Split the physical memory segment if it spans two or more free
472 #ifdef VM_FREELIST_LOWMEM
473 if (paddr < VM_LOWMEM_BOUNDARY && end > VM_LOWMEM_BOUNDARY) {
474 vm_phys_create_seg(paddr, VM_LOWMEM_BOUNDARY);
475 paddr = VM_LOWMEM_BOUNDARY;
478 #ifdef VM_FREELIST_DMA32
479 if (paddr < VM_DMA32_BOUNDARY && end > VM_DMA32_BOUNDARY) {
480 vm_phys_create_seg(paddr, VM_DMA32_BOUNDARY);
481 paddr = VM_DMA32_BOUNDARY;
484 vm_phys_create_seg(paddr, end);
488 * Initialize the physical memory allocator.
490 * Requires that vm_page_array is initialized!
495 struct vm_freelist *fl;
496 struct vm_phys_seg *end_seg, *prev_seg, *seg, *tmp_seg;
498 int dom, flind, freelist, oind, pind, segind;
501 * Compute the number of free lists, and generate the mapping from the
502 * manifest constants VM_FREELIST_* to the free list indices.
504 * Initially, the entries of vm_freelist_to_flind[] are set to either
505 * 0 or 1 to indicate which free lists should be created.
508 for (segind = vm_phys_nsegs - 1; segind >= 0; segind--) {
509 seg = &vm_phys_segs[segind];
510 #ifdef VM_FREELIST_LOWMEM
511 if (seg->end <= VM_LOWMEM_BOUNDARY)
512 vm_freelist_to_flind[VM_FREELIST_LOWMEM] = 1;
515 #ifdef VM_FREELIST_DMA32
517 #ifdef VM_DMA32_NPAGES_THRESHOLD
519 * Create the DMA32 free list only if the amount of
520 * physical memory above physical address 4G exceeds the
523 npages > VM_DMA32_NPAGES_THRESHOLD &&
525 seg->end <= VM_DMA32_BOUNDARY)
526 vm_freelist_to_flind[VM_FREELIST_DMA32] = 1;
530 npages += atop(seg->end - seg->start);
531 vm_freelist_to_flind[VM_FREELIST_DEFAULT] = 1;
534 /* Change each entry into a running total of the free lists. */
535 for (freelist = 1; freelist < VM_NFREELIST; freelist++) {
536 vm_freelist_to_flind[freelist] +=
537 vm_freelist_to_flind[freelist - 1];
539 vm_nfreelists = vm_freelist_to_flind[VM_NFREELIST - 1];
540 KASSERT(vm_nfreelists > 0, ("vm_phys_init: no free lists"));
541 /* Change each entry into a free list index. */
542 for (freelist = 0; freelist < VM_NFREELIST; freelist++)
543 vm_freelist_to_flind[freelist]--;
546 * Initialize the first_page and free_queues fields of each physical
549 #ifdef VM_PHYSSEG_SPARSE
552 for (segind = 0; segind < vm_phys_nsegs; segind++) {
553 seg = &vm_phys_segs[segind];
554 #ifdef VM_PHYSSEG_SPARSE
555 seg->first_page = &vm_page_array[npages];
556 npages += atop(seg->end - seg->start);
558 seg->first_page = PHYS_TO_VM_PAGE(seg->start);
560 #ifdef VM_FREELIST_LOWMEM
561 if (seg->end <= VM_LOWMEM_BOUNDARY) {
562 flind = vm_freelist_to_flind[VM_FREELIST_LOWMEM];
564 ("vm_phys_init: LOWMEM flind < 0"));
567 #ifdef VM_FREELIST_DMA32
568 if (seg->end <= VM_DMA32_BOUNDARY) {
569 flind = vm_freelist_to_flind[VM_FREELIST_DMA32];
571 ("vm_phys_init: DMA32 flind < 0"));
575 flind = vm_freelist_to_flind[VM_FREELIST_DEFAULT];
577 ("vm_phys_init: DEFAULT flind < 0"));
579 seg->free_queues = &vm_phys_free_queues[seg->domain][flind];
583 * Coalesce physical memory segments that are contiguous and share the
584 * same per-domain free queues.
586 prev_seg = vm_phys_segs;
587 seg = &vm_phys_segs[1];
588 end_seg = &vm_phys_segs[vm_phys_nsegs];
589 while (seg < end_seg) {
590 if (prev_seg->end == seg->start &&
591 prev_seg->free_queues == seg->free_queues) {
592 prev_seg->end = seg->end;
593 KASSERT(prev_seg->domain == seg->domain,
594 ("vm_phys_init: free queues cannot span domains"));
597 for (tmp_seg = seg; tmp_seg < end_seg; tmp_seg++)
598 *tmp_seg = *(tmp_seg + 1);
606 * Initialize the free queues.
608 for (dom = 0; dom < vm_ndomains; dom++) {
609 for (flind = 0; flind < vm_nfreelists; flind++) {
610 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
611 fl = vm_phys_free_queues[dom][flind][pind];
612 for (oind = 0; oind < VM_NFREEORDER; oind++)
613 TAILQ_INIT(&fl[oind].pl);
618 rw_init(&vm_phys_fictitious_reg_lock, "vmfctr");
622 * Register info about the NUMA topology of the system.
624 * Invoked by platform-dependent code prior to vm_phys_init().
627 vm_phys_register_domains(int ndomains, struct mem_affinity *affinity,
634 * For now the only override value that we support is 1, which
635 * effectively disables NUMA-awareness in the allocators.
637 TUNABLE_INT_FETCH("vm.numa.disabled", &numa_disabled);
642 vm_ndomains = ndomains;
643 mem_affinity = affinity;
644 mem_locality = locality;
647 for (i = 0; i < vm_ndomains; i++)
648 DOMAINSET_SET(i, &all_domains);
657 * Split a contiguous, power of two-sized set of physical pages.
659 * When this function is called by a page allocation function, the caller
660 * should request insertion at the head unless the order [order, oind) queues
661 * are known to be empty. The objective being to reduce the likelihood of
662 * long-term fragmentation by promoting contemporaneous allocation and
663 * (hopefully) deallocation.
666 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order,
671 while (oind > order) {
673 m_buddy = &m[1 << oind];
674 KASSERT(m_buddy->order == VM_NFREEORDER,
675 ("vm_phys_split_pages: page %p has unexpected order %d",
676 m_buddy, m_buddy->order));
677 vm_freelist_add(fl, m_buddy, oind, tail);
682 * Add the physical pages [m, m + npages) at the end of a power-of-two aligned
683 * and sized set to the specified free list.
685 * When this function is called by a page allocation function, the caller
686 * should request insertion at the head unless the lower-order queues are
687 * known to be empty. The objective being to reduce the likelihood of long-
688 * term fragmentation by promoting contemporaneous allocation and (hopefully)
691 * The physical page m's buddy must not be free.
694 vm_phys_enq_range(vm_page_t m, u_int npages, struct vm_freelist *fl, int tail)
699 KASSERT(npages > 0, ("vm_phys_enq_range: npages is 0"));
700 KASSERT(((VM_PAGE_TO_PHYS(m) + npages * PAGE_SIZE) &
701 ((PAGE_SIZE << (fls(npages) - 1)) - 1)) == 0,
702 ("vm_phys_enq_range: page %p and npages %u are misaligned",
705 KASSERT(m->order == VM_NFREEORDER,
706 ("vm_phys_enq_range: page %p has unexpected order %d",
708 order = ffs(npages) - 1;
709 KASSERT(order < VM_NFREEORDER,
710 ("vm_phys_enq_range: order %d is out of range", order));
711 vm_freelist_add(fl, m, order, tail);
715 } while (npages > 0);
719 * Set the pool for a contiguous, power of two-sized set of physical pages.
722 vm_phys_set_pool(int pool, vm_page_t m, int order)
726 for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
731 * Tries to allocate the specified number of pages from the specified pool
732 * within the specified domain. Returns the actual number of allocated pages
733 * and a pointer to each page through the array ma[].
735 * The returned pages may not be physically contiguous. However, in contrast
736 * to performing multiple, back-to-back calls to vm_phys_alloc_pages(..., 0),
737 * calling this function once to allocate the desired number of pages will
738 * avoid wasted time in vm_phys_split_pages().
740 * The free page queues for the specified domain must be locked.
743 vm_phys_alloc_npages(int domain, int pool, int npages, vm_page_t ma[])
745 struct vm_freelist *alt, *fl;
747 int avail, end, flind, freelist, i, need, oind, pind;
749 KASSERT(domain >= 0 && domain < vm_ndomains,
750 ("vm_phys_alloc_npages: domain %d is out of range", domain));
751 KASSERT(pool < VM_NFREEPOOL,
752 ("vm_phys_alloc_npages: pool %d is out of range", pool));
753 KASSERT(npages <= 1 << (VM_NFREEORDER - 1),
754 ("vm_phys_alloc_npages: npages %d is out of range", npages));
755 vm_domain_free_assert_locked(VM_DOMAIN(domain));
757 for (freelist = 0; freelist < VM_NFREELIST; freelist++) {
758 flind = vm_freelist_to_flind[freelist];
761 fl = vm_phys_free_queues[domain][flind][pool];
762 for (oind = 0; oind < VM_NFREEORDER; oind++) {
763 while ((m = TAILQ_FIRST(&fl[oind].pl)) != NULL) {
764 vm_freelist_rem(fl, m, oind);
766 need = imin(npages - i, avail);
767 for (end = i + need; i < end;)
771 * Return excess pages to fl. Its
772 * order [0, oind) queues are empty.
774 vm_phys_enq_range(m, avail - need, fl,
777 } else if (i == npages)
781 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
782 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
783 alt = vm_phys_free_queues[domain][flind][pind];
784 while ((m = TAILQ_FIRST(&alt[oind].pl)) !=
786 vm_freelist_rem(alt, m, oind);
787 vm_phys_set_pool(pool, m, oind);
789 need = imin(npages - i, avail);
790 for (end = i + need; i < end;)
794 * Return excess pages to fl.
795 * Its order [0, oind) queues
798 vm_phys_enq_range(m, avail -
801 } else if (i == npages)
811 * Allocate a contiguous, power of two-sized set of physical pages
812 * from the free lists.
814 * The free page queues must be locked.
817 vm_phys_alloc_pages(int domain, int pool, int order)
822 for (freelist = 0; freelist < VM_NFREELIST; freelist++) {
823 m = vm_phys_alloc_freelist_pages(domain, freelist, pool, order);
831 * Allocate a contiguous, power of two-sized set of physical pages from the
832 * specified free list. The free list must be specified using one of the
833 * manifest constants VM_FREELIST_*.
835 * The free page queues must be locked.
838 vm_phys_alloc_freelist_pages(int domain, int freelist, int pool, int order)
840 struct vm_freelist *alt, *fl;
842 int oind, pind, flind;
844 KASSERT(domain >= 0 && domain < vm_ndomains,
845 ("vm_phys_alloc_freelist_pages: domain %d is out of range",
847 KASSERT(freelist < VM_NFREELIST,
848 ("vm_phys_alloc_freelist_pages: freelist %d is out of range",
850 KASSERT(pool < VM_NFREEPOOL,
851 ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
852 KASSERT(order < VM_NFREEORDER,
853 ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
855 flind = vm_freelist_to_flind[freelist];
856 /* Check if freelist is present */
860 vm_domain_free_assert_locked(VM_DOMAIN(domain));
861 fl = &vm_phys_free_queues[domain][flind][pool][0];
862 for (oind = order; oind < VM_NFREEORDER; oind++) {
863 m = TAILQ_FIRST(&fl[oind].pl);
865 vm_freelist_rem(fl, m, oind);
866 /* The order [order, oind) queues are empty. */
867 vm_phys_split_pages(m, oind, fl, order, 1);
873 * The given pool was empty. Find the largest
874 * contiguous, power-of-two-sized set of pages in any
875 * pool. Transfer these pages to the given pool, and
876 * use them to satisfy the allocation.
878 for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
879 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
880 alt = &vm_phys_free_queues[domain][flind][pind][0];
881 m = TAILQ_FIRST(&alt[oind].pl);
883 vm_freelist_rem(alt, m, oind);
884 vm_phys_set_pool(pool, m, oind);
885 /* The order [order, oind) queues are empty. */
886 vm_phys_split_pages(m, oind, fl, order, 1);
895 * Find the vm_page corresponding to the given physical address.
898 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
900 struct vm_phys_seg *seg;
903 for (segind = 0; segind < vm_phys_nsegs; segind++) {
904 seg = &vm_phys_segs[segind];
905 if (pa >= seg->start && pa < seg->end)
906 return (&seg->first_page[atop(pa - seg->start)]);
912 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
914 struct vm_phys_fictitious_seg tmp, *seg;
921 rw_rlock(&vm_phys_fictitious_reg_lock);
922 seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
923 rw_runlock(&vm_phys_fictitious_reg_lock);
927 m = &seg->first_page[atop(pa - seg->start)];
928 KASSERT((m->flags & PG_FICTITIOUS) != 0, ("%p not fictitious", m));
934 vm_phys_fictitious_init_range(vm_page_t range, vm_paddr_t start,
935 long page_count, vm_memattr_t memattr)
939 bzero(range, page_count * sizeof(*range));
940 for (i = 0; i < page_count; i++) {
941 vm_page_initfake(&range[i], start + PAGE_SIZE * i, memattr);
942 range[i].oflags &= ~VPO_UNMANAGED;
943 range[i].busy_lock = VPB_UNBUSIED;
948 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
949 vm_memattr_t memattr)
951 struct vm_phys_fictitious_seg *seg;
954 #ifdef VM_PHYSSEG_DENSE
960 ("Start of segment isn't less than end (start: %jx end: %jx)",
961 (uintmax_t)start, (uintmax_t)end));
963 page_count = (end - start) / PAGE_SIZE;
965 #ifdef VM_PHYSSEG_DENSE
968 if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
969 fp = &vm_page_array[pi - first_page];
970 if ((pe - first_page) > vm_page_array_size) {
972 * We have a segment that starts inside
973 * of vm_page_array, but ends outside of it.
975 * Use vm_page_array pages for those that are
976 * inside of the vm_page_array range, and
977 * allocate the remaining ones.
979 dpage_count = vm_page_array_size - (pi - first_page);
980 vm_phys_fictitious_init_range(fp, start, dpage_count,
982 page_count -= dpage_count;
983 start += ptoa(dpage_count);
987 * We can allocate the full range from vm_page_array,
988 * so there's no need to register the range in the tree.
990 vm_phys_fictitious_init_range(fp, start, page_count, memattr);
992 } else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
994 * We have a segment that ends inside of vm_page_array,
995 * but starts outside of it.
997 fp = &vm_page_array[0];
998 dpage_count = pe - first_page;
999 vm_phys_fictitious_init_range(fp, ptoa(first_page), dpage_count,
1001 end -= ptoa(dpage_count);
1002 page_count -= dpage_count;
1004 } else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
1006 * Trying to register a fictitious range that expands before
1007 * and after vm_page_array.
1013 fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
1015 #ifdef VM_PHYSSEG_DENSE
1018 vm_phys_fictitious_init_range(fp, start, page_count, memattr);
1020 seg = malloc(sizeof(*seg), M_FICT_PAGES, M_WAITOK | M_ZERO);
1023 seg->first_page = fp;
1025 rw_wlock(&vm_phys_fictitious_reg_lock);
1026 RB_INSERT(fict_tree, &vm_phys_fictitious_tree, seg);
1027 rw_wunlock(&vm_phys_fictitious_reg_lock);
1033 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
1035 struct vm_phys_fictitious_seg *seg, tmp;
1036 #ifdef VM_PHYSSEG_DENSE
1040 KASSERT(start < end,
1041 ("Start of segment isn't less than end (start: %jx end: %jx)",
1042 (uintmax_t)start, (uintmax_t)end));
1044 #ifdef VM_PHYSSEG_DENSE
1047 if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
1048 if ((pe - first_page) <= vm_page_array_size) {
1050 * This segment was allocated using vm_page_array
1051 * only, there's nothing to do since those pages
1052 * were never added to the tree.
1057 * We have a segment that starts inside
1058 * of vm_page_array, but ends outside of it.
1060 * Calculate how many pages were added to the
1061 * tree and free them.
1063 start = ptoa(first_page + vm_page_array_size);
1064 } else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
1066 * We have a segment that ends inside of vm_page_array,
1067 * but starts outside of it.
1069 end = ptoa(first_page);
1070 } else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
1071 /* Since it's not possible to register such a range, panic. */
1073 "Unregistering not registered fictitious range [%#jx:%#jx]",
1074 (uintmax_t)start, (uintmax_t)end);
1080 rw_wlock(&vm_phys_fictitious_reg_lock);
1081 seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
1082 if (seg->start != start || seg->end != end) {
1083 rw_wunlock(&vm_phys_fictitious_reg_lock);
1085 "Unregistering not registered fictitious range [%#jx:%#jx]",
1086 (uintmax_t)start, (uintmax_t)end);
1088 RB_REMOVE(fict_tree, &vm_phys_fictitious_tree, seg);
1089 rw_wunlock(&vm_phys_fictitious_reg_lock);
1090 free(seg->first_page, M_FICT_PAGES);
1091 free(seg, M_FICT_PAGES);
1095 * Free a contiguous, power of two-sized set of physical pages.
1097 * The free page queues must be locked.
1100 vm_phys_free_pages(vm_page_t m, int order)
1102 struct vm_freelist *fl;
1103 struct vm_phys_seg *seg;
1107 KASSERT(m->order == VM_NFREEORDER,
1108 ("vm_phys_free_pages: page %p has unexpected order %d",
1110 KASSERT(m->pool < VM_NFREEPOOL,
1111 ("vm_phys_free_pages: page %p has unexpected pool %d",
1113 KASSERT(order < VM_NFREEORDER,
1114 ("vm_phys_free_pages: order %d is out of range", order));
1115 seg = &vm_phys_segs[m->segind];
1116 vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
1117 if (order < VM_NFREEORDER - 1) {
1118 pa = VM_PAGE_TO_PHYS(m);
1120 pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
1121 if (pa < seg->start || pa >= seg->end)
1123 m_buddy = &seg->first_page[atop(pa - seg->start)];
1124 if (m_buddy->order != order)
1126 fl = (*seg->free_queues)[m_buddy->pool];
1127 vm_freelist_rem(fl, m_buddy, order);
1128 if (m_buddy->pool != m->pool)
1129 vm_phys_set_pool(m->pool, m_buddy, order);
1131 pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
1132 m = &seg->first_page[atop(pa - seg->start)];
1133 } while (order < VM_NFREEORDER - 1);
1135 fl = (*seg->free_queues)[m->pool];
1136 vm_freelist_add(fl, m, order, 1);
1140 * Return the largest possible order of a set of pages starting at m.
1143 max_order(vm_page_t m)
1147 * Unsigned "min" is used here so that "order" is assigned
1148 * "VM_NFREEORDER - 1" when "m"'s physical address is zero
1149 * or the low-order bits of its physical address are zero
1150 * because the size of a physical address exceeds the size of
1153 return (min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
1154 VM_NFREEORDER - 1));
1158 * Free a contiguous, arbitrarily sized set of physical pages, without
1159 * merging across set boundaries.
1161 * The free page queues must be locked.
1164 vm_phys_enqueue_contig(vm_page_t m, u_long npages)
1166 struct vm_freelist *fl;
1167 struct vm_phys_seg *seg;
1172 * Avoid unnecessary coalescing by freeing the pages in the largest
1173 * possible power-of-two-sized subsets.
1175 vm_domain_free_assert_locked(vm_pagequeue_domain(m));
1176 seg = &vm_phys_segs[m->segind];
1177 fl = (*seg->free_queues)[m->pool];
1179 /* Free blocks of increasing size. */
1180 while ((order = max_order(m)) < VM_NFREEORDER - 1 &&
1181 m + (1 << order) <= m_end) {
1182 KASSERT(seg == &vm_phys_segs[m->segind],
1183 ("%s: page range [%p,%p) spans multiple segments",
1184 __func__, m_end - npages, m));
1185 vm_freelist_add(fl, m, order, 1);
1188 /* Free blocks of maximum size. */
1189 while (m + (1 << order) <= m_end) {
1190 KASSERT(seg == &vm_phys_segs[m->segind],
1191 ("%s: page range [%p,%p) spans multiple segments",
1192 __func__, m_end - npages, m));
1193 vm_freelist_add(fl, m, order, 1);
1196 /* Free blocks of diminishing size. */
1198 KASSERT(seg == &vm_phys_segs[m->segind],
1199 ("%s: page range [%p,%p) spans multiple segments",
1200 __func__, m_end - npages, m));
1201 order = flsl(m_end - m) - 1;
1202 vm_freelist_add(fl, m, order, 1);
1208 * Free a contiguous, arbitrarily sized set of physical pages.
1210 * The free page queues must be locked.
1213 vm_phys_free_contig(vm_page_t m, u_long npages)
1215 int order_start, order_end;
1216 vm_page_t m_start, m_end;
1218 vm_domain_free_assert_locked(vm_pagequeue_domain(m));
1221 order_start = max_order(m_start);
1222 if (order_start < VM_NFREEORDER - 1)
1223 m_start += 1 << order_start;
1225 order_end = max_order(m_end);
1226 if (order_end < VM_NFREEORDER - 1)
1227 m_end -= 1 << order_end;
1229 * Avoid unnecessary coalescing by freeing the pages at the start and
1230 * end of the range last.
1232 if (m_start < m_end)
1233 vm_phys_enqueue_contig(m_start, m_end - m_start);
1234 if (order_start < VM_NFREEORDER - 1)
1235 vm_phys_free_pages(m, order_start);
1236 if (order_end < VM_NFREEORDER - 1)
1237 vm_phys_free_pages(m_end, order_end);
1241 * Scan physical memory between the specified addresses "low" and "high" for a
1242 * run of contiguous physical pages that satisfy the specified conditions, and
1243 * return the lowest page in the run. The specified "alignment" determines
1244 * the alignment of the lowest physical page in the run. If the specified
1245 * "boundary" is non-zero, then the run of physical pages cannot span a
1246 * physical address that is a multiple of "boundary".
1248 * "npages" must be greater than zero. Both "alignment" and "boundary" must
1249 * be a power of two.
1252 vm_phys_scan_contig(int domain, u_long npages, vm_paddr_t low, vm_paddr_t high,
1253 u_long alignment, vm_paddr_t boundary, int options)
1256 vm_page_t m_end, m_run, m_start;
1257 struct vm_phys_seg *seg;
1260 KASSERT(npages > 0, ("npages is 0"));
1261 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1262 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1265 for (segind = 0; segind < vm_phys_nsegs; segind++) {
1266 seg = &vm_phys_segs[segind];
1267 if (seg->domain != domain)
1269 if (seg->start >= high)
1271 if (low >= seg->end)
1273 if (low <= seg->start)
1274 m_start = seg->first_page;
1276 m_start = &seg->first_page[atop(low - seg->start)];
1277 if (high < seg->end)
1281 if (pa_end - VM_PAGE_TO_PHYS(m_start) < ptoa(npages))
1283 m_end = &seg->first_page[atop(pa_end - seg->start)];
1284 m_run = vm_page_scan_contig(npages, m_start, m_end,
1285 alignment, boundary, options);
1293 * Search for the given physical page "m" in the free lists. If the search
1294 * succeeds, remove "m" from the free lists and return true. Otherwise, return
1295 * false, indicating that "m" is not in the free lists.
1297 * The free page queues must be locked.
1300 vm_phys_unfree_page(vm_page_t m)
1302 struct vm_freelist *fl;
1303 struct vm_phys_seg *seg;
1304 vm_paddr_t pa, pa_half;
1305 vm_page_t m_set, m_tmp;
1309 * First, find the contiguous, power of two-sized set of free
1310 * physical pages containing the given physical page "m" and
1311 * assign it to "m_set".
1313 seg = &vm_phys_segs[m->segind];
1314 vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
1315 for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
1316 order < VM_NFREEORDER - 1; ) {
1318 pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
1319 if (pa >= seg->start)
1320 m_set = &seg->first_page[atop(pa - seg->start)];
1324 if (m_set->order < order)
1326 if (m_set->order == VM_NFREEORDER)
1328 KASSERT(m_set->order < VM_NFREEORDER,
1329 ("vm_phys_unfree_page: page %p has unexpected order %d",
1330 m_set, m_set->order));
1333 * Next, remove "m_set" from the free lists. Finally, extract
1334 * "m" from "m_set" using an iterative algorithm: While "m_set"
1335 * is larger than a page, shrink "m_set" by returning the half
1336 * of "m_set" that does not contain "m" to the free lists.
1338 fl = (*seg->free_queues)[m_set->pool];
1339 order = m_set->order;
1340 vm_freelist_rem(fl, m_set, order);
1343 pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
1344 if (m->phys_addr < pa_half)
1345 m_tmp = &seg->first_page[atop(pa_half - seg->start)];
1348 m_set = &seg->first_page[atop(pa_half - seg->start)];
1350 vm_freelist_add(fl, m_tmp, order, 0);
1352 KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
1357 * Allocate a contiguous set of physical pages of the given size
1358 * "npages" from the free lists. All of the physical pages must be at
1359 * or above the given physical address "low" and below the given
1360 * physical address "high". The given value "alignment" determines the
1361 * alignment of the first physical page in the set. If the given value
1362 * "boundary" is non-zero, then the set of physical pages cannot cross
1363 * any physical address boundary that is a multiple of that value. Both
1364 * "alignment" and "boundary" must be a power of two.
1367 vm_phys_alloc_contig(int domain, u_long npages, vm_paddr_t low, vm_paddr_t high,
1368 u_long alignment, vm_paddr_t boundary)
1370 vm_paddr_t pa_end, pa_start;
1372 struct vm_phys_seg *seg;
1375 KASSERT(npages > 0, ("npages is 0"));
1376 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1377 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1378 vm_domain_free_assert_locked(VM_DOMAIN(domain));
1382 for (segind = vm_phys_nsegs - 1; segind >= 0; segind--) {
1383 seg = &vm_phys_segs[segind];
1384 if (seg->start >= high || seg->domain != domain)
1386 if (low >= seg->end)
1388 if (low <= seg->start)
1389 pa_start = seg->start;
1392 if (high < seg->end)
1396 if (pa_end - pa_start < ptoa(npages))
1398 m_run = vm_phys_alloc_seg_contig(seg, npages, low, high,
1399 alignment, boundary);
1407 * Allocate a run of contiguous physical pages from the free list for the
1408 * specified segment.
1411 vm_phys_alloc_seg_contig(struct vm_phys_seg *seg, u_long npages,
1412 vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
1414 struct vm_freelist *fl;
1415 vm_paddr_t pa, pa_end, size;
1418 int oind, order, pind;
1420 KASSERT(npages > 0, ("npages is 0"));
1421 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1422 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1423 vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
1424 /* Compute the queue that is the best fit for npages. */
1425 order = flsl(npages - 1);
1426 /* Search for a run satisfying the specified conditions. */
1427 size = npages << PAGE_SHIFT;
1428 for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER;
1430 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1431 fl = (*seg->free_queues)[pind];
1432 TAILQ_FOREACH(m_ret, &fl[oind].pl, listq) {
1434 * Is the size of this allocation request
1435 * larger than the largest block size?
1437 if (order >= VM_NFREEORDER) {
1439 * Determine if a sufficient number of
1440 * subsequent blocks to satisfy the
1441 * allocation request are free.
1443 pa = VM_PAGE_TO_PHYS(m_ret);
1448 pa += 1 << (PAGE_SHIFT +
1454 m = &seg->first_page[atop(pa -
1456 if (m->order != VM_NFREEORDER -
1460 /* If not, go to the next block. */
1466 * Determine if the blocks are within the
1467 * given range, satisfy the given alignment,
1468 * and do not cross the given boundary.
1470 pa = VM_PAGE_TO_PHYS(m_ret);
1472 if (pa >= low && pa_end <= high &&
1473 vm_addr_ok(pa, size, alignment, boundary))
1480 for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
1481 fl = (*seg->free_queues)[m->pool];
1482 vm_freelist_rem(fl, m, oind);
1483 if (m->pool != VM_FREEPOOL_DEFAULT)
1484 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, oind);
1486 /* Return excess pages to the free lists. */
1487 npages_end = roundup2(npages, 1 << oind);
1488 if (npages < npages_end) {
1489 fl = (*seg->free_queues)[VM_FREEPOOL_DEFAULT];
1490 vm_phys_enq_range(&m_ret[npages], npages_end - npages, fl, 0);
1496 * Return the index of the first unused slot which may be the terminating
1500 vm_phys_avail_count(void)
1504 for (i = 0; phys_avail[i + 1]; i += 2)
1506 if (i > PHYS_AVAIL_ENTRIES)
1507 panic("Improperly terminated phys_avail %d entries", i);
1513 * Assert that a phys_avail entry is valid.
1516 vm_phys_avail_check(int i)
1518 if (phys_avail[i] & PAGE_MASK)
1519 panic("Unaligned phys_avail[%d]: %#jx", i,
1520 (intmax_t)phys_avail[i]);
1521 if (phys_avail[i+1] & PAGE_MASK)
1522 panic("Unaligned phys_avail[%d + 1]: %#jx", i,
1523 (intmax_t)phys_avail[i]);
1524 if (phys_avail[i + 1] < phys_avail[i])
1525 panic("phys_avail[%d] start %#jx < end %#jx", i,
1526 (intmax_t)phys_avail[i], (intmax_t)phys_avail[i+1]);
1530 * Return the index of an overlapping phys_avail entry or -1.
1534 vm_phys_avail_find(vm_paddr_t pa)
1538 for (i = 0; phys_avail[i + 1]; i += 2)
1539 if (phys_avail[i] <= pa && phys_avail[i + 1] > pa)
1546 * Return the index of the largest entry.
1549 vm_phys_avail_largest(void)
1551 vm_paddr_t sz, largesz;
1557 for (i = 0; phys_avail[i + 1]; i += 2) {
1558 sz = vm_phys_avail_size(i);
1569 vm_phys_avail_size(int i)
1572 return (phys_avail[i + 1] - phys_avail[i]);
1576 * Split an entry at the address 'pa'. Return zero on success or errno.
1579 vm_phys_avail_split(vm_paddr_t pa, int i)
1583 vm_phys_avail_check(i);
1584 if (pa <= phys_avail[i] || pa >= phys_avail[i + 1])
1585 panic("vm_phys_avail_split: invalid address");
1586 cnt = vm_phys_avail_count();
1587 if (cnt >= PHYS_AVAIL_ENTRIES)
1589 memmove(&phys_avail[i + 2], &phys_avail[i],
1590 (cnt - i) * sizeof(phys_avail[0]));
1591 phys_avail[i + 1] = pa;
1592 phys_avail[i + 2] = pa;
1593 vm_phys_avail_check(i);
1594 vm_phys_avail_check(i+2);
1600 * Check if a given physical address can be included as part of a crash dump.
1603 vm_phys_is_dumpable(vm_paddr_t pa)
1608 if ((m = vm_phys_paddr_to_vm_page(pa)) != NULL)
1609 return ((m->flags & PG_NODUMP) == 0);
1611 for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) {
1612 if (pa >= dump_avail[i] && pa < dump_avail[i + 1])
1619 vm_phys_early_add_seg(vm_paddr_t start, vm_paddr_t end)
1621 struct vm_phys_seg *seg;
1623 if (vm_phys_early_nsegs == -1)
1624 panic("%s: called after initialization", __func__);
1625 if (vm_phys_early_nsegs == nitems(vm_phys_early_segs))
1626 panic("%s: ran out of early segments", __func__);
1628 seg = &vm_phys_early_segs[vm_phys_early_nsegs++];
1634 * This routine allocates NUMA node specific memory before the page
1635 * allocator is bootstrapped.
1638 vm_phys_early_alloc(int domain, size_t alloc_size)
1640 int i, mem_index, biggestone;
1641 vm_paddr_t pa, mem_start, mem_end, size, biggestsize, align;
1643 KASSERT(domain == -1 || (domain >= 0 && domain < vm_ndomains),
1644 ("%s: invalid domain index %d", __func__, domain));
1647 * Search the mem_affinity array for the biggest address
1648 * range in the desired domain. This is used to constrain
1649 * the phys_avail selection below.
1656 if (mem_affinity != NULL) {
1658 size = mem_affinity[i].end - mem_affinity[i].start;
1661 if (domain != -1 && mem_affinity[i].domain != domain)
1663 if (size > biggestsize) {
1668 mem_start = mem_affinity[mem_index].start;
1669 mem_end = mem_affinity[mem_index].end;
1674 * Now find biggest physical segment in within the desired
1679 for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1680 /* skip regions that are out of range */
1681 if (phys_avail[i+1] - alloc_size < mem_start ||
1682 phys_avail[i+1] > mem_end)
1684 size = vm_phys_avail_size(i);
1685 if (size > biggestsize) {
1690 alloc_size = round_page(alloc_size);
1693 * Grab single pages from the front to reduce fragmentation.
1695 if (alloc_size == PAGE_SIZE) {
1696 pa = phys_avail[biggestone];
1697 phys_avail[biggestone] += PAGE_SIZE;
1698 vm_phys_avail_check(biggestone);
1703 * Naturally align large allocations.
1705 align = phys_avail[biggestone + 1] & (alloc_size - 1);
1706 if (alloc_size + align > biggestsize)
1707 panic("cannot find a large enough size\n");
1709 vm_phys_avail_split(phys_avail[biggestone + 1] - align,
1711 /* Wasting memory. */
1712 phys_avail[biggestone + 1] -= align;
1714 phys_avail[biggestone + 1] -= alloc_size;
1715 vm_phys_avail_check(biggestone);
1716 pa = phys_avail[biggestone + 1];
1721 vm_phys_early_startup(void)
1723 struct vm_phys_seg *seg;
1726 for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1727 phys_avail[i] = round_page(phys_avail[i]);
1728 phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
1731 for (i = 0; i < vm_phys_early_nsegs; i++) {
1732 seg = &vm_phys_early_segs[i];
1733 vm_phys_add_seg(seg->start, seg->end);
1735 vm_phys_early_nsegs = -1;
1738 /* Force phys_avail to be split by domain. */
1739 if (mem_affinity != NULL) {
1742 for (i = 0; mem_affinity[i].end != 0; i++) {
1743 idx = vm_phys_avail_find(mem_affinity[i].start);
1745 phys_avail[idx] != mem_affinity[i].start)
1746 vm_phys_avail_split(mem_affinity[i].start, idx);
1747 idx = vm_phys_avail_find(mem_affinity[i].end);
1749 phys_avail[idx] != mem_affinity[i].end)
1750 vm_phys_avail_split(mem_affinity[i].end, idx);
1758 * Show the number of physical pages in each of the free lists.
1760 DB_SHOW_COMMAND(freepages, db_show_freepages)
1762 struct vm_freelist *fl;
1763 int flind, oind, pind, dom;
1765 for (dom = 0; dom < vm_ndomains; dom++) {
1766 db_printf("DOMAIN: %d\n", dom);
1767 for (flind = 0; flind < vm_nfreelists; flind++) {
1768 db_printf("FREE LIST %d:\n"
1769 "\n ORDER (SIZE) | NUMBER"
1771 for (pind = 0; pind < VM_NFREEPOOL; pind++)
1772 db_printf(" | POOL %d", pind);
1774 for (pind = 0; pind < VM_NFREEPOOL; pind++)
1775 db_printf("-- -- ");
1777 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
1778 db_printf(" %2.2d (%6.6dK)", oind,
1779 1 << (PAGE_SHIFT - 10 + oind));
1780 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1781 fl = vm_phys_free_queues[dom][flind][pind];
1782 db_printf(" | %6.6d", fl[oind].lcnt);