vm_kern.c revision 32937
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 1.40 1998/01/22 17:30:35 dyson Exp $ 65 */ 66 67/* 68 * Kernel memory management. 69 */ 70 71#include <sys/param.h> 72#include <sys/systm.h> 73#include <sys/proc.h> 74#include <sys/malloc.h> 75#include <sys/syslog.h> 76 77#include <vm/vm.h> 78#include <vm/vm_param.h> 79#include <vm/vm_prot.h> 80#include <sys/lock.h> 81#include <vm/pmap.h> 82#include <vm/vm_map.h> 83#include <vm/vm_object.h> 84#include <vm/vm_page.h> 85#include <vm/vm_pageout.h> 86#include <vm/vm_extern.h> 87 88vm_map_t kernel_map=0; 89vm_map_t kmem_map=0; 90vm_map_t exec_map=0; 91vm_map_t clean_map=0; 92vm_map_t u_map=0; 93vm_map_t buffer_map=0; 94vm_map_t mb_map=0; 95int mb_map_full=0; 96vm_map_t io_map=0; 97vm_map_t phys_map=0; 98 99/* 100 * kmem_alloc_pageable: 101 * 102 * Allocate pageable memory to the kernel's address map. 103 * "map" must be kernel_map or a submap of kernel_map. 104 */ 105 106vm_offset_t 107kmem_alloc_pageable(map, size) 108 vm_map_t map; 109 register vm_size_t size; 110{ 111 vm_offset_t addr; 112 register int result; 113 114 size = round_page(size); 115 addr = vm_map_min(map); 116 result = vm_map_find(map, NULL, (vm_offset_t) 0, 117 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 118 if (result != KERN_SUCCESS) { 119 return (0); 120 } 121 return (addr); 122} 123 124/* 125 * Allocate wired-down memory in the kernel's address map 126 * or a submap. 127 */ 128vm_offset_t 129kmem_alloc(map, size) 130 register vm_map_t map; 131 register vm_size_t size; 132{ 133 vm_offset_t addr; 134 register vm_offset_t offset; 135 vm_offset_t i; 136 137 size = round_page(size); 138 139 /* 140 * Use the kernel object for wired-down kernel pages. Assume that no 141 * region of the kernel object is referenced more than once. 142 */ 143 144 /* 145 * Locate sufficient space in the map. This will give us the final 146 * virtual address for the new memory, and thus will tell us the 147 * offset within the kernel map. 148 */ 149 vm_map_lock(map); 150 if (vm_map_findspace(map, 0, size, &addr)) { 151 vm_map_unlock(map); 152 return (0); 153 } 154 offset = addr - VM_MIN_KERNEL_ADDRESS; 155 vm_object_reference(kernel_object); 156 vm_map_insert(map, kernel_object, offset, addr, addr + size, 157 VM_PROT_ALL, VM_PROT_ALL, 0); 158 vm_map_unlock(map); 159 160 /* 161 * Guarantee that there are pages already in this object before 162 * calling vm_map_pageable. This is to prevent the following 163 * scenario: 164 * 165 * 1) Threads have swapped out, so that there is a pager for the 166 * kernel_object. 2) The kmsg zone is empty, and so we are 167 * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 168 * there is no page, but there is a pager, so we call 169 * pager_data_request. But the kmsg zone is empty, so we must 170 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 171 * we get the data back from the pager, it will be (very stale) 172 * non-zero data. kmem_alloc is defined to return zero-filled memory. 173 * 174 * We're intentionally not activating the pages we allocate to prevent a 175 * race with page-out. vm_map_pageable will wire the pages. 176 */ 177 178 for (i = 0; i < size; i += PAGE_SIZE) { 179 vm_page_t mem; 180 181 while ((mem = vm_page_alloc(kernel_object, 182 OFF_TO_IDX(offset + i), VM_ALLOC_ZERO)) == NULL) { 183 VM_WAIT; 184 } 185 if ((mem->flags & PG_ZERO) == 0) 186 vm_page_zero_fill(mem); 187 mem->flags &= ~(PG_BUSY|PG_ZERO); 188 mem->valid = VM_PAGE_BITS_ALL; 189 } 190 191 /* 192 * And finally, mark the data as non-pageable. 193 */ 194 195 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 196 197 return (addr); 198} 199 200/* 201 * kmem_free: 202 * 203 * Release a region of kernel virtual memory allocated 204 * with kmem_alloc, and return the physical pages 205 * associated with that region. 206 */ 207void 208kmem_free(map, addr, size) 209 vm_map_t map; 210 register vm_offset_t addr; 211 vm_size_t size; 212{ 213 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 214} 215 216/* 217 * kmem_suballoc: 218 * 219 * Allocates a map to manage a subrange 220 * of the kernel virtual address space. 221 * 222 * Arguments are as follows: 223 * 224 * parent Map to take range from 225 * size Size of range to find 226 * min, max Returned endpoints of map 227 * pageable Can the region be paged 228 */ 229vm_map_t 230kmem_suballoc(parent, min, max, size) 231 register vm_map_t parent; 232 vm_offset_t *min, *max; 233 register vm_size_t size; 234{ 235 register int ret; 236 vm_map_t result; 237 238 size = round_page(size); 239 240 *min = (vm_offset_t) vm_map_min(parent); 241 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 242 min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 243 if (ret != KERN_SUCCESS) { 244 printf("kmem_suballoc: bad status return of %d.\n", ret); 245 panic("kmem_suballoc"); 246 } 247 *max = *min + size; 248 pmap_reference(vm_map_pmap(parent)); 249 result = vm_map_create(vm_map_pmap(parent), *min, *max); 250 if (result == NULL) 251 panic("kmem_suballoc: cannot create submap"); 252 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 253 panic("kmem_suballoc: unable to change range to submap"); 254 return (result); 255} 256 257/* 258 * Allocate wired-down memory in the kernel's address map for the higher 259 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 260 * kmem_alloc() because we may need to allocate memory at interrupt 261 * level where we cannot block (canwait == FALSE). 262 * 263 * This routine has its own private kernel submap (kmem_map) and object 264 * (kmem_object). This, combined with the fact that only malloc uses 265 * this routine, ensures that we will never block in map or object waits. 266 * 267 * Note that this still only works in a uni-processor environment and 268 * when called at splhigh(). 269 * 270 * We don't worry about expanding the map (adding entries) since entries 271 * for wired maps are statically allocated. 272 */ 273vm_offset_t 274kmem_malloc(map, size, waitflag) 275 register vm_map_t map; 276 register vm_size_t size; 277 boolean_t waitflag; 278{ 279 register vm_offset_t offset, i; 280 vm_map_entry_t entry; 281 vm_offset_t addr; 282 vm_page_t m; 283 284 if (map != kmem_map && map != mb_map) 285 panic("kmem_malloc: map != {kmem,mb}_map"); 286 287 size = round_page(size); 288 addr = vm_map_min(map); 289 290 /* 291 * Locate sufficient space in the map. This will give us the final 292 * virtual address for the new memory, and thus will tell us the 293 * offset within the kernel map. 294 */ 295 vm_map_lock(map); 296 if (vm_map_findspace(map, 0, size, &addr)) { 297 vm_map_unlock(map); 298 if (map == mb_map) { 299 mb_map_full = TRUE; 300 log(LOG_ERR, "Out of mbuf clusters - increase maxusers!\n"); 301 return (0); 302 } 303 if (waitflag == M_WAITOK) 304 panic("kmem_malloc: kmem_map too small"); 305 return (0); 306 } 307 offset = addr - VM_MIN_KERNEL_ADDRESS; 308 vm_object_reference(kmem_object); 309 vm_map_insert(map, kmem_object, offset, addr, addr + size, 310 VM_PROT_ALL, VM_PROT_ALL, 0); 311 312 for (i = 0; i < size; i += PAGE_SIZE) { 313retry: 314 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), 315 (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 316 317 /* 318 * Ran out of space, free everything up and return. Don't need 319 * to lock page queues here as we know that the pages we got 320 * aren't on any queues. 321 */ 322 if (m == NULL) { 323 if (waitflag == M_WAITOK) { 324 VM_WAIT; 325 goto retry; 326 } 327 while (i != 0) { 328 i -= PAGE_SIZE; 329 m = vm_page_lookup(kmem_object, 330 OFF_TO_IDX(offset + i)); 331 vm_page_free(m); 332 } 333 vm_map_delete(map, addr, addr + size); 334 vm_map_unlock(map); 335 return (0); 336 } 337 m->flags &= ~PG_ZERO; 338 m->valid = VM_PAGE_BITS_ALL; 339 } 340 341 /* 342 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 343 * be able to extend the previous entry so there will be a new entry 344 * exactly corresponding to this address range and it will have 345 * wired_count == 0. 346 */ 347 if (!vm_map_lookup_entry(map, addr, &entry) || 348 entry->start != addr || entry->end != addr + size || 349 entry->wired_count) 350 panic("kmem_malloc: entry not found or misaligned"); 351 entry->wired_count++; 352 353 vm_map_simplify_entry(map, entry); 354 355 /* 356 * Loop thru pages, entering them in the pmap. (We cannot add them to 357 * the wired count without wrapping the vm_page_queue_lock in 358 * splimp...) 359 */ 360 for (i = 0; i < size; i += PAGE_SIZE) { 361 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i)); 362 vm_page_wire(m); 363 PAGE_WAKEUP(m); 364 pmap_enter(kernel_pmap, addr + i, VM_PAGE_TO_PHYS(m), 365 VM_PROT_ALL, 1); 366 m->flags |= PG_MAPPED|PG_WRITEABLE; 367 } 368 vm_map_unlock(map); 369 370 return (addr); 371} 372 373/* 374 * kmem_alloc_wait 375 * 376 * Allocates pageable memory from a sub-map of the kernel. If the submap 377 * has no room, the caller sleeps waiting for more memory in the submap. 378 * 379 */ 380vm_offset_t 381kmem_alloc_wait(map, size) 382 vm_map_t map; 383 vm_size_t size; 384{ 385 vm_offset_t addr; 386 387 size = round_page(size); 388 389 for (;;) { 390 /* 391 * To make this work for more than one map, use the map's lock 392 * to lock out sleepers/wakers. 393 */ 394 vm_map_lock(map); 395 if (vm_map_findspace(map, 0, size, &addr) == 0) 396 break; 397 /* no space now; see if we can ever get space */ 398 if (vm_map_max(map) - vm_map_min(map) < size) { 399 vm_map_unlock(map); 400 return (0); 401 } 402 vm_map_unlock(map); 403 tsleep(map, PVM, "kmaw", 0); 404 } 405 vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0); 406 vm_map_unlock(map); 407 return (addr); 408} 409 410/* 411 * kmem_free_wakeup 412 * 413 * Returns memory to a submap of the kernel, and wakes up any processes 414 * waiting for memory in that map. 415 */ 416void 417kmem_free_wakeup(map, addr, size) 418 vm_map_t map; 419 vm_offset_t addr; 420 vm_size_t size; 421{ 422 vm_map_lock(map); 423 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 424 wakeup(map); 425 vm_map_unlock(map); 426} 427 428/* 429 * Create the kernel map; insert a mapping covering kernel text, data, bss, 430 * and all space allocated thus far (`boostrap' data). The new map will thus 431 * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 432 * the range between `start' and `end' as free. 433 */ 434void 435kmem_init(start, end) 436 vm_offset_t start, end; 437{ 438 register vm_map_t m; 439 440 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end); 441 vm_map_lock(m); 442 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 443 kernel_map = m; 444 kernel_map->system_map = 1; 445 (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 446 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0); 447 /* ... and ending with the completion of the above `insert' */ 448 vm_map_unlock(m); 449} 450