| 1 | /* |
| 2 | * (MPSAFE) |
| 3 | * |
| 4 | * Copyright (c) 1991, 1993 |
| 5 | * The Regents of the University of California. All rights reserved. |
| 6 | * |
| 7 | * This code is derived from software contributed to Berkeley by |
| 8 | * The Mach Operating System project at Carnegie-Mellon University. |
| 9 | * |
| 10 | * Redistribution and use in source and binary forms, with or without |
| 11 | * modification, are permitted provided that the following conditions |
| 12 | * are met: |
| 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. All advertising materials mentioning features or use of this software |
| 19 | * must display the following acknowledgement: |
| 20 | * This product includes software developed by the University of |
| 21 | * California, Berkeley and its contributors. |
| 22 | * 4. Neither the name of the University nor the names of its contributors |
| 23 | * may be used to endorse or promote products derived from this software |
| 24 | * without specific prior written permission. |
| 25 | * |
| 26 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 27 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 28 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 29 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 30 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 31 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 32 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 33 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 34 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 35 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 36 | * SUCH DAMAGE. |
| 37 | * |
| 38 | * from: @(#)vm_pager.c 8.6 (Berkeley) 1/12/94 |
| 39 | * |
| 40 | * |
| 41 | * Copyright (c) 1987, 1990 Carnegie-Mellon University. |
| 42 | * All rights reserved. |
| 43 | * |
| 44 | * Authors: Avadis Tevanian, Jr., Michael Wayne Young |
| 45 | * |
| 46 | * Permission to use, copy, modify and distribute this software and |
| 47 | * its documentation is hereby granted, provided that both the copyright |
| 48 | * notice and this permission notice appear in all copies of the |
| 49 | * software, derivative works or modified versions, and any portions |
| 50 | * thereof, and that both notices appear in supporting documentation. |
| 51 | * |
| 52 | * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" |
| 53 | * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND |
| 54 | * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. |
| 55 | * |
| 56 | * Carnegie Mellon requests users of this software to return to |
| 57 | * |
| 58 | * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU |
| 59 | * School of Computer Science |
| 60 | * Carnegie Mellon University |
| 61 | * Pittsburgh PA 15213-3890 |
| 62 | * |
| 63 | * any improvements or extensions that they make and grant Carnegie the |
| 64 | * rights to redistribute these changes. |
| 65 | * |
| 66 | * $FreeBSD: src/sys/vm/vm_pager.c,v 1.54.2.2 2001/11/18 07:11:00 dillon Exp $ |
| 67 | * $DragonFly: src/sys/vm/vm_pager.c,v 1.24 2007/11/06 03:50:01 dillon Exp $ |
| 68 | */ |
| 69 | |
| 70 | /* |
| 71 | * Paging space routine stubs. Emulates a matchmaker-like interface |
| 72 | * for builtin pagers. |
| 73 | */ |
| 74 | |
| 75 | #include <sys/param.h> |
| 76 | #include <sys/systm.h> |
| 77 | #include <sys/kernel.h> |
| 78 | #include <sys/vnode.h> |
| 79 | #include <sys/buf.h> |
| 80 | #include <sys/ucred.h> |
| 81 | #include <sys/malloc.h> |
| 82 | #include <sys/dsched.h> |
| 83 | #include <sys/proc.h> |
| 84 | #include <sys/sysctl.h> |
| 85 | #include <sys/thread2.h> |
| 86 | |
| 87 | #include <vm/vm.h> |
| 88 | #include <vm/vm_param.h> |
| 89 | #include <vm/vm_kern.h> |
| 90 | #include <vm/vm_object.h> |
| 91 | #include <vm/vm_page.h> |
| 92 | #include <vm/vm_pager.h> |
| 93 | #include <vm/vm_extern.h> |
| 94 | |
| 95 | #include <sys/buf2.h> |
| 96 | |
| 97 | MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data"); |
| 98 | |
| 99 | extern struct pagerops defaultpagerops; |
| 100 | extern struct pagerops swappagerops; |
| 101 | extern struct pagerops vnodepagerops; |
| 102 | extern struct pagerops devicepagerops; |
| 103 | extern struct pagerops physpagerops; |
| 104 | |
| 105 | int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ |
| 106 | |
| 107 | static int dead_pager_getpage (vm_object_t, vm_page_t *, int); |
| 108 | static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *); |
| 109 | static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t); |
| 110 | static void dead_pager_dealloc (vm_object_t); |
| 111 | |
| 112 | /* |
| 113 | * No requirements. |
| 114 | */ |
| 115 | static int |
| 116 | dead_pager_getpage(vm_object_t obj, vm_page_t *mpp, int seqaccess) |
| 117 | { |
| 118 | return VM_PAGER_FAIL; |
| 119 | } |
| 120 | |
| 121 | /* |
| 122 | * No requirements. |
| 123 | */ |
| 124 | static void |
| 125 | dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags, |
| 126 | int *rtvals) |
| 127 | { |
| 128 | int i; |
| 129 | |
| 130 | for (i = 0; i < count; i++) { |
| 131 | rtvals[i] = VM_PAGER_AGAIN; |
| 132 | } |
| 133 | } |
| 134 | |
| 135 | /* |
| 136 | * No requirements. |
| 137 | */ |
| 138 | static int |
| 139 | dead_pager_haspage(vm_object_t object, vm_pindex_t pindex) |
| 140 | { |
| 141 | return FALSE; |
| 142 | } |
| 143 | |
| 144 | /* |
| 145 | * No requirements. |
| 146 | */ |
| 147 | static void |
| 148 | dead_pager_dealloc(vm_object_t object) |
| 149 | { |
| 150 | KKASSERT(object->swblock_count == 0); |
| 151 | return; |
| 152 | } |
| 153 | |
| 154 | static struct pagerops deadpagerops = { |
| 155 | dead_pager_dealloc, |
| 156 | dead_pager_getpage, |
| 157 | dead_pager_putpages, |
| 158 | dead_pager_haspage |
| 159 | }; |
| 160 | |
| 161 | struct pagerops *pagertab[] = { |
| 162 | &defaultpagerops, /* OBJT_DEFAULT */ |
| 163 | &swappagerops, /* OBJT_SWAP */ |
| 164 | &vnodepagerops, /* OBJT_VNODE */ |
| 165 | &devicepagerops, /* OBJT_DEVICE */ |
| 166 | &physpagerops, /* OBJT_PHYS */ |
| 167 | &deadpagerops /* OBJT_DEAD */ |
| 168 | }; |
| 169 | |
| 170 | int npagers = sizeof(pagertab) / sizeof(pagertab[0]); |
| 171 | |
| 172 | /* |
| 173 | * Kernel address space for mapping pages. |
| 174 | * Used by pagers where KVAs are needed for IO. |
| 175 | * |
| 176 | * XXX needs to be large enough to support the number of pending async |
| 177 | * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size |
| 178 | * (MAXPHYS == 64k) if you want to get the most efficiency. |
| 179 | */ |
| 180 | #define PAGER_MAP_SIZE (8 * 1024 * 1024) |
| 181 | |
| 182 | TAILQ_HEAD(swqueue, buf); |
| 183 | |
| 184 | int pager_map_size = PAGER_MAP_SIZE; |
| 185 | struct vm_map pager_map; |
| 186 | |
| 187 | static int bswneeded_raw; |
| 188 | static int bswneeded_kva; |
| 189 | static int nswbuf_raw; |
| 190 | static struct buf *swbuf_raw; |
| 191 | static vm_offset_t swapbkva; /* swap buffers kva */ |
| 192 | static struct swqueue bswlist_raw; /* without kva */ |
| 193 | static struct swqueue bswlist_kva; /* with kva */ |
| 194 | static struct spinlock bswspin = SPINLOCK_INITIALIZER(&bswspin); |
| 195 | static int pbuf_raw_count; |
| 196 | static int pbuf_kva_count; |
| 197 | |
| 198 | SYSCTL_INT(_vfs, OID_AUTO, pbuf_raw_count, CTLFLAG_RD, &pbuf_raw_count, 0, |
| 199 | "Kernel virtual address space reservations"); |
| 200 | SYSCTL_INT(_vfs, OID_AUTO, pbuf_kva_count, CTLFLAG_RD, &pbuf_kva_count, 0, |
| 201 | "Kernel raw address space reservations"); |
| 202 | |
| 203 | /* |
| 204 | * Initialize the swap buffer list. |
| 205 | * |
| 206 | * Called from the low level boot code only. |
| 207 | */ |
| 208 | static void |
| 209 | vm_pager_init(void *arg __unused) |
| 210 | { |
| 211 | TAILQ_INIT(&bswlist_raw); |
| 212 | TAILQ_INIT(&bswlist_kva); |
| 213 | } |
| 214 | SYSINIT(vm_mem, SI_BOOT1_VM, SI_ORDER_SECOND, vm_pager_init, NULL) |
| 215 | |
| 216 | /* |
| 217 | * Called from the low level boot code only. |
| 218 | */ |
| 219 | void |
| 220 | vm_pager_bufferinit(void) |
| 221 | { |
| 222 | struct buf *bp; |
| 223 | int i; |
| 224 | |
| 225 | /* |
| 226 | * Reserve KVM space for pbuf data. |
| 227 | */ |
| 228 | swapbkva = kmem_alloc_pageable(&pager_map, nswbuf * MAXPHYS); |
| 229 | if (!swapbkva) |
| 230 | panic("Not enough pager_map VM space for physical buffers"); |
| 231 | |
| 232 | /* |
| 233 | * Initial pbuf setup. These pbufs have KVA reservations. |
| 234 | */ |
| 235 | bp = swbuf; |
| 236 | for (i = 0; i < nswbuf; ++i, ++bp) { |
| 237 | bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva; |
| 238 | bp->b_kvasize = MAXPHYS; |
| 239 | BUF_LOCKINIT(bp); |
| 240 | buf_dep_init(bp); |
| 241 | TAILQ_INSERT_HEAD(&bswlist_kva, bp, b_freelist); |
| 242 | ++pbuf_kva_count; |
| 243 | } |
| 244 | |
| 245 | /* |
| 246 | * Initial pbuf setup. These pbufs do not have KVA reservations, |
| 247 | * so we can have a lot more of them. These are typically used |
| 248 | * to massage low level buf/bio requests. |
| 249 | */ |
| 250 | nswbuf_raw = nbuf * 2; |
| 251 | swbuf_raw = (void *)kmem_alloc(&kernel_map, |
| 252 | round_page(nswbuf_raw * sizeof(struct buf))); |
| 253 | bp = swbuf_raw; |
| 254 | for (i = 0; i < nswbuf_raw; ++i, ++bp) { |
| 255 | BUF_LOCKINIT(bp); |
| 256 | buf_dep_init(bp); |
| 257 | TAILQ_INSERT_HEAD(&bswlist_raw, bp, b_freelist); |
| 258 | ++pbuf_raw_count; |
| 259 | } |
| 260 | |
| 261 | /* |
| 262 | * Allow the clustering code to use half of our pbufs. |
| 263 | */ |
| 264 | cluster_pbuf_freecnt = nswbuf / 2; |
| 265 | } |
| 266 | |
| 267 | /* |
| 268 | * No requirements. |
| 269 | */ |
| 270 | void |
| 271 | vm_pager_deallocate(vm_object_t object) |
| 272 | { |
| 273 | (*pagertab[object->type]->pgo_dealloc) (object); |
| 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 | */ |
| 292 | void |
| 293 | vm_pager_sync(void) |
| 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 | |
| 304 | /* |
| 305 | * Initialize a physical buffer. |
| 306 | * |
| 307 | * No requirements. |
| 308 | */ |
| 309 | static void |
| 310 | initpbuf(struct buf *bp) |
| 311 | { |
| 312 | bp->b_qindex = 0; /* BQUEUE_NONE */ |
| 313 | bp->b_data = bp->b_kvabase; /* NULL if pbuf sans kva */ |
| 314 | bp->b_flags = B_PAGING; |
| 315 | bp->b_cmd = BUF_CMD_DONE; |
| 316 | bp->b_error = 0; |
| 317 | bp->b_bcount = 0; |
| 318 | bp->b_bufsize = MAXPHYS; |
| 319 | initbufbio(bp); |
| 320 | xio_init(&bp->b_xio); |
| 321 | BUF_LOCK(bp, LK_EXCLUSIVE); |
| 322 | } |
| 323 | |
| 324 | /* |
| 325 | * Allocate a physical buffer |
| 326 | * |
| 327 | * There are a limited number (nswbuf) of physical buffers. We need |
| 328 | * to make sure that no single subsystem is able to hog all of them, |
| 329 | * so each subsystem implements a counter which is typically initialized |
| 330 | * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and |
| 331 | * increments it on release, and blocks if the counter hits zero. A |
| 332 | * subsystem may initialize the counter to -1 to disable the feature, |
| 333 | * but it must still be sure to match up all uses of getpbuf() with |
| 334 | * relpbuf() using the same variable. |
| 335 | * |
| 336 | * NOTE: pfreecnt can be NULL, but this 'feature' will be removed |
| 337 | * relatively soon when the rest of the subsystems get smart about it. XXX |
| 338 | * |
| 339 | * Physical buffers can be with or without KVA space reserved. There |
| 340 | * are severe limitations on the ones with KVA reserved, and fewer |
| 341 | * limitations on the ones without. getpbuf() gets one without, |
| 342 | * getpbuf_kva() gets one with. |
| 343 | * |
| 344 | * No requirements. |
| 345 | */ |
| 346 | struct buf * |
| 347 | getpbuf(int *pfreecnt) |
| 348 | { |
| 349 | struct buf *bp; |
| 350 | |
| 351 | spin_lock(&bswspin); |
| 352 | |
| 353 | for (;;) { |
| 354 | if (pfreecnt) { |
| 355 | while (*pfreecnt == 0) |
| 356 | ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0); |
| 357 | } |
| 358 | |
| 359 | /* get a bp from the swap buffer header pool */ |
| 360 | if ((bp = TAILQ_FIRST(&bswlist_raw)) != NULL) |
| 361 | break; |
| 362 | bswneeded_raw = 1; |
| 363 | ssleep(&bswneeded_raw, &bswspin, 0, "wswbuf1", 0); |
| 364 | /* loop in case someone else grabbed one */ |
| 365 | } |
| 366 | TAILQ_REMOVE(&bswlist_raw, bp, b_freelist); |
| 367 | --pbuf_raw_count; |
| 368 | if (pfreecnt) |
| 369 | --*pfreecnt; |
| 370 | |
| 371 | spin_unlock(&bswspin); |
| 372 | |
| 373 | initpbuf(bp); |
| 374 | KKASSERT(dsched_is_clear_buf_priv(bp)); |
| 375 | |
| 376 | return (bp); |
| 377 | } |
| 378 | |
| 379 | struct buf * |
| 380 | getpbuf_kva(int *pfreecnt) |
| 381 | { |
| 382 | struct buf *bp; |
| 383 | |
| 384 | spin_lock(&bswspin); |
| 385 | |
| 386 | for (;;) { |
| 387 | if (pfreecnt) { |
| 388 | while (*pfreecnt == 0) |
| 389 | ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0); |
| 390 | } |
| 391 | |
| 392 | /* get a bp from the swap buffer header pool */ |
| 393 | if ((bp = TAILQ_FIRST(&bswlist_kva)) != NULL) |
| 394 | break; |
| 395 | bswneeded_kva = 1; |
| 396 | ssleep(&bswneeded_kva, &bswspin, 0, "wswbuf1", 0); |
| 397 | /* loop in case someone else grabbed one */ |
| 398 | } |
| 399 | TAILQ_REMOVE(&bswlist_kva, bp, b_freelist); |
| 400 | --pbuf_kva_count; |
| 401 | if (pfreecnt) |
| 402 | --*pfreecnt; |
| 403 | |
| 404 | spin_unlock(&bswspin); |
| 405 | |
| 406 | initpbuf(bp); |
| 407 | KKASSERT(dsched_is_clear_buf_priv(bp)); |
| 408 | |
| 409 | return (bp); |
| 410 | } |
| 411 | |
| 412 | /* |
| 413 | * Allocate a physical buffer, if one is available. |
| 414 | * |
| 415 | * Note that there is no NULL hack here - all subsystems using this |
| 416 | * call understand how to use pfreecnt. |
| 417 | * |
| 418 | * No requirements. |
| 419 | */ |
| 420 | struct buf * |
| 421 | trypbuf(int *pfreecnt) |
| 422 | { |
| 423 | struct buf *bp; |
| 424 | |
| 425 | spin_lock(&bswspin); |
| 426 | |
| 427 | if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist_raw)) == NULL) { |
| 428 | spin_unlock(&bswspin); |
| 429 | return NULL; |
| 430 | } |
| 431 | TAILQ_REMOVE(&bswlist_raw, bp, b_freelist); |
| 432 | --pbuf_raw_count; |
| 433 | --*pfreecnt; |
| 434 | |
| 435 | spin_unlock(&bswspin); |
| 436 | |
| 437 | initpbuf(bp); |
| 438 | |
| 439 | return bp; |
| 440 | } |
| 441 | |
| 442 | struct buf * |
| 443 | trypbuf_kva(int *pfreecnt) |
| 444 | { |
| 445 | struct buf *bp; |
| 446 | |
| 447 | spin_lock(&bswspin); |
| 448 | |
| 449 | if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist_kva)) == NULL) { |
| 450 | spin_unlock(&bswspin); |
| 451 | return NULL; |
| 452 | } |
| 453 | TAILQ_REMOVE(&bswlist_kva, bp, b_freelist); |
| 454 | --pbuf_kva_count; |
| 455 | --*pfreecnt; |
| 456 | |
| 457 | spin_unlock(&bswspin); |
| 458 | |
| 459 | initpbuf(bp); |
| 460 | |
| 461 | return bp; |
| 462 | } |
| 463 | |
| 464 | /* |
| 465 | * Release a physical buffer |
| 466 | * |
| 467 | * NOTE: pfreecnt can be NULL, but this 'feature' will be removed |
| 468 | * relatively soon when the rest of the subsystems get smart about it. XXX |
| 469 | * |
| 470 | * No requirements. |
| 471 | */ |
| 472 | void |
| 473 | relpbuf(struct buf *bp, int *pfreecnt) |
| 474 | { |
| 475 | int wake_bsw_kva = 0; |
| 476 | int wake_bsw_raw = 0; |
| 477 | int wake_freecnt = 0; |
| 478 | |
| 479 | KKASSERT(bp->b_flags & B_PAGING); |
| 480 | dsched_exit_buf(bp); |
| 481 | |
| 482 | BUF_UNLOCK(bp); |
| 483 | |
| 484 | spin_lock(&bswspin); |
| 485 | if (bp->b_kvabase) { |
| 486 | TAILQ_INSERT_HEAD(&bswlist_kva, bp, b_freelist); |
| 487 | ++pbuf_kva_count; |
| 488 | } else { |
| 489 | TAILQ_INSERT_HEAD(&bswlist_raw, bp, b_freelist); |
| 490 | ++pbuf_raw_count; |
| 491 | } |
| 492 | if (bswneeded_kva) { |
| 493 | bswneeded_kva = 0; |
| 494 | wake_bsw_kva = 1; |
| 495 | } |
| 496 | if (bswneeded_raw) { |
| 497 | bswneeded_raw = 0; |
| 498 | wake_bsw_raw = 1; |
| 499 | } |
| 500 | if (pfreecnt) { |
| 501 | if (++*pfreecnt == 1) |
| 502 | wake_freecnt = 1; |
| 503 | } |
| 504 | spin_unlock(&bswspin); |
| 505 | |
| 506 | if (wake_bsw_kva) |
| 507 | wakeup(&bswneeded_kva); |
| 508 | if (wake_bsw_raw) |
| 509 | wakeup(&bswneeded_raw); |
| 510 | if (wake_freecnt) |
| 511 | wakeup(pfreecnt); |
| 512 | } |