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 static pgo_dealloc_t dead_pager_dealloc;
92 static pgo_getpage_t dead_pager_getpage;
93 static pgo_putpages_t dead_pager_putpages;
94 static pgo_haspage_t dead_pager_haspage;
96 static struct pagerops deadpagerops = {
97 .pgo_dealloc = dead_pager_dealloc,
98 .pgo_getpage = dead_pager_getpage,
99 .pgo_putpages = dead_pager_putpages,
100 .pgo_haspage = dead_pager_haspage
103 extern struct pagerops defaultpagerops;
104 extern struct pagerops swappagerops;
105 extern struct pagerops vnodepagerops;
106 extern struct pagerops devicepagerops;
107 extern struct pagerops physpagerops;
113 dead_pager_getpage(vm_object_t obj, vm_pindex_t pindex,
114 vm_page_t *mpp, int seqaccess)
116 return VM_PAGER_FAIL;
123 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
128 for (i = 0; i < count; i++) {
129 rtvals[i] = VM_PAGER_AGAIN;
137 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex)
146 dead_pager_dealloc(vm_object_t object)
148 KKASSERT(object->swblock_count == 0);
152 struct pagerops *pagertab[] = {
153 &defaultpagerops, /* OBJT_DEFAULT */
154 &swappagerops, /* OBJT_SWAP */
155 &vnodepagerops, /* OBJT_VNODE */
156 &devicepagerops, /* OBJT_DEVICE */
157 &devicepagerops, /* OBJT_MGTDEVICE */
158 &physpagerops, /* OBJT_PHYS */
159 &deadpagerops /* OBJT_DEAD */
162 int npagers = NELEM(pagertab);
165 * Kernel address space for mapping pages.
166 * Used by pagers where KVAs are needed for IO.
168 * XXX needs to be large enough to support the number of pending async
169 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
170 * (MAXPHYS == 64k) if you want to get the most efficiency.
172 #define PAGER_MAP_SIZE (8 * 1024 * 1024)
175 #define BSWHMASK (BSWHSIZE - 1)
177 TAILQ_HEAD(swqueue, buf);
179 int pager_map_size = PAGER_MAP_SIZE;
181 static struct vm_map pager_map_store;
182 struct vm_map *pager_map = &pager_map_store;
184 static vm_offset_t swapbkva_mem; /* swap buffers kva */
185 static vm_offset_t swapbkva_kva; /* swap buffers kva */
186 static struct swqueue bswlist_mem[BSWHSIZE]; /* with preallocated memory */
187 static struct swqueue bswlist_kva[BSWHSIZE]; /* with kva */
188 static struct swqueue bswlist_raw[BSWHSIZE]; /* without kva */
189 static struct spinlock bswspin_mem[BSWHSIZE];
190 static struct spinlock bswspin_kva[BSWHSIZE];
191 static struct spinlock bswspin_raw[BSWHSIZE];
192 static int pbuf_raw_count;
193 static int pbuf_kva_count;
194 static int pbuf_mem_count;
196 SYSCTL_INT(_vm, OID_AUTO, pbuf_raw_count, CTLFLAG_RD, &pbuf_raw_count, 0,
197 "Kernel pbuf raw reservations");
198 SYSCTL_INT(_vm, OID_AUTO, pbuf_kva_count, CTLFLAG_RD, &pbuf_kva_count, 0,
199 "Kernel pbuf kva reservations");
200 SYSCTL_INT(_vm, OID_AUTO, pbuf_mem_count, CTLFLAG_RD, &pbuf_mem_count, 0,
201 "Kernel pbuf mem reservations");
204 * Initialize the swap buffer list.
206 * Called from the low level boot code only.
209 vm_pager_init(void *arg __unused)
213 for (i = 0; i < BSWHSIZE; ++i) {
214 TAILQ_INIT(&bswlist_mem[i]);
215 TAILQ_INIT(&bswlist_kva[i]);
216 TAILQ_INIT(&bswlist_raw[i]);
217 spin_init(&bswspin_mem[i], "bswmem");
218 spin_init(&bswspin_kva[i], "bswkva");
219 spin_init(&bswspin_raw[i], "bswraw");
222 SYSINIT(vm_mem, SI_BOOT1_VM, SI_ORDER_SECOND, vm_pager_init, NULL);
225 * Called from the low level boot code only.
229 vm_pager_bufferinit(void *dummy __unused)
235 * Reserve KVM space for pbuf data.
237 swapbkva_mem = kmem_alloc_pageable(pager_map, nswbuf_mem * MAXPHYS,
240 panic("Not enough pager_map VM space for physical buffers");
241 swapbkva_kva = kmem_alloc_pageable(pager_map, nswbuf_kva * MAXPHYS,
244 panic("Not enough pager_map VM space for physical buffers");
247 * Initial pbuf setup.
249 * mem - These pbufs have permanently allocated memory
250 * kva - These pbufs have unallocated kva reservations
251 * raw - These pbufs have no kva reservations
255 * Buffers with pre-allocated kernel memory can be convenient for
256 * copyin/copyout because no SMP page invalidation or other pmap
257 * operations are needed.
260 for (i = 0; i < nswbuf_mem; ++i, ++bp) {
265 bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva_mem;
266 bp->b_kvasize = MAXPHYS;
267 bp->b_swindex = i & BSWHMASK;
268 bp->b_cpumask = smp_active_mask;
271 TAILQ_INSERT_HEAD(&bswlist_mem[i & BSWHMASK], bp, b_freelist);
272 atomic_add_int(&pbuf_mem_count, 1);
273 bp->b_data = bp->b_kvabase;
274 bp->b_bcount = MAXPHYS;
275 bp->b_xio.xio_pages = bp->b_xio.xio_internal_pages;
277 pg = (vm_offset_t)bp->b_kvabase >> PAGE_SHIFT;
278 vm_object_hold(kernel_object);
279 for (j = 0; j < MAXPHYS / PAGE_SIZE; ++j) {
280 m = vm_page_alloc(kernel_object, pg, VM_ALLOC_NORMAL |
283 bp->b_xio.xio_internal_pages[j] = m;
285 /* early boot, no other cpus running yet */
286 pmap_kenter_noinval(pg * PAGE_SIZE, VM_PAGE_TO_PHYS(m));
287 cpu_invlpg((void *)(pg * PAGE_SIZE));
291 vm_object_drop(kernel_object);
292 bp->b_xio.xio_npages = j;
296 * Buffers with pre-assigned KVA bases. The KVA has no memory pages
297 * assigned to it. Saves the caller from having to reserve KVA for
301 for (i = 0; i < nswbuf_kva; ++i, ++bp) {
302 bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva_kva;
303 bp->b_kvasize = MAXPHYS;
304 bp->b_swindex = i & BSWHMASK;
307 TAILQ_INSERT_HEAD(&bswlist_kva[i & BSWHMASK], bp, b_freelist);
308 atomic_add_int(&pbuf_kva_count, 1);
312 * RAW buffers with no KVA mappings.
314 * NOTE: We use KM_NOTLBSYNC here to reduce unnecessary IPIs
315 * during startup, which can really slow down emulated
318 nswbuf_raw = nbuf * 2;
319 swbuf_raw = (void *)kmem_alloc3(kernel_map,
320 round_page(nswbuf_raw * sizeof(struct buf)),
325 for (i = 0; i < nswbuf_raw; ++i, ++bp) {
326 bp->b_swindex = i & BSWHMASK;
329 TAILQ_INSERT_HEAD(&bswlist_raw[i & BSWHMASK], bp, b_freelist);
330 atomic_add_int(&pbuf_raw_count, 1);
334 SYSINIT(do_vmpg, SI_BOOT2_MACHDEP, SI_ORDER_FIRST, vm_pager_bufferinit, NULL);
340 vm_pager_deallocate(vm_object_t object)
342 (*pagertab[object->type]->pgo_dealloc) (object);
346 * vm_pager_get_pages() - inline, see vm/vm_pager.h
347 * vm_pager_put_pages() - inline, see vm/vm_pager.h
348 * vm_pager_has_page() - inline, see vm/vm_pager.h
349 * vm_pager_page_inserted() - inline, see vm/vm_pager.h
350 * vm_pager_page_removed() - inline, see vm/vm_pager.h
354 * Search the specified pager object list for an object with the
355 * specified handle. If an object with the specified handle is found,
356 * increase its reference count and return it. Otherwise, return NULL.
358 * The pager object list must be locked.
361 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
365 TAILQ_FOREACH(object, pg_list, pager_object_entry) {
366 if (object->handle == handle) {
367 VM_OBJECT_LOCK(object);
368 if ((object->flags & OBJ_DEAD) == 0) {
369 vm_object_reference_locked(object);
370 VM_OBJECT_UNLOCK(object);
373 VM_OBJECT_UNLOCK(object);
380 * Initialize a physical buffer.
385 initpbuf(struct buf *bp)
387 bp->b_qindex = 0; /* BQUEUE_NONE */
388 bp->b_data = bp->b_kvabase; /* NULL if pbuf sans kva */
389 bp->b_flags = B_PAGING;
390 bp->b_cmd = BUF_CMD_DONE;
393 bp->b_bufsize = MAXPHYS;
395 xio_init(&bp->b_xio);
396 BUF_LOCK(bp, LK_EXCLUSIVE);
400 * Allocate a physical buffer
402 * If (pfreecnt != NULL) then *pfreecnt will be decremented on return and
403 * the function will block while it is <= 0.
405 * Physical buffers can be with or without KVA space reserved. There
406 * are severe limitations on the ones with KVA reserved, and fewer
407 * limitations on the ones without. getpbuf() gets one without,
408 * getpbuf_kva() gets one with.
413 getpbuf(int *pfreecnt)
420 while (pfreecnt && *pfreecnt <= 0) {
421 tsleep_interlock(pfreecnt, 0);
422 if ((int)atomic_fetchadd_int(pfreecnt, 0) <= 0)
423 tsleep(pfreecnt, PINTERLOCKED, "wswbuf0", 0);
425 if (pbuf_raw_count <= 0) {
426 tsleep_interlock(&pbuf_raw_count, 0);
427 if ((int)atomic_fetchadd_int(&pbuf_raw_count, 0) <= 0)
428 tsleep(&pbuf_raw_count, PINTERLOCKED,
432 iter = mycpuid & BSWHMASK;
433 for (loops = BSWHSIZE; loops; --loops) {
434 if (TAILQ_FIRST(&bswlist_raw[iter]) == NULL) {
435 iter = (iter + 1) & BSWHMASK;
438 spin_lock(&bswspin_raw[iter]);
439 if ((bp = TAILQ_FIRST(&bswlist_raw[iter])) == NULL) {
440 spin_unlock(&bswspin_raw[iter]);
441 iter = (iter + 1) & BSWHMASK;
444 TAILQ_REMOVE(&bswlist_raw[iter], bp, b_freelist);
445 atomic_add_int(&pbuf_raw_count, -1);
447 atomic_add_int(pfreecnt, -1);
448 spin_unlock(&bswspin_raw[iter]);
458 getpbuf_kva(int *pfreecnt)
465 while (pfreecnt && *pfreecnt <= 0) {
466 tsleep_interlock(pfreecnt, 0);
467 if ((int)atomic_fetchadd_int(pfreecnt, 0) <= 0)
468 tsleep(pfreecnt, PINTERLOCKED, "wswbuf2", 0);
470 if (pbuf_kva_count <= 0) {
471 tsleep_interlock(&pbuf_kva_count, 0);
472 if ((int)atomic_fetchadd_int(&pbuf_kva_count, 0) <= 0)
473 tsleep(&pbuf_kva_count, PINTERLOCKED,
477 iter = mycpuid & BSWHMASK;
478 for (loops = BSWHSIZE; loops; --loops) {
479 if (TAILQ_FIRST(&bswlist_kva[iter]) == NULL) {
480 iter = (iter + 1) & BSWHMASK;
483 spin_lock(&bswspin_kva[iter]);
484 if ((bp = TAILQ_FIRST(&bswlist_kva[iter])) == NULL) {
485 spin_unlock(&bswspin_kva[iter]);
486 iter = (iter + 1) & BSWHMASK;
489 TAILQ_REMOVE(&bswlist_kva[iter], bp, b_freelist);
490 atomic_add_int(&pbuf_kva_count, -1);
492 atomic_add_int(pfreecnt, -1);
493 spin_unlock(&bswspin_kva[iter]);
503 * Allocate a pbuf with kernel memory already preallocated. Caller must
504 * not change the mapping.
507 getpbuf_mem(int *pfreecnt)
514 while (pfreecnt && *pfreecnt <= 0) {
515 tsleep_interlock(pfreecnt, 0);
516 if ((int)atomic_fetchadd_int(pfreecnt, 0) <= 0)
517 tsleep(pfreecnt, PINTERLOCKED, "wswbuf4", 0);
519 if (pbuf_mem_count <= 0) {
520 tsleep_interlock(&pbuf_mem_count, 0);
521 if ((int)atomic_fetchadd_int(&pbuf_mem_count, 0) <= 0)
522 tsleep(&pbuf_mem_count, PINTERLOCKED,
526 iter = mycpuid & BSWHMASK;
527 for (loops = BSWHSIZE; loops; --loops) {
528 if (TAILQ_FIRST(&bswlist_mem[iter]) == NULL) {
529 iter = (iter + 1) & BSWHMASK;
532 spin_lock(&bswspin_mem[iter]);
533 if ((bp = TAILQ_FIRST(&bswlist_mem[iter])) == NULL) {
534 spin_unlock(&bswspin_mem[iter]);
535 iter = (iter + 1) & BSWHMASK;
538 TAILQ_REMOVE(&bswlist_mem[iter], bp, b_freelist);
539 atomic_add_int(&pbuf_mem_count, -1);
541 atomic_add_int(pfreecnt, -1);
542 spin_unlock(&bswspin_mem[iter]);
552 * Allocate a physical buffer, if one is available.
554 * Note that there is no NULL hack here - all subsystems using this
555 * call are required to use a non-NULL pfreecnt.
560 trypbuf(int *pfreecnt)
563 int iter = mycpuid & BSWHMASK;
566 for (loops = BSWHSIZE; loops; --loops) {
567 if (*pfreecnt <= 0 || TAILQ_FIRST(&bswlist_raw[iter]) == NULL) {
568 iter = (iter + 1) & BSWHMASK;
571 spin_lock(&bswspin_raw[iter]);
572 if (*pfreecnt <= 0 ||
573 (bp = TAILQ_FIRST(&bswlist_raw[iter])) == NULL) {
574 spin_unlock(&bswspin_raw[iter]);
575 iter = (iter + 1) & BSWHMASK;
578 TAILQ_REMOVE(&bswlist_raw[iter], bp, b_freelist);
579 atomic_add_int(&pbuf_raw_count, -1);
580 atomic_add_int(pfreecnt, -1);
582 spin_unlock(&bswspin_raw[iter]);
592 trypbuf_kva(int *pfreecnt)
595 int iter = mycpuid & BSWHMASK;
598 for (loops = BSWHSIZE; loops; --loops) {
599 if (*pfreecnt <= 0 || TAILQ_FIRST(&bswlist_kva[iter]) == NULL) {
600 iter = (iter + 1) & BSWHMASK;
603 spin_lock(&bswspin_kva[iter]);
604 if (*pfreecnt <= 0 ||
605 (bp = TAILQ_FIRST(&bswlist_kva[iter])) == NULL) {
606 spin_unlock(&bswspin_kva[iter]);
607 iter = (iter + 1) & BSWHMASK;
610 TAILQ_REMOVE(&bswlist_kva[iter], bp, b_freelist);
611 atomic_add_int(&pbuf_kva_count, -1);
612 atomic_add_int(pfreecnt, -1);
614 spin_unlock(&bswspin_kva[iter]);
624 * Release a physical buffer
626 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed
627 * relatively soon when the rest of the subsystems get smart about it. XXX
632 relpbuf(struct buf *bp, int *pfreecnt)
636 int iter = bp->b_swindex;
638 KKASSERT(bp->b_flags & B_PAGING);
643 if (bp >= swbuf_mem && bp < &swbuf_mem[nswbuf_mem]) {
644 KKASSERT(bp->b_kvabase);
645 spin_lock(&bswspin_mem[iter]);
646 TAILQ_INSERT_HEAD(&bswlist_mem[iter], bp, b_freelist);
647 if (atomic_fetchadd_int(&pbuf_mem_count, 1) == nswbuf_mem / 4)
650 if (atomic_fetchadd_int(pfreecnt, 1) == 1)
653 spin_unlock(&bswspin_mem[iter]);
655 wakeup(&pbuf_mem_count);
656 } else if (bp >= swbuf_kva && bp < &swbuf_kva[nswbuf_kva]) {
657 KKASSERT(bp->b_kvabase);
658 CPUMASK_ASSZERO(bp->b_cpumask);
659 spin_lock(&bswspin_kva[iter]);
660 TAILQ_INSERT_HEAD(&bswlist_kva[iter], bp, b_freelist);
661 if (atomic_fetchadd_int(&pbuf_kva_count, 1) == nswbuf_kva / 4)
664 if (atomic_fetchadd_int(pfreecnt, 1) == 1)
667 spin_unlock(&bswspin_kva[iter]);
669 wakeup(&pbuf_kva_count);
671 KKASSERT(bp->b_kvabase == NULL);
672 KKASSERT(bp >= swbuf_raw && bp < &swbuf_raw[nswbuf_raw]);
673 CPUMASK_ASSZERO(bp->b_cpumask);
674 spin_lock(&bswspin_raw[iter]);
675 TAILQ_INSERT_HEAD(&bswlist_raw[iter], bp, b_freelist);
676 if (atomic_fetchadd_int(&pbuf_raw_count, 1) == nswbuf_raw / 4)
679 if (atomic_fetchadd_int(pfreecnt, 1) == 1)
682 spin_unlock(&bswspin_raw[iter]);
684 wakeup(&pbuf_raw_count);
691 pbuf_adjcount(int *pfreecnt, int n)
694 atomic_add_int(pfreecnt, n);