vm_kern.c revision 7066
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 1.11 1995/02/21 01:22:45 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 78#include <vm/vm.h> 79#include <vm/vm_page.h> 80#include <vm/vm_pageout.h> 81#include <vm/vm_kern.h> 82 83vm_map_t buffer_map; 84vm_map_t kernel_map; 85vm_map_t kmem_map; 86vm_map_t mb_map; 87vm_map_t io_map; 88vm_map_t clean_map; 89vm_map_t pager_map; 90vm_map_t phys_map; 91vm_map_t exec_map; 92vm_map_t u_map; 93extern int mb_map_full; 94 95/* 96 * kmem_alloc_pageable: 97 * 98 * Allocate pageable memory to the kernel's address map. 99 * map must be "kernel_map" below. 100 */ 101 102vm_offset_t 103kmem_alloc_pageable(map, size) 104 vm_map_t map; 105 register vm_size_t size; 106{ 107 vm_offset_t addr; 108 register int result; 109 110#if 0 111 if (map != kernel_map) 112 panic("kmem_alloc_pageable: not called with kernel_map"); 113#endif 114 115 size = round_page(size); 116 117 addr = vm_map_min(map); 118 result = vm_map_find(map, NULL, (vm_offset_t) 0, 119 &addr, size, TRUE); 120 if (result != KERN_SUCCESS) { 121 return (0); 122 } 123 return (addr); 124} 125 126/* 127 * Allocate wired-down memory in the kernel's address map 128 * or a submap. 129 */ 130vm_offset_t 131kmem_alloc(map, size) 132 register vm_map_t map; 133 register vm_size_t size; 134{ 135 vm_offset_t addr; 136 register vm_offset_t offset; 137 vm_offset_t i; 138 139 size = round_page(size); 140 141 /* 142 * Use the kernel object for wired-down kernel pages. Assume that no 143 * region of the kernel object is referenced more than once. 144 */ 145 146 /* 147 * Locate sufficient space in the map. This will give us the final 148 * virtual address for the new memory, and thus will tell us the 149 * offset within the kernel map. 150 */ 151 vm_map_lock(map); 152 if (vm_map_findspace(map, 0, size, &addr)) { 153 vm_map_unlock(map); 154 return (0); 155 } 156 offset = addr - VM_MIN_KERNEL_ADDRESS; 157 vm_object_reference(kernel_object); 158 vm_map_insert(map, kernel_object, offset, addr, addr + size); 159 vm_map_unlock(map); 160 161 /* 162 * Guarantee that there are pages already in this object before 163 * calling vm_map_pageable. This is to prevent the following 164 * scenario: 165 * 166 * 1) Threads have swapped out, so that there is a pager for the 167 * kernel_object. 2) The kmsg zone is empty, and so we are 168 * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 169 * there is no page, but there is a pager, so we call 170 * pager_data_request. But the kmsg zone is empty, so we must 171 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 172 * we get the data back from the pager, it will be (very stale) 173 * non-zero data. kmem_alloc is defined to return zero-filled memory. 174 * 175 * We're intentionally not activating the pages we allocate to prevent a 176 * race with page-out. vm_map_pageable will wire the pages. 177 */ 178 179 vm_object_lock(kernel_object); 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)) == NULL) { 184 vm_object_unlock(kernel_object); 185 VM_WAIT; 186 vm_object_lock(kernel_object); 187 } 188 vm_page_zero_fill(mem); 189 mem->flags &= ~PG_BUSY; 190 mem->valid = VM_PAGE_BITS_ALL; 191 } 192 vm_object_unlock(kernel_object); 193 194 /* 195 * And finally, mark the data as non-pageable. 196 */ 197 198 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 199 200 /* 201 * Try to coalesce the map 202 */ 203 vm_map_simplify(map, addr); 204 205 return (addr); 206} 207 208/* 209 * kmem_free: 210 * 211 * Release a region of kernel virtual memory allocated 212 * with kmem_alloc, and return the physical pages 213 * associated with that region. 214 */ 215void 216kmem_free(map, addr, size) 217 vm_map_t map; 218 register vm_offset_t addr; 219 vm_size_t size; 220{ 221 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 222} 223 224/* 225 * kmem_suballoc: 226 * 227 * Allocates a map to manage a subrange 228 * of the kernel virtual address space. 229 * 230 * Arguments are as follows: 231 * 232 * parent Map to take range from 233 * size Size of range to find 234 * min, max Returned endpoints of map 235 * pageable Can the region be paged 236 */ 237vm_map_t 238kmem_suballoc(parent, min, max, size, pageable) 239 register vm_map_t parent; 240 vm_offset_t *min, *max; 241 register vm_size_t size; 242 boolean_t pageable; 243{ 244 register int ret; 245 vm_map_t result; 246 247 size = round_page(size); 248 249 *min = (vm_offset_t) vm_map_min(parent); 250 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 251 min, size, TRUE); 252 if (ret != KERN_SUCCESS) { 253 printf("kmem_suballoc: bad status return of %d.\n", ret); 254 panic("kmem_suballoc"); 255 } 256 *max = *min + size; 257 pmap_reference(vm_map_pmap(parent)); 258 result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 259 if (result == NULL) 260 panic("kmem_suballoc: cannot create submap"); 261 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 262 panic("kmem_suballoc: unable to change range to submap"); 263 return (result); 264} 265 266/* 267 * Allocate wired-down memory in the kernel's address map for the higher 268 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 269 * kmem_alloc() because we may need to allocate memory at interrupt 270 * level where we cannot block (canwait == FALSE). 271 * 272 * This routine has its own private kernel submap (kmem_map) and object 273 * (kmem_object). This, combined with the fact that only malloc uses 274 * this routine, ensures that we will never block in map or object waits. 275 * 276 * Note that this still only works in a uni-processor environment and 277 * when called at splhigh(). 278 * 279 * We don't worry about expanding the map (adding entries) since entries 280 * for wired maps are statically allocated. 281 */ 282vm_offset_t 283kmem_malloc(map, size, waitflag) 284 register vm_map_t map; 285 register vm_size_t size; 286 boolean_t waitflag; 287{ 288 register vm_offset_t offset, i; 289 vm_map_entry_t entry; 290 vm_offset_t addr; 291 vm_page_t m; 292 293 if (map != kmem_map && map != mb_map) 294 panic("kmem_malloc: map != {kmem,mb}_map"); 295 296 size = round_page(size); 297 addr = vm_map_min(map); 298 299 /* 300 * Locate sufficient space in the map. This will give us the final 301 * virtual address for the new memory, and thus will tell us the 302 * offset within the kernel map. 303 */ 304 vm_map_lock(map); 305 if (vm_map_findspace(map, 0, size, &addr)) { 306 vm_map_unlock(map); 307 if (map == mb_map) { 308 mb_map_full = TRUE; 309 log(LOG_ERR, "mb_map full\n"); 310 return (0); 311 } 312 if (waitflag == M_WAITOK) 313 panic("kmem_malloc: kmem_map too small"); 314 return (0); 315 } 316 offset = addr - vm_map_min(kmem_map); 317 vm_object_reference(kmem_object); 318 vm_map_insert(map, kmem_object, offset, addr, addr + size); 319 320 /* 321 * If we can wait, just mark the range as wired (will fault pages as 322 * necessary). 323 */ 324 if (waitflag == M_WAITOK) { 325 vm_map_unlock(map); 326 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, 327 FALSE); 328 vm_map_simplify(map, addr); 329 return (addr); 330 } 331 /* 332 * If we cannot wait then we must allocate all memory up front, 333 * pulling it off the active queue to prevent pageout. 334 */ 335 vm_object_lock(kmem_object); 336 for (i = 0; i < size; i += PAGE_SIZE) { 337 m = vm_page_alloc(kmem_object, offset + i, 338 (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 339 340 /* 341 * Ran out of space, free everything up and return. Don't need 342 * to lock page queues here as we know that the pages we got 343 * aren't on any queues. 344 */ 345 if (m == NULL) { 346 while (i != 0) { 347 i -= PAGE_SIZE; 348 m = vm_page_lookup(kmem_object, offset + i); 349 vm_page_free(m); 350 } 351 vm_object_unlock(kmem_object); 352 vm_map_delete(map, addr, addr + size); 353 vm_map_unlock(map); 354 return (0); 355 } 356#if 0 357 vm_page_zero_fill(m); 358#endif 359 m->flags &= ~PG_BUSY; 360 m->valid = VM_PAGE_BITS_ALL; 361 } 362 vm_object_unlock(kmem_object); 363 364 /* 365 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 366 * be able to extend the previous entry so there will be a new entry 367 * exactly corresponding to this address range and it will have 368 * wired_count == 0. 369 */ 370 if (!vm_map_lookup_entry(map, addr, &entry) || 371 entry->start != addr || entry->end != addr + size || 372 entry->wired_count) 373 panic("kmem_malloc: entry not found or misaligned"); 374 entry->wired_count++; 375 376 /* 377 * Loop thru pages, entering them in the pmap. (We cannot add them to 378 * the wired count without wrapping the vm_page_queue_lock in 379 * splimp...) 380 */ 381 for (i = 0; i < size; i += PAGE_SIZE) { 382 vm_object_lock(kmem_object); 383 m = vm_page_lookup(kmem_object, offset + i); 384 vm_object_unlock(kmem_object); 385 pmap_kenter(addr + i, VM_PAGE_TO_PHYS(m)); 386 } 387 vm_map_unlock(map); 388 389 vm_map_simplify(map, addr); 390 return (addr); 391} 392 393/* 394 * kmem_alloc_wait 395 * 396 * Allocates pageable memory from a sub-map of the kernel. If the submap 397 * has no room, the caller sleeps waiting for more memory in the submap. 398 * 399 */ 400vm_offset_t 401kmem_alloc_wait(map, size) 402 vm_map_t map; 403 vm_size_t size; 404{ 405 vm_offset_t addr; 406 407 size = round_page(size); 408 409 for (;;) { 410 /* 411 * To make this work for more than one map, use the map's lock 412 * to lock out sleepers/wakers. 413 */ 414 vm_map_lock(map); 415 if (vm_map_findspace(map, 0, size, &addr) == 0) 416 break; 417 /* no space now; see if we can ever get space */ 418 if (vm_map_max(map) - vm_map_min(map) < size) { 419 vm_map_unlock(map); 420 return (0); 421 } 422 assert_wait((int) map, TRUE); 423 vm_map_unlock(map); 424 thread_block("kmaw"); 425 } 426 vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size); 427 vm_map_unlock(map); 428 return (addr); 429} 430 431/* 432 * kmem_free_wakeup 433 * 434 * Returns memory to a submap of the kernel, and wakes up any threads 435 * waiting for memory in that map. 436 */ 437void 438kmem_free_wakeup(map, addr, size) 439 vm_map_t map; 440 vm_offset_t addr; 441 vm_size_t size; 442{ 443 vm_map_lock(map); 444 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 445 thread_wakeup((int) map); 446 vm_map_unlock(map); 447} 448 449/* 450 * Create the kernel map; insert a mapping covering kernel text, data, bss, 451 * and all space allocated thus far (`boostrap' data). The new map will thus 452 * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 453 * the range between `start' and `end' as free. 454 */ 455void 456kmem_init(start, end) 457 vm_offset_t start, end; 458{ 459 register vm_map_t m; 460 461 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 462 vm_map_lock(m); 463 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 464 kernel_map = m; 465 (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 466 VM_MIN_KERNEL_ADDRESS, start); 467 /* ... and ending with the completion of the above `insert' */ 468 vm_map_unlock(m); 469} 470