vm_kern.c revision 12662
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 1.17 1995/11/13 10:53:38 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 91vm_map_t buffer_map; 92vm_map_t kernel_map; 93vm_map_t kmem_map; 94vm_map_t mb_map; 95int mb_map_full; 96vm_map_t io_map; 97vm_map_t clean_map; 98vm_map_t phys_map; 99vm_map_t exec_map; 100vm_map_t u_map; 101 102/* 103 * kmem_alloc_pageable: 104 * 105 * Allocate pageable memory to the kernel's address map. 106 * "map" must be kernel_map or a submap of kernel_map. 107 */ 108 109vm_offset_t 110kmem_alloc_pageable(map, size) 111 vm_map_t map; 112 register vm_size_t size; 113{ 114 vm_offset_t addr; 115 register int result; 116 117 size = round_page(size); 118 addr = vm_map_min(map); 119 result = vm_map_find(map, NULL, (vm_offset_t) 0, 120 &addr, size, TRUE); 121 if (result != KERN_SUCCESS) { 122 return (0); 123 } 124 return (addr); 125} 126 127/* 128 * Allocate wired-down memory in the kernel's address map 129 * or a submap. 130 */ 131vm_offset_t 132kmem_alloc(map, size) 133 register vm_map_t map; 134 register vm_size_t size; 135{ 136 vm_offset_t addr; 137 register vm_offset_t offset; 138 vm_offset_t i; 139 140 size = round_page(size); 141 142 /* 143 * Use the kernel object for wired-down kernel pages. Assume that no 144 * region of the kernel object is referenced more than once. 145 */ 146 147 /* 148 * Locate sufficient space in the map. This will give us the final 149 * virtual address for the new memory, and thus will tell us the 150 * offset within the kernel map. 151 */ 152 vm_map_lock(map); 153 if (vm_map_findspace(map, 0, size, &addr)) { 154 vm_map_unlock(map); 155 return (0); 156 } 157 offset = addr - VM_MIN_KERNEL_ADDRESS; 158 vm_object_reference(kernel_object); 159 vm_map_insert(map, kernel_object, offset, addr, addr + size); 160 vm_map_unlock(map); 161 162 /* 163 * Guarantee that there are pages already in this object before 164 * calling vm_map_pageable. This is to prevent the following 165 * scenario: 166 * 167 * 1) Threads have swapped out, so that there is a pager for the 168 * kernel_object. 2) The kmsg zone is empty, and so we are 169 * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 170 * there is no page, but there is a pager, so we call 171 * pager_data_request. But the kmsg zone is empty, so we must 172 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 173 * we get the data back from the pager, it will be (very stale) 174 * non-zero data. kmem_alloc is defined to return zero-filled memory. 175 * 176 * We're intentionally not activating the pages we allocate to prevent a 177 * race with page-out. vm_map_pageable will wire the pages. 178 */ 179 180 for (i = 0; i < size; i += PAGE_SIZE) { 181 vm_page_t mem; 182 183 while ((mem = vm_page_alloc(kernel_object, offset + i, (VM_ALLOC_NORMAL|VM_ALLOC_ZERO))) == NULL) { 184 VM_WAIT; 185 } 186 if ((mem->flags & PG_ZERO) == 0) 187 vm_page_zero_fill(mem); 188 mem->flags &= ~(PG_BUSY|PG_ZERO); 189 mem->valid = VM_PAGE_BITS_ALL; 190 } 191 192 /* 193 * And finally, mark the data as non-pageable. 194 */ 195 196 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 197 198 /* 199 * Try to coalesce the map 200 */ 201 vm_map_simplify(map, addr); 202 203 return (addr); 204} 205 206/* 207 * kmem_free: 208 * 209 * Release a region of kernel virtual memory allocated 210 * with kmem_alloc, and return the physical pages 211 * associated with that region. 212 */ 213void 214kmem_free(map, addr, size) 215 vm_map_t map; 216 register vm_offset_t addr; 217 vm_size_t size; 218{ 219 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 220} 221 222/* 223 * kmem_suballoc: 224 * 225 * Allocates a map to manage a subrange 226 * of the kernel virtual address space. 227 * 228 * Arguments are as follows: 229 * 230 * parent Map to take range from 231 * size Size of range to find 232 * min, max Returned endpoints of map 233 * pageable Can the region be paged 234 */ 235vm_map_t 236kmem_suballoc(parent, min, max, size, pageable) 237 register vm_map_t parent; 238 vm_offset_t *min, *max; 239 register vm_size_t size; 240 boolean_t pageable; 241{ 242 register int ret; 243 vm_map_t result; 244 245 size = round_page(size); 246 247 *min = (vm_offset_t) vm_map_min(parent); 248 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 249 min, size, TRUE); 250 if (ret != KERN_SUCCESS) { 251 printf("kmem_suballoc: bad status return of %d.\n", ret); 252 panic("kmem_suballoc"); 253 } 254 *max = *min + size; 255 pmap_reference(vm_map_pmap(parent)); 256 result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 257 if (result == NULL) 258 panic("kmem_suballoc: cannot create submap"); 259 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 260 panic("kmem_suballoc: unable to change range to submap"); 261 return (result); 262} 263 264/* 265 * Allocate wired-down memory in the kernel's address map for the higher 266 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 267 * kmem_alloc() because we may need to allocate memory at interrupt 268 * level where we cannot block (canwait == FALSE). 269 * 270 * This routine has its own private kernel submap (kmem_map) and object 271 * (kmem_object). This, combined with the fact that only malloc uses 272 * this routine, ensures that we will never block in map or object waits. 273 * 274 * Note that this still only works in a uni-processor environment and 275 * when called at splhigh(). 276 * 277 * We don't worry about expanding the map (adding entries) since entries 278 * for wired maps are statically allocated. 279 */ 280vm_offset_t 281kmem_malloc(map, size, waitflag) 282 register vm_map_t map; 283 register vm_size_t size; 284 boolean_t waitflag; 285{ 286 register vm_offset_t offset, i; 287 vm_map_entry_t entry; 288 vm_offset_t addr; 289 vm_page_t m; 290 291 if (map != kmem_map && map != mb_map) 292 panic("kmem_malloc: map != {kmem,mb}_map"); 293 294 size = round_page(size); 295 addr = vm_map_min(map); 296 297 /* 298 * Locate sufficient space in the map. This will give us the final 299 * virtual address for the new memory, and thus will tell us the 300 * offset within the kernel map. 301 */ 302 vm_map_lock(map); 303 if (vm_map_findspace(map, 0, size, &addr)) { 304 vm_map_unlock(map); 305 if (map == mb_map) { 306 mb_map_full = TRUE; 307 log(LOG_ERR, "mb_map full\n"); 308 return (0); 309 } 310 if (waitflag == M_WAITOK) 311 panic("kmem_malloc: kmem_map too small"); 312 return (0); 313 } 314 offset = addr - vm_map_min(kmem_map); 315 vm_object_reference(kmem_object); 316 vm_map_insert(map, kmem_object, offset, addr, addr + size); 317 318 /* 319 * If we can wait, just mark the range as wired (will fault pages as 320 * necessary). 321 */ 322 if (waitflag == M_WAITOK) { 323 vm_map_unlock(map); 324 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, 325 FALSE); 326 vm_map_simplify(map, addr); 327 return (addr); 328 } 329 /* 330 * If we cannot wait then we must allocate all memory up front, 331 * pulling it off the active queue to prevent pageout. 332 */ 333 for (i = 0; i < size; i += PAGE_SIZE) { 334 m = vm_page_alloc(kmem_object, offset + i, 335 (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 336 337 /* 338 * Ran out of space, free everything up and return. Don't need 339 * to lock page queues here as we know that the pages we got 340 * aren't on any queues. 341 */ 342 if (m == NULL) { 343 while (i != 0) { 344 i -= PAGE_SIZE; 345 m = vm_page_lookup(kmem_object, offset + i); 346 vm_page_free(m); 347 } 348 vm_map_delete(map, addr, addr + size); 349 vm_map_unlock(map); 350 return (0); 351 } 352 m->flags &= ~(PG_BUSY|PG_ZERO); 353 m->valid = VM_PAGE_BITS_ALL; 354 } 355 356 /* 357 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 358 * be able to extend the previous entry so there will be a new entry 359 * exactly corresponding to this address range and it will have 360 * wired_count == 0. 361 */ 362 if (!vm_map_lookup_entry(map, addr, &entry) || 363 entry->start != addr || entry->end != addr + size || 364 entry->wired_count) 365 panic("kmem_malloc: entry not found or misaligned"); 366 entry->wired_count++; 367 368 /* 369 * Loop thru pages, entering them in the pmap. (We cannot add them to 370 * the wired count without wrapping the vm_page_queue_lock in 371 * splimp...) 372 */ 373 for (i = 0; i < size; i += PAGE_SIZE) { 374 m = vm_page_lookup(kmem_object, offset + i); 375 pmap_kenter(addr + i, VM_PAGE_TO_PHYS(m)); 376 } 377 vm_map_unlock(map); 378 379 vm_map_simplify(map, addr); 380 return (addr); 381} 382 383/* 384 * kmem_alloc_wait 385 * 386 * Allocates pageable memory from a sub-map of the kernel. If the submap 387 * has no room, the caller sleeps waiting for more memory in the submap. 388 * 389 */ 390vm_offset_t 391kmem_alloc_wait(map, size) 392 vm_map_t map; 393 vm_size_t size; 394{ 395 vm_offset_t addr; 396 397 size = round_page(size); 398 399 for (;;) { 400 /* 401 * To make this work for more than one map, use the map's lock 402 * to lock out sleepers/wakers. 403 */ 404 vm_map_lock(map); 405 if (vm_map_findspace(map, 0, size, &addr) == 0) 406 break; 407 /* no space now; see if we can ever get space */ 408 if (vm_map_max(map) - vm_map_min(map) < size) { 409 vm_map_unlock(map); 410 return (0); 411 } 412 vm_map_unlock(map); 413 tsleep(map, PVM, "kmaw", 0); 414 } 415 vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size); 416 vm_map_unlock(map); 417 return (addr); 418} 419 420/* 421 * kmem_free_wakeup 422 * 423 * Returns memory to a submap of the kernel, and wakes up any processes 424 * waiting for memory in that map. 425 */ 426void 427kmem_free_wakeup(map, addr, size) 428 vm_map_t map; 429 vm_offset_t addr; 430 vm_size_t size; 431{ 432 vm_map_lock(map); 433 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 434 wakeup(map); 435 vm_map_unlock(map); 436} 437 438/* 439 * Create the kernel map; insert a mapping covering kernel text, data, bss, 440 * and all space allocated thus far (`boostrap' data). The new map will thus 441 * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 442 * the range between `start' and `end' as free. 443 */ 444void 445kmem_init(start, end) 446 vm_offset_t start, end; 447{ 448 register vm_map_t m; 449 450 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 451 vm_map_lock(m); 452 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 453 kernel_map = m; 454 (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 455 VM_MIN_KERNEL_ADDRESS, start); 456 /* ... and ending with the completion of the above `insert' */ 457 vm_map_unlock(m); 458} 459