4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_pager.c 8.6 (Berkeley) 1/12/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
62 * $FreeBSD: src/sys/vm/vm_pager.c,v 1.54.2.2 2001/11/18 07:11:00 dillon Exp $
66 * Paging space routine stubs. Emulates a matchmaker-like interface
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/kernel.h>
73 #include <sys/vnode.h>
75 #include <sys/ucred.h>
76 #include <sys/dsched.h>
78 #include <sys/sysctl.h>
81 #include <vm/vm_param.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_pager.h>
86 #include <vm/vm_extern.h>
89 #include <vm/vm_page2.h>
91 extern struct pagerops defaultpagerops;
92 extern struct pagerops swappagerops;
93 extern struct pagerops vnodepagerops;
94 extern struct pagerops devicepagerops;
95 extern struct pagerops physpagerops;
97 static int dead_pager_getpage (vm_object_t, vm_page_t *, int);
98 static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *);
99 static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t);
100 static void dead_pager_dealloc (vm_object_t);
106 dead_pager_getpage(vm_object_t obj, vm_page_t *mpp, int seqaccess)
108 return VM_PAGER_FAIL;
115 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
120 for (i = 0; i < count; i++) {
121 rtvals[i] = VM_PAGER_AGAIN;
129 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex)
138 dead_pager_dealloc(vm_object_t object)
140 KKASSERT(object->swblock_count == 0);
144 static struct pagerops deadpagerops = {
151 struct pagerops *pagertab[] = {
152 &defaultpagerops, /* OBJT_DEFAULT */
153 &swappagerops, /* OBJT_SWAP */
154 &vnodepagerops, /* OBJT_VNODE */
155 &devicepagerops, /* OBJT_DEVICE */
156 &devicepagerops, /* OBJT_MGTDEVICE */
157 &physpagerops, /* OBJT_PHYS */
158 &deadpagerops /* OBJT_DEAD */
161 int npagers = NELEM(pagertab);
164 * Kernel address space for mapping pages.
165 * Used by pagers where KVAs are needed for IO.
167 * XXX needs to be large enough to support the number of pending async
168 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
169 * (MAXPHYS == 64k) if you want to get the most efficiency.
171 #define PAGER_MAP_SIZE (8 * 1024 * 1024)
174 #define BSWHMASK (BSWHSIZE - 1)
176 TAILQ_HEAD(swqueue, buf);
178 int pager_map_size = PAGER_MAP_SIZE;
179 struct vm_map pager_map;
181 static vm_offset_t swapbkva_mem; /* swap buffers kva */
182 static vm_offset_t swapbkva_kva; /* swap buffers kva */
183 static struct swqueue bswlist_mem[BSWHSIZE]; /* with preallocated memory */
184 static struct swqueue bswlist_kva[BSWHSIZE]; /* with kva */
185 static struct swqueue bswlist_raw[BSWHSIZE]; /* without kva */
186 static struct spinlock bswspin_mem[BSWHSIZE];
187 static struct spinlock bswspin_kva[BSWHSIZE];
188 static struct spinlock bswspin_raw[BSWHSIZE];
189 static int pbuf_raw_count;
190 static int pbuf_kva_count;
191 static int pbuf_mem_count;
193 SYSCTL_INT(_vm, OID_AUTO, pbuf_raw_count, CTLFLAG_RD, &pbuf_raw_count, 0,
194 "Kernel pbuf raw reservations");
195 SYSCTL_INT(_vm, OID_AUTO, pbuf_kva_count, CTLFLAG_RD, &pbuf_kva_count, 0,
196 "Kernel pbuf kva reservations");
197 SYSCTL_INT(_vm, OID_AUTO, pbuf_mem_count, CTLFLAG_RD, &pbuf_mem_count, 0,
198 "Kernel pbuf mem reservations");
201 * Initialize the swap buffer list.
203 * Called from the low level boot code only.
206 vm_pager_init(void *arg __unused)
210 for (i = 0; i < BSWHSIZE; ++i) {
211 TAILQ_INIT(&bswlist_mem[i]);
212 TAILQ_INIT(&bswlist_kva[i]);
213 TAILQ_INIT(&bswlist_raw[i]);
214 spin_init(&bswspin_mem[i], "bswmem");
215 spin_init(&bswspin_kva[i], "bswkva");
216 spin_init(&bswspin_raw[i], "bswraw");
219 SYSINIT(vm_mem, SI_BOOT1_VM, SI_ORDER_SECOND, vm_pager_init, NULL);
222 * Called from the low level boot code only.
226 vm_pager_bufferinit(void *dummy __unused)
232 * Reserve KVM space for pbuf data.
234 swapbkva_mem = kmem_alloc_pageable(&pager_map, nswbuf_mem * MAXPHYS,
237 panic("Not enough pager_map VM space for physical buffers");
238 swapbkva_kva = kmem_alloc_pageable(&pager_map, nswbuf_kva * MAXPHYS,
241 panic("Not enough pager_map VM space for physical buffers");
244 * Initial pbuf setup.
246 * mem - These pbufs have permanently allocated memory
247 * kva - These pbufs have unallocated kva reservations
248 * raw - These pbufs have no kva reservations
252 * Buffers with pre-allocated kernel memory can be convenient for
253 * copyin/copyout because no SMP page invalidation or other pmap
254 * operations are needed.
257 for (i = 0; i < nswbuf_mem; ++i, ++bp) {
262 bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva_mem;
263 bp->b_kvasize = MAXPHYS;
264 bp->b_swindex = i & BSWHMASK;
265 bp->b_cpumask = smp_active_mask;
268 TAILQ_INSERT_HEAD(&bswlist_mem[i & BSWHMASK], bp, b_freelist);
269 atomic_add_int(&pbuf_mem_count, 1);
270 bp->b_data = bp->b_kvabase;
271 bp->b_bcount = MAXPHYS;
272 bp->b_xio.xio_pages = bp->b_xio.xio_internal_pages;
274 pg = (vm_offset_t)bp->b_kvabase >> PAGE_SHIFT;
275 vm_object_hold(&kernel_object);
276 for (j = 0; j < MAXPHYS / PAGE_SIZE; ++j) {
277 m = vm_page_alloc(&kernel_object, pg, VM_ALLOC_NORMAL |
280 bp->b_xio.xio_internal_pages[j] = m;
282 /* early boot, no other cpus running yet */
283 pmap_kenter_noinval(pg * PAGE_SIZE, VM_PAGE_TO_PHYS(m));
284 cpu_invlpg((void *)(pg * PAGE_SIZE));
288 vm_object_drop(&kernel_object);
289 bp->b_xio.xio_npages = j;
293 * Buffers with pre-assigned KVA bases. The KVA has no memory pages
294 * assigned to it. Saves the caller from having to reserve KVA for
298 for (i = 0; i < nswbuf_kva; ++i, ++bp) {
299 bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva_kva;
300 bp->b_kvasize = MAXPHYS;
301 bp->b_swindex = i & BSWHMASK;
304 TAILQ_INSERT_HEAD(&bswlist_kva[i & BSWHMASK], bp, b_freelist);
305 atomic_add_int(&pbuf_kva_count, 1);
309 * RAW buffers with no KVA mappings.
311 * NOTE: We use KM_NOTLBSYNC here to reduce unnecessary IPIs
312 * during startup, which can really slow down emulated
315 nswbuf_raw = nbuf * 2;
316 swbuf_raw = (void *)kmem_alloc3(&kernel_map,
317 round_page(nswbuf_raw * sizeof(struct buf)),
322 for (i = 0; i < nswbuf_raw; ++i, ++bp) {
323 bp->b_swindex = i & BSWHMASK;
326 TAILQ_INSERT_HEAD(&bswlist_raw[i & BSWHMASK], bp, b_freelist);
327 atomic_add_int(&pbuf_raw_count, 1);
331 SYSINIT(do_vmpg, SI_BOOT2_MACHDEP, SI_ORDER_FIRST, vm_pager_bufferinit, NULL);
337 vm_pager_deallocate(vm_object_t object)
339 (*pagertab[object->type]->pgo_dealloc) (object);
343 * vm_pager_get_pages() - inline, see vm/vm_pager.h
344 * vm_pager_put_pages() - inline, see vm/vm_pager.h
345 * vm_pager_has_page() - inline, see vm/vm_pager.h
346 * vm_pager_page_inserted() - inline, see vm/vm_pager.h
347 * vm_pager_page_removed() - inline, see vm/vm_pager.h
351 * Search the specified pager object list for an object with the
352 * specified handle. If an object with the specified handle is found,
353 * increase its reference count and return it. Otherwise, return NULL.
355 * The pager object list must be locked.
358 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
362 TAILQ_FOREACH(object, pg_list, pager_object_entry) {
363 if (object->handle == handle) {
364 VM_OBJECT_LOCK(object);
365 if ((object->flags & OBJ_DEAD) == 0) {
366 vm_object_reference_locked(object);
367 VM_OBJECT_UNLOCK(object);
370 VM_OBJECT_UNLOCK(object);
377 * Initialize a physical buffer.
382 initpbuf(struct buf *bp)
384 bp->b_qindex = 0; /* BQUEUE_NONE */
385 bp->b_data = bp->b_kvabase; /* NULL if pbuf sans kva */
386 bp->b_flags = B_PAGING;
387 bp->b_cmd = BUF_CMD_DONE;
390 bp->b_bufsize = MAXPHYS;
392 xio_init(&bp->b_xio);
393 BUF_LOCK(bp, LK_EXCLUSIVE);
397 * Allocate a physical buffer
399 * If (pfreecnt != NULL) then *pfreecnt will be decremented on return and
400 * the function will block while it is <= 0.
402 * Physical buffers can be with or without KVA space reserved. There
403 * are severe limitations on the ones with KVA reserved, and fewer
404 * limitations on the ones without. getpbuf() gets one without,
405 * getpbuf_kva() gets one with.
410 getpbuf(int *pfreecnt)
417 while (pfreecnt && *pfreecnt <= 0) {
418 tsleep_interlock(pfreecnt, 0);
419 if ((int)atomic_fetchadd_int(pfreecnt, 0) <= 0)
420 tsleep(pfreecnt, PINTERLOCKED, "wswbuf0", 0);
422 if (pbuf_raw_count <= 0) {
423 tsleep_interlock(&pbuf_raw_count, 0);
424 if ((int)atomic_fetchadd_int(&pbuf_raw_count, 0) <= 0)
425 tsleep(&pbuf_raw_count, PINTERLOCKED,
429 iter = mycpuid & BSWHMASK;
430 for (loops = BSWHSIZE; loops; --loops) {
431 if (TAILQ_FIRST(&bswlist_raw[iter]) == NULL) {
432 iter = (iter + 1) & BSWHMASK;
435 spin_lock(&bswspin_raw[iter]);
436 if ((bp = TAILQ_FIRST(&bswlist_raw[iter])) == NULL) {
437 spin_unlock(&bswspin_raw[iter]);
438 iter = (iter + 1) & BSWHMASK;
441 TAILQ_REMOVE(&bswlist_raw[iter], bp, b_freelist);
442 atomic_add_int(&pbuf_raw_count, -1);
444 atomic_add_int(pfreecnt, -1);
445 spin_unlock(&bswspin_raw[iter]);
455 getpbuf_kva(int *pfreecnt)
462 while (pfreecnt && *pfreecnt <= 0) {
463 tsleep_interlock(pfreecnt, 0);
464 if ((int)atomic_fetchadd_int(pfreecnt, 0) <= 0)
465 tsleep(pfreecnt, PINTERLOCKED, "wswbuf2", 0);
467 if (pbuf_kva_count <= 0) {
468 tsleep_interlock(&pbuf_kva_count, 0);
469 if ((int)atomic_fetchadd_int(&pbuf_kva_count, 0) <= 0)
470 tsleep(&pbuf_kva_count, PINTERLOCKED,
474 iter = mycpuid & BSWHMASK;
475 for (loops = BSWHSIZE; loops; --loops) {
476 if (TAILQ_FIRST(&bswlist_kva[iter]) == NULL) {
477 iter = (iter + 1) & BSWHMASK;
480 spin_lock(&bswspin_kva[iter]);
481 if ((bp = TAILQ_FIRST(&bswlist_kva[iter])) == NULL) {
482 spin_unlock(&bswspin_kva[iter]);
483 iter = (iter + 1) & BSWHMASK;
486 TAILQ_REMOVE(&bswlist_kva[iter], bp, b_freelist);
487 atomic_add_int(&pbuf_kva_count, -1);
489 atomic_add_int(pfreecnt, -1);
490 spin_unlock(&bswspin_kva[iter]);
500 * Allocate a pbuf with kernel memory already preallocated. Caller must
501 * not change the mapping.
504 getpbuf_mem(int *pfreecnt)
511 while (pfreecnt && *pfreecnt <= 0) {
512 tsleep_interlock(pfreecnt, 0);
513 if ((int)atomic_fetchadd_int(pfreecnt, 0) <= 0)
514 tsleep(pfreecnt, PINTERLOCKED, "wswbuf4", 0);
516 if (pbuf_mem_count <= 0) {
517 tsleep_interlock(&pbuf_mem_count, 0);
518 if ((int)atomic_fetchadd_int(&pbuf_mem_count, 0) <= 0)
519 tsleep(&pbuf_mem_count, PINTERLOCKED,
523 iter = mycpuid & BSWHMASK;
524 for (loops = BSWHSIZE; loops; --loops) {
525 if (TAILQ_FIRST(&bswlist_mem[iter]) == NULL) {
526 iter = (iter + 1) & BSWHMASK;
529 spin_lock(&bswspin_mem[iter]);
530 if ((bp = TAILQ_FIRST(&bswlist_mem[iter])) == NULL) {
531 spin_unlock(&bswspin_mem[iter]);
532 iter = (iter + 1) & BSWHMASK;
535 TAILQ_REMOVE(&bswlist_mem[iter], bp, b_freelist);
536 atomic_add_int(&pbuf_mem_count, -1);
538 atomic_add_int(pfreecnt, -1);
539 spin_unlock(&bswspin_mem[iter]);
549 * Allocate a physical buffer, if one is available.
551 * Note that there is no NULL hack here - all subsystems using this
552 * call are required to use a non-NULL pfreecnt.
557 trypbuf(int *pfreecnt)
560 int iter = mycpuid & BSWHMASK;
563 for (loops = BSWHSIZE; loops; --loops) {
564 if (*pfreecnt <= 0 || TAILQ_FIRST(&bswlist_raw[iter]) == NULL) {
565 iter = (iter + 1) & BSWHMASK;
568 spin_lock(&bswspin_raw[iter]);
569 if (*pfreecnt <= 0 ||
570 (bp = TAILQ_FIRST(&bswlist_raw[iter])) == NULL) {
571 spin_unlock(&bswspin_raw[iter]);
572 iter = (iter + 1) & BSWHMASK;
575 TAILQ_REMOVE(&bswlist_raw[iter], bp, b_freelist);
576 atomic_add_int(&pbuf_raw_count, -1);
577 atomic_add_int(pfreecnt, -1);
579 spin_unlock(&bswspin_raw[iter]);
589 trypbuf_kva(int *pfreecnt)
592 int iter = mycpuid & BSWHMASK;
595 for (loops = BSWHSIZE; loops; --loops) {
596 if (*pfreecnt <= 0 || TAILQ_FIRST(&bswlist_kva[iter]) == NULL) {
597 iter = (iter + 1) & BSWHMASK;
600 spin_lock(&bswspin_kva[iter]);
601 if (*pfreecnt <= 0 ||
602 (bp = TAILQ_FIRST(&bswlist_kva[iter])) == NULL) {
603 spin_unlock(&bswspin_kva[iter]);
604 iter = (iter + 1) & BSWHMASK;
607 TAILQ_REMOVE(&bswlist_kva[iter], bp, b_freelist);
608 atomic_add_int(&pbuf_kva_count, -1);
609 atomic_add_int(pfreecnt, -1);
611 spin_unlock(&bswspin_kva[iter]);
621 * Release a physical buffer
623 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed
624 * relatively soon when the rest of the subsystems get smart about it. XXX
629 relpbuf(struct buf *bp, int *pfreecnt)
633 int iter = bp->b_swindex;
635 KKASSERT(bp->b_flags & B_PAGING);
640 if (bp >= swbuf_mem && bp < &swbuf_mem[nswbuf_mem]) {
641 KKASSERT(bp->b_kvabase);
642 spin_lock(&bswspin_mem[iter]);
643 TAILQ_INSERT_HEAD(&bswlist_mem[iter], bp, b_freelist);
644 if (atomic_fetchadd_int(&pbuf_mem_count, 1) == nswbuf_mem / 4)
647 if (atomic_fetchadd_int(pfreecnt, 1) == 1)
650 spin_unlock(&bswspin_mem[iter]);
652 wakeup(&pbuf_mem_count);
653 } else if (bp >= swbuf_kva && bp < &swbuf_kva[nswbuf_kva]) {
654 KKASSERT(bp->b_kvabase);
655 CPUMASK_ASSZERO(bp->b_cpumask);
656 spin_lock(&bswspin_kva[iter]);
657 TAILQ_INSERT_HEAD(&bswlist_kva[iter], bp, b_freelist);
658 if (atomic_fetchadd_int(&pbuf_kva_count, 1) == nswbuf_kva / 4)
661 if (atomic_fetchadd_int(pfreecnt, 1) == 1)
664 spin_unlock(&bswspin_kva[iter]);
666 wakeup(&pbuf_kva_count);
668 KKASSERT(bp->b_kvabase == NULL);
669 KKASSERT(bp >= swbuf_raw && bp < &swbuf_raw[nswbuf_raw]);
670 CPUMASK_ASSZERO(bp->b_cpumask);
671 spin_lock(&bswspin_raw[iter]);
672 TAILQ_INSERT_HEAD(&bswlist_raw[iter], bp, b_freelist);
673 if (atomic_fetchadd_int(&pbuf_raw_count, 1) == nswbuf_raw / 4)
676 if (atomic_fetchadd_int(pfreecnt, 1) == 1)
679 spin_unlock(&bswspin_raw[iter]);
681 wakeup(&pbuf_raw_count);
688 pbuf_adjcount(int *pfreecnt, int n)
691 atomic_add_int(pfreecnt, n);