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 boolean_t 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 &devicepagerops, /* OBJT_MGTDEVICE */ 159 &physpagerops, /* OBJT_PHYS */ 160 &deadpagerops /* OBJT_DEAD */ 161 }; 162 163 int npagers = NELEM(pagertab); 164 165 /* 166 * Kernel address space for mapping pages. 167 * Used by pagers where KVAs are needed for IO. 168 * 169 * XXX needs to be large enough to support the number of pending async 170 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size 171 * (MAXPHYS == 64k) if you want to get the most efficiency. 172 */ 173 #define PAGER_MAP_SIZE (8 * 1024 * 1024) 174 175 TAILQ_HEAD(swqueue, buf); 176 177 int pager_map_size = PAGER_MAP_SIZE; 178 struct vm_map pager_map; 179 180 static int bswneeded_raw; 181 static int bswneeded_kva; 182 static long nswbuf_raw; 183 static struct buf *swbuf_raw; 184 static vm_offset_t swapbkva; /* swap buffers kva */ 185 static struct swqueue bswlist_raw; /* without kva */ 186 static struct swqueue bswlist_kva; /* with kva */ 187 static struct spinlock bswspin = SPINLOCK_INITIALIZER(&bswspin, "bswspin"); 188 static int pbuf_raw_count; 189 static int pbuf_kva_count; 190 191 SYSCTL_INT(_vfs, OID_AUTO, pbuf_raw_count, CTLFLAG_RD, &pbuf_raw_count, 0, 192 "Kernel virtual address space reservations"); 193 SYSCTL_INT(_vfs, OID_AUTO, pbuf_kva_count, CTLFLAG_RD, &pbuf_kva_count, 0, 194 "Kernel raw address space reservations"); 195 196 /* 197 * Initialize the swap buffer list. 198 * 199 * Called from the low level boot code only. 200 */ 201 static void 202 vm_pager_init(void *arg __unused) 203 { 204 TAILQ_INIT(&bswlist_raw); 205 TAILQ_INIT(&bswlist_kva); 206 } 207 SYSINIT(vm_mem, SI_BOOT1_VM, SI_ORDER_SECOND, vm_pager_init, NULL) 208 209 /* 210 * Called from the low level boot code only. 211 */ 212 static 213 void 214 vm_pager_bufferinit(void *dummy __unused) 215 { 216 struct buf *bp; 217 long i; 218 219 /* 220 * Reserve KVM space for pbuf data. 221 */ 222 swapbkva = kmem_alloc_pageable(&pager_map, nswbuf * MAXPHYS); 223 if (!swapbkva) 224 panic("Not enough pager_map VM space for physical buffers"); 225 226 /* 227 * Initial pbuf setup. These pbufs have KVA reservations. 228 */ 229 bp = swbuf; 230 for (i = 0; i < nswbuf; ++i, ++bp) { 231 bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva; 232 bp->b_kvasize = MAXPHYS; 233 BUF_LOCKINIT(bp); 234 buf_dep_init(bp); 235 TAILQ_INSERT_HEAD(&bswlist_kva, bp, b_freelist); 236 ++pbuf_kva_count; 237 } 238 239 /* 240 * Initial pbuf setup. These pbufs do not have KVA reservations, 241 * so we can have a lot more of them. These are typically used 242 * to massage low level buf/bio requests. 243 * 244 * NOTE: We use KM_NOTLBSYNC here to reduce unnecessary IPIs 245 * during startup, which can really slow down emulated 246 * systems. 247 */ 248 nswbuf_raw = nbuf * 2; 249 swbuf_raw = (void *)kmem_alloc3(&kernel_map, 250 round_page(nswbuf_raw * sizeof(struct buf)), 251 KM_NOTLBSYNC); 252 smp_invltlb(); 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 SYSINIT(do_vmpg, SI_BOOT2_MACHDEP, SI_ORDER_FIRST, vm_pager_bufferinit, NULL); 268 269 /* 270 * No requirements. 271 */ 272 void 273 vm_pager_deallocate(vm_object_t object) 274 { 275 (*pagertab[object->type]->pgo_dealloc) (object); 276 } 277 278 /* 279 * vm_pager_get_pages() - inline, see vm/vm_pager.h 280 * vm_pager_put_pages() - inline, see vm/vm_pager.h 281 * vm_pager_has_page() - inline, see vm/vm_pager.h 282 * vm_pager_page_inserted() - inline, see vm/vm_pager.h 283 * vm_pager_page_removed() - inline, see vm/vm_pager.h 284 */ 285 286 /* 287 * Search the specified pager object list for an object with the 288 * specified handle. If an object with the specified handle is found, 289 * increase its reference count and return it. Otherwise, return NULL. 290 * 291 * The pager object list must be locked. 292 */ 293 vm_object_t 294 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle) 295 { 296 vm_object_t object; 297 298 TAILQ_FOREACH(object, pg_list, pager_object_list) { 299 if (object->handle == handle) { 300 VM_OBJECT_LOCK(object); 301 if ((object->flags & OBJ_DEAD) == 0) { 302 vm_object_reference_locked(object); 303 VM_OBJECT_UNLOCK(object); 304 break; 305 } 306 VM_OBJECT_UNLOCK(object); 307 } 308 } 309 return (object); 310 } 311 312 /* 313 * Initialize a physical buffer. 314 * 315 * No requirements. 316 */ 317 static void 318 initpbuf(struct buf *bp) 319 { 320 bp->b_qindex = 0; /* BQUEUE_NONE */ 321 bp->b_data = bp->b_kvabase; /* NULL if pbuf sans kva */ 322 bp->b_flags = B_PAGING; 323 bp->b_cmd = BUF_CMD_DONE; 324 bp->b_error = 0; 325 bp->b_bcount = 0; 326 bp->b_bufsize = MAXPHYS; 327 initbufbio(bp); 328 xio_init(&bp->b_xio); 329 BUF_LOCK(bp, LK_EXCLUSIVE); 330 } 331 332 /* 333 * Allocate a physical buffer 334 * 335 * There are a limited number (nswbuf) of physical buffers. We need 336 * to make sure that no single subsystem is able to hog all of them, 337 * so each subsystem implements a counter which is typically initialized 338 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and 339 * increments it on release, and blocks if the counter hits zero. A 340 * subsystem may initialize the counter to -1 to disable the feature, 341 * but it must still be sure to match up all uses of getpbuf() with 342 * relpbuf() using the same variable. 343 * 344 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 345 * relatively soon when the rest of the subsystems get smart about it. XXX 346 * 347 * Physical buffers can be with or without KVA space reserved. There 348 * are severe limitations on the ones with KVA reserved, and fewer 349 * limitations on the ones without. getpbuf() gets one without, 350 * getpbuf_kva() gets one with. 351 * 352 * No requirements. 353 */ 354 struct buf * 355 getpbuf(int *pfreecnt) 356 { 357 struct buf *bp; 358 359 spin_lock(&bswspin); 360 361 for (;;) { 362 if (pfreecnt) { 363 while (*pfreecnt == 0) 364 ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0); 365 } 366 367 /* get a bp from the swap buffer header pool */ 368 if ((bp = TAILQ_FIRST(&bswlist_raw)) != NULL) 369 break; 370 bswneeded_raw = 1; 371 ssleep(&bswneeded_raw, &bswspin, 0, "wswbuf1", 0); 372 /* loop in case someone else grabbed one */ 373 } 374 TAILQ_REMOVE(&bswlist_raw, bp, b_freelist); 375 --pbuf_raw_count; 376 if (pfreecnt) 377 --*pfreecnt; 378 379 spin_unlock(&bswspin); 380 381 initpbuf(bp); 382 KKASSERT(dsched_is_clear_buf_priv(bp)); 383 384 return (bp); 385 } 386 387 struct buf * 388 getpbuf_kva(int *pfreecnt) 389 { 390 struct buf *bp; 391 392 spin_lock(&bswspin); 393 394 for (;;) { 395 if (pfreecnt) { 396 while (*pfreecnt == 0) 397 ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0); 398 } 399 400 /* get a bp from the swap buffer header pool */ 401 if ((bp = TAILQ_FIRST(&bswlist_kva)) != NULL) 402 break; 403 bswneeded_kva = 1; 404 ssleep(&bswneeded_kva, &bswspin, 0, "wswbuf1", 0); 405 /* loop in case someone else grabbed one */ 406 } 407 TAILQ_REMOVE(&bswlist_kva, bp, b_freelist); 408 --pbuf_kva_count; 409 if (pfreecnt) 410 --*pfreecnt; 411 412 spin_unlock(&bswspin); 413 414 initpbuf(bp); 415 KKASSERT(dsched_is_clear_buf_priv(bp)); 416 417 return (bp); 418 } 419 420 /* 421 * Allocate a physical buffer, if one is available. 422 * 423 * Note that there is no NULL hack here - all subsystems using this 424 * call understand how to use pfreecnt. 425 * 426 * No requirements. 427 */ 428 struct buf * 429 trypbuf(int *pfreecnt) 430 { 431 struct buf *bp; 432 433 spin_lock(&bswspin); 434 435 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist_raw)) == NULL) { 436 spin_unlock(&bswspin); 437 return NULL; 438 } 439 TAILQ_REMOVE(&bswlist_raw, bp, b_freelist); 440 --pbuf_raw_count; 441 --*pfreecnt; 442 443 spin_unlock(&bswspin); 444 445 initpbuf(bp); 446 447 return bp; 448 } 449 450 struct buf * 451 trypbuf_kva(int *pfreecnt) 452 { 453 struct buf *bp; 454 455 spin_lock(&bswspin); 456 457 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist_kva)) == NULL) { 458 spin_unlock(&bswspin); 459 return NULL; 460 } 461 TAILQ_REMOVE(&bswlist_kva, bp, b_freelist); 462 --pbuf_kva_count; 463 --*pfreecnt; 464 465 spin_unlock(&bswspin); 466 467 initpbuf(bp); 468 469 return bp; 470 } 471 472 /* 473 * Release a physical buffer 474 * 475 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed 476 * relatively soon when the rest of the subsystems get smart about it. XXX 477 * 478 * No requirements. 479 */ 480 void 481 relpbuf(struct buf *bp, int *pfreecnt) 482 { 483 int wake_bsw_kva = 0; 484 int wake_bsw_raw = 0; 485 int wake_freecnt = 0; 486 487 KKASSERT(bp->b_flags & B_PAGING); 488 dsched_exit_buf(bp); 489 490 BUF_UNLOCK(bp); 491 492 spin_lock(&bswspin); 493 if (bp->b_kvabase) { 494 TAILQ_INSERT_HEAD(&bswlist_kva, bp, b_freelist); 495 ++pbuf_kva_count; 496 } else { 497 TAILQ_INSERT_HEAD(&bswlist_raw, bp, b_freelist); 498 ++pbuf_raw_count; 499 } 500 if (bswneeded_kva) { 501 bswneeded_kva = 0; 502 wake_bsw_kva = 1; 503 } 504 if (bswneeded_raw) { 505 bswneeded_raw = 0; 506 wake_bsw_raw = 1; 507 } 508 if (pfreecnt) { 509 if (++*pfreecnt == 1) 510 wake_freecnt = 1; 511 } 512 spin_unlock(&bswspin); 513 514 if (wake_bsw_kva) 515 wakeup(&bswneeded_kva); 516 if (wake_bsw_raw) 517 wakeup(&bswneeded_raw); 518 if (wake_freecnt) 519 wakeup(pfreecnt); 520 } 521 522 void 523 pbuf_adjcount(int *pfreecnt, int n) 524 { 525 if (n) { 526 spin_lock(&bswspin); 527 *pfreecnt += n; 528 spin_unlock(&bswspin); 529 wakeup(pfreecnt); 530 } 531 } 532