kernel - simplify vm pager ops, add pre-faulting for zero-fill pages.
[dragonfly.git] / sys / vm / vm_pager.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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 the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * from: @(#)vm_pager.c 8.6 (Berkeley) 1/12/94
37 *
38 *
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
41 *
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43 *
44 * Permission to use, copy, modify and distribute this software and
45 * its documentation is hereby granted, provided that both the copyright
46 * notice and this permission notice appear in all copies of the
47 * software, derivative works or modified versions, and any portions
48 * thereof, and that both notices appear in supporting documentation.
49 *
50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53 *
54 * Carnegie Mellon requests users of this software to return to
55 *
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
60 *
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
63 *
64 * $FreeBSD: src/sys/vm/vm_pager.c,v 1.54.2.2 2001/11/18 07:11:00 dillon Exp $
408357d8 65 * $DragonFly: src/sys/vm/vm_pager.c,v 1.24 2007/11/06 03:50:01 dillon Exp $
984263bc
MD
66 */
67
68/*
69 * Paging space routine stubs. Emulates a matchmaker-like interface
70 * for builtin pagers.
71 */
72
73#include <sys/param.h>
74#include <sys/systm.h>
75#include <sys/kernel.h>
76#include <sys/vnode.h>
77#include <sys/buf.h>
78#include <sys/ucred.h>
79#include <sys/malloc.h>
80#include <sys/proc.h>
cdd46d2e 81#include <sys/thread2.h>
984263bc
MD
82
83#include <vm/vm.h>
84#include <vm/vm_param.h>
85#include <vm/vm_object.h>
86#include <vm/vm_page.h>
87#include <vm/vm_pager.h>
88#include <vm/vm_extern.h>
89
3020e3be
MD
90#include <sys/buf2.h>
91
984263bc
MD
92MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data");
93
94extern struct pagerops defaultpagerops;
95extern struct pagerops swappagerops;
96extern struct pagerops vnodepagerops;
97extern struct pagerops devicepagerops;
98extern struct pagerops physpagerops;
99
100int cluster_pbuf_freecnt = -1; /* unlimited to begin with */
101
1b9d3514 102static int dead_pager_getpage (vm_object_t, vm_page_t *, int);
57f7b636 103static vm_object_t dead_pager_alloc (void *, off_t, vm_prot_t, off_t);
1388df65 104static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *);
1b9d3514 105static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t);
1388df65 106static void dead_pager_dealloc (vm_object_t);
984263bc
MD
107
108static int
1b9d3514 109dead_pager_getpage(vm_object_t obj, vm_page_t *mpp, int seqaccess)
984263bc
MD
110{
111 return VM_PAGER_FAIL;
112}
113
114static vm_object_t
57f7b636 115dead_pager_alloc(void *handle, off_t size, vm_prot_t prot, off_t off)
984263bc
MD
116{
117 return NULL;
118}
119
120static void
57e43348
MD
121dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
122 int *rtvals)
984263bc
MD
123{
124 int i;
125
126 for (i = 0; i < count; i++) {
127 rtvals[i] = VM_PAGER_AGAIN;
128 }
129}
130
131static int
1b9d3514 132dead_pager_haspage(vm_object_t object, vm_pindex_t pindex)
984263bc 133{
984263bc
MD
134 return FALSE;
135}
136
137static void
57e43348 138dead_pager_dealloc(vm_object_t object)
984263bc
MD
139{
140 return;
141}
142
143static struct pagerops deadpagerops = {
144 NULL,
145 dead_pager_alloc,
146 dead_pager_dealloc,
1b9d3514 147 dead_pager_getpage,
984263bc
MD
148 dead_pager_putpages,
149 dead_pager_haspage,
150 NULL
151};
152
153struct pagerops *pagertab[] = {
154 &defaultpagerops, /* OBJT_DEFAULT */
155 &swappagerops, /* OBJT_SWAP */
156 &vnodepagerops, /* OBJT_VNODE */
157 &devicepagerops, /* OBJT_DEVICE */
158 &physpagerops, /* OBJT_PHYS */
159 &deadpagerops /* OBJT_DEAD */
160};
161
162int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
163
164/*
165 * Kernel address space for mapping pages.
166 * Used by pagers where KVAs are needed for IO.
167 *
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.
171 */
172#define PAGER_MAP_SIZE (8 * 1024 * 1024)
173
174int pager_map_size = PAGER_MAP_SIZE;
e4846942
MD
175struct vm_map pager_map;
176
984263bc
MD
177static int bswneeded;
178static vm_offset_t swapbkva; /* swap buffers kva */
bb3e0be6 179static TAILQ_HEAD(swqueue, buf) bswlist;
a50fe973 180static struct spinlock bswspin = SPINLOCK_INITIALIZER(&bswspin);
984263bc
MD
181
182void
57e43348 183vm_pager_init(void)
984263bc
MD
184{
185 struct pagerops **pgops;
186
bb3e0be6
HP
187 /*
188 * Initialize the swap buffer list.
189 */
190 TAILQ_INIT(&bswlist);
191
984263bc
MD
192 /*
193 * Initialize known pagers
194 */
195 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
196 if (pgops && ((*pgops)->pgo_init != NULL))
197 (*(*pgops)->pgo_init) ();
198}
199
200void
57e43348 201vm_pager_bufferinit(void)
984263bc
MD
202{
203 struct buf *bp;
204 int i;
205
984263bc 206 /*
312dcd01
MD
207 * Reserve KVM space for pbuf data.
208 */
e4846942 209 swapbkva = kmem_alloc_pageable(&pager_map, nswbuf * MAXPHYS);
312dcd01
MD
210 if (!swapbkva)
211 panic("Not enough pager_map VM space for physical buffers");
212
213 /*
214 * Initial pbuf setup.
984263bc 215 */
312dcd01
MD
216 bp = swbuf;
217 for (i = 0; i < nswbuf; ++i, ++bp) {
973c11b9 218 bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva;
312dcd01 219 bp->b_kvasize = MAXPHYS;
984263bc
MD
220 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
221 BUF_LOCKINIT(bp);
408357d8 222 buf_dep_init(bp);
984263bc
MD
223 }
224
312dcd01
MD
225 /*
226 * Allow the clustering code to use half of our pbufs.
227 */
984263bc 228 cluster_pbuf_freecnt = nswbuf / 2;
984263bc
MD
229}
230
231/*
232 * Allocate an instance of a pager of the given type.
233 * Size, protection and offset parameters are passed in for pagers that
234 * need to perform page-level validation (e.g. the device pager).
235 */
236vm_object_t
57f7b636
MD
237vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
238 vm_prot_t prot, off_t off)
984263bc
MD
239{
240 struct pagerops *ops;
241
242 ops = pagertab[type];
243 if (ops)
244 return ((*ops->pgo_alloc) (handle, size, prot, off));
245 return (NULL);
246}
247
248void
57e43348 249vm_pager_deallocate(vm_object_t object)
984263bc
MD
250{
251 (*pagertab[object->type]->pgo_dealloc) (object);
252}
253
254/*
255 * vm_pager_strategy:
256 *
257 * called with no specific spl
258 * Execute strategy routine directly to pager.
259 */
260
261void
81b5c339 262vm_pager_strategy(vm_object_t object, struct bio *bio)
984263bc 263{
81b5c339
MD
264 struct buf *bp;
265
984263bc 266 if (pagertab[object->type]->pgo_strategy) {
81b5c339 267 (*pagertab[object->type]->pgo_strategy)(object, bio);
984263bc 268 } else {
81b5c339 269 bp = bio->bio_buf;
984263bc
MD
270 bp->b_flags |= B_ERROR;
271 bp->b_error = ENXIO;
81b5c339 272 biodone(bio);
984263bc
MD
273 }
274}
275
276/*
277 * vm_pager_get_pages() - inline, see vm/vm_pager.h
278 * vm_pager_put_pages() - inline, see vm/vm_pager.h
279 * vm_pager_has_page() - inline, see vm/vm_pager.h
280 * vm_pager_page_inserted() - inline, see vm/vm_pager.h
281 * vm_pager_page_removed() - inline, see vm/vm_pager.h
282 */
283
284#if 0
285/*
286 * vm_pager_sync:
287 *
288 * Called by pageout daemon before going back to sleep.
289 * Gives pagers a chance to clean up any completed async pageing
290 * operations.
291 */
292void
57e43348 293vm_pager_sync(void)
984263bc
MD
294{
295 struct pagerops **pgops;
296
297 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
298 if (pgops && ((*pgops)->pgo_sync != NULL))
299 (*(*pgops)->pgo_sync) ();
300}
301
302#endif
303
984263bc 304vm_object_t
57e43348 305vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
984263bc 306{
5f910b2f 307 vm_object_t object;
984263bc
MD
308
309 for (object = TAILQ_FIRST(pg_list); object != NULL; object = TAILQ_NEXT(object,pager_object_list))
310 if (object->handle == handle)
311 return (object);
312 return (NULL);
313}
314
315/*
312dcd01 316 * Initialize a physical buffer.
984263bc 317 */
984263bc
MD
318static void
319initpbuf(struct buf *bp)
320{
b3098c79 321 bp->b_qindex = 0; /* BQUEUE_NONE */
312dcd01 322 bp->b_data = bp->b_kvabase;
a8f169e2 323 bp->b_flags = B_PAGING;
10f3fee5 324 bp->b_cmd = BUF_CMD_DONE;
984263bc 325 bp->b_error = 0;
9a71d53f
MD
326 bp->b_bcount = 0;
327 bp->b_bufsize = MAXPHYS;
81b5c339 328 initbufbio(bp);
54f51aeb 329 xio_init(&bp->b_xio);
984263bc
MD
330 BUF_LOCK(bp, LK_EXCLUSIVE);
331}
332
333/*
a50fe973 334 * Allocate a physical buffer
984263bc
MD
335 *
336 * There are a limited number (nswbuf) of physical buffers. We need
337 * to make sure that no single subsystem is able to hog all of them,
338 * so each subsystem implements a counter which is typically initialized
339 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and
340 * increments it on release, and blocks if the counter hits zero. A
341 * subsystem may initialize the counter to -1 to disable the feature,
342 * but it must still be sure to match up all uses of getpbuf() with
343 * relpbuf() using the same variable.
344 *
345 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed
346 * relatively soon when the rest of the subsystems get smart about it. XXX
a50fe973
MD
347 *
348 * MPSAFE
984263bc
MD
349 */
350struct buf *
57e43348 351getpbuf(int *pfreecnt)
984263bc 352{
984263bc
MD
353 struct buf *bp;
354
a50fe973 355 spin_lock_wr(&bswspin);
984263bc
MD
356
357 for (;;) {
358 if (pfreecnt) {
a50fe973 359 while (*pfreecnt == 0)
e590ee86 360 ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0);
984263bc
MD
361 }
362
363 /* get a bp from the swap buffer header pool */
364 if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
365 break;
984263bc 366 bswneeded = 1;
e590ee86 367 ssleep(&bswneeded, &bswspin, 0, "wswbuf1", 0);
984263bc
MD
368 /* loop in case someone else grabbed one */
369 }
370 TAILQ_REMOVE(&bswlist, bp, b_freelist);
371 if (pfreecnt)
372 --*pfreecnt;
a50fe973
MD
373
374 spin_unlock_wr(&bswspin);
984263bc
MD
375
376 initpbuf(bp);
377 return bp;
378}
379
380/*
a50fe973 381 * Allocate a physical buffer, if one is available.
984263bc
MD
382 *
383 * Note that there is no NULL hack here - all subsystems using this
384 * call understand how to use pfreecnt.
a50fe973
MD
385 *
386 * MPSAFE
984263bc
MD
387 */
388struct buf *
57e43348 389trypbuf(int *pfreecnt)
984263bc 390{
984263bc
MD
391 struct buf *bp;
392
a50fe973
MD
393 spin_lock_wr(&bswspin);
394
984263bc 395 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
a50fe973 396 spin_unlock_wr(&bswspin);
984263bc
MD
397 return NULL;
398 }
399 TAILQ_REMOVE(&bswlist, bp, b_freelist);
984263bc
MD
400 --*pfreecnt;
401
a50fe973 402 spin_unlock_wr(&bswspin);
984263bc
MD
403
404 initpbuf(bp);
405
406 return bp;
407}
408
409/*
a50fe973 410 * Release a physical buffer
984263bc
MD
411 *
412 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed
413 * relatively soon when the rest of the subsystems get smart about it. XXX
a50fe973
MD
414 *
415 * MPSAFE
984263bc
MD
416 */
417void
57e43348 418relpbuf(struct buf *bp, int *pfreecnt)
984263bc 419{
a50fe973
MD
420 int wake_bsw = 0;
421 int wake_freecnt = 0;
984263bc 422
a8f169e2 423 KKASSERT(bp->b_flags & B_PAGING);
984263bc 424
a50fe973 425 spin_lock_wr(&bswspin);
984263bc 426
a50fe973
MD
427 BUF_UNLOCK(bp);
428 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
984263bc
MD
429 if (bswneeded) {
430 bswneeded = 0;
a50fe973 431 wake_bsw = 1;
984263bc
MD
432 }
433 if (pfreecnt) {
434 if (++*pfreecnt == 1)
a50fe973 435 wake_freecnt = 1;
984263bc 436 }
a50fe973
MD
437
438 spin_unlock_wr(&bswspin);
439
440 if (wake_bsw)
441 wakeup(&bswneeded);
442 if (wake_freecnt)
443 wakeup(pfreecnt);
984263bc
MD
444}
445