sys/vm/vm_pager.c

   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 $
  65  * $DragonFly: src/sys/vm/vm_pager.c,v 1.24 2007/11/06 03:50:01 dillon Exp $
  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>
  81 #include <sys/thread2.h>
  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
  90 #include <sys/buf2.h>
  91
  92 MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data");
  93
  94 extern struct pagerops defaultpagerops;
  95 extern struct pagerops swappagerops;
  96 extern struct pagerops vnodepagerops;
  97 extern struct pagerops devicepagerops;
  98 extern struct pagerops physpagerops;
  99
 100 int cluster_pbuf_freecnt = -1;  /* unlimited to begin with */
 101
 102 static int dead_pager_getpage (vm_object_t, vm_page_t *, int);
 103 static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *);
 104 static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t);
 105 static void dead_pager_dealloc (vm_object_t);
 106
 107 static int
 108 dead_pager_getpage(vm_object_t obj, vm_page_t *mpp, int seqaccess)
 109 {
 110         return VM_PAGER_FAIL;
 111 }
 112
 113 static void
 114 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
 115                     int *rtvals)
 116 {
 117         int i;
 118
 119         for (i = 0; i < count; i++) {
 120                 rtvals[i] = VM_PAGER_AGAIN;
 121         }
 122 }
 123
 124 static int
 125 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex)
 126 {
 127         return FALSE;
 128 }
 129
 130 static void
 131 dead_pager_dealloc(vm_object_t object)
 132 {
 133         KKASSERT(object->swblock_count == 0);
 134         return;
 135 }
 136
 137 static struct pagerops deadpagerops = {
 138         dead_pager_dealloc,
 139         dead_pager_getpage,
 140         dead_pager_putpages,
 141         dead_pager_haspage
 142 };
 143
 144 struct pagerops *pagertab[] = {
 145         &defaultpagerops,       /* OBJT_DEFAULT */
 146         &swappagerops,          /* OBJT_SWAP */
 147         &vnodepagerops,         /* OBJT_VNODE */
 148         &devicepagerops,        /* OBJT_DEVICE */
 149         &physpagerops,          /* OBJT_PHYS */
 150         &deadpagerops           /* OBJT_DEAD */
 151 };
 152
 153 int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
 154
 155 /*
 156  * Kernel address space for mapping pages.
 157  * Used by pagers where KVAs are needed for IO.
 158  *
 159  * XXX needs to be large enough to support the number of pending async
 160  * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
 161  * (MAXPHYS == 64k) if you want to get the most efficiency.
 162  */
 163 #define PAGER_MAP_SIZE  (8 * 1024 * 1024)
 164
 165 int pager_map_size = PAGER_MAP_SIZE;
 166 struct vm_map pager_map;
 167
 168 static int bswneeded;
 169 static vm_offset_t swapbkva;            /* swap buffers kva */
 170 static TAILQ_HEAD(swqueue, buf) bswlist;
 171 static struct spinlock bswspin = SPINLOCK_INITIALIZER(&bswspin);
 172
 173 static void
 174 vm_pager_init(void *arg __unused)
 175 {
 176         /*
 177          * Initialize the swap buffer list.
 178          */
 179         TAILQ_INIT(&bswlist);
 180 }
 181 SYSINIT(vm_mem, SI_BOOT1_VM, SI_ORDER_SECOND, vm_pager_init, NULL)
 182
 183 void
 184 vm_pager_bufferinit(void)
 185 {
 186         struct buf *bp;
 187         int i;
 188
 189         /*
 190          * Reserve KVM space for pbuf data.
 191          */
 192         swapbkva = kmem_alloc_pageable(&pager_map, nswbuf * MAXPHYS);
 193         if (!swapbkva)
 194                 panic("Not enough pager_map VM space for physical buffers");
 195
 196         /*
 197          * Initial pbuf setup.
 198          */
 199         bp = swbuf;
 200         for (i = 0; i < nswbuf; ++i, ++bp) {
 201                 bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva;
 202                 bp->b_kvasize = MAXPHYS;
 203                 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
 204                 BUF_LOCKINIT(bp);
 205                 buf_dep_init(bp);
 206         }
 207
 208         /*
 209          * Allow the clustering code to use half of our pbufs.
 210          */
 211         cluster_pbuf_freecnt = nswbuf / 2;
 212 }
 213
 214 void
 215 vm_pager_deallocate(vm_object_t object)
 216 {
 217         (*pagertab[object->type]->pgo_dealloc) (object);
 218 }
 219
 220 /*
 221  * vm_pager_get_pages() - inline, see vm/vm_pager.h
 222  * vm_pager_put_pages() - inline, see vm/vm_pager.h
 223  * vm_pager_has_page() - inline, see vm/vm_pager.h
 224  * vm_pager_page_inserted() - inline, see vm/vm_pager.h
 225  * vm_pager_page_removed() - inline, see vm/vm_pager.h
 226  */
 227
 228 #if 0
 229 /*
 230  *      vm_pager_sync:
 231  *
 232  *      Called by pageout daemon before going back to sleep.
 233  *      Gives pagers a chance to clean up any completed async pageing
 234  *      operations.
 235  */
 236 void
 237 vm_pager_sync(void)
 238 {
 239         struct pagerops **pgops;
 240
 241         for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
 242                 if (pgops && ((*pgops)->pgo_sync != NULL))
 243                         (*(*pgops)->pgo_sync) ();
 244 }
 245
 246 #endif
 247
 248 /*
 249  * Initialize a physical buffer.
 250  */
 251 static void
 252 initpbuf(struct buf *bp)
 253 {
 254         bp->b_qindex = 0; /* BQUEUE_NONE */
 255         bp->b_data = bp->b_kvabase;
 256         bp->b_flags = B_PAGING;
 257         bp->b_cmd = BUF_CMD_DONE;
 258         bp->b_error = 0;
 259         bp->b_bcount = 0;
 260         bp->b_bufsize = MAXPHYS;
 261         initbufbio(bp);
 262         xio_init(&bp->b_xio);
 263         BUF_LOCK(bp, LK_EXCLUSIVE);
 264 }
 265
 266 /*
 267  * Allocate a physical buffer
 268  *
 269  *      There are a limited number (nswbuf) of physical buffers.  We need
 270  *      to make sure that no single subsystem is able to hog all of them,
 271  *      so each subsystem implements a counter which is typically initialized
 272  *      to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
 273  *      increments it on release, and blocks if the counter hits zero.  A
 274  *      subsystem may initialize the counter to -1 to disable the feature,
 275  *      but it must still be sure to match up all uses of getpbuf() with
 276  *      relpbuf() using the same variable.
 277  *
 278  *      NOTE: pfreecnt can be NULL, but this 'feature' will be removed
 279  *      relatively soon when the rest of the subsystems get smart about it. XXX
 280  *
 281  * MPSAFE
 282  */
 283 struct buf *
 284 getpbuf(int *pfreecnt)
 285 {
 286         struct buf *bp;
 287
 288         spin_lock_wr(&bswspin);
 289
 290         for (;;) {
 291                 if (pfreecnt) {
 292                         while (*pfreecnt == 0)
 293                                 ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0);
 294                 }
 295
 296                 /* get a bp from the swap buffer header pool */
 297                 if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
 298                         break;
 299                 bswneeded = 1;
 300                 ssleep(&bswneeded, &bswspin, 0, "wswbuf1", 0);
 301                 /* loop in case someone else grabbed one */
 302         }
 303         TAILQ_REMOVE(&bswlist, bp, b_freelist);
 304         if (pfreecnt)
 305                 --*pfreecnt;
 306
 307         spin_unlock_wr(&bswspin);
 308
 309         initpbuf(bp);
 310         return bp;
 311 }
 312
 313 /*
 314  * Allocate a physical buffer, if one is available.
 315  *
 316  *      Note that there is no NULL hack here - all subsystems using this
 317  *      call understand how to use pfreecnt.
 318  *
 319  * MPSAFE
 320  */
 321 struct buf *
 322 trypbuf(int *pfreecnt)
 323 {
 324         struct buf *bp;
 325
 326         spin_lock_wr(&bswspin);
 327
 328         if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
 329                 spin_unlock_wr(&bswspin);
 330                 return NULL;
 331         }
 332         TAILQ_REMOVE(&bswlist, bp, b_freelist);
 333         --*pfreecnt;
 334
 335         spin_unlock_wr(&bswspin);
 336
 337         initpbuf(bp);
 338
 339         return bp;
 340 }
 341
 342 /*
 343  * Release a physical buffer
 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
 347  *
 348  * MPSAFE
 349  */
 350 void
 351 relpbuf(struct buf *bp, int *pfreecnt)
 352 {
 353         int wake_bsw = 0;
 354         int wake_freecnt = 0;
 355
 356         KKASSERT(bp->b_flags & B_PAGING);
 357
 358         spin_lock_wr(&bswspin);
 359
 360         BUF_UNLOCK(bp);
 361         TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
 362         if (bswneeded) {
 363                 bswneeded = 0;
 364                 wake_bsw = 1;
 365         }
 366         if (pfreecnt) {
 367                 if (++*pfreecnt == 1)
 368                         wake_freecnt = 1;
 369         }
 370
 371         spin_unlock_wr(&bswspin);
 372
 373         if (wake_bsw)
 374                 wakeup(&bswneeded);
 375         if (wake_freecnt)
 376                 wakeup(pfreecnt);
 377 }
 378