vm_kern.c revision 12726
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_kern.c 8.3 (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 * $Id: vm_kern.c,v 1.18 1995/12/07 12:48:13 davidg Exp $ 65 */ 66 67/* 68 * Kernel memory management. 69 */ 70 71#include <sys/param.h> 72#include <sys/systm.h> 73#include <sys/kernel.h> 74#include <sys/proc.h> 75#include <sys/malloc.h> 76#include <sys/syslog.h> 77#include <sys/queue.h> 78#include <sys/vmmeter.h> 79 80#include <vm/vm.h> 81#include <vm/vm_param.h> 82#include <vm/vm_prot.h> 83#include <vm/lock.h> 84#include <vm/pmap.h> 85#include <vm/vm_map.h> 86#include <vm/vm_object.h> 87#include <vm/vm_page.h> 88#include <vm/vm_pageout.h> 89#include <vm/vm_kern.h> 90#include <vm/vm_extern.h> 91 92vm_map_t buffer_map; 93vm_map_t kernel_map; 94vm_map_t kmem_map; 95vm_map_t mb_map; 96int mb_map_full; 97vm_map_t io_map; 98vm_map_t clean_map; 99vm_map_t phys_map; 100vm_map_t exec_map; 101vm_map_t u_map; 102 103/* 104 * kmem_alloc_pageable: 105 * 106 * Allocate pageable memory to the kernel's address map. 107 * "map" must be kernel_map or a submap of kernel_map. 108 */ 109 110vm_offset_t 111kmem_alloc_pageable(map, size) 112 vm_map_t map; 113 register vm_size_t size; 114{ 115 vm_offset_t addr; 116 register int result; 117 118 size = round_page(size); 119 addr = vm_map_min(map); 120 result = vm_map_find(map, NULL, (vm_offset_t) 0, 121 &addr, size, TRUE); 122 if (result != KERN_SUCCESS) { 123 return (0); 124 } 125 return (addr); 126} 127 128/* 129 * Allocate wired-down memory in the kernel's address map 130 * or a submap. 131 */ 132vm_offset_t 133kmem_alloc(map, size) 134 register vm_map_t map; 135 register vm_size_t size; 136{ 137 vm_offset_t addr; 138 register vm_offset_t offset; 139 vm_offset_t i; 140 141 size = round_page(size); 142 143 /* 144 * Use the kernel object for wired-down kernel pages. Assume that no 145 * region of the kernel object is referenced more than once. 146 */ 147 148 /* 149 * Locate sufficient space in the map. This will give us the final 150 * virtual address for the new memory, and thus will tell us the 151 * offset within the kernel map. 152 */ 153 vm_map_lock(map); 154 if (vm_map_findspace(map, 0, size, &addr)) { 155 vm_map_unlock(map); 156 return (0); 157 } 158 offset = addr - VM_MIN_KERNEL_ADDRESS; 159 vm_object_reference(kernel_object); 160 vm_map_insert(map, kernel_object, offset, addr, addr + size); 161 vm_map_unlock(map); 162 163 /* 164 * Guarantee that there are pages already in this object before 165 * calling vm_map_pageable. This is to prevent the following 166 * scenario: 167 * 168 * 1) Threads have swapped out, so that there is a pager for the 169 * kernel_object. 2) The kmsg zone is empty, and so we are 170 * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 171 * there is no page, but there is a pager, so we call 172 * pager_data_request. But the kmsg zone is empty, so we must 173 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 174 * we get the data back from the pager, it will be (very stale) 175 * non-zero data. kmem_alloc is defined to return zero-filled memory. 176 * 177 * We're intentionally not activating the pages we allocate to prevent a 178 * race with page-out. vm_map_pageable will wire the pages. 179 */ 180 181 for (i = 0; i < size; i += PAGE_SIZE) { 182 vm_page_t mem; 183 184 while ((mem = vm_page_alloc(kernel_object, offset + i, (VM_ALLOC_NORMAL|VM_ALLOC_ZERO))) == NULL) { 185 VM_WAIT; 186 } 187 if ((mem->flags & PG_ZERO) == 0) 188 vm_page_zero_fill(mem); 189 mem->flags &= ~(PG_BUSY|PG_ZERO); 190 mem->valid = VM_PAGE_BITS_ALL; 191 } 192 193 /* 194 * And finally, mark the data as non-pageable. 195 */ 196 197 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 198 199 /* 200 * Try to coalesce the map 201 */ 202 vm_map_simplify(map, addr); 203 204 return (addr); 205} 206 207/* 208 * kmem_free: 209 * 210 * Release a region of kernel virtual memory allocated 211 * with kmem_alloc, and return the physical pages 212 * associated with that region. 213 */ 214void 215kmem_free(map, addr, size) 216 vm_map_t map; 217 register vm_offset_t addr; 218 vm_size_t size; 219{ 220 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 221} 222 223/* 224 * kmem_suballoc: 225 * 226 * Allocates a map to manage a subrange 227 * of the kernel virtual address space. 228 * 229 * Arguments are as follows: 230 * 231 * parent Map to take range from 232 * size Size of range to find 233 * min, max Returned endpoints of map 234 * pageable Can the region be paged 235 */ 236vm_map_t 237kmem_suballoc(parent, min, max, size, pageable) 238 register vm_map_t parent; 239 vm_offset_t *min, *max; 240 register vm_size_t size; 241 boolean_t pageable; 242{ 243 register int ret; 244 vm_map_t result; 245 246 size = round_page(size); 247 248 *min = (vm_offset_t) vm_map_min(parent); 249 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 250 min, size, TRUE); 251 if (ret != KERN_SUCCESS) { 252 printf("kmem_suballoc: bad status return of %d.\n", ret); 253 panic("kmem_suballoc"); 254 } 255 *max = *min + size; 256 pmap_reference(vm_map_pmap(parent)); 257 result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 258 if (result == NULL) 259 panic("kmem_suballoc: cannot create submap"); 260 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 261 panic("kmem_suballoc: unable to change range to submap"); 262 return (result); 263} 264 265/* 266 * Allocate wired-down memory in the kernel's address map for the higher 267 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 268 * kmem_alloc() because we may need to allocate memory at interrupt 269 * level where we cannot block (canwait == FALSE). 270 * 271 * This routine has its own private kernel submap (kmem_map) and object 272 * (kmem_object). This, combined with the fact that only malloc uses 273 * this routine, ensures that we will never block in map or object waits. 274 * 275 * Note that this still only works in a uni-processor environment and 276 * when called at splhigh(). 277 * 278 * We don't worry about expanding the map (adding entries) since entries 279 * for wired maps are statically allocated. 280 */ 281vm_offset_t 282kmem_malloc(map, size, waitflag) 283 register vm_map_t map; 284 register vm_size_t size; 285 boolean_t waitflag; 286{ 287 register vm_offset_t offset, i; 288 vm_map_entry_t entry; 289 vm_offset_t addr; 290 vm_page_t m; 291 292 if (map != kmem_map && map != mb_map) 293 panic("kmem_malloc: map != {kmem,mb}_map"); 294 295 size = round_page(size); 296 addr = vm_map_min(map); 297 298 /* 299 * Locate sufficient space in the map. This will give us the final 300 * virtual address for the new memory, and thus will tell us the 301 * offset within the kernel map. 302 */ 303 vm_map_lock(map); 304 if (vm_map_findspace(map, 0, size, &addr)) { 305 vm_map_unlock(map); 306 if (map == mb_map) { 307 mb_map_full = TRUE; 308 log(LOG_ERR, "mb_map full\n"); 309 return (0); 310 } 311 if (waitflag == M_WAITOK) 312 panic("kmem_malloc: kmem_map too small"); 313 return (0); 314 } 315 offset = addr - vm_map_min(kmem_map); 316 vm_object_reference(kmem_object); 317 vm_map_insert(map, kmem_object, offset, addr, addr + size); 318 319 /* 320 * If we can wait, just mark the range as wired (will fault pages as 321 * necessary). 322 */ 323 if (waitflag == M_WAITOK) { 324 vm_map_unlock(map); 325 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, 326 FALSE); 327 vm_map_simplify(map, addr); 328 return (addr); 329 } 330 /* 331 * If we cannot wait then we must allocate all memory up front, 332 * pulling it off the active queue to prevent pageout. 333 */ 334 for (i = 0; i < size; i += PAGE_SIZE) { 335 m = vm_page_alloc(kmem_object, offset + i, 336 (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 337 338 /* 339 * Ran out of space, free everything up and return. Don't need 340 * to lock page queues here as we know that the pages we got 341 * aren't on any queues. 342 */ 343 if (m == NULL) { 344 while (i != 0) { 345 i -= PAGE_SIZE; 346 m = vm_page_lookup(kmem_object, offset + i); 347 vm_page_free(m); 348 } 349 vm_map_delete(map, addr, addr + size); 350 vm_map_unlock(map); 351 return (0); 352 } 353 m->flags &= ~(PG_BUSY|PG_ZERO); 354 m->valid = VM_PAGE_BITS_ALL; 355 } 356 357 /* 358 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 359 * be able to extend the previous entry so there will be a new entry 360 * exactly corresponding to this address range and it will have 361 * wired_count == 0. 362 */ 363 if (!vm_map_lookup_entry(map, addr, &entry) || 364 entry->start != addr || entry->end != addr + size || 365 entry->wired_count) 366 panic("kmem_malloc: entry not found or misaligned"); 367 entry->wired_count++; 368 369 /* 370 * Loop thru pages, entering them in the pmap. (We cannot add them to 371 * the wired count without wrapping the vm_page_queue_lock in 372 * splimp...) 373 */ 374 for (i = 0; i < size; i += PAGE_SIZE) { 375 m = vm_page_lookup(kmem_object, offset + i); 376 pmap_kenter(addr + i, VM_PAGE_TO_PHYS(m)); 377 } 378 vm_map_unlock(map); 379 380 vm_map_simplify(map, addr); 381 return (addr); 382} 383 384/* 385 * kmem_alloc_wait 386 * 387 * Allocates pageable memory from a sub-map of the kernel. If the submap 388 * has no room, the caller sleeps waiting for more memory in the submap. 389 * 390 */ 391vm_offset_t 392kmem_alloc_wait(map, size) 393 vm_map_t map; 394 vm_size_t size; 395{ 396 vm_offset_t addr; 397 398 size = round_page(size); 399 400 for (;;) { 401 /* 402 * To make this work for more than one map, use the map's lock 403 * to lock out sleepers/wakers. 404 */ 405 vm_map_lock(map); 406 if (vm_map_findspace(map, 0, size, &addr) == 0) 407 break; 408 /* no space now; see if we can ever get space */ 409 if (vm_map_max(map) - vm_map_min(map) < size) { 410 vm_map_unlock(map); 411 return (0); 412 } 413 vm_map_unlock(map); 414 tsleep(map, PVM, "kmaw", 0); 415 } 416 vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size); 417 vm_map_unlock(map); 418 return (addr); 419} 420 421/* 422 * kmem_free_wakeup 423 * 424 * Returns memory to a submap of the kernel, and wakes up any processes 425 * waiting for memory in that map. 426 */ 427void 428kmem_free_wakeup(map, addr, size) 429 vm_map_t map; 430 vm_offset_t addr; 431 vm_size_t size; 432{ 433 vm_map_lock(map); 434 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 435 wakeup(map); 436 vm_map_unlock(map); 437} 438 439/* 440 * Create the kernel map; insert a mapping covering kernel text, data, bss, 441 * and all space allocated thus far (`boostrap' data). The new map will thus 442 * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 443 * the range between `start' and `end' as free. 444 */ 445void 446kmem_init(start, end) 447 vm_offset_t start, end; 448{ 449 register vm_map_t m; 450 451 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 452 vm_map_lock(m); 453 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 454 kernel_map = m; 455 (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 456 VM_MIN_KERNEL_ADDRESS, start); 457 /* ... and ending with the completion of the above `insert' */ 458 vm_map_unlock(m); 459} 460