xref: /freebsd/sys/vm/vm_phys.c (revision a98ff317388a00b992f1bf8404dee596f9383f5e)
1 /*-
2  * Copyright (c) 2002-2006 Rice University
3  * Copyright (c) 2007 Alan L. Cox <alc@cs.rice.edu>
4  * All rights reserved.
5  *
6  * This software was developed for the FreeBSD Project by Alan L. Cox,
7  * Olivier Crameri, Peter Druschel, Sitaram Iyer, and Juan Navarro.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
21  * A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT
22  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
24  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
25  * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
26  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY
28  * WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  *	Physical memory system implementation
34  *
35  * Any external functions defined by this module are only to be used by the
36  * virtual memory system.
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_ddb.h"
43 #include "opt_vm.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/lock.h>
48 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mutex.h>
51 #if MAXMEMDOM > 1
52 #include <sys/proc.h>
53 #endif
54 #include <sys/queue.h>
55 #include <sys/sbuf.h>
56 #include <sys/sysctl.h>
57 #include <sys/vmmeter.h>
58 
59 #include <ddb/ddb.h>
60 
61 #include <vm/vm.h>
62 #include <vm/vm_param.h>
63 #include <vm/vm_kern.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_page.h>
66 #include <vm/vm_phys.h>
67 
68 struct vm_freelist {
69 	struct pglist pl;
70 	int lcnt;
71 };
72 
73 struct vm_phys_seg {
74 	vm_paddr_t	start;
75 	vm_paddr_t	end;
76 	vm_page_t	first_page;
77 	int		domain;
78 	struct vm_freelist (*free_queues)[VM_NFREEPOOL][VM_NFREEORDER];
79 };
80 
81 struct mem_affinity *mem_affinity;
82 
83 int vm_ndomains = 1;
84 
85 static struct vm_phys_seg vm_phys_segs[VM_PHYSSEG_MAX];
86 
87 static int vm_phys_nsegs;
88 
89 #define VM_PHYS_FICTITIOUS_NSEGS	8
90 static struct vm_phys_fictitious_seg {
91 	vm_paddr_t	start;
92 	vm_paddr_t	end;
93 	vm_page_t	first_page;
94 } vm_phys_fictitious_segs[VM_PHYS_FICTITIOUS_NSEGS];
95 static struct mtx vm_phys_fictitious_reg_mtx;
96 MALLOC_DEFINE(M_FICT_PAGES, "", "");
97 
98 static struct vm_freelist
99     vm_phys_free_queues[MAXMEMDOM][VM_NFREELIST][VM_NFREEPOOL][VM_NFREEORDER];
100 
101 static int vm_nfreelists = VM_FREELIST_DEFAULT + 1;
102 
103 static int cnt_prezero;
104 SYSCTL_INT(_vm_stats_misc, OID_AUTO, cnt_prezero, CTLFLAG_RD,
105     &cnt_prezero, 0, "The number of physical pages prezeroed at idle time");
106 
107 static int sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS);
108 SYSCTL_OID(_vm, OID_AUTO, phys_free, CTLTYPE_STRING | CTLFLAG_RD,
109     NULL, 0, sysctl_vm_phys_free, "A", "Phys Free Info");
110 
111 static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS);
112 SYSCTL_OID(_vm, OID_AUTO, phys_segs, CTLTYPE_STRING | CTLFLAG_RD,
113     NULL, 0, sysctl_vm_phys_segs, "A", "Phys Seg Info");
114 
115 SYSCTL_INT(_vm, OID_AUTO, ndomains, CTLFLAG_RD,
116     &vm_ndomains, 0, "Number of physical memory domains available.");
117 
118 static vm_page_t vm_phys_alloc_domain_pages(int domain, int flind, int pool,
119     int order);
120 static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind,
121     int domain);
122 static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind);
123 static int vm_phys_paddr_to_segind(vm_paddr_t pa);
124 static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl,
125     int order);
126 
127 static __inline int
128 vm_rr_selectdomain(void)
129 {
130 #if MAXMEMDOM > 1
131 	struct thread *td;
132 
133 	td = curthread;
134 
135 	td->td_dom_rr_idx++;
136 	td->td_dom_rr_idx %= vm_ndomains;
137 	return (td->td_dom_rr_idx);
138 #else
139 	return (0);
140 #endif
141 }
142 
143 /*
144  * Outputs the state of the physical memory allocator, specifically,
145  * the amount of physical memory in each free list.
146  */
147 static int
148 sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS)
149 {
150 	struct sbuf sbuf;
151 	struct vm_freelist *fl;
152 	int dom, error, flind, oind, pind;
153 
154 	error = sysctl_wire_old_buffer(req, 0);
155 	if (error != 0)
156 		return (error);
157 	sbuf_new_for_sysctl(&sbuf, NULL, 128 * vm_ndomains, req);
158 	for (dom = 0; dom < vm_ndomains; dom++) {
159 		sbuf_printf(&sbuf,"DOMAIN: %d\n", dom);
160 		for (flind = 0; flind < vm_nfreelists; flind++) {
161 			sbuf_printf(&sbuf, "FREE LIST %d:\n"
162 			    "\n  ORDER (SIZE)  |  NUMBER"
163 			    "\n              ", flind);
164 			for (pind = 0; pind < VM_NFREEPOOL; pind++)
165 				sbuf_printf(&sbuf, "  |  POOL %d", pind);
166 			sbuf_printf(&sbuf, "\n--            ");
167 			for (pind = 0; pind < VM_NFREEPOOL; pind++)
168 				sbuf_printf(&sbuf, "-- --      ");
169 			sbuf_printf(&sbuf, "--\n");
170 			for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
171 				sbuf_printf(&sbuf, "  %2d (%6dK)", oind,
172 				    1 << (PAGE_SHIFT - 10 + oind));
173 				for (pind = 0; pind < VM_NFREEPOOL; pind++) {
174 				fl = vm_phys_free_queues[dom][flind][pind];
175 					sbuf_printf(&sbuf, "  |  %6.6d",
176 					    fl[oind].lcnt);
177 				}
178 				sbuf_printf(&sbuf, "\n");
179 			}
180 			sbuf_printf(&sbuf, "\n");
181 		}
182 		sbuf_printf(&sbuf, "\n");
183 	}
184 	error = sbuf_finish(&sbuf);
185 	sbuf_delete(&sbuf);
186 	return (error);
187 }
188 
189 /*
190  * Outputs the set of physical memory segments.
191  */
192 static int
193 sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS)
194 {
195 	struct sbuf sbuf;
196 	struct vm_phys_seg *seg;
197 	int error, segind;
198 
199 	error = sysctl_wire_old_buffer(req, 0);
200 	if (error != 0)
201 		return (error);
202 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
203 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
204 		sbuf_printf(&sbuf, "\nSEGMENT %d:\n\n", segind);
205 		seg = &vm_phys_segs[segind];
206 		sbuf_printf(&sbuf, "start:     %#jx\n",
207 		    (uintmax_t)seg->start);
208 		sbuf_printf(&sbuf, "end:       %#jx\n",
209 		    (uintmax_t)seg->end);
210 		sbuf_printf(&sbuf, "domain:    %d\n", seg->domain);
211 		sbuf_printf(&sbuf, "free list: %p\n", seg->free_queues);
212 	}
213 	error = sbuf_finish(&sbuf);
214 	sbuf_delete(&sbuf);
215 	return (error);
216 }
217 
218 static void
219 vm_freelist_add(struct vm_freelist *fl, vm_page_t m, int order, int tail)
220 {
221 
222 	m->order = order;
223 	if (tail)
224 		TAILQ_INSERT_TAIL(&fl[order].pl, m, pageq);
225 	else
226 		TAILQ_INSERT_HEAD(&fl[order].pl, m, pageq);
227 	fl[order].lcnt++;
228 }
229 
230 static void
231 vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order)
232 {
233 
234 	TAILQ_REMOVE(&fl[order].pl, m, pageq);
235 	fl[order].lcnt--;
236 	m->order = VM_NFREEORDER;
237 }
238 
239 /*
240  * Create a physical memory segment.
241  */
242 static void
243 _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind, int domain)
244 {
245 	struct vm_phys_seg *seg;
246 #ifdef VM_PHYSSEG_SPARSE
247 	long pages;
248 	int segind;
249 
250 	pages = 0;
251 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
252 		seg = &vm_phys_segs[segind];
253 		pages += atop(seg->end - seg->start);
254 	}
255 #endif
256 	KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX,
257 	    ("vm_phys_create_seg: increase VM_PHYSSEG_MAX"));
258 	KASSERT(domain < vm_ndomains,
259 	    ("vm_phys_create_seg: invalid domain provided"));
260 	seg = &vm_phys_segs[vm_phys_nsegs++];
261 	seg->start = start;
262 	seg->end = end;
263 	seg->domain = domain;
264 #ifdef VM_PHYSSEG_SPARSE
265 	seg->first_page = &vm_page_array[pages];
266 #else
267 	seg->first_page = PHYS_TO_VM_PAGE(start);
268 #endif
269 	seg->free_queues = &vm_phys_free_queues[domain][flind];
270 }
271 
272 static void
273 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind)
274 {
275 	int i;
276 
277 	if (mem_affinity == NULL) {
278 		_vm_phys_create_seg(start, end, flind, 0);
279 		return;
280 	}
281 
282 	for (i = 0;; i++) {
283 		if (mem_affinity[i].end == 0)
284 			panic("Reached end of affinity info");
285 		if (mem_affinity[i].end <= start)
286 			continue;
287 		if (mem_affinity[i].start > start)
288 			panic("No affinity info for start %jx",
289 			    (uintmax_t)start);
290 		if (mem_affinity[i].end >= end) {
291 			_vm_phys_create_seg(start, end, flind,
292 			    mem_affinity[i].domain);
293 			break;
294 		}
295 		_vm_phys_create_seg(start, mem_affinity[i].end, flind,
296 		    mem_affinity[i].domain);
297 		start = mem_affinity[i].end;
298 	}
299 }
300 
301 /*
302  * Initialize the physical memory allocator.
303  */
304 void
305 vm_phys_init(void)
306 {
307 	struct vm_freelist *fl;
308 	int dom, flind, i, oind, pind;
309 
310 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
311 #ifdef	VM_FREELIST_ISADMA
312 		if (phys_avail[i] < 16777216) {
313 			if (phys_avail[i + 1] > 16777216) {
314 				vm_phys_create_seg(phys_avail[i], 16777216,
315 				    VM_FREELIST_ISADMA);
316 				vm_phys_create_seg(16777216, phys_avail[i + 1],
317 				    VM_FREELIST_DEFAULT);
318 			} else {
319 				vm_phys_create_seg(phys_avail[i],
320 				    phys_avail[i + 1], VM_FREELIST_ISADMA);
321 			}
322 			if (VM_FREELIST_ISADMA >= vm_nfreelists)
323 				vm_nfreelists = VM_FREELIST_ISADMA + 1;
324 		} else
325 #endif
326 #ifdef	VM_FREELIST_HIGHMEM
327 		if (phys_avail[i + 1] > VM_HIGHMEM_ADDRESS) {
328 			if (phys_avail[i] < VM_HIGHMEM_ADDRESS) {
329 				vm_phys_create_seg(phys_avail[i],
330 				    VM_HIGHMEM_ADDRESS, VM_FREELIST_DEFAULT);
331 				vm_phys_create_seg(VM_HIGHMEM_ADDRESS,
332 				    phys_avail[i + 1], VM_FREELIST_HIGHMEM);
333 			} else {
334 				vm_phys_create_seg(phys_avail[i],
335 				    phys_avail[i + 1], VM_FREELIST_HIGHMEM);
336 			}
337 			if (VM_FREELIST_HIGHMEM >= vm_nfreelists)
338 				vm_nfreelists = VM_FREELIST_HIGHMEM + 1;
339 		} else
340 #endif
341 		vm_phys_create_seg(phys_avail[i], phys_avail[i + 1],
342 		    VM_FREELIST_DEFAULT);
343 	}
344 	for (dom = 0; dom < vm_ndomains; dom++) {
345 		for (flind = 0; flind < vm_nfreelists; flind++) {
346 			for (pind = 0; pind < VM_NFREEPOOL; pind++) {
347 				fl = vm_phys_free_queues[dom][flind][pind];
348 				for (oind = 0; oind < VM_NFREEORDER; oind++)
349 					TAILQ_INIT(&fl[oind].pl);
350 			}
351 		}
352 	}
353 	mtx_init(&vm_phys_fictitious_reg_mtx, "vmfctr", NULL, MTX_DEF);
354 }
355 
356 /*
357  * Split a contiguous, power of two-sized set of physical pages.
358  */
359 static __inline void
360 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order)
361 {
362 	vm_page_t m_buddy;
363 
364 	while (oind > order) {
365 		oind--;
366 		m_buddy = &m[1 << oind];
367 		KASSERT(m_buddy->order == VM_NFREEORDER,
368 		    ("vm_phys_split_pages: page %p has unexpected order %d",
369 		    m_buddy, m_buddy->order));
370 		vm_freelist_add(fl, m_buddy, oind, 0);
371         }
372 }
373 
374 /*
375  * Initialize a physical page and add it to the free lists.
376  */
377 void
378 vm_phys_add_page(vm_paddr_t pa)
379 {
380 	vm_page_t m;
381 
382 	cnt.v_page_count++;
383 	m = vm_phys_paddr_to_vm_page(pa);
384 	m->phys_addr = pa;
385 	m->queue = PQ_NONE;
386 	m->segind = vm_phys_paddr_to_segind(pa);
387 	m->flags = PG_FREE;
388 	KASSERT(m->order == VM_NFREEORDER,
389 	    ("vm_phys_add_page: page %p has unexpected order %d",
390 	    m, m->order));
391 	m->pool = VM_FREEPOOL_DEFAULT;
392 	pmap_page_init(m);
393 	mtx_lock(&vm_page_queue_free_mtx);
394 	cnt.v_free_count++;
395 	vm_phys_free_pages(m, 0);
396 	mtx_unlock(&vm_page_queue_free_mtx);
397 }
398 
399 /*
400  * Allocate a contiguous, power of two-sized set of physical pages
401  * from the free lists.
402  *
403  * The free page queues must be locked.
404  */
405 vm_page_t
406 vm_phys_alloc_pages(int pool, int order)
407 {
408 	vm_page_t m;
409 	int dom, domain, flind;
410 
411 	KASSERT(pool < VM_NFREEPOOL,
412 	    ("vm_phys_alloc_pages: pool %d is out of range", pool));
413 	KASSERT(order < VM_NFREEORDER,
414 	    ("vm_phys_alloc_pages: order %d is out of range", order));
415 
416 	for (dom = 0; dom < vm_ndomains; dom++) {
417 		domain = vm_rr_selectdomain();
418 		for (flind = 0; flind < vm_nfreelists; flind++) {
419 			m = vm_phys_alloc_domain_pages(domain, flind, pool,
420 			    order);
421 			if (m != NULL)
422 				return (m);
423 		}
424 	}
425 	return (NULL);
426 }
427 
428 /*
429  * Find and dequeue a free page on the given free list, with the
430  * specified pool and order
431  */
432 vm_page_t
433 vm_phys_alloc_freelist_pages(int flind, int pool, int order)
434 {
435 	vm_page_t m;
436 	int dom, domain;
437 
438 	KASSERT(flind < VM_NFREELIST,
439 	    ("vm_phys_alloc_freelist_pages: freelist %d is out of range", flind));
440 	KASSERT(pool < VM_NFREEPOOL,
441 	    ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
442 	KASSERT(order < VM_NFREEORDER,
443 	    ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
444 
445 	for (dom = 0; dom < vm_ndomains; dom++) {
446 		domain = vm_rr_selectdomain();
447 		m = vm_phys_alloc_domain_pages(domain, flind, pool, order);
448 		if (m != NULL)
449 			return (m);
450 	}
451 	return (NULL);
452 }
453 
454 static vm_page_t
455 vm_phys_alloc_domain_pages(int domain, int flind, int pool, int order)
456 {
457 	struct vm_freelist *fl;
458 	struct vm_freelist *alt;
459 	int oind, pind;
460 	vm_page_t m;
461 
462 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
463 	fl = &vm_phys_free_queues[domain][flind][pool][0];
464 	for (oind = order; oind < VM_NFREEORDER; oind++) {
465 		m = TAILQ_FIRST(&fl[oind].pl);
466 		if (m != NULL) {
467 			vm_freelist_rem(fl, m, oind);
468 			vm_phys_split_pages(m, oind, fl, order);
469 			return (m);
470 		}
471 	}
472 
473 	/*
474 	 * The given pool was empty.  Find the largest
475 	 * contiguous, power-of-two-sized set of pages in any
476 	 * pool.  Transfer these pages to the given pool, and
477 	 * use them to satisfy the allocation.
478 	 */
479 	for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
480 		for (pind = 0; pind < VM_NFREEPOOL; pind++) {
481 			alt = &vm_phys_free_queues[domain][flind][pind][0];
482 			m = TAILQ_FIRST(&alt[oind].pl);
483 			if (m != NULL) {
484 				vm_freelist_rem(alt, m, oind);
485 				vm_phys_set_pool(pool, m, oind);
486 				vm_phys_split_pages(m, oind, fl, order);
487 				return (m);
488 			}
489 		}
490 	}
491 	return (NULL);
492 }
493 
494 /*
495  * Find the vm_page corresponding to the given physical address.
496  */
497 vm_page_t
498 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
499 {
500 	struct vm_phys_seg *seg;
501 	int segind;
502 
503 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
504 		seg = &vm_phys_segs[segind];
505 		if (pa >= seg->start && pa < seg->end)
506 			return (&seg->first_page[atop(pa - seg->start)]);
507 	}
508 	return (NULL);
509 }
510 
511 vm_page_t
512 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
513 {
514 	struct vm_phys_fictitious_seg *seg;
515 	vm_page_t m;
516 	int segind;
517 
518 	m = NULL;
519 	for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
520 		seg = &vm_phys_fictitious_segs[segind];
521 		if (pa >= seg->start && pa < seg->end) {
522 			m = &seg->first_page[atop(pa - seg->start)];
523 			KASSERT((m->flags & PG_FICTITIOUS) != 0,
524 			    ("%p not fictitious", m));
525 			break;
526 		}
527 	}
528 	return (m);
529 }
530 
531 int
532 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
533     vm_memattr_t memattr)
534 {
535 	struct vm_phys_fictitious_seg *seg;
536 	vm_page_t fp;
537 	long i, page_count;
538 	int segind;
539 #ifdef VM_PHYSSEG_DENSE
540 	long pi;
541 	boolean_t malloced;
542 #endif
543 
544 	page_count = (end - start) / PAGE_SIZE;
545 
546 #ifdef VM_PHYSSEG_DENSE
547 	pi = atop(start);
548 	if (pi >= first_page && atop(end) < vm_page_array_size) {
549 		fp = &vm_page_array[pi - first_page];
550 		malloced = FALSE;
551 	} else
552 #endif
553 	{
554 		fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
555 		    M_WAITOK | M_ZERO);
556 #ifdef VM_PHYSSEG_DENSE
557 		malloced = TRUE;
558 #endif
559 	}
560 	for (i = 0; i < page_count; i++) {
561 		vm_page_initfake(&fp[i], start + PAGE_SIZE * i, memattr);
562 		fp[i].oflags &= ~(VPO_BUSY | VPO_UNMANAGED);
563 	}
564 	mtx_lock(&vm_phys_fictitious_reg_mtx);
565 	for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
566 		seg = &vm_phys_fictitious_segs[segind];
567 		if (seg->start == 0 && seg->end == 0) {
568 			seg->start = start;
569 			seg->end = end;
570 			seg->first_page = fp;
571 			mtx_unlock(&vm_phys_fictitious_reg_mtx);
572 			return (0);
573 		}
574 	}
575 	mtx_unlock(&vm_phys_fictitious_reg_mtx);
576 #ifdef VM_PHYSSEG_DENSE
577 	if (malloced)
578 #endif
579 		free(fp, M_FICT_PAGES);
580 	return (EBUSY);
581 }
582 
583 void
584 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
585 {
586 	struct vm_phys_fictitious_seg *seg;
587 	vm_page_t fp;
588 	int segind;
589 #ifdef VM_PHYSSEG_DENSE
590 	long pi;
591 #endif
592 
593 #ifdef VM_PHYSSEG_DENSE
594 	pi = atop(start);
595 #endif
596 
597 	mtx_lock(&vm_phys_fictitious_reg_mtx);
598 	for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
599 		seg = &vm_phys_fictitious_segs[segind];
600 		if (seg->start == start && seg->end == end) {
601 			seg->start = seg->end = 0;
602 			fp = seg->first_page;
603 			seg->first_page = NULL;
604 			mtx_unlock(&vm_phys_fictitious_reg_mtx);
605 #ifdef VM_PHYSSEG_DENSE
606 			if (pi < first_page || atop(end) >= vm_page_array_size)
607 #endif
608 				free(fp, M_FICT_PAGES);
609 			return;
610 		}
611 	}
612 	mtx_unlock(&vm_phys_fictitious_reg_mtx);
613 	KASSERT(0, ("Unregistering not registered fictitious range"));
614 }
615 
616 /*
617  * Find the segment containing the given physical address.
618  */
619 static int
620 vm_phys_paddr_to_segind(vm_paddr_t pa)
621 {
622 	struct vm_phys_seg *seg;
623 	int segind;
624 
625 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
626 		seg = &vm_phys_segs[segind];
627 		if (pa >= seg->start && pa < seg->end)
628 			return (segind);
629 	}
630 	panic("vm_phys_paddr_to_segind: paddr %#jx is not in any segment" ,
631 	    (uintmax_t)pa);
632 }
633 
634 /*
635  * Free a contiguous, power of two-sized set of physical pages.
636  *
637  * The free page queues must be locked.
638  */
639 void
640 vm_phys_free_pages(vm_page_t m, int order)
641 {
642 	struct vm_freelist *fl;
643 	struct vm_phys_seg *seg;
644 	vm_paddr_t pa;
645 	vm_page_t m_buddy;
646 
647 	KASSERT(m->order == VM_NFREEORDER,
648 	    ("vm_phys_free_pages: page %p has unexpected order %d",
649 	    m, m->order));
650 	KASSERT(m->pool < VM_NFREEPOOL,
651 	    ("vm_phys_free_pages: page %p has unexpected pool %d",
652 	    m, m->pool));
653 	KASSERT(order < VM_NFREEORDER,
654 	    ("vm_phys_free_pages: order %d is out of range", order));
655 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
656 	seg = &vm_phys_segs[m->segind];
657 	if (order < VM_NFREEORDER - 1) {
658 		pa = VM_PAGE_TO_PHYS(m);
659 		do {
660 			pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
661 			if (pa < seg->start || pa >= seg->end)
662 				break;
663 			m_buddy = &seg->first_page[atop(pa - seg->start)];
664 			if (m_buddy->order != order)
665 				break;
666 			fl = (*seg->free_queues)[m_buddy->pool];
667 			vm_freelist_rem(fl, m_buddy, order);
668 			if (m_buddy->pool != m->pool)
669 				vm_phys_set_pool(m->pool, m_buddy, order);
670 			order++;
671 			pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
672 			m = &seg->first_page[atop(pa - seg->start)];
673 		} while (order < VM_NFREEORDER - 1);
674 	}
675 	fl = (*seg->free_queues)[m->pool];
676 	vm_freelist_add(fl, m, order, 1);
677 }
678 
679 /*
680  * Free a contiguous, arbitrarily sized set of physical pages.
681  *
682  * The free page queues must be locked.
683  */
684 void
685 vm_phys_free_contig(vm_page_t m, u_long npages)
686 {
687 	u_int n;
688 	int order;
689 
690 	/*
691 	 * Avoid unnecessary coalescing by freeing the pages in the largest
692 	 * possible power-of-two-sized subsets.
693 	 */
694 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
695 	for (;; npages -= n) {
696 		/*
697 		 * Unsigned "min" is used here so that "order" is assigned
698 		 * "VM_NFREEORDER - 1" when "m"'s physical address is zero
699 		 * or the low-order bits of its physical address are zero
700 		 * because the size of a physical address exceeds the size of
701 		 * a long.
702 		 */
703 		order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
704 		    VM_NFREEORDER - 1);
705 		n = 1 << order;
706 		if (npages < n)
707 			break;
708 		vm_phys_free_pages(m, order);
709 		m += n;
710 	}
711 	/* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */
712 	for (; npages > 0; npages -= n) {
713 		order = flsl(npages) - 1;
714 		n = 1 << order;
715 		vm_phys_free_pages(m, order);
716 		m += n;
717 	}
718 }
719 
720 /*
721  * Set the pool for a contiguous, power of two-sized set of physical pages.
722  */
723 void
724 vm_phys_set_pool(int pool, vm_page_t m, int order)
725 {
726 	vm_page_t m_tmp;
727 
728 	for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
729 		m_tmp->pool = pool;
730 }
731 
732 /*
733  * Search for the given physical page "m" in the free lists.  If the search
734  * succeeds, remove "m" from the free lists and return TRUE.  Otherwise, return
735  * FALSE, indicating that "m" is not in the free lists.
736  *
737  * The free page queues must be locked.
738  */
739 boolean_t
740 vm_phys_unfree_page(vm_page_t m)
741 {
742 	struct vm_freelist *fl;
743 	struct vm_phys_seg *seg;
744 	vm_paddr_t pa, pa_half;
745 	vm_page_t m_set, m_tmp;
746 	int order;
747 
748 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
749 
750 	/*
751 	 * First, find the contiguous, power of two-sized set of free
752 	 * physical pages containing the given physical page "m" and
753 	 * assign it to "m_set".
754 	 */
755 	seg = &vm_phys_segs[m->segind];
756 	for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
757 	    order < VM_NFREEORDER - 1; ) {
758 		order++;
759 		pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
760 		if (pa >= seg->start)
761 			m_set = &seg->first_page[atop(pa - seg->start)];
762 		else
763 			return (FALSE);
764 	}
765 	if (m_set->order < order)
766 		return (FALSE);
767 	if (m_set->order == VM_NFREEORDER)
768 		return (FALSE);
769 	KASSERT(m_set->order < VM_NFREEORDER,
770 	    ("vm_phys_unfree_page: page %p has unexpected order %d",
771 	    m_set, m_set->order));
772 
773 	/*
774 	 * Next, remove "m_set" from the free lists.  Finally, extract
775 	 * "m" from "m_set" using an iterative algorithm: While "m_set"
776 	 * is larger than a page, shrink "m_set" by returning the half
777 	 * of "m_set" that does not contain "m" to the free lists.
778 	 */
779 	fl = (*seg->free_queues)[m_set->pool];
780 	order = m_set->order;
781 	vm_freelist_rem(fl, m_set, order);
782 	while (order > 0) {
783 		order--;
784 		pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
785 		if (m->phys_addr < pa_half)
786 			m_tmp = &seg->first_page[atop(pa_half - seg->start)];
787 		else {
788 			m_tmp = m_set;
789 			m_set = &seg->first_page[atop(pa_half - seg->start)];
790 		}
791 		vm_freelist_add(fl, m_tmp, order, 0);
792 	}
793 	KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
794 	return (TRUE);
795 }
796 
797 /*
798  * Try to zero one physical page.  Used by an idle priority thread.
799  */
800 boolean_t
801 vm_phys_zero_pages_idle(void)
802 {
803 	static struct vm_freelist *fl;
804 	static int flind, oind, pind;
805 	vm_page_t m, m_tmp;
806 	int domain;
807 
808 	domain = vm_rr_selectdomain();
809 	fl = vm_phys_free_queues[domain][0][0];
810 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
811 	for (;;) {
812 		TAILQ_FOREACH_REVERSE(m, &fl[oind].pl, pglist, pageq) {
813 			for (m_tmp = m; m_tmp < &m[1 << oind]; m_tmp++) {
814 				if ((m_tmp->flags & (PG_CACHED | PG_ZERO)) == 0) {
815 					vm_phys_unfree_page(m_tmp);
816 					cnt.v_free_count--;
817 					mtx_unlock(&vm_page_queue_free_mtx);
818 					pmap_zero_page_idle(m_tmp);
819 					m_tmp->flags |= PG_ZERO;
820 					mtx_lock(&vm_page_queue_free_mtx);
821 					cnt.v_free_count++;
822 					vm_phys_free_pages(m_tmp, 0);
823 					vm_page_zero_count++;
824 					cnt_prezero++;
825 					return (TRUE);
826 				}
827 			}
828 		}
829 		oind++;
830 		if (oind == VM_NFREEORDER) {
831 			oind = 0;
832 			pind++;
833 			if (pind == VM_NFREEPOOL) {
834 				pind = 0;
835 				flind++;
836 				if (flind == vm_nfreelists)
837 					flind = 0;
838 			}
839 			fl = vm_phys_free_queues[domain][flind][pind];
840 		}
841 	}
842 }
843 
844 /*
845  * Allocate a contiguous set of physical pages of the given size
846  * "npages" from the free lists.  All of the physical pages must be at
847  * or above the given physical address "low" and below the given
848  * physical address "high".  The given value "alignment" determines the
849  * alignment of the first physical page in the set.  If the given value
850  * "boundary" is non-zero, then the set of physical pages cannot cross
851  * any physical address boundary that is a multiple of that value.  Both
852  * "alignment" and "boundary" must be a power of two.
853  */
854 vm_page_t
855 vm_phys_alloc_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
856     u_long alignment, vm_paddr_t boundary)
857 {
858 	struct vm_freelist *fl;
859 	struct vm_phys_seg *seg;
860 	vm_paddr_t pa, pa_last, size;
861 	vm_page_t m, m_ret;
862 	u_long npages_end;
863 	int dom, domain, flind, oind, order, pind;
864 
865 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
866 	size = npages << PAGE_SHIFT;
867 	KASSERT(size != 0,
868 	    ("vm_phys_alloc_contig: size must not be 0"));
869 	KASSERT((alignment & (alignment - 1)) == 0,
870 	    ("vm_phys_alloc_contig: alignment must be a power of 2"));
871 	KASSERT((boundary & (boundary - 1)) == 0,
872 	    ("vm_phys_alloc_contig: boundary must be a power of 2"));
873 	/* Compute the queue that is the best fit for npages. */
874 	for (order = 0; (1 << order) < npages; order++);
875 	dom = 0;
876 restartdom:
877 	domain = vm_rr_selectdomain();
878 	for (flind = 0; flind < vm_nfreelists; flind++) {
879 		for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER; oind++) {
880 			for (pind = 0; pind < VM_NFREEPOOL; pind++) {
881 				fl = &vm_phys_free_queues[domain][flind][pind][0];
882 				TAILQ_FOREACH(m_ret, &fl[oind].pl, pageq) {
883 					/*
884 					 * A free list may contain physical pages
885 					 * from one or more segments.
886 					 */
887 					seg = &vm_phys_segs[m_ret->segind];
888 					if (seg->start > high ||
889 					    low >= seg->end)
890 						continue;
891 
892 					/*
893 					 * Is the size of this allocation request
894 					 * larger than the largest block size?
895 					 */
896 					if (order >= VM_NFREEORDER) {
897 						/*
898 						 * Determine if a sufficient number
899 						 * of subsequent blocks to satisfy
900 						 * the allocation request are free.
901 						 */
902 						pa = VM_PAGE_TO_PHYS(m_ret);
903 						pa_last = pa + size;
904 						for (;;) {
905 							pa += 1 << (PAGE_SHIFT + VM_NFREEORDER - 1);
906 							if (pa >= pa_last)
907 								break;
908 							if (pa < seg->start ||
909 							    pa >= seg->end)
910 								break;
911 							m = &seg->first_page[atop(pa - seg->start)];
912 							if (m->order != VM_NFREEORDER - 1)
913 								break;
914 						}
915 						/* If not, continue to the next block. */
916 						if (pa < pa_last)
917 							continue;
918 					}
919 
920 					/*
921 					 * Determine if the blocks are within the given range,
922 					 * satisfy the given alignment, and do not cross the
923 					 * given boundary.
924 					 */
925 					pa = VM_PAGE_TO_PHYS(m_ret);
926 					if (pa >= low &&
927 					    pa + size <= high &&
928 					    (pa & (alignment - 1)) == 0 &&
929 					    ((pa ^ (pa + size - 1)) & ~(boundary - 1)) == 0)
930 						goto done;
931 				}
932 			}
933 		}
934 	}
935 	if (++dom < vm_ndomains)
936 		goto restartdom;
937 	return (NULL);
938 done:
939 	for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
940 		fl = (*seg->free_queues)[m->pool];
941 		vm_freelist_rem(fl, m, m->order);
942 	}
943 	if (m_ret->pool != VM_FREEPOOL_DEFAULT)
944 		vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m_ret, oind);
945 	fl = (*seg->free_queues)[m_ret->pool];
946 	vm_phys_split_pages(m_ret, oind, fl, order);
947 	/* Return excess pages to the free lists. */
948 	npages_end = roundup2(npages, 1 << imin(oind, order));
949 	if (npages < npages_end)
950 		vm_phys_free_contig(&m_ret[npages], npages_end - npages);
951 	return (m_ret);
952 }
953 
954 #ifdef DDB
955 /*
956  * Show the number of physical pages in each of the free lists.
957  */
958 DB_SHOW_COMMAND(freepages, db_show_freepages)
959 {
960 	struct vm_freelist *fl;
961 	int flind, oind, pind, dom;
962 
963 	for (dom = 0; dom < vm_ndomains; dom++) {
964 		db_printf("DOMAIN: %d\n", dom);
965 		for (flind = 0; flind < vm_nfreelists; flind++) {
966 			db_printf("FREE LIST %d:\n"
967 			    "\n  ORDER (SIZE)  |  NUMBER"
968 			    "\n              ", flind);
969 			for (pind = 0; pind < VM_NFREEPOOL; pind++)
970 				db_printf("  |  POOL %d", pind);
971 			db_printf("\n--            ");
972 			for (pind = 0; pind < VM_NFREEPOOL; pind++)
973 				db_printf("-- --      ");
974 			db_printf("--\n");
975 			for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
976 				db_printf("  %2.2d (%6.6dK)", oind,
977 				    1 << (PAGE_SHIFT - 10 + oind));
978 				for (pind = 0; pind < VM_NFREEPOOL; pind++) {
979 				fl = vm_phys_free_queues[dom][flind][pind];
980 					db_printf("  |  %6.6d", fl[oind].lcnt);
981 				}
982 				db_printf("\n");
983 			}
984 			db_printf("\n");
985 		}
986 		db_printf("\n");
987 	}
988 }
989 #endif
990