xref: /freebsd/sys/vm/vm_phys.c (revision e72055b7feba695a760d45f01f0f8268b1cb4a74)
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/rwlock.h>
56 #include <sys/sbuf.h>
57 #include <sys/sysctl.h>
58 #include <sys/tree.h>
59 #include <sys/vmmeter.h>
60 
61 #include <ddb/ddb.h>
62 
63 #include <vm/vm.h>
64 #include <vm/vm_param.h>
65 #include <vm/vm_kern.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_page.h>
68 #include <vm/vm_phys.h>
69 
70 _Static_assert(sizeof(long) * NBBY >= VM_PHYSSEG_MAX,
71     "Too many physsegs.");
72 
73 struct mem_affinity *mem_affinity;
74 
75 int vm_ndomains = 1;
76 
77 struct vm_phys_seg vm_phys_segs[VM_PHYSSEG_MAX];
78 int vm_phys_nsegs;
79 
80 struct vm_phys_fictitious_seg;
81 static int vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *,
82     struct vm_phys_fictitious_seg *);
83 
84 RB_HEAD(fict_tree, vm_phys_fictitious_seg) vm_phys_fictitious_tree =
85     RB_INITIALIZER(_vm_phys_fictitious_tree);
86 
87 struct vm_phys_fictitious_seg {
88 	RB_ENTRY(vm_phys_fictitious_seg) node;
89 	/* Memory region data */
90 	vm_paddr_t	start;
91 	vm_paddr_t	end;
92 	vm_page_t	first_page;
93 };
94 
95 RB_GENERATE_STATIC(fict_tree, vm_phys_fictitious_seg, node,
96     vm_phys_fictitious_cmp);
97 
98 static struct rwlock vm_phys_fictitious_reg_lock;
99 MALLOC_DEFINE(M_FICT_PAGES, "vm_fictitious", "Fictitious VM pages");
100 
101 static struct vm_freelist
102     vm_phys_free_queues[MAXMEMDOM][VM_NFREELIST][VM_NFREEPOOL][VM_NFREEORDER];
103 
104 static int vm_nfreelists = VM_FREELIST_DEFAULT + 1;
105 
106 static int cnt_prezero;
107 SYSCTL_INT(_vm_stats_misc, OID_AUTO, cnt_prezero, CTLFLAG_RD,
108     &cnt_prezero, 0, "The number of physical pages prezeroed at idle time");
109 
110 static int sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS);
111 SYSCTL_OID(_vm, OID_AUTO, phys_free, CTLTYPE_STRING | CTLFLAG_RD,
112     NULL, 0, sysctl_vm_phys_free, "A", "Phys Free Info");
113 
114 static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS);
115 SYSCTL_OID(_vm, OID_AUTO, phys_segs, CTLTYPE_STRING | CTLFLAG_RD,
116     NULL, 0, sysctl_vm_phys_segs, "A", "Phys Seg Info");
117 
118 SYSCTL_INT(_vm, OID_AUTO, ndomains, CTLFLAG_RD,
119     &vm_ndomains, 0, "Number of physical memory domains available.");
120 
121 static vm_page_t vm_phys_alloc_domain_pages(int domain, int flind, int pool,
122     int order);
123 static void _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind,
124     int domain);
125 static void vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind);
126 static int vm_phys_paddr_to_segind(vm_paddr_t pa);
127 static void vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl,
128     int order);
129 
130 /*
131  * Red-black tree helpers for vm fictitious range management.
132  */
133 static inline int
134 vm_phys_fictitious_in_range(struct vm_phys_fictitious_seg *p,
135     struct vm_phys_fictitious_seg *range)
136 {
137 
138 	KASSERT(range->start != 0 && range->end != 0,
139 	    ("Invalid range passed on search for vm_fictitious page"));
140 	if (p->start >= range->end)
141 		return (1);
142 	if (p->start < range->start)
143 		return (-1);
144 
145 	return (0);
146 }
147 
148 static int
149 vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *p1,
150     struct vm_phys_fictitious_seg *p2)
151 {
152 
153 	/* Check if this is a search for a page */
154 	if (p1->end == 0)
155 		return (vm_phys_fictitious_in_range(p1, p2));
156 
157 	KASSERT(p2->end != 0,
158     ("Invalid range passed as second parameter to vm fictitious comparison"));
159 
160 	/* Searching to add a new range */
161 	if (p1->end <= p2->start)
162 		return (-1);
163 	if (p1->start >= p2->end)
164 		return (1);
165 
166 	panic("Trying to add overlapping vm fictitious ranges:\n"
167 	    "[%#jx:%#jx] and [%#jx:%#jx]", (uintmax_t)p1->start,
168 	    (uintmax_t)p1->end, (uintmax_t)p2->start, (uintmax_t)p2->end);
169 }
170 
171 static __inline int
172 vm_rr_selectdomain(void)
173 {
174 #if MAXMEMDOM > 1
175 	struct thread *td;
176 
177 	td = curthread;
178 
179 	td->td_dom_rr_idx++;
180 	td->td_dom_rr_idx %= vm_ndomains;
181 	return (td->td_dom_rr_idx);
182 #else
183 	return (0);
184 #endif
185 }
186 
187 boolean_t
188 vm_phys_domain_intersects(long mask, vm_paddr_t low, vm_paddr_t high)
189 {
190 	struct vm_phys_seg *s;
191 	int idx;
192 
193 	while ((idx = ffsl(mask)) != 0) {
194 		idx--;	/* ffsl counts from 1 */
195 		mask &= ~(1UL << idx);
196 		s = &vm_phys_segs[idx];
197 		if (low < s->end && high > s->start)
198 			return (TRUE);
199 	}
200 	return (FALSE);
201 }
202 
203 /*
204  * Outputs the state of the physical memory allocator, specifically,
205  * the amount of physical memory in each free list.
206  */
207 static int
208 sysctl_vm_phys_free(SYSCTL_HANDLER_ARGS)
209 {
210 	struct sbuf sbuf;
211 	struct vm_freelist *fl;
212 	int dom, error, flind, oind, pind;
213 
214 	error = sysctl_wire_old_buffer(req, 0);
215 	if (error != 0)
216 		return (error);
217 	sbuf_new_for_sysctl(&sbuf, NULL, 128 * vm_ndomains, req);
218 	for (dom = 0; dom < vm_ndomains; dom++) {
219 		sbuf_printf(&sbuf,"\nDOMAIN %d:\n", dom);
220 		for (flind = 0; flind < vm_nfreelists; flind++) {
221 			sbuf_printf(&sbuf, "\nFREE LIST %d:\n"
222 			    "\n  ORDER (SIZE)  |  NUMBER"
223 			    "\n              ", flind);
224 			for (pind = 0; pind < VM_NFREEPOOL; pind++)
225 				sbuf_printf(&sbuf, "  |  POOL %d", pind);
226 			sbuf_printf(&sbuf, "\n--            ");
227 			for (pind = 0; pind < VM_NFREEPOOL; pind++)
228 				sbuf_printf(&sbuf, "-- --      ");
229 			sbuf_printf(&sbuf, "--\n");
230 			for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
231 				sbuf_printf(&sbuf, "  %2d (%6dK)", oind,
232 				    1 << (PAGE_SHIFT - 10 + oind));
233 				for (pind = 0; pind < VM_NFREEPOOL; pind++) {
234 				fl = vm_phys_free_queues[dom][flind][pind];
235 					sbuf_printf(&sbuf, "  |  %6d",
236 					    fl[oind].lcnt);
237 				}
238 				sbuf_printf(&sbuf, "\n");
239 			}
240 		}
241 	}
242 	error = sbuf_finish(&sbuf);
243 	sbuf_delete(&sbuf);
244 	return (error);
245 }
246 
247 /*
248  * Outputs the set of physical memory segments.
249  */
250 static int
251 sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS)
252 {
253 	struct sbuf sbuf;
254 	struct vm_phys_seg *seg;
255 	int error, segind;
256 
257 	error = sysctl_wire_old_buffer(req, 0);
258 	if (error != 0)
259 		return (error);
260 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
261 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
262 		sbuf_printf(&sbuf, "\nSEGMENT %d:\n\n", segind);
263 		seg = &vm_phys_segs[segind];
264 		sbuf_printf(&sbuf, "start:     %#jx\n",
265 		    (uintmax_t)seg->start);
266 		sbuf_printf(&sbuf, "end:       %#jx\n",
267 		    (uintmax_t)seg->end);
268 		sbuf_printf(&sbuf, "domain:    %d\n", seg->domain);
269 		sbuf_printf(&sbuf, "free list: %p\n", seg->free_queues);
270 	}
271 	error = sbuf_finish(&sbuf);
272 	sbuf_delete(&sbuf);
273 	return (error);
274 }
275 
276 static void
277 vm_freelist_add(struct vm_freelist *fl, vm_page_t m, int order, int tail)
278 {
279 
280 	m->order = order;
281 	if (tail)
282 		TAILQ_INSERT_TAIL(&fl[order].pl, m, plinks.q);
283 	else
284 		TAILQ_INSERT_HEAD(&fl[order].pl, m, plinks.q);
285 	fl[order].lcnt++;
286 }
287 
288 static void
289 vm_freelist_rem(struct vm_freelist *fl, vm_page_t m, int order)
290 {
291 
292 	TAILQ_REMOVE(&fl[order].pl, m, plinks.q);
293 	fl[order].lcnt--;
294 	m->order = VM_NFREEORDER;
295 }
296 
297 /*
298  * Create a physical memory segment.
299  */
300 static void
301 _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind, int domain)
302 {
303 	struct vm_phys_seg *seg;
304 #ifdef VM_PHYSSEG_SPARSE
305 	long pages;
306 	int segind;
307 
308 	pages = 0;
309 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
310 		seg = &vm_phys_segs[segind];
311 		pages += atop(seg->end - seg->start);
312 	}
313 #endif
314 	KASSERT(vm_phys_nsegs < VM_PHYSSEG_MAX,
315 	    ("vm_phys_create_seg: increase VM_PHYSSEG_MAX"));
316 	KASSERT(domain < vm_ndomains,
317 	    ("vm_phys_create_seg: invalid domain provided"));
318 	seg = &vm_phys_segs[vm_phys_nsegs++];
319 	seg->start = start;
320 	seg->end = end;
321 	seg->domain = domain;
322 #ifdef VM_PHYSSEG_SPARSE
323 	seg->first_page = &vm_page_array[pages];
324 #else
325 	seg->first_page = PHYS_TO_VM_PAGE(start);
326 #endif
327 	seg->free_queues = &vm_phys_free_queues[domain][flind];
328 }
329 
330 static void
331 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int flind)
332 {
333 	int i;
334 
335 	if (mem_affinity == NULL) {
336 		_vm_phys_create_seg(start, end, flind, 0);
337 		return;
338 	}
339 
340 	for (i = 0;; i++) {
341 		if (mem_affinity[i].end == 0)
342 			panic("Reached end of affinity info");
343 		if (mem_affinity[i].end <= start)
344 			continue;
345 		if (mem_affinity[i].start > start)
346 			panic("No affinity info for start %jx",
347 			    (uintmax_t)start);
348 		if (mem_affinity[i].end >= end) {
349 			_vm_phys_create_seg(start, end, flind,
350 			    mem_affinity[i].domain);
351 			break;
352 		}
353 		_vm_phys_create_seg(start, mem_affinity[i].end, flind,
354 		    mem_affinity[i].domain);
355 		start = mem_affinity[i].end;
356 	}
357 }
358 
359 /*
360  * Initialize the physical memory allocator.
361  */
362 void
363 vm_phys_init(void)
364 {
365 	struct vm_freelist *fl;
366 	int dom, flind, i, oind, pind;
367 
368 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
369 #ifdef	VM_FREELIST_ISADMA
370 		if (phys_avail[i] < 16777216) {
371 			if (phys_avail[i + 1] > 16777216) {
372 				vm_phys_create_seg(phys_avail[i], 16777216,
373 				    VM_FREELIST_ISADMA);
374 				vm_phys_create_seg(16777216, phys_avail[i + 1],
375 				    VM_FREELIST_DEFAULT);
376 			} else {
377 				vm_phys_create_seg(phys_avail[i],
378 				    phys_avail[i + 1], VM_FREELIST_ISADMA);
379 			}
380 			if (VM_FREELIST_ISADMA >= vm_nfreelists)
381 				vm_nfreelists = VM_FREELIST_ISADMA + 1;
382 		} else
383 #endif
384 #ifdef	VM_FREELIST_HIGHMEM
385 		if (phys_avail[i + 1] > VM_HIGHMEM_ADDRESS) {
386 			if (phys_avail[i] < VM_HIGHMEM_ADDRESS) {
387 				vm_phys_create_seg(phys_avail[i],
388 				    VM_HIGHMEM_ADDRESS, VM_FREELIST_DEFAULT);
389 				vm_phys_create_seg(VM_HIGHMEM_ADDRESS,
390 				    phys_avail[i + 1], VM_FREELIST_HIGHMEM);
391 			} else {
392 				vm_phys_create_seg(phys_avail[i],
393 				    phys_avail[i + 1], VM_FREELIST_HIGHMEM);
394 			}
395 			if (VM_FREELIST_HIGHMEM >= vm_nfreelists)
396 				vm_nfreelists = VM_FREELIST_HIGHMEM + 1;
397 		} else
398 #endif
399 		vm_phys_create_seg(phys_avail[i], phys_avail[i + 1],
400 		    VM_FREELIST_DEFAULT);
401 	}
402 	for (dom = 0; dom < vm_ndomains; dom++) {
403 		for (flind = 0; flind < vm_nfreelists; flind++) {
404 			for (pind = 0; pind < VM_NFREEPOOL; pind++) {
405 				fl = vm_phys_free_queues[dom][flind][pind];
406 				for (oind = 0; oind < VM_NFREEORDER; oind++)
407 					TAILQ_INIT(&fl[oind].pl);
408 			}
409 		}
410 	}
411 	rw_init(&vm_phys_fictitious_reg_lock, "vmfctr");
412 }
413 
414 /*
415  * Split a contiguous, power of two-sized set of physical pages.
416  */
417 static __inline void
418 vm_phys_split_pages(vm_page_t m, int oind, struct vm_freelist *fl, int order)
419 {
420 	vm_page_t m_buddy;
421 
422 	while (oind > order) {
423 		oind--;
424 		m_buddy = &m[1 << oind];
425 		KASSERT(m_buddy->order == VM_NFREEORDER,
426 		    ("vm_phys_split_pages: page %p has unexpected order %d",
427 		    m_buddy, m_buddy->order));
428 		vm_freelist_add(fl, m_buddy, oind, 0);
429         }
430 }
431 
432 /*
433  * Initialize a physical page and add it to the free lists.
434  */
435 void
436 vm_phys_add_page(vm_paddr_t pa)
437 {
438 	vm_page_t m;
439 	struct vm_domain *vmd;
440 
441 	vm_cnt.v_page_count++;
442 	m = vm_phys_paddr_to_vm_page(pa);
443 	m->phys_addr = pa;
444 	m->queue = PQ_NONE;
445 	m->segind = vm_phys_paddr_to_segind(pa);
446 	vmd = vm_phys_domain(m);
447 	vmd->vmd_page_count++;
448 	vmd->vmd_segs |= 1UL << m->segind;
449 	KASSERT(m->order == VM_NFREEORDER,
450 	    ("vm_phys_add_page: page %p has unexpected order %d",
451 	    m, m->order));
452 	m->pool = VM_FREEPOOL_DEFAULT;
453 	pmap_page_init(m);
454 	mtx_lock(&vm_page_queue_free_mtx);
455 	vm_phys_freecnt_adj(m, 1);
456 	vm_phys_free_pages(m, 0);
457 	mtx_unlock(&vm_page_queue_free_mtx);
458 }
459 
460 /*
461  * Allocate a contiguous, power of two-sized set of physical pages
462  * from the free lists.
463  *
464  * The free page queues must be locked.
465  */
466 vm_page_t
467 vm_phys_alloc_pages(int pool, int order)
468 {
469 	vm_page_t m;
470 	int dom, domain, flind;
471 
472 	KASSERT(pool < VM_NFREEPOOL,
473 	    ("vm_phys_alloc_pages: pool %d is out of range", pool));
474 	KASSERT(order < VM_NFREEORDER,
475 	    ("vm_phys_alloc_pages: order %d is out of range", order));
476 
477 	for (dom = 0; dom < vm_ndomains; dom++) {
478 		domain = vm_rr_selectdomain();
479 		for (flind = 0; flind < vm_nfreelists; flind++) {
480 			m = vm_phys_alloc_domain_pages(domain, flind, pool,
481 			    order);
482 			if (m != NULL)
483 				return (m);
484 		}
485 	}
486 	return (NULL);
487 }
488 
489 /*
490  * Find and dequeue a free page on the given free list, with the
491  * specified pool and order
492  */
493 vm_page_t
494 vm_phys_alloc_freelist_pages(int flind, int pool, int order)
495 {
496 	vm_page_t m;
497 	int dom, domain;
498 
499 	KASSERT(flind < VM_NFREELIST,
500 	    ("vm_phys_alloc_freelist_pages: freelist %d is out of range", flind));
501 	KASSERT(pool < VM_NFREEPOOL,
502 	    ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
503 	KASSERT(order < VM_NFREEORDER,
504 	    ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
505 
506 	for (dom = 0; dom < vm_ndomains; dom++) {
507 		domain = vm_rr_selectdomain();
508 		m = vm_phys_alloc_domain_pages(domain, flind, pool, order);
509 		if (m != NULL)
510 			return (m);
511 	}
512 	return (NULL);
513 }
514 
515 static vm_page_t
516 vm_phys_alloc_domain_pages(int domain, int flind, int pool, int order)
517 {
518 	struct vm_freelist *fl;
519 	struct vm_freelist *alt;
520 	int oind, pind;
521 	vm_page_t m;
522 
523 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
524 	fl = &vm_phys_free_queues[domain][flind][pool][0];
525 	for (oind = order; oind < VM_NFREEORDER; oind++) {
526 		m = TAILQ_FIRST(&fl[oind].pl);
527 		if (m != NULL) {
528 			vm_freelist_rem(fl, m, oind);
529 			vm_phys_split_pages(m, oind, fl, order);
530 			return (m);
531 		}
532 	}
533 
534 	/*
535 	 * The given pool was empty.  Find the largest
536 	 * contiguous, power-of-two-sized set of pages in any
537 	 * pool.  Transfer these pages to the given pool, and
538 	 * use them to satisfy the allocation.
539 	 */
540 	for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
541 		for (pind = 0; pind < VM_NFREEPOOL; pind++) {
542 			alt = &vm_phys_free_queues[domain][flind][pind][0];
543 			m = TAILQ_FIRST(&alt[oind].pl);
544 			if (m != NULL) {
545 				vm_freelist_rem(alt, m, oind);
546 				vm_phys_set_pool(pool, m, oind);
547 				vm_phys_split_pages(m, oind, fl, order);
548 				return (m);
549 			}
550 		}
551 	}
552 	return (NULL);
553 }
554 
555 /*
556  * Find the vm_page corresponding to the given physical address.
557  */
558 vm_page_t
559 vm_phys_paddr_to_vm_page(vm_paddr_t pa)
560 {
561 	struct vm_phys_seg *seg;
562 	int segind;
563 
564 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
565 		seg = &vm_phys_segs[segind];
566 		if (pa >= seg->start && pa < seg->end)
567 			return (&seg->first_page[atop(pa - seg->start)]);
568 	}
569 	return (NULL);
570 }
571 
572 vm_page_t
573 vm_phys_fictitious_to_vm_page(vm_paddr_t pa)
574 {
575 	struct vm_phys_fictitious_seg tmp, *seg;
576 	vm_page_t m;
577 
578 	m = NULL;
579 	tmp.start = pa;
580 	tmp.end = 0;
581 
582 	rw_rlock(&vm_phys_fictitious_reg_lock);
583 	seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
584 	rw_runlock(&vm_phys_fictitious_reg_lock);
585 	if (seg == NULL)
586 		return (NULL);
587 
588 	m = &seg->first_page[atop(pa - seg->start)];
589 	KASSERT((m->flags & PG_FICTITIOUS) != 0, ("%p not fictitious", m));
590 
591 	return (m);
592 }
593 
594 static inline void
595 vm_phys_fictitious_init_range(vm_page_t range, vm_paddr_t start,
596     long page_count, vm_memattr_t memattr)
597 {
598 	long i;
599 
600 	for (i = 0; i < page_count; i++) {
601 		vm_page_initfake(&range[i], start + PAGE_SIZE * i, memattr);
602 		range[i].oflags &= ~VPO_UNMANAGED;
603 		range[i].busy_lock = VPB_UNBUSIED;
604 	}
605 }
606 
607 int
608 vm_phys_fictitious_reg_range(vm_paddr_t start, vm_paddr_t end,
609     vm_memattr_t memattr)
610 {
611 	struct vm_phys_fictitious_seg *seg;
612 	vm_page_t fp;
613 	long page_count;
614 #ifdef VM_PHYSSEG_DENSE
615 	long pi, pe;
616 	long dpage_count;
617 #endif
618 
619 	KASSERT(start < end,
620 	    ("Start of segment isn't less than end (start: %jx end: %jx)",
621 	    (uintmax_t)start, (uintmax_t)end));
622 
623 	page_count = (end - start) / PAGE_SIZE;
624 
625 #ifdef VM_PHYSSEG_DENSE
626 	pi = atop(start);
627 	pe = atop(end);
628 	if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
629 		fp = &vm_page_array[pi - first_page];
630 		if ((pe - first_page) > vm_page_array_size) {
631 			/*
632 			 * We have a segment that starts inside
633 			 * of vm_page_array, but ends outside of it.
634 			 *
635 			 * Use vm_page_array pages for those that are
636 			 * inside of the vm_page_array range, and
637 			 * allocate the remaining ones.
638 			 */
639 			dpage_count = vm_page_array_size - (pi - first_page);
640 			vm_phys_fictitious_init_range(fp, start, dpage_count,
641 			    memattr);
642 			page_count -= dpage_count;
643 			start += ptoa(dpage_count);
644 			goto alloc;
645 		}
646 		/*
647 		 * We can allocate the full range from vm_page_array,
648 		 * so there's no need to register the range in the tree.
649 		 */
650 		vm_phys_fictitious_init_range(fp, start, page_count, memattr);
651 		return (0);
652 	} else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
653 		/*
654 		 * We have a segment that ends inside of vm_page_array,
655 		 * but starts outside of it.
656 		 */
657 		fp = &vm_page_array[0];
658 		dpage_count = pe - first_page;
659 		vm_phys_fictitious_init_range(fp, ptoa(first_page), dpage_count,
660 		    memattr);
661 		end -= ptoa(dpage_count);
662 		page_count -= dpage_count;
663 		goto alloc;
664 	} else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
665 		/*
666 		 * Trying to register a fictitious range that expands before
667 		 * and after vm_page_array.
668 		 */
669 		return (EINVAL);
670 	} else {
671 alloc:
672 #endif
673 		fp = malloc(page_count * sizeof(struct vm_page), M_FICT_PAGES,
674 		    M_WAITOK | M_ZERO);
675 #ifdef VM_PHYSSEG_DENSE
676 	}
677 #endif
678 	vm_phys_fictitious_init_range(fp, start, page_count, memattr);
679 
680 	seg = malloc(sizeof(*seg), M_FICT_PAGES, M_WAITOK | M_ZERO);
681 	seg->start = start;
682 	seg->end = end;
683 	seg->first_page = fp;
684 
685 	rw_wlock(&vm_phys_fictitious_reg_lock);
686 	RB_INSERT(fict_tree, &vm_phys_fictitious_tree, seg);
687 	rw_wunlock(&vm_phys_fictitious_reg_lock);
688 
689 	return (0);
690 }
691 
692 void
693 vm_phys_fictitious_unreg_range(vm_paddr_t start, vm_paddr_t end)
694 {
695 	struct vm_phys_fictitious_seg *seg, tmp;
696 #ifdef VM_PHYSSEG_DENSE
697 	long pi, pe;
698 #endif
699 
700 	KASSERT(start < end,
701 	    ("Start of segment isn't less than end (start: %jx end: %jx)",
702 	    (uintmax_t)start, (uintmax_t)end));
703 
704 #ifdef VM_PHYSSEG_DENSE
705 	pi = atop(start);
706 	pe = atop(end);
707 	if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
708 		if ((pe - first_page) <= vm_page_array_size) {
709 			/*
710 			 * This segment was allocated using vm_page_array
711 			 * only, there's nothing to do since those pages
712 			 * were never added to the tree.
713 			 */
714 			return;
715 		}
716 		/*
717 		 * We have a segment that starts inside
718 		 * of vm_page_array, but ends outside of it.
719 		 *
720 		 * Calculate how many pages were added to the
721 		 * tree and free them.
722 		 */
723 		start = ptoa(first_page + vm_page_array_size);
724 	} else if (pe > first_page && (pe - first_page) < vm_page_array_size) {
725 		/*
726 		 * We have a segment that ends inside of vm_page_array,
727 		 * but starts outside of it.
728 		 */
729 		end = ptoa(first_page);
730 	} else if (pi < first_page && pe > (first_page + vm_page_array_size)) {
731 		/* Since it's not possible to register such a range, panic. */
732 		panic(
733 		    "Unregistering not registered fictitious range [%#jx:%#jx]",
734 		    (uintmax_t)start, (uintmax_t)end);
735 	}
736 #endif
737 	tmp.start = start;
738 	tmp.end = 0;
739 
740 	rw_wlock(&vm_phys_fictitious_reg_lock);
741 	seg = RB_FIND(fict_tree, &vm_phys_fictitious_tree, &tmp);
742 	if (seg->start != start || seg->end != end) {
743 		rw_wunlock(&vm_phys_fictitious_reg_lock);
744 		panic(
745 		    "Unregistering not registered fictitious range [%#jx:%#jx]",
746 		    (uintmax_t)start, (uintmax_t)end);
747 	}
748 	RB_REMOVE(fict_tree, &vm_phys_fictitious_tree, seg);
749 	rw_wunlock(&vm_phys_fictitious_reg_lock);
750 	free(seg->first_page, M_FICT_PAGES);
751 	free(seg, M_FICT_PAGES);
752 }
753 
754 /*
755  * Find the segment containing the given physical address.
756  */
757 static int
758 vm_phys_paddr_to_segind(vm_paddr_t pa)
759 {
760 	struct vm_phys_seg *seg;
761 	int segind;
762 
763 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
764 		seg = &vm_phys_segs[segind];
765 		if (pa >= seg->start && pa < seg->end)
766 			return (segind);
767 	}
768 	panic("vm_phys_paddr_to_segind: paddr %#jx is not in any segment" ,
769 	    (uintmax_t)pa);
770 }
771 
772 /*
773  * Free a contiguous, power of two-sized set of physical pages.
774  *
775  * The free page queues must be locked.
776  */
777 void
778 vm_phys_free_pages(vm_page_t m, int order)
779 {
780 	struct vm_freelist *fl;
781 	struct vm_phys_seg *seg;
782 	vm_paddr_t pa;
783 	vm_page_t m_buddy;
784 
785 	KASSERT(m->order == VM_NFREEORDER,
786 	    ("vm_phys_free_pages: page %p has unexpected order %d",
787 	    m, m->order));
788 	KASSERT(m->pool < VM_NFREEPOOL,
789 	    ("vm_phys_free_pages: page %p has unexpected pool %d",
790 	    m, m->pool));
791 	KASSERT(order < VM_NFREEORDER,
792 	    ("vm_phys_free_pages: order %d is out of range", order));
793 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
794 	seg = &vm_phys_segs[m->segind];
795 	if (order < VM_NFREEORDER - 1) {
796 		pa = VM_PAGE_TO_PHYS(m);
797 		do {
798 			pa ^= ((vm_paddr_t)1 << (PAGE_SHIFT + order));
799 			if (pa < seg->start || pa >= seg->end)
800 				break;
801 			m_buddy = &seg->first_page[atop(pa - seg->start)];
802 			if (m_buddy->order != order)
803 				break;
804 			fl = (*seg->free_queues)[m_buddy->pool];
805 			vm_freelist_rem(fl, m_buddy, order);
806 			if (m_buddy->pool != m->pool)
807 				vm_phys_set_pool(m->pool, m_buddy, order);
808 			order++;
809 			pa &= ~(((vm_paddr_t)1 << (PAGE_SHIFT + order)) - 1);
810 			m = &seg->first_page[atop(pa - seg->start)];
811 		} while (order < VM_NFREEORDER - 1);
812 	}
813 	fl = (*seg->free_queues)[m->pool];
814 	vm_freelist_add(fl, m, order, 1);
815 }
816 
817 /*
818  * Free a contiguous, arbitrarily sized set of physical pages.
819  *
820  * The free page queues must be locked.
821  */
822 void
823 vm_phys_free_contig(vm_page_t m, u_long npages)
824 {
825 	u_int n;
826 	int order;
827 
828 	/*
829 	 * Avoid unnecessary coalescing by freeing the pages in the largest
830 	 * possible power-of-two-sized subsets.
831 	 */
832 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
833 	for (;; npages -= n) {
834 		/*
835 		 * Unsigned "min" is used here so that "order" is assigned
836 		 * "VM_NFREEORDER - 1" when "m"'s physical address is zero
837 		 * or the low-order bits of its physical address are zero
838 		 * because the size of a physical address exceeds the size of
839 		 * a long.
840 		 */
841 		order = min(ffsl(VM_PAGE_TO_PHYS(m) >> PAGE_SHIFT) - 1,
842 		    VM_NFREEORDER - 1);
843 		n = 1 << order;
844 		if (npages < n)
845 			break;
846 		vm_phys_free_pages(m, order);
847 		m += n;
848 	}
849 	/* The residual "npages" is less than "1 << (VM_NFREEORDER - 1)". */
850 	for (; npages > 0; npages -= n) {
851 		order = flsl(npages) - 1;
852 		n = 1 << order;
853 		vm_phys_free_pages(m, order);
854 		m += n;
855 	}
856 }
857 
858 /*
859  * Set the pool for a contiguous, power of two-sized set of physical pages.
860  */
861 void
862 vm_phys_set_pool(int pool, vm_page_t m, int order)
863 {
864 	vm_page_t m_tmp;
865 
866 	for (m_tmp = m; m_tmp < &m[1 << order]; m_tmp++)
867 		m_tmp->pool = pool;
868 }
869 
870 /*
871  * Search for the given physical page "m" in the free lists.  If the search
872  * succeeds, remove "m" from the free lists and return TRUE.  Otherwise, return
873  * FALSE, indicating that "m" is not in the free lists.
874  *
875  * The free page queues must be locked.
876  */
877 boolean_t
878 vm_phys_unfree_page(vm_page_t m)
879 {
880 	struct vm_freelist *fl;
881 	struct vm_phys_seg *seg;
882 	vm_paddr_t pa, pa_half;
883 	vm_page_t m_set, m_tmp;
884 	int order;
885 
886 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
887 
888 	/*
889 	 * First, find the contiguous, power of two-sized set of free
890 	 * physical pages containing the given physical page "m" and
891 	 * assign it to "m_set".
892 	 */
893 	seg = &vm_phys_segs[m->segind];
894 	for (m_set = m, order = 0; m_set->order == VM_NFREEORDER &&
895 	    order < VM_NFREEORDER - 1; ) {
896 		order++;
897 		pa = m->phys_addr & (~(vm_paddr_t)0 << (PAGE_SHIFT + order));
898 		if (pa >= seg->start)
899 			m_set = &seg->first_page[atop(pa - seg->start)];
900 		else
901 			return (FALSE);
902 	}
903 	if (m_set->order < order)
904 		return (FALSE);
905 	if (m_set->order == VM_NFREEORDER)
906 		return (FALSE);
907 	KASSERT(m_set->order < VM_NFREEORDER,
908 	    ("vm_phys_unfree_page: page %p has unexpected order %d",
909 	    m_set, m_set->order));
910 
911 	/*
912 	 * Next, remove "m_set" from the free lists.  Finally, extract
913 	 * "m" from "m_set" using an iterative algorithm: While "m_set"
914 	 * is larger than a page, shrink "m_set" by returning the half
915 	 * of "m_set" that does not contain "m" to the free lists.
916 	 */
917 	fl = (*seg->free_queues)[m_set->pool];
918 	order = m_set->order;
919 	vm_freelist_rem(fl, m_set, order);
920 	while (order > 0) {
921 		order--;
922 		pa_half = m_set->phys_addr ^ (1 << (PAGE_SHIFT + order));
923 		if (m->phys_addr < pa_half)
924 			m_tmp = &seg->first_page[atop(pa_half - seg->start)];
925 		else {
926 			m_tmp = m_set;
927 			m_set = &seg->first_page[atop(pa_half - seg->start)];
928 		}
929 		vm_freelist_add(fl, m_tmp, order, 0);
930 	}
931 	KASSERT(m_set == m, ("vm_phys_unfree_page: fatal inconsistency"));
932 	return (TRUE);
933 }
934 
935 /*
936  * Try to zero one physical page.  Used by an idle priority thread.
937  */
938 boolean_t
939 vm_phys_zero_pages_idle(void)
940 {
941 	static struct vm_freelist *fl;
942 	static int flind, oind, pind;
943 	vm_page_t m, m_tmp;
944 	int domain;
945 
946 	domain = vm_rr_selectdomain();
947 	fl = vm_phys_free_queues[domain][0][0];
948 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
949 	for (;;) {
950 		TAILQ_FOREACH_REVERSE(m, &fl[oind].pl, pglist, plinks.q) {
951 			for (m_tmp = m; m_tmp < &m[1 << oind]; m_tmp++) {
952 				if ((m_tmp->flags & (PG_CACHED | PG_ZERO)) == 0) {
953 					vm_phys_unfree_page(m_tmp);
954 					vm_phys_freecnt_adj(m, -1);
955 					mtx_unlock(&vm_page_queue_free_mtx);
956 					pmap_zero_page_idle(m_tmp);
957 					m_tmp->flags |= PG_ZERO;
958 					mtx_lock(&vm_page_queue_free_mtx);
959 					vm_phys_freecnt_adj(m, 1);
960 					vm_phys_free_pages(m_tmp, 0);
961 					vm_page_zero_count++;
962 					cnt_prezero++;
963 					return (TRUE);
964 				}
965 			}
966 		}
967 		oind++;
968 		if (oind == VM_NFREEORDER) {
969 			oind = 0;
970 			pind++;
971 			if (pind == VM_NFREEPOOL) {
972 				pind = 0;
973 				flind++;
974 				if (flind == vm_nfreelists)
975 					flind = 0;
976 			}
977 			fl = vm_phys_free_queues[domain][flind][pind];
978 		}
979 	}
980 }
981 
982 /*
983  * Allocate a contiguous set of physical pages of the given size
984  * "npages" from the free lists.  All of the physical pages must be at
985  * or above the given physical address "low" and below the given
986  * physical address "high".  The given value "alignment" determines the
987  * alignment of the first physical page in the set.  If the given value
988  * "boundary" is non-zero, then the set of physical pages cannot cross
989  * any physical address boundary that is a multiple of that value.  Both
990  * "alignment" and "boundary" must be a power of two.
991  */
992 vm_page_t
993 vm_phys_alloc_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
994     u_long alignment, vm_paddr_t boundary)
995 {
996 	struct vm_freelist *fl;
997 	struct vm_phys_seg *seg;
998 	vm_paddr_t pa, pa_last, size;
999 	vm_page_t m, m_ret;
1000 	u_long npages_end;
1001 	int dom, domain, flind, oind, order, pind;
1002 
1003 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
1004 	size = npages << PAGE_SHIFT;
1005 	KASSERT(size != 0,
1006 	    ("vm_phys_alloc_contig: size must not be 0"));
1007 	KASSERT((alignment & (alignment - 1)) == 0,
1008 	    ("vm_phys_alloc_contig: alignment must be a power of 2"));
1009 	KASSERT((boundary & (boundary - 1)) == 0,
1010 	    ("vm_phys_alloc_contig: boundary must be a power of 2"));
1011 	/* Compute the queue that is the best fit for npages. */
1012 	for (order = 0; (1 << order) < npages; order++);
1013 	dom = 0;
1014 restartdom:
1015 	domain = vm_rr_selectdomain();
1016 	for (flind = 0; flind < vm_nfreelists; flind++) {
1017 		for (oind = min(order, VM_NFREEORDER - 1); oind < VM_NFREEORDER; oind++) {
1018 			for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1019 				fl = &vm_phys_free_queues[domain][flind][pind][0];
1020 				TAILQ_FOREACH(m_ret, &fl[oind].pl, plinks.q) {
1021 					/*
1022 					 * A free list may contain physical pages
1023 					 * from one or more segments.
1024 					 */
1025 					seg = &vm_phys_segs[m_ret->segind];
1026 					if (seg->start > high ||
1027 					    low >= seg->end)
1028 						continue;
1029 
1030 					/*
1031 					 * Is the size of this allocation request
1032 					 * larger than the largest block size?
1033 					 */
1034 					if (order >= VM_NFREEORDER) {
1035 						/*
1036 						 * Determine if a sufficient number
1037 						 * of subsequent blocks to satisfy
1038 						 * the allocation request are free.
1039 						 */
1040 						pa = VM_PAGE_TO_PHYS(m_ret);
1041 						pa_last = pa + size;
1042 						for (;;) {
1043 							pa += 1 << (PAGE_SHIFT + VM_NFREEORDER - 1);
1044 							if (pa >= pa_last)
1045 								break;
1046 							if (pa < seg->start ||
1047 							    pa >= seg->end)
1048 								break;
1049 							m = &seg->first_page[atop(pa - seg->start)];
1050 							if (m->order != VM_NFREEORDER - 1)
1051 								break;
1052 						}
1053 						/* If not, continue to the next block. */
1054 						if (pa < pa_last)
1055 							continue;
1056 					}
1057 
1058 					/*
1059 					 * Determine if the blocks are within the given range,
1060 					 * satisfy the given alignment, and do not cross the
1061 					 * given boundary.
1062 					 */
1063 					pa = VM_PAGE_TO_PHYS(m_ret);
1064 					if (pa >= low &&
1065 					    pa + size <= high &&
1066 					    (pa & (alignment - 1)) == 0 &&
1067 					    ((pa ^ (pa + size - 1)) & ~(boundary - 1)) == 0)
1068 						goto done;
1069 				}
1070 			}
1071 		}
1072 	}
1073 	if (++dom < vm_ndomains)
1074 		goto restartdom;
1075 	return (NULL);
1076 done:
1077 	for (m = m_ret; m < &m_ret[npages]; m = &m[1 << oind]) {
1078 		fl = (*seg->free_queues)[m->pool];
1079 		vm_freelist_rem(fl, m, m->order);
1080 	}
1081 	if (m_ret->pool != VM_FREEPOOL_DEFAULT)
1082 		vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m_ret, oind);
1083 	fl = (*seg->free_queues)[m_ret->pool];
1084 	vm_phys_split_pages(m_ret, oind, fl, order);
1085 	/* Return excess pages to the free lists. */
1086 	npages_end = roundup2(npages, 1 << imin(oind, order));
1087 	if (npages < npages_end)
1088 		vm_phys_free_contig(&m_ret[npages], npages_end - npages);
1089 	return (m_ret);
1090 }
1091 
1092 #ifdef DDB
1093 /*
1094  * Show the number of physical pages in each of the free lists.
1095  */
1096 DB_SHOW_COMMAND(freepages, db_show_freepages)
1097 {
1098 	struct vm_freelist *fl;
1099 	int flind, oind, pind, dom;
1100 
1101 	for (dom = 0; dom < vm_ndomains; dom++) {
1102 		db_printf("DOMAIN: %d\n", dom);
1103 		for (flind = 0; flind < vm_nfreelists; flind++) {
1104 			db_printf("FREE LIST %d:\n"
1105 			    "\n  ORDER (SIZE)  |  NUMBER"
1106 			    "\n              ", flind);
1107 			for (pind = 0; pind < VM_NFREEPOOL; pind++)
1108 				db_printf("  |  POOL %d", pind);
1109 			db_printf("\n--            ");
1110 			for (pind = 0; pind < VM_NFREEPOOL; pind++)
1111 				db_printf("-- --      ");
1112 			db_printf("--\n");
1113 			for (oind = VM_NFREEORDER - 1; oind >= 0; oind--) {
1114 				db_printf("  %2.2d (%6.6dK)", oind,
1115 				    1 << (PAGE_SHIFT - 10 + oind));
1116 				for (pind = 0; pind < VM_NFREEPOOL; pind++) {
1117 				fl = vm_phys_free_queues[dom][flind][pind];
1118 					db_printf("  |  %6.6d", fl[oind].lcnt);
1119 				}
1120 				db_printf("\n");
1121 			}
1122 			db_printf("\n");
1123 		}
1124 		db_printf("\n");
1125 	}
1126 }
1127 #endif
1128