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>
49 #include <sys/domainset.h>
51 #include <sys/kernel.h>
52 #include <sys/malloc.h>
53 #include <sys/mutex.h>
55 #include <sys/queue.h>
56 #include <sys/rwlock.h>
58 #include <sys/sysctl.h>
60 #include <sys/vmmeter.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_param.h>
67 #include <vm/vm_kern.h>
68 #include <vm/vm_object.h>
69 #include <vm/vm_page.h>
70 #include <vm/vm_phys.h>
71 #include <vm/vm_pagequeue.h>
73 _Static_assert(sizeof(long) * NBBY >= VM_PHYSSEG_MAX,
74 "Too many physsegs.");
77 struct mem_affinity __read_mostly *mem_affinity;
78 int __read_mostly *mem_locality;
81 int __read_mostly vm_ndomains = 1;
82 domainset_t __read_mostly all_domains = DOMAINSET_T_INITIALIZER(0x1);
84 struct vm_phys_seg __read_mostly vm_phys_segs[VM_PHYSSEG_MAX];
85 int __read_mostly vm_phys_nsegs;
86 static struct vm_phys_seg vm_phys_early_segs[8];
87 static int vm_phys_early_nsegs;
89 struct vm_phys_fictitious_seg;
90 static int vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *,
91 struct vm_phys_fictitious_seg *);
93 RB_HEAD(fict_tree, vm_phys_fictitious_seg) vm_phys_fictitious_tree =
94 RB_INITIALIZER(&vm_phys_fictitious_tree);
96 struct vm_phys_fictitious_seg {
97 RB_ENTRY(vm_phys_fictitious_seg) node;
98 /* Memory region data */
101 vm_page_t first_page;
104 RB_GENERATE_STATIC(fict_tree, vm_phys_fictitious_seg, node,
105 vm_phys_fictitious_cmp);
107 static struct rwlock_padalign vm_phys_fictitious_reg_lock;
108 MALLOC_DEFINE(M_FICT_PAGES, "vm_fictitious", "Fictitious VM pages");
110 static struct vm_freelist __aligned(CACHE_LINE_SIZE)
111 vm_phys_free_queues[MAXMEMDOM][VM_NFREELIST][VM_NFREEPOOL]
114 static int __read_mostly vm_nfreelists;
117 * These "avail lists" are globals used to communicate boot-time physical
118 * memory layout to other parts of the kernel. Each physically contiguous
119 * region of memory is defined by a start address at an even index and an
120 * end address at the following odd index. Each list is terminated by a
121 * pair of zero entries.
123 * dump_avail tells the dump code what regions to include in a crash dump, and
124 * phys_avail is all of the remaining physical memory that is available for
127 * Initially dump_avail and phys_avail are identical. Boot time memory
128 * allocations remove extents from phys_avail that may still be included
131 vm_paddr_t phys_avail[PHYS_AVAIL_COUNT];
132 vm_paddr_t dump_avail[PHYS_AVAIL_COUNT];
135 * Provides the mapping from VM_FREELIST_* to free list indices (flind).
137 static int __read_mostly vm_freelist_to_flind[VM_NFREELIST];
139 CTASSERT(VM_FREELIST_DEFAULT == 0);
141 #ifdef VM_FREELIST_DMA32
142 #define VM_DMA32_BOUNDARY ((vm_paddr_t)1 << 32)
146 * Enforce the assumptions made by vm_phys_add_seg() and vm_phys_init() about
147 * the ordering of the free list boundaries.
149 #if defined(VM_LOWMEM_BOUNDARY) && defined(VM_DMA32_BOUNDARY)
150 CTASSERT(VM_LOWMEM_BOUNDARY < VM_DMA32_BOUNDARY);
153 static int sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS);
154 SYSCTL_OID(_vm, OID_AUTO, phys_free,
155 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
156 sysctl_vm_phys_free, "A",
159 static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS);
160 SYSCTL_OID(_vm, OID_AUTO, phys_segs,
161 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
162 sysctl_vm_phys_segs, "A",
166 static int sysctl_vm_phys_locality(SYSCTL_HANDLER_ARGS);
167 SYSCTL_OID(_vm, OID_AUTO, phys_locality,
168 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
169 sysctl_vm_phys_locality, "A",
170 "Phys Locality Info");
173 SYSCTL_INT(_vm, OID_AUTO, ndomains, CTLFLAG_RD,
174 &vm_ndomains, 0, "Number of physical memory domains available.");
176 static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain);
177 static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end);
178 static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl,
179 int order, int tail);
182 * Red-black tree helpers for vm fictitious range management.
185 vm_phys_fictitious_in_range(struct vm_phys_fictitious_seg *p,
186 struct vm_phys_fictitious_seg *range)
189 KASSERT(range->start != 0 && range->end != 0,
190 ("Invalid range passed on search for vm_fictitious page"));
191 if (p->start >= range->end)
193 if (p->start < range->start)
200 vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *p1,
201 struct vm_phys_fictitious_seg *p2)
204 /* Check if this is a search for a page */
206 return (vm_phys_fictitious_in_range(p1, p2));
208 KASSERT(p2->end != 0,
209 ("Invalid range passed as second parameter to vm fictitious comparison"));
211 /* Searching to add a new range */
212 if (p1->end <= p2->start)
214 if (p1->start >= p2->end)
217 panic("Trying to add overlapping vm fictitious ranges:\n"
218 "[%#jx:%#jx] and [%#jx:%#jx]", (uintmax_t)p1->start,
219 (uintmax_t)p1->end, (uintmax_t)p2->start, (uintmax_t)p2->end);
223 vm_phys_domain_match(int prefer, vm_paddr_t low, vm_paddr_t high)
229 if (vm_ndomains == 1 || mem_affinity == NULL)
232 DOMAINSET_ZERO(&mask);
234 * Check for any memory that overlaps low, high.
236 for (i = 0; mem_affinity[i].end != 0; i++)
237 if (mem_affinity[i].start <= high &&
238 mem_affinity[i].end >= low)
239 DOMAINSET_SET(mem_affinity[i].domain, &mask);
240 if (prefer != -1 && DOMAINSET_ISSET(prefer, &mask))
242 if (DOMAINSET_EMPTY(&mask))
243 panic("vm_phys_domain_match: Impossible constraint");
244 return (DOMAINSET_FFS(&mask) - 1);
251 * Outputs the state of the physical memory allocator, specifically,
252 * the amount of physical memory in each free list.
255 sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS)
258 struct vm_freelist *fl;
259 int dom, error, flind, oind, pind;
261 error = sysctl_wire_old_buffer(req, 0);
264 sbuf_new_for_sysctl(&sbuf, NULL, 128 * vm_ndomains, req);
265 for (dom = 0; dom < vm_ndomains; dom++) {
266 sbuf_printf(&sbuf,"\nDOMAIN %d:\n", dom);
267 for (flind = 0; flind < vm_nfreelists; flind++) {
268 sbuf_printf(&sbuf, "\nFREE LIST %d:\n"
269 "\n ORDER (SIZE) | NUMBER"
271 for (pind = 0; pind < VM_NFREEPOOL; pind++)
272 sbuf_printf(&sbuf, " | POOL %d", pind);
273 sbuf_printf(&sbuf, "\n-- ");
274 for (pind = 0; pind < VM_NFREEPOOL; pind++)
275 sbuf_printf(&sbuf, "-- -- ");
276 sbuf_printf(&sbuf, "--\n");
277 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
278 sbuf_printf(&sbuf, " %2d (%6dK)", oind,
279 1 << (PAGE_SHIFT - 10 + oind));
280 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
281 fl = vm_phys_free_queues[dom][flind][pind];
282 sbuf_printf(&sbuf, " | %6d",
285 sbuf_printf(&sbuf, "\n");
289 error = sbuf_finish(&sbuf);
295 * Outputs the set of physical memory segments.
298 sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS)
301 struct vm_phys_seg *seg;
304 error = sysctl_wire_old_buffer(req, 0);
307 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
308 for (segind = 0; segind < vm_phys_nsegs; segind++) {
309 sbuf_printf(&sbuf, "\nSEGMENT %d:\n\n", segind);
310 seg = &vm_phys_segs[segind];
311 sbuf_printf(&sbuf, "start: %#jx\n",
312 (uintmax_t)seg->start);
313 sbuf_printf(&sbuf, "end: %#jx\n",
314 (uintmax_t)seg->end);
315 sbuf_printf(&sbuf, "domain: %d\n", seg->domain);
316 sbuf_printf(&sbuf, "free list: %p\n", seg->free_queues);
318 error = sbuf_finish(&sbuf);
324 * Return affinity, or -1 if there's no affinity information.
327 vm_phys_mem_affinity(int f, int t)
331 if (mem_locality == NULL)
333 if (f >= vm_ndomains || t >= vm_ndomains)
335 return (mem_locality[f * vm_ndomains + t]);
343 * Outputs the VM locality table.
346 sysctl_vm_phys_locality(SYSCTL_HANDLER_ARGS)
351 error = sysctl_wire_old_buffer(req, 0);
354 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
356 sbuf_printf(&sbuf, "\n");
358 for (i = 0; i < vm_ndomains; i++) {
359 sbuf_printf(&sbuf, "%d: ", i);
360 for (j = 0; j < vm_ndomains; j++) {
361 sbuf_printf(&sbuf, "%d ", vm_phys_mem_affinity(i, j));
363 sbuf_printf(&sbuf, "\n");
365 error = sbuf_finish(&sbuf);
372 vm_freelist_add(struct vm_freelist *fl, vm_page_t m, int order, int tail)
377 TAILQ_INSERT_TAIL(&fl[order].pl, m, listq);
379 TAILQ_INSERT_HEAD(&fl[order].pl, m, listq);
384 vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order)
387 TAILQ_REMOVE(&fl[order].pl, m, listq);
389 m->order = VM_NFREEORDER;
393 * Create a physical memory segment.
396 _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain)
398 struct vm_phys_seg *seg;
400 KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX,
401 ("vm_phys_create_seg: increase VM_PHYSSEG_MAX"));
402 KASSERT(domain >= 0 && domain < vm_ndomains,
403 ("vm_phys_create_seg: invalid domain provided"));
404 seg = &vm_phys_segs[vm_phys_nsegs++];
405 while (seg > vm_phys_segs && (seg - 1)->start >= end) {
411 seg->domain = domain;
415 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end)
420 if (mem_affinity == NULL) {
421 _vm_phys_create_seg(start, end, 0);
426 if (mem_affinity[i].end == 0)
427 panic("Reached end of affinity info");
428 if (mem_affinity[i].end <= start)
430 if (mem_affinity[i].start > start)
431 panic("No affinity info for start %jx",
433 if (mem_affinity[i].end >= end) {
434 _vm_phys_create_seg(start, end,
435 mem_affinity[i].domain);
438 _vm_phys_create_seg(start, mem_affinity[i].end,
439 mem_affinity[i].domain);
440 start = mem_affinity[i].end;
443 _vm_phys_create_seg(start, end, 0);
448 * Add a physical memory segment.
451 vm_phys_add_seg(vm_paddr_t start, vm_paddr_t end)
455 KASSERT((start & PAGE_MASK) == 0,
456 ("vm_phys_define_seg: start is not page aligned"));
457 KASSERT((end & PAGE_MASK) == 0,
458 ("vm_phys_define_seg: end is not page aligned"));
461 * Split the physical memory segment if it spans two or more free
465 #ifdef VM_FREELIST_LOWMEM
466 if (paddr < VM_LOWMEM_BOUNDARY && end > VM_LOWMEM_BOUNDARY) {
467 vm_phys_create_seg(paddr, VM_LOWMEM_BOUNDARY);
468 paddr = VM_LOWMEM_BOUNDARY;
471 #ifdef VM_FREELIST_DMA32
472 if (paddr < VM_DMA32_BOUNDARY && end > VM_DMA32_BOUNDARY) {
473 vm_phys_create_seg(paddr, VM_DMA32_BOUNDARY);
474 paddr = VM_DMA32_BOUNDARY;
477 vm_phys_create_seg(paddr, end);
481 * Initialize the physical memory allocator.
483 * Requires that vm_page_array is initialized!
488 struct vm_freelist *fl;
489 struct vm_phys_seg *end_seg, *prev_seg, *seg, *tmp_seg;
491 int dom, flind, freelist, oind, pind, segind;
494 * Compute the number of free lists, and generate the mapping from the
495 * manifest constants VM_FREELIST_* to the free list indices.
497 * Initially, the entries of vm_freelist_to_flind[] are set to either
498 * 0 or 1 to indicate which free lists should be created.
501 for (segind = vm_phys_nsegs - 1; segind >= 0; segind--) {
502 seg = &vm_phys_segs[segind];
503 #ifdef VM_FREELIST_LOWMEM
504 if (seg->end <= VM_LOWMEM_BOUNDARY)
505 vm_freelist_to_flind[VM_FREELIST_LOWMEM] = 1;
508 #ifdef VM_FREELIST_DMA32
510 #ifdef VM_DMA32_NPAGES_THRESHOLD
512 * Create the DMA32 free list only if the amount of
513 * physical memory above physical address 4G exceeds the
516 npages > VM_DMA32_NPAGES_THRESHOLD &&
518 seg->end <= VM_DMA32_BOUNDARY)
519 vm_freelist_to_flind[VM_FREELIST_DMA32] = 1;
523 npages += atop(seg->end - seg->start);
524 vm_freelist_to_flind[VM_FREELIST_DEFAULT] = 1;
527 /* Change each entry into a running total of the free lists. */
528 for (freelist = 1; freelist < VM_NFREELIST; freelist++) {
529 vm_freelist_to_flind[freelist] +=
530 vm_freelist_to_flind[freelist - 1];
532 vm_nfreelists = vm_freelist_to_flind[VM_NFREELIST - 1];
533 KASSERT(vm_nfreelists > 0, ("vm_phys_init: no free lists"));
534 /* Change each entry into a free list index. */
535 for (freelist = 0; freelist < VM_NFREELIST; freelist++)
536 vm_freelist_to_flind[freelist]--;
539 * Initialize the first_page and free_queues fields of each physical
542 #ifdef VM_PHYSSEG_SPARSE
545 for (segind = 0; segind < vm_phys_nsegs; segind++) {
546 seg = &vm_phys_segs[segind];
547 #ifdef VM_PHYSSEG_SPARSE
548 seg->first_page = &vm_page_array[npages];
549 npages += atop(seg->end - seg->start);
551 seg->first_page = PHYS_TO_VM_PAGE(seg->start);
553 #ifdef VM_FREELIST_LOWMEM
554 if (seg->end <= VM_LOWMEM_BOUNDARY) {
555 flind = vm_freelist_to_flind[VM_FREELIST_LOWMEM];
557 ("vm_phys_init: LOWMEM flind < 0"));
560 #ifdef VM_FREELIST_DMA32
561 if (seg->end <= VM_DMA32_BOUNDARY) {
562 flind = vm_freelist_to_flind[VM_FREELIST_DMA32];
564 ("vm_phys_init: DMA32 flind < 0"));
568 flind = vm_freelist_to_flind[VM_FREELIST_DEFAULT];
570 ("vm_phys_init: DEFAULT flind < 0"));
572 seg->free_queues = &vm_phys_free_queues[seg->domain][flind];
576 * Coalesce physical memory segments that are contiguous and share the
577 * same per-domain free queues.
579 prev_seg = vm_phys_segs;
580 seg = &vm_phys_segs[1];
581 end_seg = &vm_phys_segs[vm_phys_nsegs];
582 while (seg < end_seg) {
583 if (prev_seg->end == seg->start &&
584 prev_seg->free_queues == seg->free_queues) {
585 prev_seg->end = seg->end;
586 KASSERT(prev_seg->domain == seg->domain,
587 ("vm_phys_init: free queues cannot span domains"));
590 for (tmp_seg = seg; tmp_seg < end_seg; tmp_seg++)
591 *tmp_seg = *(tmp_seg + 1);
599 * Initialize the free queues.
601 for (dom = 0; dom < vm_ndomains; dom++) {
602 for (flind = 0; flind < vm_nfreelists; flind++) {
603 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
604 fl = vm_phys_free_queues[dom][flind][pind];
605 for (oind = 0; oind < VM_NFREEORDER; oind++)
606 TAILQ_INIT(&fl[oind].pl);
611 rw_init(&vm_phys_fictitious_reg_lock, "vmfctr");
615 * Register info about the NUMA topology of the system.
617 * Invoked by platform-dependent code prior to vm_phys_init().
620 vm_phys_register_domains(int ndomains, struct mem_affinity *affinity,
627 * For now the only override value that we support is 1, which
628 * effectively disables NUMA-awareness in the allocators.
631 TUNABLE_INT_FETCH("vm.numa.disabled", &d);
636 vm_ndomains = ndomains;
637 mem_affinity = affinity;
638 mem_locality = locality;
641 for (i = 0; i < vm_ndomains; i++)
642 DOMAINSET_SET(i, &all_domains);
651 * Split a contiguous, power of two-sized set of physical pages.
653 * When this function is called by a page allocation function, the caller
654 * should request insertion at the head unless the order [order, oind) queues
655 * are known to be empty. The objective being to reduce the likelihood of
656 * long-term fragmentation by promoting contemporaneous allocation and
657 * (hopefully) deallocation.
660 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order,
665 while (oind > order) {
667 m_buddy = &m[1 << oind];
668 KASSERT(m_buddy->order == VM_NFREEORDER,
669 ("vm_phys_split_pages: page %p has unexpected order %d",
670 m_buddy, m_buddy->order));
671 vm_freelist_add(fl, m_buddy, oind, tail);
676 * Add the physical pages [m, m + npages) at the end of a power-of-two aligned
677 * and sized set to the specified free list.
679 * When this function is called by a page allocation function, the caller
680 * should request insertion at the head unless the lower-order queues are
681 * known to be empty. The objective being to reduce the likelihood of long-
682 * term fragmentation by promoting contemporaneous allocation and (hopefully)
685 * The physical page m's buddy must not be free.
688 vm_phys_enq_range(vm_page_t m, u_int npages, struct vm_freelist *fl, int tail)
693 KASSERT(npages > 0, ("vm_phys_enq_range: npages is 0"));
694 KASSERT(((VM_PAGE_TO_PHYS(m) + npages * PAGE_SIZE) &
695 ((PAGE_SIZE << (fls(npages) - 1)) - 1)) == 0,
696 ("vm_phys_enq_range: page %p and npages %u are misaligned",
699 KASSERT(m->order == VM_NFREEORDER,
700 ("vm_phys_enq_range: page %p has unexpected order %d",
702 order = ffs(npages) - 1;
703 KASSERT(order < VM_NFREEORDER,
704 ("vm_phys_enq_range: order %d is out of range", order));
705 vm_freelist_add(fl, m, order, tail);
709 } while (npages > 0);
713 * Set the pool for a contiguous, power of two-sized set of physical pages.
716 vm_phys_set_pool(int pool, vm_page_t m, int order)
720 for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
725 * Tries to allocate the specified number of pages from the specified pool
726 * within the specified domain. Returns the actual number of allocated pages
727 * and a pointer to each page through the array ma[].
729 * The returned pages may not be physically contiguous. However, in contrast
730 * to performing multiple, back-to-back calls to vm_phys_alloc_pages(..., 0),
731 * calling this function once to allocate the desired number of pages will
732 * avoid wasted time in vm_phys_split_pages().
734 * The free page queues for the specified domain must be locked.
737 vm_phys_alloc_npages(int domain, int pool, int npages, vm_page_t ma[])
739 struct vm_freelist *alt, *fl;
741 int avail, end, flind, freelist, i, need, oind, pind;
743 KASSERT(domain >= 0 && domain < vm_ndomains,
744 ("vm_phys_alloc_npages: domain %d is out of range", domain));
745 KASSERT(pool < VM_NFREEPOOL,
746 ("vm_phys_alloc_npages: pool %d is out of range", pool));
747 KASSERT(npages <= 1 << (VM_NFREEORDER - 1),
748 ("vm_phys_alloc_npages: npages %d is out of range", npages));
749 vm_domain_free_assert_locked(VM_DOMAIN(domain));
751 for (freelist = 0; freelist < VM_NFREELIST; freelist++) {
752 flind = vm_freelist_to_flind[freelist];
755 fl = vm_phys_free_queues[domain][flind][pool];
756 for (oind = 0; oind < VM_NFREEORDER; oind++) {
757 while ((m = TAILQ_FIRST(&fl[oind].pl)) != NULL) {
758 vm_freelist_rem(fl, m, oind);
760 need = imin(npages - i, avail);
761 for (end = i + need; i < end;)
765 * Return excess pages to fl. Its
766 * order [0, oind) queues are empty.
768 vm_phys_enq_range(m, avail - need, fl,
771 } else if (i == npages)
775 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
776 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
777 alt = vm_phys_free_queues[domain][flind][pind];
778 while ((m = TAILQ_FIRST(&alt[oind].pl)) !=
780 vm_freelist_rem(alt, m, oind);
781 vm_phys_set_pool(pool, m, oind);
783 need = imin(npages - i, avail);
784 for (end = i + need; i < end;)
788 * Return excess pages to fl.
789 * Its order [0, oind) queues
792 vm_phys_enq_range(m, avail -
795 } else if (i == npages)
805 * Allocate a contiguous, power of two-sized set of physical pages
806 * from the free lists.
808 * The free page queues must be locked.
811 vm_phys_alloc_pages(int domain, int pool, int order)
816 for (freelist = 0; freelist < VM_NFREELIST; freelist++) {
817 m = vm_phys_alloc_freelist_pages(domain, freelist, pool, order);
825 * Allocate a contiguous, power of two-sized set of physical pages from the
826 * specified free list. The free list must be specified using one of the
827 * manifest constants VM_FREELIST_*.
829 * The free page queues must be locked.
832 vm_phys_alloc_freelist_pages(int domain, int freelist, int pool, int order)
834 struct vm_freelist *alt, *fl;
836 int oind, pind, flind;
838 KASSERT(domain >= 0 && domain < vm_ndomains,
839 ("vm_phys_alloc_freelist_pages: domain %d is out of range",
841 KASSERT(freelist < VM_NFREELIST,
842 ("vm_phys_alloc_freelist_pages: freelist %d is out of range",
844 KASSERT(pool < VM_NFREEPOOL,
845 ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
846 KASSERT(order < VM_NFREEORDER,
847 ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
849 flind = vm_freelist_to_flind[freelist];
850 /* Check if freelist is present */
854 vm_domain_free_assert_locked(VM_DOMAIN(domain));
855 fl = &vm_phys_free_queues[domain][flind][pool][0];
856 for (oind = order; oind < VM_NFREEORDER; oind++) {
857 m = TAILQ_FIRST(&fl[oind].pl);
859 vm_freelist_rem(fl, m, oind);
860 /* The order [order, oind) queues are empty. */
861 vm_phys_split_pages(m, oind, fl, order, 1);
867 * The given pool was empty. Find the largest
868 * contiguous, power-of-two-sized set of pages in any
869 * pool. Transfer these pages to the given pool, and
870 * use them to satisfy the allocation.
872 for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
873 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
874 alt = &vm_phys_free_queues[domain][flind][pind][0];
875 m = TAILQ_FIRST(&alt[oind].pl);
877 vm_freelist_rem(alt, m, oind);
878 vm_phys_set_pool(pool, m, oind);
879 /* The order [order, oind) queues are empty. */
880 vm_phys_split_pages(m, oind, fl, order, 1);
889 * Find the vm_page corresponding to the given physical address.
892 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
894 struct vm_phys_seg *seg;
897 for (segind = 0; segind < vm_phys_nsegs; segind++) {
898 seg = &vm_phys_segs[segind];
899 if (pa >= seg->start && pa < seg->end)
900 return (&seg->first_page[atop(pa - seg->start)]);
906 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
908 struct vm_phys_fictitious_seg tmp, *seg;
915 rw_rlock(&vm_phys_fictitious_reg_lock);
916 seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
917 rw_runlock(&vm_phys_fictitious_reg_lock);
921 m = &seg->first_page[atop(pa - seg->start)];
922 KASSERT((m->flags & PG_FICTITIOUS) != 0, ("%p not fictitious", m));
928 vm_phys_fictitious_init_range(vm_page_t range, vm_paddr_t start,
929 long page_count, vm_memattr_t memattr)
933 bzero(range, page_count * sizeof(*range));
934 for (i = 0; i < page_count; i++) {
935 vm_page_initfake(&range[i], start + PAGE_SIZE * i, memattr);
936 range[i].oflags &= ~VPO_UNMANAGED;
937 range[i].busy_lock = VPB_UNBUSIED;
942 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
943 vm_memattr_t memattr)
945 struct vm_phys_fictitious_seg *seg;
948 #ifdef VM_PHYSSEG_DENSE
954 ("Start of segment isn't less than end (start: %jx end: %jx)",
955 (uintmax_t)start, (uintmax_t)end));
957 page_count = (end - start) / PAGE_SIZE;
959 #ifdef VM_PHYSSEG_DENSE
962 if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
963 fp = &vm_page_array[pi - first_page];
964 if ((pe - first_page) > vm_page_array_size) {
966 * We have a segment that starts inside
967 * of vm_page_array, but ends outside of it.
969 * Use vm_page_array pages for those that are
970 * inside of the vm_page_array range, and
971 * allocate the remaining ones.
973 dpage_count = vm_page_array_size - (pi - first_page);
974 vm_phys_fictitious_init_range(fp, start, dpage_count,
976 page_count -= dpage_count;
977 start += ptoa(dpage_count);
981 * We can allocate the full range from vm_page_array,
982 * so there's no need to register the range in the tree.
984 vm_phys_fictitious_init_range(fp, start, page_count, memattr);
986 } else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
988 * We have a segment that ends inside of vm_page_array,
989 * but starts outside of it.
991 fp = &vm_page_array[0];
992 dpage_count = pe - first_page;
993 vm_phys_fictitious_init_range(fp, ptoa(first_page), dpage_count,
995 end -= ptoa(dpage_count);
996 page_count -= dpage_count;
998 } else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
1000 * Trying to register a fictitious range that expands before
1001 * and after vm_page_array.
1007 fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
1009 #ifdef VM_PHYSSEG_DENSE
1012 vm_phys_fictitious_init_range(fp, start, page_count, memattr);
1014 seg = malloc(sizeof(*seg), M_FICT_PAGES, M_WAITOK | M_ZERO);
1017 seg->first_page = fp;
1019 rw_wlock(&vm_phys_fictitious_reg_lock);
1020 RB_INSERT(fict_tree, &vm_phys_fictitious_tree, seg);
1021 rw_wunlock(&vm_phys_fictitious_reg_lock);
1027 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
1029 struct vm_phys_fictitious_seg *seg, tmp;
1030 #ifdef VM_PHYSSEG_DENSE
1034 KASSERT(start < end,
1035 ("Start of segment isn't less than end (start: %jx end: %jx)",
1036 (uintmax_t)start, (uintmax_t)end));
1038 #ifdef VM_PHYSSEG_DENSE
1041 if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
1042 if ((pe - first_page) <= vm_page_array_size) {
1044 * This segment was allocated using vm_page_array
1045 * only, there's nothing to do since those pages
1046 * were never added to the tree.
1051 * We have a segment that starts inside
1052 * of vm_page_array, but ends outside of it.
1054 * Calculate how many pages were added to the
1055 * tree and free them.
1057 start = ptoa(first_page + vm_page_array_size);
1058 } else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
1060 * We have a segment that ends inside of vm_page_array,
1061 * but starts outside of it.
1063 end = ptoa(first_page);
1064 } else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
1065 /* Since it's not possible to register such a range, panic. */
1067 "Unregistering not registered fictitious range [%#jx:%#jx]",
1068 (uintmax_t)start, (uintmax_t)end);
1074 rw_wlock(&vm_phys_fictitious_reg_lock);
1075 seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
1076 if (seg->start != start || seg->end != end) {
1077 rw_wunlock(&vm_phys_fictitious_reg_lock);
1079 "Unregistering not registered fictitious range [%#jx:%#jx]",
1080 (uintmax_t)start, (uintmax_t)end);
1082 RB_REMOVE(fict_tree, &vm_phys_fictitious_tree, seg);
1083 rw_wunlock(&vm_phys_fictitious_reg_lock);
1084 free(seg->first_page, M_FICT_PAGES);
1085 free(seg, M_FICT_PAGES);
1089 * Free a contiguous, power of two-sized set of physical pages.
1091 * The free page queues must be locked.
1094 vm_phys_free_pages(vm_page_t m, int order)
1096 struct vm_freelist *fl;
1097 struct vm_phys_seg *seg;
1101 KASSERT(m->order == VM_NFREEORDER,
1102 ("vm_phys_free_pages: page %p has unexpected order %d",
1104 KASSERT(m->pool < VM_NFREEPOOL,
1105 ("vm_phys_free_pages: page %p has unexpected pool %d",
1107 KASSERT(order < VM_NFREEORDER,
1108 ("vm_phys_free_pages: order %d is out of range", order));
1109 seg = &vm_phys_segs[m->segind];
1110 vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
1111 if (order < VM_NFREEORDER - 1) {
1112 pa = VM_PAGE_TO_PHYS(m);
1114 pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
1115 if (pa < seg->start || pa >= seg->end)
1117 m_buddy = &seg->first_page[atop(pa - seg->start)];
1118 if (m_buddy->order != order)
1120 fl = (*seg->free_queues)[m_buddy->pool];
1121 vm_freelist_rem(fl, m_buddy, order);
1122 if (m_buddy->pool != m->pool)
1123 vm_phys_set_pool(m->pool, m_buddy, order);
1125 pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
1126 m = &seg->first_page[atop(pa - seg->start)];
1127 } while (order < VM_NFREEORDER - 1);
1129 fl = (*seg->free_queues)[m->pool];
1130 vm_freelist_add(fl, m, order, 1);
1134 * Return the largest possible order of a set of pages starting at m.
1137 max_order(vm_page_t m)
1141 * Unsigned "min" is used here so that "order" is assigned
1142 * "VM_NFREEORDER - 1" when "m"'s physical address is zero
1143 * or the low-order bits of its physical address are zero
1144 * because the size of a physical address exceeds the size of
1147 return (min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
1148 VM_NFREEORDER - 1));
1152 * Free a contiguous, arbitrarily sized set of physical pages, without
1153 * merging across set boundaries.
1155 * The free page queues must be locked.
1158 vm_phys_enqueue_contig(vm_page_t m, u_long npages)
1160 struct vm_freelist *fl;
1161 struct vm_phys_seg *seg;
1166 * Avoid unnecessary coalescing by freeing the pages in the largest
1167 * possible power-of-two-sized subsets.
1169 vm_domain_free_assert_locked(vm_pagequeue_domain(m));
1170 seg = &vm_phys_segs[m->segind];
1171 fl = (*seg->free_queues)[m->pool];
1173 /* Free blocks of increasing size. */
1174 while ((order = max_order(m)) < VM_NFREEORDER - 1 &&
1175 m + (1 << order) <= m_end) {
1176 KASSERT(seg == &vm_phys_segs[m->segind],
1177 ("%s: page range [%p,%p) spans multiple segments",
1178 __func__, m_end - npages, m));
1179 vm_freelist_add(fl, m, order, 1);
1182 /* Free blocks of maximum size. */
1183 while (m + (1 << order) <= m_end) {
1184 KASSERT(seg == &vm_phys_segs[m->segind],
1185 ("%s: page range [%p,%p) spans multiple segments",
1186 __func__, m_end - npages, m));
1187 vm_freelist_add(fl, m, order, 1);
1190 /* Free blocks of diminishing size. */
1192 KASSERT(seg == &vm_phys_segs[m->segind],
1193 ("%s: page range [%p,%p) spans multiple segments",
1194 __func__, m_end - npages, m));
1195 order = flsl(m_end - m) - 1;
1196 vm_freelist_add(fl, m, order, 1);
1202 * Free a contiguous, arbitrarily sized set of physical pages.
1204 * The free page queues must be locked.
1207 vm_phys_free_contig(vm_page_t m, u_long npages)
1209 int order_start, order_end;
1210 vm_page_t m_start, m_end;
1212 vm_domain_free_assert_locked(vm_pagequeue_domain(m));
1215 order_start = max_order(m_start);
1216 if (order_start < VM_NFREEORDER - 1)
1217 m_start += 1 << order_start;
1219 order_end = max_order(m_end);
1220 if (order_end < VM_NFREEORDER - 1)
1221 m_end -= 1 << order_end;
1223 * Avoid unnecessary coalescing by freeing the pages at the start and
1224 * end of the range last.
1226 if (m_start < m_end)
1227 vm_phys_enqueue_contig(m_start, m_end - m_start);
1228 if (order_start < VM_NFREEORDER - 1)
1229 vm_phys_free_pages(m, order_start);
1230 if (order_end < VM_NFREEORDER - 1)
1231 vm_phys_free_pages(m_end, order_end);
1235 * Scan physical memory between the specified addresses "low" and "high" for a
1236 * run of contiguous physical pages that satisfy the specified conditions, and
1237 * return the lowest page in the run. The specified "alignment" determines
1238 * the alignment of the lowest physical page in the run. If the specified
1239 * "boundary" is non-zero, then the run of physical pages cannot span a
1240 * physical address that is a multiple of "boundary".
1242 * "npages" must be greater than zero. Both "alignment" and "boundary" must
1243 * be a power of two.
1246 vm_phys_scan_contig(int domain, u_long npages, vm_paddr_t low, vm_paddr_t high,
1247 u_long alignment, vm_paddr_t boundary, int options)
1250 vm_page_t m_end, m_run, m_start;
1251 struct vm_phys_seg *seg;
1254 KASSERT(npages > 0, ("npages is 0"));
1255 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1256 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1259 for (segind = 0; segind < vm_phys_nsegs; segind++) {
1260 seg = &vm_phys_segs[segind];
1261 if (seg->domain != domain)
1263 if (seg->start >= high)
1265 if (low >= seg->end)
1267 if (low <= seg->start)
1268 m_start = seg->first_page;
1270 m_start = &seg->first_page[atop(low - seg->start)];
1271 if (high < seg->end)
1275 if (pa_end - VM_PAGE_TO_PHYS(m_start) < ptoa(npages))
1277 m_end = &seg->first_page[atop(pa_end - seg->start)];
1278 m_run = vm_page_scan_contig(npages, m_start, m_end,
1279 alignment, boundary, options);
1287 * Search for the given physical page "m" in the free lists. If the search
1288 * succeeds, remove "m" from the free lists and return TRUE. Otherwise, return
1289 * FALSE, indicating that "m" is not in the free lists.
1291 * The free page queues must be locked.
1294 vm_phys_unfree_page(vm_page_t m)
1296 struct vm_freelist *fl;
1297 struct vm_phys_seg *seg;
1298 vm_paddr_t pa, pa_half;
1299 vm_page_t m_set, m_tmp;
1303 * First, find the contiguous, power of two-sized set of free
1304 * physical pages containing the given physical page "m" and
1305 * assign it to "m_set".
1307 seg = &vm_phys_segs[m->segind];
1308 vm_domain_free_assert_locked(VM_DOMAIN(seg->domain));
1309 for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
1310 order < VM_NFREEORDER - 1; ) {
1312 pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
1313 if (pa >= seg->start)
1314 m_set = &seg->first_page[atop(pa - seg->start)];
1318 if (m_set->order < order)
1320 if (m_set->order == VM_NFREEORDER)
1322 KASSERT(m_set->order < VM_NFREEORDER,
1323 ("vm_phys_unfree_page: page %p has unexpected order %d",
1324 m_set, m_set->order));
1327 * Next, remove "m_set" from the free lists. Finally, extract
1328 * "m" from "m_set" using an iterative algorithm: While "m_set"
1329 * is larger than a page, shrink "m_set" by returning the half
1330 * of "m_set" that does not contain "m" to the free lists.
1332 fl = (*seg->free_queues)[m_set->pool];
1333 order = m_set->order;
1334 vm_freelist_rem(fl, m_set, order);
1337 pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
1338 if (m->phys_addr < pa_half)
1339 m_tmp = &seg->first_page[atop(pa_half - seg->start)];
1342 m_set = &seg->first_page[atop(pa_half - seg->start)];
1344 vm_freelist_add(fl, m_tmp, order, 0);
1346 KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
1351 * Allocate a run of contiguous physical pages from the specified free list
1355 vm_phys_alloc_queues_contig(
1356 struct vm_freelist (*queues)[VM_NFREEPOOL][VM_NFREEORDER_MAX],
1357 u_long npages, vm_paddr_t low, vm_paddr_t high,
1358 u_long alignment, vm_paddr_t boundary)
1360 struct vm_phys_seg *seg;
1361 struct vm_freelist *fl;
1362 vm_paddr_t pa, pa_end, size;
1365 int oind, order, pind;
1367 KASSERT(npages > 0, ("npages is 0"));
1368 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1369 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1370 /* Compute the queue that is the best fit for npages. */
1371 order = flsl(npages - 1);
1372 /* Search for a run satisfying the specified conditions. */
1373 size = npages << PAGE_SHIFT;
1374 for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER;
1376 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1377 fl = (*queues)[pind];
1378 TAILQ_FOREACH(m_ret, &fl[oind].pl, listq) {
1380 * Determine if the address range starting at pa
1381 * is within the given range, satisfies the
1382 * given alignment, and does not cross the given
1385 pa = VM_PAGE_TO_PHYS(m_ret);
1387 if (pa < low || pa_end > high ||
1388 !vm_addr_ok(pa, size, alignment, boundary))
1392 * Is the size of this allocation request
1393 * no more than the largest block size?
1395 if (order < VM_NFREEORDER)
1399 * Determine if the address range is valid
1400 * (without overflow in pa_end calculation)
1401 * and fits within the segment.
1403 seg = &vm_phys_segs[m_ret->segind];
1404 if (pa_end < pa || seg->end < pa_end)
1408 * Determine if a series of free oind-blocks
1409 * starting here can satisfy the allocation
1414 (PAGE_SHIFT + VM_NFREEORDER - 1);
1417 } while (VM_NFREEORDER - 1 == seg->first_page[
1418 atop(pa - seg->start)].order);
1421 * Determine if an additional series of free
1422 * blocks of diminishing size can help to
1423 * satisfy the allocation request.
1426 m = &seg->first_page[
1427 atop(pa - seg->start)];
1428 if (m->order == VM_NFREEORDER ||
1429 pa + (2 << (PAGE_SHIFT + m->order))
1432 pa += 1 << (PAGE_SHIFT + m->order);
1441 for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
1442 fl = (*queues)[m->pool];
1444 vm_freelist_rem(fl, m, oind);
1445 if (m->pool != VM_FREEPOOL_DEFAULT)
1446 vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, oind);
1448 /* Return excess pages to the free lists. */
1449 npages_end = roundup2(npages, 1 << oind);
1450 if (npages < npages_end) {
1451 fl = (*queues)[VM_FREEPOOL_DEFAULT];
1452 vm_phys_enq_range(&m_ret[npages], npages_end - npages, fl, 0);
1458 * Allocate a contiguous set of physical pages of the given size
1459 * "npages" from the free lists. All of the physical pages must be at
1460 * or above the given physical address "low" and below the given
1461 * physical address "high". The given value "alignment" determines the
1462 * alignment of the first physical page in the set. If the given value
1463 * "boundary" is non-zero, then the set of physical pages cannot cross
1464 * any physical address boundary that is a multiple of that value. Both
1465 * "alignment" and "boundary" must be a power of two.
1468 vm_phys_alloc_contig(int domain, u_long npages, vm_paddr_t low, vm_paddr_t high,
1469 u_long alignment, vm_paddr_t boundary)
1471 vm_paddr_t pa_end, pa_start;
1473 struct vm_phys_seg *seg;
1474 struct vm_freelist (*queues)[VM_NFREEPOOL][VM_NFREEORDER_MAX];
1477 KASSERT(npages > 0, ("npages is 0"));
1478 KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
1479 KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
1480 vm_domain_free_assert_locked(VM_DOMAIN(domain));
1485 for (segind = vm_phys_nsegs - 1; segind >= 0; segind--) {
1486 seg = &vm_phys_segs[segind];
1487 if (seg->start >= high || seg->domain != domain)
1489 if (low >= seg->end)
1491 if (low <= seg->start)
1492 pa_start = seg->start;
1495 if (high < seg->end)
1499 if (pa_end - pa_start < ptoa(npages))
1502 * If a previous segment led to a search using
1503 * the same free lists as would this segment, then
1504 * we've actually already searched within this
1507 if (seg->free_queues == queues)
1509 queues = seg->free_queues;
1510 m_run = vm_phys_alloc_queues_contig(queues, npages,
1511 low, high, alignment, boundary);
1519 * Return the index of the first unused slot which may be the terminating
1523 vm_phys_avail_count(void)
1527 for (i = 0; phys_avail[i + 1]; i += 2)
1529 if (i > PHYS_AVAIL_ENTRIES)
1530 panic("Improperly terminated phys_avail %d entries", i);
1536 * Assert that a phys_avail entry is valid.
1539 vm_phys_avail_check(int i)
1541 if (phys_avail[i] & PAGE_MASK)
1542 panic("Unaligned phys_avail[%d]: %#jx", i,
1543 (intmax_t)phys_avail[i]);
1544 if (phys_avail[i+1] & PAGE_MASK)
1545 panic("Unaligned phys_avail[%d + 1]: %#jx", i,
1546 (intmax_t)phys_avail[i]);
1547 if (phys_avail[i + 1] < phys_avail[i])
1548 panic("phys_avail[%d] start %#jx < end %#jx", i,
1549 (intmax_t)phys_avail[i], (intmax_t)phys_avail[i+1]);
1553 * Return the index of an overlapping phys_avail entry or -1.
1557 vm_phys_avail_find(vm_paddr_t pa)
1561 for (i = 0; phys_avail[i + 1]; i += 2)
1562 if (phys_avail[i] <= pa && phys_avail[i + 1] > pa)
1569 * Return the index of the largest entry.
1572 vm_phys_avail_largest(void)
1574 vm_paddr_t sz, largesz;
1580 for (i = 0; phys_avail[i + 1]; i += 2) {
1581 sz = vm_phys_avail_size(i);
1592 vm_phys_avail_size(int i)
1595 return (phys_avail[i + 1] - phys_avail[i]);
1599 * Split an entry at the address 'pa'. Return zero on success or errno.
1602 vm_phys_avail_split(vm_paddr_t pa, int i)
1606 vm_phys_avail_check(i);
1607 if (pa <= phys_avail[i] || pa >= phys_avail[i + 1])
1608 panic("vm_phys_avail_split: invalid address");
1609 cnt = vm_phys_avail_count();
1610 if (cnt >= PHYS_AVAIL_ENTRIES)
1612 memmove(&phys_avail[i + 2], &phys_avail[i],
1613 (cnt - i) * sizeof(phys_avail[0]));
1614 phys_avail[i + 1] = pa;
1615 phys_avail[i + 2] = pa;
1616 vm_phys_avail_check(i);
1617 vm_phys_avail_check(i+2);
1623 * Check if a given physical address can be included as part of a crash dump.
1626 vm_phys_is_dumpable(vm_paddr_t pa)
1631 if ((m = vm_phys_paddr_to_vm_page(pa)) != NULL)
1632 return ((m->flags & PG_NODUMP) == 0);
1634 for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) {
1635 if (pa >= dump_avail[i] && pa < dump_avail[i + 1])
1642 vm_phys_early_add_seg(vm_paddr_t start, vm_paddr_t end)
1644 struct vm_phys_seg *seg;
1646 if (vm_phys_early_nsegs == -1)
1647 panic("%s: called after initialization", __func__);
1648 if (vm_phys_early_nsegs == nitems(vm_phys_early_segs))
1649 panic("%s: ran out of early segments", __func__);
1651 seg = &vm_phys_early_segs[vm_phys_early_nsegs++];
1657 * This routine allocates NUMA node specific memory before the page
1658 * allocator is bootstrapped.
1661 vm_phys_early_alloc(int domain, size_t alloc_size)
1667 vm_paddr_t pa, mem_start, mem_end, size, biggestsize, align;
1669 KASSERT(domain == -1 || (domain >= 0 && domain < vm_ndomains),
1670 ("%s: invalid domain index %d", __func__, domain));
1673 * Search the mem_affinity array for the biggest address
1674 * range in the desired domain. This is used to constrain
1675 * the phys_avail selection below.
1682 if (mem_affinity != NULL) {
1684 size = mem_affinity[i].end - mem_affinity[i].start;
1687 if (domain != -1 && mem_affinity[i].domain != domain)
1689 if (size > biggestsize) {
1694 mem_start = mem_affinity[mem_index].start;
1695 mem_end = mem_affinity[mem_index].end;
1700 * Now find biggest physical segment in within the desired
1705 for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1706 /* skip regions that are out of range */
1707 if (phys_avail[i+1] - alloc_size < mem_start ||
1708 phys_avail[i+1] > mem_end)
1710 size = vm_phys_avail_size(i);
1711 if (size > biggestsize) {
1716 alloc_size = round_page(alloc_size);
1719 * Grab single pages from the front to reduce fragmentation.
1721 if (alloc_size == PAGE_SIZE) {
1722 pa = phys_avail[biggestone];
1723 phys_avail[biggestone] += PAGE_SIZE;
1724 vm_phys_avail_check(biggestone);
1729 * Naturally align large allocations.
1731 align = phys_avail[biggestone + 1] & (alloc_size - 1);
1732 if (alloc_size + align > biggestsize)
1733 panic("cannot find a large enough size\n");
1735 vm_phys_avail_split(phys_avail[biggestone + 1] - align,
1737 /* Wasting memory. */
1738 phys_avail[biggestone + 1] -= align;
1740 phys_avail[biggestone + 1] -= alloc_size;
1741 vm_phys_avail_check(biggestone);
1742 pa = phys_avail[biggestone + 1];
1747 vm_phys_early_startup(void)
1749 struct vm_phys_seg *seg;
1752 for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1753 phys_avail[i] = round_page(phys_avail[i]);
1754 phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
1757 for (i = 0; i < vm_phys_early_nsegs; i++) {
1758 seg = &vm_phys_early_segs[i];
1759 vm_phys_add_seg(seg->start, seg->end);
1761 vm_phys_early_nsegs = -1;
1764 /* Force phys_avail to be split by domain. */
1765 if (mem_affinity != NULL) {
1768 for (i = 0; mem_affinity[i].end != 0; i++) {
1769 idx = vm_phys_avail_find(mem_affinity[i].start);
1771 phys_avail[idx] != mem_affinity[i].start)
1772 vm_phys_avail_split(mem_affinity[i].start, idx);
1773 idx = vm_phys_avail_find(mem_affinity[i].end);
1775 phys_avail[idx] != mem_affinity[i].end)
1776 vm_phys_avail_split(mem_affinity[i].end, idx);
1784 * Show the number of physical pages in each of the free lists.
1786 DB_SHOW_COMMAND(freepages, db_show_freepages)
1788 struct vm_freelist *fl;
1789 int flind, oind, pind, dom;
1791 for (dom = 0; dom < vm_ndomains; dom++) {
1792 db_printf("DOMAIN: %d\n", dom);
1793 for (flind = 0; flind < vm_nfreelists; flind++) {
1794 db_printf("FREE LIST %d:\n"
1795 "\n ORDER (SIZE) | NUMBER"
1797 for (pind = 0; pind < VM_NFREEPOOL; pind++)
1798 db_printf(" | POOL %d", pind);
1800 for (pind = 0; pind < VM_NFREEPOOL; pind++)
1801 db_printf("-- -- ");
1803 for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
1804 db_printf(" %2.2d (%6.6dK)", oind,
1805 1 << (PAGE_SHIFT - 10 + oind));
1806 for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1807 fl = vm_phys_free_queues[dom][flind][pind];
1808 db_printf(" | %6.6d", fl[oind].lcnt);