xref: /freebsd/sys/vm/vm_phys.c (revision 55bce0c1203e70d8b62a3dedc9235ab39660c6f4)
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 		pmap_page_init(&fp[i]);
563 		fp[i].oflags &= ~(VPO_BUSY | VPO_UNMANAGED);
564 	}
565 	mtx_lock(&vm_phys_fictitious_reg_mtx);
566 	for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
567 		seg = &vm_phys_fictitious_segs[segind];
568 		if (seg->start == 0 && seg->end == 0) {
569 			seg->start = start;
570 			seg->end = end;
571 			seg->first_page = fp;
572 			mtx_unlock(&vm_phys_fictitious_reg_mtx);
573 			return (0);
574 		}
575 	}
576 	mtx_unlock(&vm_phys_fictitious_reg_mtx);
577 #ifdef VM_PHYSSEG_DENSE
578 	if (malloced)
579 #endif
580 		free(fp, M_FICT_PAGES);
581 	return (EBUSY);
582 }
583 
584 void
585 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
586 {
587 	struct vm_phys_fictitious_seg *seg;
588 	vm_page_t fp;
589 	int segind;
590 #ifdef VM_PHYSSEG_DENSE
591 	long pi;
592 #endif
593 
594 #ifdef VM_PHYSSEG_DENSE
595 	pi = atop(start);
596 #endif
597 
598 	mtx_lock(&vm_phys_fictitious_reg_mtx);
599 	for (segind = 0; segind < VM_PHYS_FICTITIOUS_NSEGS; segind++) {
600 		seg = &vm_phys_fictitious_segs[segind];
601 		if (seg->start == start && seg->end == end) {
602 			seg->start = seg->end = 0;
603 			fp = seg->first_page;
604 			seg->first_page = NULL;
605 			mtx_unlock(&vm_phys_fictitious_reg_mtx);
606 #ifdef VM_PHYSSEG_DENSE
607 			if (pi < first_page || atop(end) >= vm_page_array_size)
608 #endif
609 				free(fp, M_FICT_PAGES);
610 			return;
611 		}
612 	}
613 	mtx_unlock(&vm_phys_fictitious_reg_mtx);
614 	KASSERT(0, ("Unregistering not registered fictitious range"));
615 }
616 
617 /*
618  * Find the segment containing the given physical address.
619  */
620 static int
621 vm_phys_paddr_to_segind(vm_paddr_t pa)
622 {
623 	struct vm_phys_seg *seg;
624 	int segind;
625 
626 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
627 		seg = &vm_phys_segs[segind];
628 		if (pa >= seg->start && pa < seg->end)
629 			return (segind);
630 	}
631 	panic("vm_phys_paddr_to_segind: paddr %#jx is not in any segment" ,
632 	    (uintmax_t)pa);
633 }
634 
635 /*
636  * Free a contiguous, power of two-sized set of physical pages.
637  *
638  * The free page queues must be locked.
639  */
640 void
641 vm_phys_free_pages(vm_page_t m, int order)
642 {
643 	struct vm_freelist *fl;
644 	struct vm_phys_seg *seg;
645 	vm_paddr_t pa;
646 	vm_page_t m_buddy;
647 
648 	KASSERT(m->order == VM_NFREEORDER,
649 	    ("vm_phys_free_pages: page %p has unexpected order %d",
650 	    m, m->order));
651 	KASSERT(m->pool < VM_NFREEPOOL,
652 	    ("vm_phys_free_pages: page %p has unexpected pool %d",
653 	    m, m->pool));
654 	KASSERT(order < VM_NFREEORDER,
655 	    ("vm_phys_free_pages: order %d is out of range", order));
656 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
657 	seg = &vm_phys_segs[m->segind];
658 	if (order < VM_NFREEORDER - 1) {
659 		pa = VM_PAGE_TO_PHYS(m);
660 		do {
661 			pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
662 			if (pa < seg->start || pa >= seg->end)
663 				break;
664 			m_buddy = &seg->first_page[atop(pa - seg->start)];
665 			if (m_buddy->order != order)
666 				break;
667 			fl = (*seg->free_queues)[m_buddy->pool];
668 			vm_freelist_rem(fl, m_buddy, order);
669 			if (m_buddy->pool != m->pool)
670 				vm_phys_set_pool(m->pool, m_buddy, order);
671 			order++;
672 			pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
673 			m = &seg->first_page[atop(pa - seg->start)];
674 		} while (order < VM_NFREEORDER - 1);
675 	}
676 	fl = (*seg->free_queues)[m->pool];
677 	vm_freelist_add(fl, m, order, 1);
678 }
679 
680 /*
681  * Free a contiguous, arbitrarily sized set of physical pages.
682  *
683  * The free page queues must be locked.
684  */
685 void
686 vm_phys_free_contig(vm_page_t m, u_long npages)
687 {
688 	u_int n;
689 	int order;
690 
691 	/*
692 	 * Avoid unnecessary coalescing by freeing the pages in the largest
693 	 * possible power-of-two-sized subsets.
694 	 */
695 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
696 	for (;; npages -= n) {
697 		/*
698 		 * Unsigned "min" is used here so that "order" is assigned
699 		 * "VM_NFREEORDER - 1" when "m"'s physical address is zero
700 		 * or the low-order bits of its physical address are zero
701 		 * because the size of a physical address exceeds the size of
702 		 * a long.
703 		 */
704 		order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
705 		    VM_NFREEORDER - 1);
706 		n = 1 << order;
707 		if (npages < n)
708 			break;
709 		vm_phys_free_pages(m, order);
710 		m += n;
711 	}
712 	/* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */
713 	for (; npages > 0; npages -= n) {
714 		order = flsl(npages) - 1;
715 		n = 1 << order;
716 		vm_phys_free_pages(m, order);
717 		m += n;
718 	}
719 }
720 
721 /*
722  * Set the pool for a contiguous, power of two-sized set of physical pages.
723  */
724 void
725 vm_phys_set_pool(int pool, vm_page_t m, int order)
726 {
727 	vm_page_t m_tmp;
728 
729 	for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
730 		m_tmp->pool = pool;
731 }
732 
733 /*
734  * Search for the given physical page "m" in the free lists.  If the search
735  * succeeds, remove "m" from the free lists and return TRUE.  Otherwise, return
736  * FALSE, indicating that "m" is not in the free lists.
737  *
738  * The free page queues must be locked.
739  */
740 boolean_t
741 vm_phys_unfree_page(vm_page_t m)
742 {
743 	struct vm_freelist *fl;
744 	struct vm_phys_seg *seg;
745 	vm_paddr_t pa, pa_half;
746 	vm_page_t m_set, m_tmp;
747 	int order;
748 
749 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
750 
751 	/*
752 	 * First, find the contiguous, power of two-sized set of free
753 	 * physical pages containing the given physical page "m" and
754 	 * assign it to "m_set".
755 	 */
756 	seg = &vm_phys_segs[m->segind];
757 	for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
758 	    order < VM_NFREEORDER - 1; ) {
759 		order++;
760 		pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
761 		if (pa >= seg->start)
762 			m_set = &seg->first_page[atop(pa - seg->start)];
763 		else
764 			return (FALSE);
765 	}
766 	if (m_set->order < order)
767 		return (FALSE);
768 	if (m_set->order == VM_NFREEORDER)
769 		return (FALSE);
770 	KASSERT(m_set->order < VM_NFREEORDER,
771 	    ("vm_phys_unfree_page: page %p has unexpected order %d",
772 	    m_set, m_set->order));
773 
774 	/*
775 	 * Next, remove "m_set" from the free lists.  Finally, extract
776 	 * "m" from "m_set" using an iterative algorithm: While "m_set"
777 	 * is larger than a page, shrink "m_set" by returning the half
778 	 * of "m_set" that does not contain "m" to the free lists.
779 	 */
780 	fl = (*seg->free_queues)[m_set->pool];
781 	order = m_set->order;
782 	vm_freelist_rem(fl, m_set, order);
783 	while (order > 0) {
784 		order--;
785 		pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
786 		if (m->phys_addr < pa_half)
787 			m_tmp = &seg->first_page[atop(pa_half - seg->start)];
788 		else {
789 			m_tmp = m_set;
790 			m_set = &seg->first_page[atop(pa_half - seg->start)];
791 		}
792 		vm_freelist_add(fl, m_tmp, order, 0);
793 	}
794 	KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
795 	return (TRUE);
796 }
797 
798 /*
799  * Try to zero one physical page.  Used by an idle priority thread.
800  */
801 boolean_t
802 vm_phys_zero_pages_idle(void)
803 {
804 	static struct vm_freelist *fl;
805 	static int flind, oind, pind;
806 	vm_page_t m, m_tmp;
807 	int domain;
808 
809 	domain = vm_rr_selectdomain();
810 	fl = vm_phys_free_queues[domain][0][0];
811 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
812 	for (;;) {
813 		TAILQ_FOREACH_REVERSE(m, &fl[oind].pl, pglist, pageq) {
814 			for (m_tmp = m; m_tmp < &m[1 << oind]; m_tmp++) {
815 				if ((m_tmp->flags & (PG_CACHED | PG_ZERO)) == 0) {
816 					vm_phys_unfree_page(m_tmp);
817 					cnt.v_free_count--;
818 					mtx_unlock(&vm_page_queue_free_mtx);
819 					pmap_zero_page_idle(m_tmp);
820 					m_tmp->flags |= PG_ZERO;
821 					mtx_lock(&vm_page_queue_free_mtx);
822 					cnt.v_free_count++;
823 					vm_phys_free_pages(m_tmp, 0);
824 					vm_page_zero_count++;
825 					cnt_prezero++;
826 					return (TRUE);
827 				}
828 			}
829 		}
830 		oind++;
831 		if (oind == VM_NFREEORDER) {
832 			oind = 0;
833 			pind++;
834 			if (pind == VM_NFREEPOOL) {
835 				pind = 0;
836 				flind++;
837 				if (flind == vm_nfreelists)
838 					flind = 0;
839 			}
840 			fl = vm_phys_free_queues[domain][flind][pind];
841 		}
842 	}
843 }
844 
845 /*
846  * Allocate a contiguous set of physical pages of the given size
847  * "npages" from the free lists.  All of the physical pages must be at
848  * or above the given physical address "low" and below the given
849  * physical address "high".  The given value "alignment" determines the
850  * alignment of the first physical page in the set.  If the given value
851  * "boundary" is non-zero, then the set of physical pages cannot cross
852  * any physical address boundary that is a multiple of that value.  Both
853  * "alignment" and "boundary" must be a power of two.
854  */
855 vm_page_t
856 vm_phys_alloc_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
857     u_long alignment, vm_paddr_t boundary)
858 {
859 	struct vm_freelist *fl;
860 	struct vm_phys_seg *seg;
861 	vm_paddr_t pa, pa_last, size;
862 	vm_page_t m, m_ret;
863 	u_long npages_end;
864 	int dom, domain, flind, oind, order, pind;
865 
866 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
867 	size = npages << PAGE_SHIFT;
868 	KASSERT(size != 0,
869 	    ("vm_phys_alloc_contig: size must not be 0"));
870 	KASSERT((alignment & (alignment - 1)) == 0,
871 	    ("vm_phys_alloc_contig: alignment must be a power of 2"));
872 	KASSERT((boundary & (boundary - 1)) == 0,
873 	    ("vm_phys_alloc_contig: boundary must be a power of 2"));
874 	/* Compute the queue that is the best fit for npages. */
875 	for (order = 0; (1 << order) < npages; order++);
876 	dom = 0;
877 restartdom:
878 	domain = vm_rr_selectdomain();
879 	for (flind = 0; flind < vm_nfreelists; flind++) {
880 		for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER; oind++) {
881 			for (pind = 0; pind < VM_NFREEPOOL; pind++) {
882 				fl = &vm_phys_free_queues[domain][flind][pind][0];
883 				TAILQ_FOREACH(m_ret, &fl[oind].pl, pageq) {
884 					/*
885 					 * A free list may contain physical pages
886 					 * from one or more segments.
887 					 */
888 					seg = &vm_phys_segs[m_ret->segind];
889 					if (seg->start > high ||
890 					    low >= seg->end)
891 						continue;
892 
893 					/*
894 					 * Is the size of this allocation request
895 					 * larger than the largest block size?
896 					 */
897 					if (order >= VM_NFREEORDER) {
898 						/*
899 						 * Determine if a sufficient number
900 						 * of subsequent blocks to satisfy
901 						 * the allocation request are free.
902 						 */
903 						pa = VM_PAGE_TO_PHYS(m_ret);
904 						pa_last = pa + size;
905 						for (;;) {
906 							pa += 1 << (PAGE_SHIFT + VM_NFREEORDER - 1);
907 							if (pa >= pa_last)
908 								break;
909 							if (pa < seg->start ||
910 							    pa >= seg->end)
911 								break;
912 							m = &seg->first_page[atop(pa - seg->start)];
913 							if (m->order != VM_NFREEORDER - 1)
914 								break;
915 						}
916 						/* If not, continue to the next block. */
917 						if (pa < pa_last)
918 							continue;
919 					}
920 
921 					/*
922 					 * Determine if the blocks are within the given range,
923 					 * satisfy the given alignment, and do not cross the
924 					 * given boundary.
925 					 */
926 					pa = VM_PAGE_TO_PHYS(m_ret);
927 					if (pa >= low &&
928 					    pa + size <= high &&
929 					    (pa & (alignment - 1)) == 0 &&
930 					    ((pa ^ (pa + size - 1)) & ~(boundary - 1)) == 0)
931 						goto done;
932 				}
933 			}
934 		}
935 	}
936 	if (++dom < vm_ndomains)
937 		goto restartdom;
938 	return (NULL);
939 done:
940 	for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
941 		fl = (*seg->free_queues)[m->pool];
942 		vm_freelist_rem(fl, m, m->order);
943 	}
944 	if (m_ret->pool != VM_FREEPOOL_DEFAULT)
945 		vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m_ret, oind);
946 	fl = (*seg->free_queues)[m_ret->pool];
947 	vm_phys_split_pages(m_ret, oind, fl, order);
948 	/* Return excess pages to the free lists. */
949 	npages_end = roundup2(npages, 1 << imin(oind, order));
950 	if (npages < npages_end)
951 		vm_phys_free_contig(&m_ret[npages], npages_end - npages);
952 	return (m_ret);
953 }
954 
955 #ifdef DDB
956 /*
957  * Show the number of physical pages in each of the free lists.
958  */
959 DB_SHOW_COMMAND(freepages, db_show_freepages)
960 {
961 	struct vm_freelist *fl;
962 	int flind, oind, pind, dom;
963 
964 	for (dom = 0; dom < vm_ndomains; dom++) {
965 		db_printf("DOMAIN: %d\n", dom);
966 		for (flind = 0; flind < vm_nfreelists; flind++) {
967 			db_printf("FREE LIST %d:\n"
968 			    "\n  ORDER (SIZE)  |  NUMBER"
969 			    "\n              ", flind);
970 			for (pind = 0; pind < VM_NFREEPOOL; pind++)
971 				db_printf("  |  POOL %d", pind);
972 			db_printf("\n--            ");
973 			for (pind = 0; pind < VM_NFREEPOOL; pind++)
974 				db_printf("-- --      ");
975 			db_printf("--\n");
976 			for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
977 				db_printf("  %2.2d (%6.6dK)", oind,
978 				    1 << (PAGE_SHIFT - 10 + oind));
979 				for (pind = 0; pind < VM_NFREEPOOL; pind++) {
980 				fl = vm_phys_free_queues[dom][flind][pind];
981 					db_printf("  |  %6.6d", fl[oind].lcnt);
982 				}
983 				db_printf("\n");
984 			}
985 			db_printf("\n");
986 		}
987 		db_printf("\n");
988 	}
989 }
990 #endif
991