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_BUSY | PG_FICTITIOUS | PG_MARKER;
155 marker.queue = queue;
156 marker.wire_count = 1;
158 vm_page_queues_spin_lock(queue);
159 TAILQ_INSERT_HEAD(&pq->pl, &marker, pageq);
160 vm_page_queues_spin_unlock(queue);
163 * Iterate the queue. Note that the vm_page spinlock must be
164 * acquired before the pageq spinlock so it's easiest to simply
165 * not hold it in the loop iteration.
167 while (count-- > 0 && (m = TAILQ_NEXT(&marker, pageq)) != NULL) {
168 vm_page_and_queue_spin_lock(m);
169 if (m != TAILQ_NEXT(&marker, pageq)) {
170 vm_page_and_queue_spin_unlock(m);
174 KKASSERT(m->queue == queue);
176 TAILQ_REMOVE(&pq->pl, &marker, pageq);
177 TAILQ_INSERT_AFTER(&pq->pl, m, &marker, pageq);
179 if (m->flags & PG_MARKER) {
180 vm_page_and_queue_spin_unlock(m);
183 if (vm_page_busy_try(m, TRUE)) {
184 vm_page_and_queue_spin_unlock(m);
187 vm_page_and_queue_spin_unlock(m);
190 * We've successfully busied the page
192 if (m->queue - m->pc != queue) {
196 if ((object = m->object) == NULL) {
200 vm_page_test_dirty(m);
202 vm_object_hold(object);
203 KKASSERT(m->object == object);
205 if (object->type == OBJT_VNODE) {
207 vn_lock(object->handle, LK_EXCLUSIVE|LK_RETRY);
208 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
209 vn_unlock(((struct vnode *)object->handle));
210 } else if (object->type == OBJT_SWAP ||
211 object->type == OBJT_DEFAULT) {
213 vm_pageout_flush(&m_tmp, 1, 0);
217 vm_object_drop(object);
218 } else if (m->hold_count == 0) {
226 * Scrap our local marker
228 vm_page_queues_spin_lock(queue);
229 TAILQ_REMOVE(&pq->pl, &marker, pageq);
230 vm_page_queues_spin_unlock(queue);
234 * vm_contig_pg_alloc:
236 * Allocate contiguous pages from the VM. This function does not
237 * map the allocated pages into the kernel map, otherwise it is
238 * impossible to make large allocations (i.e. >2G).
240 * Malloc()'s data structures have been used for collection of
241 * statistics and for allocations of less than a page.
244 vm_contig_pg_alloc(unsigned long size, vm_paddr_t low, vm_paddr_t high,
245 unsigned long alignment, unsigned long boundary, int mflags)
249 vm_page_t pga = vm_page_array;
253 size = round_page(size);
255 panic("vm_contig_pg_alloc: size must not be 0");
256 if ((alignment & (alignment - 1)) != 0)
257 panic("vm_contig_pg_alloc: alignment must be a power of 2");
258 if ((boundary & (boundary - 1)) != 0)
259 panic("vm_contig_pg_alloc: boundary must be a power of 2");
264 * Three passes (0, 1, 2). Each pass scans the VM page list for
265 * free or cached pages. After each pass if the entire scan failed
266 * we attempt to flush inactive pages and reset the start index back
267 * to 0. For passes 1 and 2 we also attempt to flush active pages.
269 for (pass = 0; pass < 3; pass++) {
271 * Find first page in array that is free, within range,
272 * aligned, and such that the boundary won't be crossed.
275 for (i = start; i < vmstats.v_page_count; i++) {
277 phys = VM_PAGE_TO_PHYS(m);
278 pqtype = m->queue - m->pc;
279 if (((pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) &&
280 (phys >= low) && (phys < high) &&
281 ((phys & (alignment - 1)) == 0) &&
282 (((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0) &&
283 m->busy == 0 && m->wire_count == 0 &&
284 m->hold_count == 0 && (m->flags & PG_BUSY) == 0
292 * If we cannot find the page in the given range, or we have
293 * crossed the boundary, call the vm_contig_pg_clean() function
294 * for flushing out the queues, and returning it back to
297 if ((i == vmstats.v_page_count) ||
298 ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) {
301 * Best effort flush of all inactive pages.
302 * This is quite quick, for now stall all
303 * callers, even if they've specified M_NOWAIT.
305 vm_contig_pg_clean(PQ_INACTIVE,
306 vmstats.v_inactive_count);
309 * Best effort flush of active pages.
311 * This is very, very slow.
312 * Only do this if the caller has agreed to M_WAITOK.
314 * If enough pages are flushed, we may succeed on
315 * next (final) pass, if not the caller, contigmalloc(),
316 * will fail in the index < 0 case.
318 if (pass > 0 && (mflags & M_WAITOK)) {
319 vm_contig_pg_clean(PQ_ACTIVE,
320 vmstats.v_active_count);
324 * We're already too high in the address space
325 * to succeed, reset to 0 for the next iteration.
328 continue; /* next pass */
333 * Check successive pages for contiguous and free.
335 * (still in critical section)
337 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) {
339 pqtype = m->queue - m->pc;
340 if ((VM_PAGE_TO_PHYS(&m[0]) !=
341 (VM_PAGE_TO_PHYS(&m[-1]) + PAGE_SIZE)) ||
342 ((pqtype != PQ_FREE) && (pqtype != PQ_CACHE)) ||
343 m->busy || m->wire_count ||
344 m->hold_count || (m->flags & PG_BUSY)
352 * Try to allocate the pages.
354 * (still in critical section)
356 for (i = start; i < (start + size / PAGE_SIZE); i++) {
359 if (vm_page_busy_try(m, TRUE)) {
360 vm_contig_pg_free(start,
361 (i - start) * PAGE_SIZE);
365 pqtype = m->queue - m->pc;
366 if (pqtype == PQ_CACHE) {
369 continue; /* retry the page */
371 if (pqtype != PQ_FREE) {
373 vm_contig_pg_free(start,
374 (i - start) * PAGE_SIZE);
378 KKASSERT(m->object == NULL);
379 vm_page_unqueue_nowakeup(m);
380 m->valid = VM_PAGE_BITS_ALL;
381 if (m->flags & PG_ZERO)
382 vm_page_zero_count--;
383 KASSERT(m->dirty == 0,
384 ("vm_contig_pg_alloc: page %p was dirty", m));
385 KKASSERT(m->wire_count == 0);
386 KKASSERT(m->busy == 0);
389 * Clear all flags except PG_BUSY, PG_ZERO, and
390 * PG_WANTED, then unbusy the now allocated page.
392 vm_page_flag_clear(m, ~(PG_BUSY|PG_ZERO|PG_WANTED));
397 * Our job is done, return the index page of vm_page_array.
399 return (start); /* aka &pga[start] */
411 * Remove pages previously allocated by vm_contig_pg_alloc, and
412 * assume all references to the pages have been removed, and that
413 * it is OK to add them back to the free list.
415 * Caller must ensure no races on the page range in question.
416 * No other requirements.
419 vm_contig_pg_free(int start, u_long size)
421 vm_page_t pga = vm_page_array;
425 size = round_page(size);
427 panic("vm_contig_pg_free: size must not be 0");
429 for (i = start; i < (start + size / PAGE_SIZE); i++) {
431 vm_page_busy_wait(m, FALSE, "cpgfr");
439 * Map previously allocated (vm_contig_pg_alloc) range of pages from
440 * vm_page_array[] into the KVA. Once mapped, the pages are part of
441 * the Kernel, and are to free'ed with kmem_free(&kernel_map, addr, size).
446 vm_contig_pg_kmap(int start, u_long size, vm_map_t map, int flags)
448 vm_offset_t addr, tmp_addr;
449 vm_page_t pga = vm_page_array;
452 size = round_page(size);
454 panic("vm_contig_pg_kmap: size must not be 0");
457 * We've found a contiguous chunk that meets our requirements.
458 * Allocate KVM, and assign phys pages and return a kernel VM
461 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
463 if (vm_map_findspace(map, vm_map_min(map), size, PAGE_SIZE, 0, &addr) !=
466 * XXX We almost never run out of kernel virtual
467 * space, so we don't make the allocated memory
471 vm_map_entry_release(count);
476 * kernel_object maps 1:1 to kernel_map.
478 vm_object_hold(&kernel_object);
479 vm_object_reference_locked(&kernel_object);
480 vm_map_insert(map, &count,
481 &kernel_object, addr,
484 VM_PROT_ALL, VM_PROT_ALL,
487 vm_map_entry_release(count);
490 for (i = start; i < (start + size / PAGE_SIZE); i++) {
491 vm_page_t m = &pga[i];
492 if (vm_page_insert(m, &kernel_object, OFF_TO_IDX(tmp_addr)) ==
494 panic("vm_contig_pg_kmap: page already exists @%p",
495 (void *)(intptr_t)tmp_addr);
497 if ((flags & M_ZERO) && !(m->flags & PG_ZERO))
498 pmap_zero_page(VM_PAGE_TO_PHYS(m));
500 tmp_addr += PAGE_SIZE;
502 vm_map_wire(map, addr, addr + size, 0);
504 vm_object_drop(&kernel_object);
514 unsigned long size, /* should be size_t here and for malloc() */
515 struct malloc_type *type,
519 unsigned long alignment,
520 unsigned long boundary)
522 return contigmalloc_map(size, type, flags, low, high, alignment,
523 boundary, &kernel_map);
531 unsigned long size, /* should be size_t here and for malloc() */
532 struct malloc_type *type,
536 unsigned long alignment,
537 unsigned long boundary,
543 index = vm_contig_pg_alloc(size, low, high, alignment, boundary, flags);
545 kprintf("contigmalloc_map: failed size %lu low=%llx "
546 "high=%llx align=%lu boundary=%lu flags=%08x\n",
547 size, (long long)low, (long long)high,
548 alignment, boundary, flags);
552 rv = (void *)vm_contig_pg_kmap(index, size, map, flags);
554 vm_contig_pg_free(index, size);
563 contigfree(void *addr, unsigned long size, struct malloc_type *type)
565 kmem_free(&kernel_map, (vm_offset_t)addr, size);
572 vm_page_alloc_contig(
576 vm_offset_t alignment)
578 return ((vm_offset_t)contigmalloc_map(size, M_DEVBUF, M_NOWAIT, low,
579 high, alignment, 0ul, &kernel_map));