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. 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. 21 * 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 32 * SUCH DAMAGE. 33 * 34 * from: @(#)vm_pager.c 8.6 (Berkeley) 1/12/94 35 * 36 * 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 41 * 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. 47 * 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. 51 * 52 * Carnegie Mellon requests users of this software to return to 53 * 54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55 * School of Computer Science 56 * Carnegie Mellon University 57 * Pittsburgh PA 15213-3890 58 * 59 * any improvements or extensions that they make and grant Carnegie the 60 * rights to redistribute these changes. 61 * 62 * $FreeBSD: src/sys/vm/vm_pager.c,v 1.54.2.2 2001/11/18 07:11:00 dillon Exp $ 63 */ 64 65 /* 66 * Paging space routine stubs. Emulates a matchmaker-like interface 67 * for builtin pagers. 68 */ 69 70 #include <sys/param.h> 71 #include <sys/systm.h> 72 #include <sys/kernel.h> 73 #include <sys/vnode.h> 74 #include <sys/buf.h> 75 #include <sys/ucred.h> 76 #include <sys/dsched.h> 77 #include <sys/proc.h> 78 #include <sys/sysctl.h> 79 #include <sys/thread2.h> 80 81 #include <vm/vm.h> 82 #include <vm/vm_param.h> 83 #include <vm/vm_kern.h> 84 #include <vm/vm_object.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_pager.h> 87 #include <vm/vm_extern.h> 88 89 #include <sys/buf2.h> 90 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; 96 97 int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ 98 99 static int dead_pager_getpage (vm_object_t, vm_page_t *, int); 100 static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *); 101 static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t); 102 static void dead_pager_dealloc (vm_object_t); 103 104 /* 105 * No requirements. 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 /* 114 * No requirements. 115 */ 116 static void 117 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags, 118 int *rtvals) 119 { 120 int i; 121 122 for (i = 0; i < count; i++) { 123 rtvals[i] = VM_PAGER_AGAIN; 124 } 125 } 126 127 /* 128 * No requirements. 129 */ 130 static int 131 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex) 132 { 133 return FALSE; 134 } 135 136 /* 137 * No requirements. 138 */ 139 static void 140 dead_pager_dealloc(vm_object_t object) 141 { 142 KKASSERT(object->swblock_count == 0); 143 return; 144 } 145 146 static struct pagerops deadpagerops = { 147 dead_pager_dealloc, 148 dead_pager_getpage, 149 dead_pager_putpages, 150 dead_pager_haspage 151 }; 152 153 struct 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 }; 161 162 int npagers = NELEM(pagertab); 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) 173 174 TAILQ_HEAD(swqueue, buf); 175 176 int pager_map_size = PAGER_MAP_SIZE; 177 struct vm_map pager_map; 178 179 static int bswneeded_raw; 180 static int bswneeded_kva; 181 static long nswbuf_raw; 182 static struct buf *swbuf_raw; 183 static vm_offset_t swapbkva; /* swap buffers kva */ 184 static struct swqueue bswlist_raw; /* without kva */ 185 static struct swqueue bswlist_kva; /* with kva */ 186 static struct spinlock bswspin = SPINLOCK_INITIALIZER(&bswspin); 187 static int pbuf_raw_count; 188 static int pbuf_kva_count; 189 190 SYSCTL_INT(_vfs, OID_AUTO, pbuf_raw_count, CTLFLAG_RD, &pbuf_raw_count, 0, 191 "Kernel virtual address space reservations"); 192 SYSCTL_INT(_vfs, OID_AUTO, pbuf_kva_count, CTLFLAG_RD, &pbuf_kva_count, 0, 193 "Kernel raw address space reservations"); 194 195 /* 196 * Initialize the swap buffer list. 197 * 198 * Called from the low level boot code only. 199 */ 200 static void 201 vm_pager_init(void *arg __unused) 202 { 203 TAILQ_INIT(&bswlist_raw); 204 TAILQ_INIT(&bswlist_kva); 205 } 206 SYSINIT(vm_mem, SI_BOOT1_VM, SI_ORDER_SECOND, vm_pager_init, NULL) 207 208 /* 209 * Called from the low level boot code only. 210 */ 211 void 212 vm_pager_bufferinit(void) 213 { 214 struct buf *bp; 215 long i; 216 217 /* 218 * Reserve KVM space for pbuf data. 219 */ 220 swapbkva = kmem_alloc_pageable(&pager_map, nswbuf * MAXPHYS); 221 if (!swapbkva) 222 panic("Not enough pager_map VM space for physical buffers"); 223 224 /* 225 * Initial pbuf setup. These pbufs have KVA reservations. 226 */ 227 bp = swbuf; 228 for (i = 0; i < nswbuf; ++i, ++bp) { 229 bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva; 230 bp->b_kvasize = MAXPHYS; 231 BUF_LOCKINIT(bp); 232 buf_dep_init(bp); 233 TAILQ_INSERT_HEAD(&bswlist_kva, bp, b_freelist); 234 ++pbuf_kva_count; 235 } 236 237 /* 238 * Initial pbuf setup. These pbufs do not have KVA reservations, 239 * so we can have a lot more of them. These are typically used 240 * to massage low level buf/bio requests. 241 */ 242 nswbuf_raw = nbuf * 2; 243 swbuf_raw = (void *)kmem_alloc(&kernel_map, 244 round_page(nswbuf_raw * sizeof(struct buf))); 245 bp = swbuf_raw; 246 for (i = 0; i < nswbuf_raw; ++i, ++bp) { 247 BUF_LOCKINIT(bp); 248 buf_dep_init(bp); 249 TAILQ_INSERT_HEAD(&bswlist_raw, bp, b_freelist); 250 ++pbuf_raw_count; 251 } 252 253 /* 254 * Allow the clustering code to use half of our pbufs. 255 */ 256 cluster_pbuf_freecnt = nswbuf / 2; 257 } 258 259 /* 260 * No requirements. 261 */ 262 void 263 vm_pager_deallocate(vm_object_t object) 264 { 265 (*pagertab[object->type]->pgo_dealloc) (object); 266 } 267 268 /* 269 * vm_pager_get_pages() - inline, see vm/vm_pager.h 270 * vm_pager_put_pages() - inline, see vm/vm_pager.h 271 * vm_pager_has_page() - inline, see vm/vm_pager.h 272 * vm_pager_page_inserted() - inline, see vm/vm_pager.h 273 * vm_pager_page_removed() - inline, see vm/vm_pager.h 274 */ 275 276 #if 0 277 /* 278 * vm_pager_sync: 279 * 280 * Called by pageout daemon before going back to sleep. 281 * Gives pagers a chance to clean up any completed async pageing 282 * operations. 283 */ 284 void 285 vm_pager_sync(void) 286 { 287 struct pagerops **pgops; 288 289 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 290 if (pgops && ((*pgops)->pgo_sync != NULL)) 291 (*(*pgops)->pgo_sync) (); 292 } 293 294 #endif 295 296 /* 297 * Initialize a physical buffer. 298 * 299 * No requirements. 300 */ 301 static void 302 initpbuf(struct buf *bp) 303 { 304 bp->b_qindex = 0; /* BQUEUE_NONE */ 305 bp->b_data = bp->b_kvabase; /* NULL if pbuf sans kva */ 306 bp->b_flags = B_PAGING; 307 bp->b_cmd = BUF_CMD_DONE; 308 bp->b_error = 0; 309 bp->b_bcount = 0; 310 bp->b_bufsize = MAXPHYS; 311 initbufbio(bp); 312 xio_init(&bp->b_xio); 313 BUF_LOCK(bp, LK_EXCLUSIVE); 314 } 315 316 /* 317 * Allocate a physical buffer 318 * 319 * There are a limited number (nswbuf) of physical buffers. We need 320 * to make sure that no single subsystem is able to hog all of them, 321 * so each subsystem implements a counter which is typically initialized 322 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 323 * increments it on release, and blocks if the counter hits zero. A 324 * subsystem may initialize the counter to -1 to disable the feature, 325 * but it must still be sure to match up all uses of getpbuf() with 326 * relpbuf() using the same variable. 327 * 328 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 329 * relatively soon when the rest of the subsystems get smart about it. XXX 330 * 331 * Physical buffers can be with or without KVA space reserved. There 332 * are severe limitations on the ones with KVA reserved, and fewer 333 * limitations on the ones without. getpbuf() gets one without, 334 * getpbuf_kva() gets one with. 335 * 336 * No requirements. 337 */ 338 struct buf * 339 getpbuf(int *pfreecnt) 340 { 341 struct buf *bp; 342 343 spin_lock(&bswspin); 344 345 for (;;) { 346 if (pfreecnt) { 347 while (*pfreecnt == 0) 348 ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0); 349 } 350 351 /* get a bp from the swap buffer header pool */ 352 if ((bp = TAILQ_FIRST(&bswlist_raw)) != NULL) 353 break; 354 bswneeded_raw = 1; 355 ssleep(&bswneeded_raw, &bswspin, 0, "wswbuf1", 0); 356 /* loop in case someone else grabbed one */ 357 } 358 TAILQ_REMOVE(&bswlist_raw, bp, b_freelist); 359 --pbuf_raw_count; 360 if (pfreecnt) 361 --*pfreecnt; 362 363 spin_unlock(&bswspin); 364 365 initpbuf(bp); 366 KKASSERT(dsched_is_clear_buf_priv(bp)); 367 368 return (bp); 369 } 370 371 struct buf * 372 getpbuf_kva(int *pfreecnt) 373 { 374 struct buf *bp; 375 376 spin_lock(&bswspin); 377 378 for (;;) { 379 if (pfreecnt) { 380 while (*pfreecnt == 0) 381 ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0); 382 } 383 384 /* get a bp from the swap buffer header pool */ 385 if ((bp = TAILQ_FIRST(&bswlist_kva)) != NULL) 386 break; 387 bswneeded_kva = 1; 388 ssleep(&bswneeded_kva, &bswspin, 0, "wswbuf1", 0); 389 /* loop in case someone else grabbed one */ 390 } 391 TAILQ_REMOVE(&bswlist_kva, bp, b_freelist); 392 --pbuf_kva_count; 393 if (pfreecnt) 394 --*pfreecnt; 395 396 spin_unlock(&bswspin); 397 398 initpbuf(bp); 399 KKASSERT(dsched_is_clear_buf_priv(bp)); 400 401 return (bp); 402 } 403 404 /* 405 * Allocate a physical buffer, if one is available. 406 * 407 * Note that there is no NULL hack here - all subsystems using this 408 * call understand how to use pfreecnt. 409 * 410 * No requirements. 411 */ 412 struct buf * 413 trypbuf(int *pfreecnt) 414 { 415 struct buf *bp; 416 417 spin_lock(&bswspin); 418 419 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist_raw)) == NULL) { 420 spin_unlock(&bswspin); 421 return NULL; 422 } 423 TAILQ_REMOVE(&bswlist_raw, bp, b_freelist); 424 --pbuf_raw_count; 425 --*pfreecnt; 426 427 spin_unlock(&bswspin); 428 429 initpbuf(bp); 430 431 return bp; 432 } 433 434 struct buf * 435 trypbuf_kva(int *pfreecnt) 436 { 437 struct buf *bp; 438 439 spin_lock(&bswspin); 440 441 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist_kva)) == NULL) { 442 spin_unlock(&bswspin); 443 return NULL; 444 } 445 TAILQ_REMOVE(&bswlist_kva, bp, b_freelist); 446 --pbuf_kva_count; 447 --*pfreecnt; 448 449 spin_unlock(&bswspin); 450 451 initpbuf(bp); 452 453 return bp; 454 } 455 456 /* 457 * Release a physical buffer 458 * 459 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 460 * relatively soon when the rest of the subsystems get smart about it. XXX 461 * 462 * No requirements. 463 */ 464 void 465 relpbuf(struct buf *bp, int *pfreecnt) 466 { 467 int wake_bsw_kva = 0; 468 int wake_bsw_raw = 0; 469 int wake_freecnt = 0; 470 471 KKASSERT(bp->b_flags & B_PAGING); 472 dsched_exit_buf(bp); 473 474 BUF_UNLOCK(bp); 475 476 spin_lock(&bswspin); 477 if (bp->b_kvabase) { 478 TAILQ_INSERT_HEAD(&bswlist_kva, bp, b_freelist); 479 ++pbuf_kva_count; 480 } else { 481 TAILQ_INSERT_HEAD(&bswlist_raw, bp, b_freelist); 482 ++pbuf_raw_count; 483 } 484 if (bswneeded_kva) { 485 bswneeded_kva = 0; 486 wake_bsw_kva = 1; 487 } 488 if (bswneeded_raw) { 489 bswneeded_raw = 0; 490 wake_bsw_raw = 1; 491 } 492 if (pfreecnt) { 493 if (++*pfreecnt == 1) 494 wake_freecnt = 1; 495 } 496 spin_unlock(&bswspin); 497 498 if (wake_bsw_kva) 499 wakeup(&bswneeded_kva); 500 if (wake_bsw_raw) 501 wakeup(&bswneeded_raw); 502 if (wake_freecnt) 503 wakeup(pfreecnt); 504 } 505 506 void 507 pbuf_adjcount(int *pfreecnt, int n) 508 { 509 if (n) { 510 spin_lock(&bswspin); 511 *pfreecnt += n; 512 spin_unlock(&bswspin); 513 wakeup(pfreecnt); 514 } 515 } 516