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 */ 68 69 /* 70 * Paging space routine stubs. Emulates a matchmaker-like interface 71 * for builtin pagers. 72 */ 73 74 #include <sys/param.h> 75 #include <sys/systm.h> 76 #include <sys/kernel.h> 77 #include <sys/vnode.h> 78 #include <sys/buf.h> 79 #include <sys/ucred.h> 80 #include <sys/malloc.h> 81 #include <sys/dsched.h> 82 #include <sys/proc.h> 83 #include <sys/sysctl.h> 84 #include <sys/thread2.h> 85 86 #include <vm/vm.h> 87 #include <vm/vm_param.h> 88 #include <vm/vm_kern.h> 89 #include <vm/vm_object.h> 90 #include <vm/vm_page.h> 91 #include <vm/vm_pager.h> 92 #include <vm/vm_extern.h> 93 94 #include <sys/buf2.h> 95 96 MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data"); 97 98 extern struct pagerops defaultpagerops; 99 extern struct pagerops swappagerops; 100 extern struct pagerops vnodepagerops; 101 extern struct pagerops devicepagerops; 102 extern struct pagerops physpagerops; 103 104 int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ 105 106 static int dead_pager_getpage (vm_object_t, vm_page_t *, int); 107 static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *); 108 static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t); 109 static void dead_pager_dealloc (vm_object_t); 110 111 /* 112 * No requirements. 113 */ 114 static int 115 dead_pager_getpage(vm_object_t obj, vm_page_t *mpp, int seqaccess) 116 { 117 return VM_PAGER_FAIL; 118 } 119 120 /* 121 * No requirements. 122 */ 123 static void 124 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags, 125 int *rtvals) 126 { 127 int i; 128 129 for (i = 0; i < count; i++) { 130 rtvals[i] = VM_PAGER_AGAIN; 131 } 132 } 133 134 /* 135 * No requirements. 136 */ 137 static int 138 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex) 139 { 140 return FALSE; 141 } 142 143 /* 144 * No requirements. 145 */ 146 static void 147 dead_pager_dealloc(vm_object_t object) 148 { 149 KKASSERT(object->swblock_count == 0); 150 return; 151 } 152 153 static struct pagerops deadpagerops = { 154 dead_pager_dealloc, 155 dead_pager_getpage, 156 dead_pager_putpages, 157 dead_pager_haspage 158 }; 159 160 struct pagerops *pagertab[] = { 161 &defaultpagerops, /* OBJT_DEFAULT */ 162 &swappagerops, /* OBJT_SWAP */ 163 &vnodepagerops, /* OBJT_VNODE */ 164 &devicepagerops, /* OBJT_DEVICE */ 165 &physpagerops, /* OBJT_PHYS */ 166 &deadpagerops /* OBJT_DEAD */ 167 }; 168 169 int npagers = NELEM(pagertab); 170 171 /* 172 * Kernel address space for mapping pages. 173 * Used by pagers where KVAs are needed for IO. 174 * 175 * XXX needs to be large enough to support the number of pending async 176 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size 177 * (MAXPHYS == 64k) if you want to get the most efficiency. 178 */ 179 #define PAGER_MAP_SIZE (8 * 1024 * 1024) 180 181 TAILQ_HEAD(swqueue, buf); 182 183 int pager_map_size = PAGER_MAP_SIZE; 184 struct vm_map pager_map; 185 186 static int bswneeded_raw; 187 static int bswneeded_kva; 188 static int nswbuf_raw; 189 static struct buf *swbuf_raw; 190 static vm_offset_t swapbkva; /* swap buffers kva */ 191 static struct swqueue bswlist_raw; /* without kva */ 192 static struct swqueue bswlist_kva; /* with kva */ 193 static struct spinlock bswspin = SPINLOCK_INITIALIZER(&bswspin); 194 static int pbuf_raw_count; 195 static int pbuf_kva_count; 196 197 SYSCTL_INT(_vfs, OID_AUTO, pbuf_raw_count, CTLFLAG_RD, &pbuf_raw_count, 0, 198 "Kernel virtual address space reservations"); 199 SYSCTL_INT(_vfs, OID_AUTO, pbuf_kva_count, CTLFLAG_RD, &pbuf_kva_count, 0, 200 "Kernel raw address space reservations"); 201 202 /* 203 * Initialize the swap buffer list. 204 * 205 * Called from the low level boot code only. 206 */ 207 static void 208 vm_pager_init(void *arg __unused) 209 { 210 TAILQ_INIT(&bswlist_raw); 211 TAILQ_INIT(&bswlist_kva); 212 } 213 SYSINIT(vm_mem, SI_BOOT1_VM, SI_ORDER_SECOND, vm_pager_init, NULL) 214 215 /* 216 * Called from the low level boot code only. 217 */ 218 void 219 vm_pager_bufferinit(void) 220 { 221 struct buf *bp; 222 int i; 223 224 /* 225 * Reserve KVM space for pbuf data. 226 */ 227 swapbkva = kmem_alloc_pageable(&pager_map, nswbuf * MAXPHYS); 228 if (!swapbkva) 229 panic("Not enough pager_map VM space for physical buffers"); 230 231 /* 232 * Initial pbuf setup. These pbufs have KVA reservations. 233 */ 234 bp = swbuf; 235 for (i = 0; i < nswbuf; ++i, ++bp) { 236 bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva; 237 bp->b_kvasize = MAXPHYS; 238 BUF_LOCKINIT(bp); 239 buf_dep_init(bp); 240 TAILQ_INSERT_HEAD(&bswlist_kva, bp, b_freelist); 241 ++pbuf_kva_count; 242 } 243 244 /* 245 * Initial pbuf setup. These pbufs do not have KVA reservations, 246 * so we can have a lot more of them. These are typically used 247 * to massage low level buf/bio requests. 248 */ 249 nswbuf_raw = nbuf * 2; 250 swbuf_raw = (void *)kmem_alloc(&kernel_map, 251 round_page(nswbuf_raw * sizeof(struct buf))); 252 bp = swbuf_raw; 253 for (i = 0; i < nswbuf_raw; ++i, ++bp) { 254 BUF_LOCKINIT(bp); 255 buf_dep_init(bp); 256 TAILQ_INSERT_HEAD(&bswlist_raw, bp, b_freelist); 257 ++pbuf_raw_count; 258 } 259 260 /* 261 * Allow the clustering code to use half of our pbufs. 262 */ 263 cluster_pbuf_freecnt = nswbuf / 2; 264 } 265 266 /* 267 * No requirements. 268 */ 269 void 270 vm_pager_deallocate(vm_object_t object) 271 { 272 (*pagertab[object->type]->pgo_dealloc) (object); 273 } 274 275 /* 276 * vm_pager_get_pages() - inline, see vm/vm_pager.h 277 * vm_pager_put_pages() - inline, see vm/vm_pager.h 278 * vm_pager_has_page() - inline, see vm/vm_pager.h 279 * vm_pager_page_inserted() - inline, see vm/vm_pager.h 280 * vm_pager_page_removed() - inline, see vm/vm_pager.h 281 */ 282 283 #if 0 284 /* 285 * vm_pager_sync: 286 * 287 * Called by pageout daemon before going back to sleep. 288 * Gives pagers a chance to clean up any completed async pageing 289 * operations. 290 */ 291 void 292 vm_pager_sync(void) 293 { 294 struct pagerops **pgops; 295 296 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++) 297 if (pgops && ((*pgops)->pgo_sync != NULL)) 298 (*(*pgops)->pgo_sync) (); 299 } 300 301 #endif 302 303 /* 304 * Initialize a physical buffer. 305 * 306 * No requirements. 307 */ 308 static void 309 initpbuf(struct buf *bp) 310 { 311 bp->b_qindex = 0; /* BQUEUE_NONE */ 312 bp->b_data = bp->b_kvabase; /* NULL if pbuf sans kva */ 313 bp->b_flags = B_PAGING; 314 bp->b_cmd = BUF_CMD_DONE; 315 bp->b_error = 0; 316 bp->b_bcount = 0; 317 bp->b_bufsize = MAXPHYS; 318 initbufbio(bp); 319 xio_init(&bp->b_xio); 320 BUF_LOCK(bp, LK_EXCLUSIVE); 321 } 322 323 /* 324 * Allocate a physical buffer 325 * 326 * There are a limited number (nswbuf) of physical buffers. We need 327 * to make sure that no single subsystem is able to hog all of them, 328 * so each subsystem implements a counter which is typically initialized 329 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 330 * increments it on release, and blocks if the counter hits zero. A 331 * subsystem may initialize the counter to -1 to disable the feature, 332 * but it must still be sure to match up all uses of getpbuf() with 333 * relpbuf() using the same variable. 334 * 335 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 336 * relatively soon when the rest of the subsystems get smart about it. XXX 337 * 338 * Physical buffers can be with or without KVA space reserved. There 339 * are severe limitations on the ones with KVA reserved, and fewer 340 * limitations on the ones without. getpbuf() gets one without, 341 * getpbuf_kva() gets one with. 342 * 343 * No requirements. 344 */ 345 struct buf * 346 getpbuf(int *pfreecnt) 347 { 348 struct buf *bp; 349 350 spin_lock(&bswspin); 351 352 for (;;) { 353 if (pfreecnt) { 354 while (*pfreecnt == 0) 355 ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0); 356 } 357 358 /* get a bp from the swap buffer header pool */ 359 if ((bp = TAILQ_FIRST(&bswlist_raw)) != NULL) 360 break; 361 bswneeded_raw = 1; 362 ssleep(&bswneeded_raw, &bswspin, 0, "wswbuf1", 0); 363 /* loop in case someone else grabbed one */ 364 } 365 TAILQ_REMOVE(&bswlist_raw, bp, b_freelist); 366 --pbuf_raw_count; 367 if (pfreecnt) 368 --*pfreecnt; 369 370 spin_unlock(&bswspin); 371 372 initpbuf(bp); 373 KKASSERT(dsched_is_clear_buf_priv(bp)); 374 375 return (bp); 376 } 377 378 struct buf * 379 getpbuf_kva(int *pfreecnt) 380 { 381 struct buf *bp; 382 383 spin_lock(&bswspin); 384 385 for (;;) { 386 if (pfreecnt) { 387 while (*pfreecnt == 0) 388 ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0); 389 } 390 391 /* get a bp from the swap buffer header pool */ 392 if ((bp = TAILQ_FIRST(&bswlist_kva)) != NULL) 393 break; 394 bswneeded_kva = 1; 395 ssleep(&bswneeded_kva, &bswspin, 0, "wswbuf1", 0); 396 /* loop in case someone else grabbed one */ 397 } 398 TAILQ_REMOVE(&bswlist_kva, bp, b_freelist); 399 --pbuf_kva_count; 400 if (pfreecnt) 401 --*pfreecnt; 402 403 spin_unlock(&bswspin); 404 405 initpbuf(bp); 406 KKASSERT(dsched_is_clear_buf_priv(bp)); 407 408 return (bp); 409 } 410 411 /* 412 * Allocate a physical buffer, if one is available. 413 * 414 * Note that there is no NULL hack here - all subsystems using this 415 * call understand how to use pfreecnt. 416 * 417 * No requirements. 418 */ 419 struct buf * 420 trypbuf(int *pfreecnt) 421 { 422 struct buf *bp; 423 424 spin_lock(&bswspin); 425 426 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist_raw)) == NULL) { 427 spin_unlock(&bswspin); 428 return NULL; 429 } 430 TAILQ_REMOVE(&bswlist_raw, bp, b_freelist); 431 --pbuf_raw_count; 432 --*pfreecnt; 433 434 spin_unlock(&bswspin); 435 436 initpbuf(bp); 437 438 return bp; 439 } 440 441 struct buf * 442 trypbuf_kva(int *pfreecnt) 443 { 444 struct buf *bp; 445 446 spin_lock(&bswspin); 447 448 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist_kva)) == NULL) { 449 spin_unlock(&bswspin); 450 return NULL; 451 } 452 TAILQ_REMOVE(&bswlist_kva, bp, b_freelist); 453 --pbuf_kva_count; 454 --*pfreecnt; 455 456 spin_unlock(&bswspin); 457 458 initpbuf(bp); 459 460 return bp; 461 } 462 463 /* 464 * Release a physical buffer 465 * 466 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 467 * relatively soon when the rest of the subsystems get smart about it. XXX 468 * 469 * No requirements. 470 */ 471 void 472 relpbuf(struct buf *bp, int *pfreecnt) 473 { 474 int wake_bsw_kva = 0; 475 int wake_bsw_raw = 0; 476 int wake_freecnt = 0; 477 478 KKASSERT(bp->b_flags & B_PAGING); 479 dsched_exit_buf(bp); 480 481 BUF_UNLOCK(bp); 482 483 spin_lock(&bswspin); 484 if (bp->b_kvabase) { 485 TAILQ_INSERT_HEAD(&bswlist_kva, bp, b_freelist); 486 ++pbuf_kva_count; 487 } else { 488 TAILQ_INSERT_HEAD(&bswlist_raw, bp, b_freelist); 489 ++pbuf_raw_count; 490 } 491 if (bswneeded_kva) { 492 bswneeded_kva = 0; 493 wake_bsw_kva = 1; 494 } 495 if (bswneeded_raw) { 496 bswneeded_raw = 0; 497 wake_bsw_raw = 1; 498 } 499 if (pfreecnt) { 500 if (++*pfreecnt == 1) 501 wake_freecnt = 1; 502 } 503 spin_unlock(&bswspin); 504 505 if (wake_bsw_kva) 506 wakeup(&bswneeded_kva); 507 if (wake_bsw_raw) 508 wakeup(&bswneeded_raw); 509 if (wake_freecnt) 510 wakeup(pfreecnt); 511 } 512