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 1.15 2006/01/09 15:14:49 corecode 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 <vm/vm_page2.h>
124 * vm_contig_pg_clean:
126 * Do a thorough cleanup of the specified 'queue', which can be either
127 * PQ_ACTIVE or PQ_INACTIVE by doing a walkthrough. If the page is not
128 * marked dirty, it is shoved into the page cache, provided no one has
129 * currently aqcuired it, otherwise localized action per object type
130 * is taken for cleanup:
132 * In the OBJT_VNODE case, the whole page range is cleaned up
133 * using the vm_object_page_clean() routine, by specyfing a
134 * start and end of '0'.
136 * Otherwise if the object is of any other type, the generic
137 * pageout (daemon) flush routine is invoked.
139 * We must be in a critical section.
142 vm_contig_pg_clean(int queue)
145 vm_page_t m, m_tmp, next;
147 for (m = TAILQ_FIRST(&vm_page_queues[queue].pl); m != NULL; m = next) {
148 KASSERT(m->queue == queue,
149 ("vm_contig_clean: page %p's queue is not %d",
151 next = TAILQ_NEXT(m, pageq);
153 if (vm_page_sleep_busy(m, TRUE, "vpctw0"))
156 vm_page_test_dirty(m);
159 if (object->type == OBJT_VNODE) {
160 vn_lock(object->handle,
161 LK_EXCLUSIVE | LK_RETRY, curthread);
162 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
163 VOP_UNLOCK(((struct vnode *)object->handle),
166 } else if (object->type == OBJT_SWAP ||
167 object->type == OBJT_DEFAULT) {
169 vm_pageout_flush(&m_tmp, 1, 0);
173 if ((m->dirty == 0) && (m->busy == 0) && (m->hold_count == 0))
180 * vm_contig_pg_flush:
182 * Attempt to flush (count) pages from the given page queue. This may or
183 * may not succeed. Take up to <count> passes and delay 1/20 of a second
187 vm_contig_pg_flush(int queue, int count)
190 if (!vm_contig_pg_clean(queue))
196 * vm_contig_pg_alloc:
198 * Allocate contiguous pages from the VM. This function does not
199 * map the allocated pages into the kernel map, otherwise it is
200 * impossible to make large allocations (i.e. >2G).
202 * Malloc()'s data structures have been used for collection of
203 * statistics and for allocations of less than a page.
206 vm_contig_pg_alloc(unsigned long size, vm_paddr_t low, vm_paddr_t high,
207 unsigned long alignment, unsigned long boundary, int mflags)
211 vm_page_t pga = vm_page_array;
215 size = round_page(size);
217 panic("vm_contig_pg_alloc: size must not be 0");
218 if ((alignment & (alignment - 1)) != 0)
219 panic("vm_contig_pg_alloc: alignment must be a power of 2");
220 if ((boundary & (boundary - 1)) != 0)
221 panic("vm_contig_pg_alloc: boundary must be a power of 2");
227 * Three passes (0, 1, 2). Each pass scans the VM page list for
228 * free or cached pages. After each pass if the entire scan failed
229 * we attempt to flush inactive pages and reset the start index back
230 * to 0. For passes 1 and 2 we also attempt to flush active pages.
232 for (pass = 0; pass < 3; pass++) {
234 * Find first page in array that is free, within range,
235 * aligned, and such that the boundary won't be crossed.
238 for (i = start; i < vmstats.v_page_count; i++) {
240 phys = VM_PAGE_TO_PHYS(m);
241 pqtype = m->queue - m->pc;
242 if (((pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) &&
243 (phys >= low) && (phys < high) &&
244 ((phys & (alignment - 1)) == 0) &&
245 (((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0) &&
246 m->busy == 0 && m->wire_count == 0 &&
247 m->hold_count == 0 && (m->flags & PG_BUSY) == 0
255 * If we cannot find the page in the given range, or we have
256 * crossed the boundary, call the vm_contig_pg_clean() function
257 * for flushing out the queues, and returning it back to
260 if ((i == vmstats.v_page_count) ||
261 ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) {
264 * Best effort flush of all inactive pages.
265 * This is quite quick, for now stall all
266 * callers, even if they've specified M_NOWAIT.
268 vm_contig_pg_flush(PQ_INACTIVE,
269 vmstats.v_inactive_count);
271 crit_exit(); /* give interrupts a chance */
275 * Best effort flush of active pages.
277 * This is very, very slow.
278 * Only do this if the caller has agreed to M_WAITOK.
280 * If enough pages are flushed, we may succeed on
281 * next (final) pass, if not the caller, contigmalloc(),
282 * will fail in the index < 0 case.
284 if (pass > 0 && (mflags & M_WAITOK)) {
285 vm_contig_pg_flush (PQ_ACTIVE,
286 vmstats.v_active_count);
290 * We're already too high in the address space
291 * to succeed, reset to 0 for the next iteration.
294 crit_exit(); /* give interrupts a chance */
296 continue; /* next pass */
301 * Check successive pages for contiguous and free.
303 * (still in critical section)
305 for (i = start + 1; i < (start + size / PAGE_SIZE); i++) {
307 pqtype = m->queue - m->pc;
308 if ((VM_PAGE_TO_PHYS(&m[0]) !=
309 (VM_PAGE_TO_PHYS(&m[-1]) + PAGE_SIZE)) ||
310 ((pqtype != PQ_FREE) && (pqtype != PQ_CACHE)) ||
311 m->busy || m->wire_count ||
312 m->hold_count || (m->flags & PG_BUSY)
320 * (still in critical section)
322 for (i = start; i < (start + size / PAGE_SIZE); i++) {
324 pqtype = m->queue - m->pc;
325 if (pqtype == PQ_CACHE) {
329 KKASSERT(m->object == NULL);
330 vm_page_unqueue_nowakeup(m);
331 m->valid = VM_PAGE_BITS_ALL;
332 if (m->flags & PG_ZERO)
333 vm_page_zero_count--;
334 /* Don't clear the PG_ZERO flag, we'll need it later. */
336 KASSERT(m->dirty == 0,
337 ("vm_contig_pg_alloc: page %p was dirty", m));
343 * Our job is done, return the index page of vm_page_array.
346 return (start); /* aka &pga[start] */
359 * Remove pages previously allocated by vm_contig_pg_alloc, and
360 * assume all references to the pages have been removed, and that
361 * it is OK to add them back to the free list.
364 vm_contig_pg_free(int start, u_long size)
366 vm_page_t pga = vm_page_array;
370 size = round_page(size);
372 panic("vm_contig_pg_free: size must not be 0");
374 for (i = start; i < (start + size / PAGE_SIZE); i++) {
384 * Map previously allocated (vm_contig_pg_alloc) range of pages from
385 * vm_page_array[] into the KVA. Once mapped, the pages are part of
386 * the Kernel, and are to free'ed with kmem_free(kernel_map, addr, size).
389 vm_contig_pg_kmap(int start, u_long size, vm_map_t map, int flags)
391 vm_offset_t addr, tmp_addr;
392 vm_page_t pga = vm_page_array;
395 size = round_page(size);
397 panic("vm_contig_pg_kmap: size must not be 0");
402 * We've found a contiguous chunk that meets our requirements.
403 * Allocate KVM, and assign phys pages and return a kernel VM
406 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
408 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr) !=
411 * XXX We almost never run out of kernel virtual
412 * space, so we don't make the allocated memory
416 vm_map_entry_release(count);
420 vm_object_reference(kernel_object);
421 vm_map_insert(map, &count,
422 kernel_object, addr - VM_MIN_KERNEL_ADDRESS,
423 addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
425 vm_map_entry_release(count);
428 for (i = start; i < (start + size / PAGE_SIZE); i++) {
429 vm_page_t m = &pga[i];
430 vm_page_insert(m, kernel_object,
431 OFF_TO_IDX(tmp_addr - VM_MIN_KERNEL_ADDRESS));
432 if ((flags & M_ZERO) && !(m->flags & PG_ZERO))
433 pmap_zero_page(VM_PAGE_TO_PHYS(m));
435 tmp_addr += PAGE_SIZE;
437 vm_map_wire(map, addr, addr + size, 0);
445 unsigned long size, /* should be size_t here and for malloc() */
446 struct malloc_type *type,
450 unsigned long alignment,
451 unsigned long boundary)
453 return contigmalloc_map(size, type, flags, low, high, alignment,
454 boundary, kernel_map);
459 unsigned long size, /* should be size_t here and for malloc() */
460 struct malloc_type *type,
464 unsigned long alignment,
465 unsigned long boundary,
471 index = vm_contig_pg_alloc(size, low, high, alignment, boundary, flags);
473 printf("contigmalloc_map: failed in index < 0 case!\n");
477 rv = (void *)vm_contig_pg_kmap(index, size, map, flags);
479 vm_contig_pg_free(index, size);
485 contigfree(void *addr, unsigned long size, struct malloc_type *type)
487 kmem_free(kernel_map, (vm_offset_t)addr, size);
491 vm_page_alloc_contig(
495 vm_offset_t alignment)
497 return ((vm_offset_t)contigmalloc_map(size, M_DEVBUF, M_NOWAIT, low,
498 high, alignment, 0ul, kernel_map));