2 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Hiten Pandya <hmp@backplane.com>.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * Copyright (c) 1991 Regents of the University of California.
37 * All rights reserved.
39 * This code is derived from software contributed to Berkeley by
40 * The Mach Operating System project at Carnegie-Mellon University.
42 * Redistribution and use in source and binary forms, with or without
43 * modification, are permitted provided that the following conditions
45 * 1. Redistributions of source code must retain the above copyright
46 * notice, this list of conditions and the following disclaimer.
47 * 2. Redistributions in binary form must reproduce the above copyright
48 * notice, this list of conditions and the following disclaimer in the
49 * documentation and/or other materials provided with the distribution.
50 * 3. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91
67 * $DragonFly: src/sys/vm/vm_contig.c,v 1.21 2006/12/28 21:24:02 dillon Exp $
71 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
72 * All rights reserved.
74 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
76 * Permission to use, copy, modify and distribute this software and
77 * its documentation is hereby granted, provided that both the copyright
78 * notice and this permission notice appear in all copies of the
79 * software, derivative works or modified versions, and any portions
80 * thereof, and that both notices appear in supporting documentation.
82 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
83 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
84 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
86 * Carnegie Mellon requests users of this software to return to
88 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
89 * School of Computer Science
90 * Carnegie Mellon University
91 * Pittsburgh PA 15213-3890
93 * any improvements or extensions that they make and grant Carnegie the
94 * rights to redistribute these changes.
98 * Contiguous memory allocation API.
101 #include <sys/param.h>
102 #include <sys/systm.h>
103 #include <sys/malloc.h>
104 #include <sys/proc.h>
105 #include <sys/lock.h>
106 #include <sys/vmmeter.h>
107 #include <sys/vnode.h>
110 #include <vm/vm_param.h>
111 #include <vm/vm_kern.h>
113 #include <vm/vm_map.h>
114 #include <vm/vm_object.h>
115 #include <vm/vm_page.h>
116 #include <vm/vm_pageout.h>
117 #include <vm/vm_pager.h>
118 #include <vm/vm_extern.h>
120 #include <sys/thread2.h>
121 #include <sys/spinlock2.h>
122 #include <vm/vm_page2.h>
124 static void vm_contig_pg_free(int start, u_long size);
127 * vm_contig_pg_clean:
129 * Do a thorough cleanup of the specified 'queue', which can be either
130 * PQ_ACTIVE or PQ_INACTIVE by doing a walkthrough. If the page is not
131 * marked dirty, it is shoved into the page cache, provided no one has
132 * currently aqcuired it, otherwise localized action per object type
133 * is taken for cleanup:
135 * In the OBJT_VNODE case, the whole page range is cleaned up
136 * using the vm_object_page_clean() routine, by specyfing a
137 * start and end of '0'.
139 * Otherwise if the object is of any other type, the generic
140 * pageout (daemon) flush routine is invoked.
143 vm_contig_pg_clean(int queue, int count)
147 struct vm_page marker;
148 struct vpgqueues *pq = &vm_page_queues[queue];
151 * Setup a local marker
153 bzero(&marker, sizeof(marker));
154 marker.flags = PG_FICTITIOUS | PG_MARKER;
155 marker.busy_count = PBUSY_LOCKED;
156 marker.queue = queue;
157 marker.wire_count = 1;
159 vm_page_queues_spin_lock(queue);
160 TAILQ_INSERT_HEAD(&pq->pl, &marker, pageq);
161 vm_page_queues_spin_unlock(queue);
164 * Iterate the queue. Note that the vm_page spinlock must be
165 * acquired before the pageq spinlock so it's easiest to simply
166 * not hold it in the loop iteration.
168 while (count-- > 0 && (m = TAILQ_NEXT(&marker, pageq)) != NULL) {
169 vm_page_and_queue_spin_lock(m);
170 if (m != TAILQ_NEXT(&marker, pageq)) {
171 vm_page_and_queue_spin_unlock(m);
175 KKASSERT(m->queue == queue);
177 TAILQ_REMOVE(&pq->pl, &marker, pageq);
178 TAILQ_INSERT_AFTER(&pq->pl, m, &marker, pageq);
180 if (m->flags & PG_MARKER) {
181 vm_page_and_queue_spin_unlock(m);
184 if (vm_page_busy_try(m, TRUE)) {
185 vm_page_and_queue_spin_unlock(m);
188 vm_page_and_queue_spin_unlock(m);
191 * We've successfully busied the page
193 if (m->queue - m->pc != queue) {
197 if (m->wire_count || m->hold_count) {
201 if ((object = m->object) == NULL) {
205 vm_page_test_dirty(m);
206 if (m->dirty || (m->flags & PG_NEED_COMMIT)) {
207 vm_object_hold(object);
208 KKASSERT(m->object == object);
210 if (object->type == OBJT_VNODE) {
212 vn_lock(object->handle, LK_EXCLUSIVE|LK_RETRY);
213 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
214 vn_unlock(((struct vnode *)object->handle));
215 } else if (object->type == OBJT_SWAP ||
216 object->type == OBJT_DEFAULT) {
218 vm_pageout_flush(&m_tmp, 1, 0);
222 vm_object_drop(object);
223 } else if (m->hold_count == 0) {
231 * Scrap our local marker
233 vm_page_queues_spin_lock(queue);
234 TAILQ_REMOVE(&pq->pl, &marker, pageq);
235 vm_page_queues_spin_unlock(queue);
239 * vm_contig_pg_alloc:
241 * Allocate contiguous pages from the VM. This function does not
242 * map the allocated pages into the kernel map, otherwise it is
243 * impossible to make large allocations (i.e. >2G).
245 * Malloc()'s data structures have been used for collection of
246 * statistics and for allocations of less than a page.
249 vm_contig_pg_alloc(unsigned long size, vm_paddr_t low, vm_paddr_t high,
250 unsigned long alignment, unsigned long boundary, int mflags)
252 int i, q, start, pass;
254 vm_page_t pga = vm_page_array;
258 size = round_page(size);
260 panic("vm_contig_pg_alloc: size must not be 0");
261 if ((alignment & (alignment - 1)) != 0)
262 panic("vm_contig_pg_alloc: alignment must be a power of 2");
263 if ((boundary & (boundary - 1)) != 0)
264 panic("vm_contig_pg_alloc: boundary must be a power of 2");
267 * See if we can get the pages from the contiguous page reserve
268 * alist. The returned pages will be allocated and wired but not
271 m = vm_page_alloc_contig(
272 low, high, alignment, boundary, size, VM_MEMATTR_DEFAULT);
274 return (m - &pga[0]);
277 * Three passes (0, 1, 2). Each pass scans the VM page list for
278 * free or cached pages. After each pass if the entire scan failed
279 * we attempt to flush inactive pages and reset the start index back
280 * to 0. For passes 1 and 2 we also attempt to flush active pages.
283 for (pass = 0; pass < 3; pass++) {
285 * Find first page in array that is free, within range,
286 * aligned, and such that the boundary won't be crossed.
289 for (i = start; i < vmstats.v_page_count; i++) {
291 phys = VM_PAGE_TO_PHYS(m);
292 pqtype = m->queue - m->pc;
293 if (((pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) &&
294 (phys >= low) && (phys < high) &&
295 ((phys & (alignment - 1)) == 0) &&
296 (((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0) &&
297 m->wire_count == 0 && m->hold_count == 0 &&
299 (PBUSY_LOCKED | PBUSY_MASK)) == 0 &&
300 (m->flags & PG_NEED_COMMIT) == 0)
307 * If we cannot find the page in the given range, or we have
308 * crossed the boundary, call the vm_contig_pg_clean() function
309 * for flushing out the queues, and returning it back to
312 if ((i == vmstats.v_page_count) ||
313 ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) {
316 * Best effort flush of all inactive pages.
317 * This is quite quick, for now stall all
318 * callers, even if they've specified M_NOWAIT.
320 for (q = 0; q < PQ_L2_SIZE; ++q) {
321 vm_contig_pg_clean(PQ_INACTIVE + q,
322 vmstats.v_inactive_count);
327 * Best effort flush of active pages.
329 * This is very, very slow.
330 * Only do this if the caller has agreed to M_WAITOK.
332 * If enough pages are flushed, we may succeed on
333 * next (final) pass, if not the caller, contigmalloc(),
334 * will fail in the index < 0 case.
336 if (pass > 0 && (mflags & M_WAITOK)) {
337 for (q = 0; q < PQ_L2_SIZE; ++q) {
338 vm_contig_pg_clean(PQ_ACTIVE + q,
339 vmstats.v_active_count);
345 * We're already too high in the address space
346 * to succeed, reset to 0 for the next iteration.
349 continue; /* next pass */
354 * Check successive pages for contiguous and free.
356 * (still in critical section)
358 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) {
360 pqtype = m->queue - m->pc;
361 if ((VM_PAGE_TO_PHYS(&m[0]) !=
362 (VM_PAGE_TO_PHYS(&m[-1]) + PAGE_SIZE)) ||
363 ((pqtype != PQ_FREE) && (pqtype != PQ_CACHE)) ||
366 (m->busy_count & (PBUSY_LOCKED | PBUSY_MASK)) ||
367 (m->flags & PG_NEED_COMMIT))
375 * Try to allocate the pages, wiring them as we go.
377 * (still in critical section)
379 for (i = start; i < (start + size / PAGE_SIZE); i++) {
382 if (vm_page_busy_try(m, TRUE)) {
383 vm_contig_pg_free(start,
384 (i - start) * PAGE_SIZE);
388 pqtype = m->queue - m->pc;
389 if (pqtype == PQ_CACHE &&
390 m->hold_count == 0 &&
391 m->wire_count == 0 &&
392 (m->flags & (PG_UNMANAGED | PG_NEED_COMMIT)) == 0) {
393 vm_page_protect(m, VM_PROT_NONE);
394 KKASSERT((m->flags & PG_MAPPED) == 0);
395 KKASSERT(m->dirty == 0);
398 continue; /* retry the page */
400 if (pqtype != PQ_FREE || m->hold_count) {
402 vm_contig_pg_free(start,
403 (i - start) * PAGE_SIZE);
407 KKASSERT((m->valid & m->dirty) == 0);
408 KKASSERT(m->wire_count == 0);
409 KKASSERT(m->object == NULL);
410 vm_page_unqueue_nowakeup(m);
411 m->valid = VM_PAGE_BITS_ALL;
412 KASSERT(m->dirty == 0,
413 ("vm_contig_pg_alloc: page %p was dirty", m));
414 KKASSERT(m->wire_count == 0);
415 KKASSERT((m->busy_count & PBUSY_MASK) == 0);
418 * Clear all flags. Then unbusy the now allocated
421 vm_page_flag_clear(m, ~PG_KEEP_NEWPAGE_MASK);
427 * Our job is done, return the index page of vm_page_array.
429 return (start); /* aka &pga[start] */
441 * Remove pages previously allocated by vm_contig_pg_alloc, and
442 * assume all references to the pages have been removed, and that
443 * it is OK to add them back to the free list.
445 * Caller must ensure no races on the page range in question.
446 * No other requirements.
449 vm_contig_pg_free(int start, u_long size)
451 vm_page_t pga = vm_page_array;
453 size = round_page(size);
455 panic("vm_contig_pg_free: size must not be 0");
458 * The pages are wired, vm_page_free_contig() determines whether they
459 * belong to the contig space or not and either frees them to that
460 * space (leaving them wired), or unwires the page and frees it to the
461 * normal PQ_FREE queue.
463 vm_page_free_contig(&pga[start], size);
469 * Map previously allocated (vm_contig_pg_alloc) range of pages from
470 * vm_page_array[] into the KVA. Once mapped, the pages are part of
471 * the Kernel, and are to free'ed with kmem_free(&kernel_map, addr, size).
476 vm_contig_pg_kmap(int start, u_long size, vm_map_t map, int flags)
480 vm_page_t pga = vm_page_array;
484 panic("vm_contig_pg_kmap: size must not be 0");
485 size = round_page(size);
486 addr = kmem_alloc_pageable(&kernel_map, size, VM_SUBSYS_CONTIG);
488 pa = VM_PAGE_TO_PHYS(&pga[start]);
489 for (offset = 0; offset < size; offset += PAGE_SIZE)
490 pmap_kenter_noinval(addr + offset, pa + offset);
491 pmap_invalidate_range(&kernel_pmap, addr, addr + size);
493 bzero((void *)addr, size);
503 unsigned long size, /* should be size_t here and for malloc() */
504 struct malloc_type *type,
508 unsigned long alignment,
509 unsigned long boundary)
511 return contigmalloc_map(size, type, flags, low, high, alignment,
512 boundary, &kernel_map);
519 contigmalloc_map(unsigned long size, struct malloc_type *type,
520 int flags, vm_paddr_t low, vm_paddr_t high,
521 unsigned long alignment, unsigned long boundary,
527 index = vm_contig_pg_alloc(size, low, high, alignment, boundary, flags);
529 kprintf("contigmalloc_map: failed size %lu low=%llx "
530 "high=%llx align=%lu boundary=%lu flags=%08x\n",
531 size, (long long)low, (long long)high,
532 alignment, boundary, flags);
536 rv = (void *)vm_contig_pg_kmap(index, size, map, flags);
538 vm_contig_pg_free(index, size);
547 contigfree(void *addr, unsigned long size, struct malloc_type *type)
553 panic("vm_contig_pg_kmap: size must not be 0");
554 size = round_page(size);
556 pa = pmap_kextract((vm_offset_t)addr);
557 pmap_qremove((vm_offset_t)addr, size / PAGE_SIZE);
558 kmem_free(&kernel_map, (vm_offset_t)addr, size);
560 m = PHYS_TO_VM_PAGE(pa);
561 vm_page_free_contig(m, size);
568 kmem_alloc_contig(vm_offset_t size, vm_paddr_t low, vm_paddr_t high,
569 vm_offset_t alignment)
571 return ((vm_offset_t)contigmalloc_map(size, M_DEVBUF, M_NOWAIT, low,
572 high, alignment, 0ul, &kernel_map));