vm_kern.c revision 12259
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 1.16 1995/09/03 20:40:39 dyson 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; 87int mb_map_full; 88vm_map_t io_map; 89vm_map_t clean_map; 90vm_map_t phys_map; 91vm_map_t exec_map; 92vm_map_t u_map; 93 94/* 95 * kmem_alloc_pageable: 96 * 97 * Allocate pageable memory to the kernel's address map. 98 * "map" must be kernel_map or a submap of kernel_map. 99 */ 100 101vm_offset_t 102kmem_alloc_pageable(map, size) 103 vm_map_t map; 104 register vm_size_t size; 105{ 106 vm_offset_t addr; 107 register int result; 108 109 size = round_page(size); 110 addr = vm_map_min(map); 111 result = vm_map_find(map, NULL, (vm_offset_t) 0, 112 &addr, size, TRUE); 113 if (result != KERN_SUCCESS) { 114 return (0); 115 } 116 return (addr); 117} 118 119/* 120 * Allocate wired-down memory in the kernel's address map 121 * or a submap. 122 */ 123vm_offset_t 124kmem_alloc(map, size) 125 register vm_map_t map; 126 register vm_size_t size; 127{ 128 vm_offset_t addr; 129 register vm_offset_t offset; 130 vm_offset_t i; 131 132 size = round_page(size); 133 134 /* 135 * Use the kernel object for wired-down kernel pages. Assume that no 136 * region of the kernel object is referenced more than once. 137 */ 138 139 /* 140 * Locate sufficient space in the map. This will give us the final 141 * virtual address for the new memory, and thus will tell us the 142 * offset within the kernel map. 143 */ 144 vm_map_lock(map); 145 if (vm_map_findspace(map, 0, size, &addr)) { 146 vm_map_unlock(map); 147 return (0); 148 } 149 offset = addr - VM_MIN_KERNEL_ADDRESS; 150 vm_object_reference(kernel_object); 151 vm_map_insert(map, kernel_object, offset, addr, addr + size); 152 vm_map_unlock(map); 153 154 /* 155 * Guarantee that there are pages already in this object before 156 * calling vm_map_pageable. This is to prevent the following 157 * scenario: 158 * 159 * 1) Threads have swapped out, so that there is a pager for the 160 * kernel_object. 2) The kmsg zone is empty, and so we are 161 * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 162 * there is no page, but there is a pager, so we call 163 * pager_data_request. But the kmsg zone is empty, so we must 164 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 165 * we get the data back from the pager, it will be (very stale) 166 * non-zero data. kmem_alloc is defined to return zero-filled memory. 167 * 168 * We're intentionally not activating the pages we allocate to prevent a 169 * race with page-out. vm_map_pageable will wire the pages. 170 */ 171 172 for (i = 0; i < size; i += PAGE_SIZE) { 173 vm_page_t mem; 174 175 while ((mem = vm_page_alloc(kernel_object, offset + i, (VM_ALLOC_NORMAL|VM_ALLOC_ZERO))) == NULL) { 176 VM_WAIT; 177 } 178 if ((mem->flags & PG_ZERO) == 0) 179 vm_page_zero_fill(mem); 180 mem->flags &= ~(PG_BUSY|PG_ZERO); 181 mem->valid = VM_PAGE_BITS_ALL; 182 } 183 184 /* 185 * And finally, mark the data as non-pageable. 186 */ 187 188 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 189 190 /* 191 * Try to coalesce the map 192 */ 193 vm_map_simplify(map, addr); 194 195 return (addr); 196} 197 198/* 199 * kmem_free: 200 * 201 * Release a region of kernel virtual memory allocated 202 * with kmem_alloc, and return the physical pages 203 * associated with that region. 204 */ 205void 206kmem_free(map, addr, size) 207 vm_map_t map; 208 register vm_offset_t addr; 209 vm_size_t size; 210{ 211 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 212} 213 214/* 215 * kmem_suballoc: 216 * 217 * Allocates a map to manage a subrange 218 * of the kernel virtual address space. 219 * 220 * Arguments are as follows: 221 * 222 * parent Map to take range from 223 * size Size of range to find 224 * min, max Returned endpoints of map 225 * pageable Can the region be paged 226 */ 227vm_map_t 228kmem_suballoc(parent, min, max, size, pageable) 229 register vm_map_t parent; 230 vm_offset_t *min, *max; 231 register vm_size_t size; 232 boolean_t pageable; 233{ 234 register int ret; 235 vm_map_t result; 236 237 size = round_page(size); 238 239 *min = (vm_offset_t) vm_map_min(parent); 240 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 241 min, size, TRUE); 242 if (ret != KERN_SUCCESS) { 243 printf("kmem_suballoc: bad status return of %d.\n", ret); 244 panic("kmem_suballoc"); 245 } 246 *max = *min + size; 247 pmap_reference(vm_map_pmap(parent)); 248 result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 249 if (result == NULL) 250 panic("kmem_suballoc: cannot create submap"); 251 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 252 panic("kmem_suballoc: unable to change range to submap"); 253 return (result); 254} 255 256/* 257 * Allocate wired-down memory in the kernel's address map for the higher 258 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 259 * kmem_alloc() because we may need to allocate memory at interrupt 260 * level where we cannot block (canwait == FALSE). 261 * 262 * This routine has its own private kernel submap (kmem_map) and object 263 * (kmem_object). This, combined with the fact that only malloc uses 264 * this routine, ensures that we will never block in map or object waits. 265 * 266 * Note that this still only works in a uni-processor environment and 267 * when called at splhigh(). 268 * 269 * We don't worry about expanding the map (adding entries) since entries 270 * for wired maps are statically allocated. 271 */ 272vm_offset_t 273kmem_malloc(map, size, waitflag) 274 register vm_map_t map; 275 register vm_size_t size; 276 boolean_t waitflag; 277{ 278 register vm_offset_t offset, i; 279 vm_map_entry_t entry; 280 vm_offset_t addr; 281 vm_page_t m; 282 283 if (map != kmem_map && map != mb_map) 284 panic("kmem_malloc: map != {kmem,mb}_map"); 285 286 size = round_page(size); 287 addr = vm_map_min(map); 288 289 /* 290 * Locate sufficient space in the map. This will give us the final 291 * virtual address for the new memory, and thus will tell us the 292 * offset within the kernel map. 293 */ 294 vm_map_lock(map); 295 if (vm_map_findspace(map, 0, size, &addr)) { 296 vm_map_unlock(map); 297 if (map == mb_map) { 298 mb_map_full = TRUE; 299 log(LOG_ERR, "mb_map full\n"); 300 return (0); 301 } 302 if (waitflag == M_WAITOK) 303 panic("kmem_malloc: kmem_map too small"); 304 return (0); 305 } 306 offset = addr - vm_map_min(kmem_map); 307 vm_object_reference(kmem_object); 308 vm_map_insert(map, kmem_object, offset, addr, addr + size); 309 310 /* 311 * If we can wait, just mark the range as wired (will fault pages as 312 * necessary). 313 */ 314 if (waitflag == M_WAITOK) { 315 vm_map_unlock(map); 316 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, 317 FALSE); 318 vm_map_simplify(map, addr); 319 return (addr); 320 } 321 /* 322 * If we cannot wait then we must allocate all memory up front, 323 * pulling it off the active queue to prevent pageout. 324 */ 325 for (i = 0; i < size; i += PAGE_SIZE) { 326 m = vm_page_alloc(kmem_object, offset + i, 327 (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 328 329 /* 330 * Ran out of space, free everything up and return. Don't need 331 * to lock page queues here as we know that the pages we got 332 * aren't on any queues. 333 */ 334 if (m == NULL) { 335 while (i != 0) { 336 i -= PAGE_SIZE; 337 m = vm_page_lookup(kmem_object, offset + i); 338 vm_page_free(m); 339 } 340 vm_map_delete(map, addr, addr + size); 341 vm_map_unlock(map); 342 return (0); 343 } 344 m->flags &= ~(PG_BUSY|PG_ZERO); 345 m->valid = VM_PAGE_BITS_ALL; 346 } 347 348 /* 349 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 350 * be able to extend the previous entry so there will be a new entry 351 * exactly corresponding to this address range and it will have 352 * wired_count == 0. 353 */ 354 if (!vm_map_lookup_entry(map, addr, &entry) || 355 entry->start != addr || entry->end != addr + size || 356 entry->wired_count) 357 panic("kmem_malloc: entry not found or misaligned"); 358 entry->wired_count++; 359 360 /* 361 * Loop thru pages, entering them in the pmap. (We cannot add them to 362 * the wired count without wrapping the vm_page_queue_lock in 363 * splimp...) 364 */ 365 for (i = 0; i < size; i += PAGE_SIZE) { 366 m = vm_page_lookup(kmem_object, offset + i); 367 pmap_kenter(addr + i, VM_PAGE_TO_PHYS(m)); 368 } 369 vm_map_unlock(map); 370 371 vm_map_simplify(map, addr); 372 return (addr); 373} 374 375/* 376 * kmem_alloc_wait 377 * 378 * Allocates pageable memory from a sub-map of the kernel. If the submap 379 * has no room, the caller sleeps waiting for more memory in the submap. 380 * 381 */ 382vm_offset_t 383kmem_alloc_wait(map, size) 384 vm_map_t map; 385 vm_size_t size; 386{ 387 vm_offset_t addr; 388 389 size = round_page(size); 390 391 for (;;) { 392 /* 393 * To make this work for more than one map, use the map's lock 394 * to lock out sleepers/wakers. 395 */ 396 vm_map_lock(map); 397 if (vm_map_findspace(map, 0, size, &addr) == 0) 398 break; 399 /* no space now; see if we can ever get space */ 400 if (vm_map_max(map) - vm_map_min(map) < size) { 401 vm_map_unlock(map); 402 return (0); 403 } 404 vm_map_unlock(map); 405 tsleep(map, PVM, "kmaw", 0); 406 } 407 vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size); 408 vm_map_unlock(map); 409 return (addr); 410} 411 412/* 413 * kmem_free_wakeup 414 * 415 * Returns memory to a submap of the kernel, and wakes up any processes 416 * waiting for memory in that map. 417 */ 418void 419kmem_free_wakeup(map, addr, size) 420 vm_map_t map; 421 vm_offset_t addr; 422 vm_size_t size; 423{ 424 vm_map_lock(map); 425 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 426 wakeup(map); 427 vm_map_unlock(map); 428} 429 430/* 431 * Create the kernel map; insert a mapping covering kernel text, data, bss, 432 * and all space allocated thus far (`boostrap' data). The new map will thus 433 * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 434 * the range between `start' and `end' as free. 435 */ 436void 437kmem_init(start, end) 438 vm_offset_t start, end; 439{ 440 register vm_map_t m; 441 442 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 443 vm_map_lock(m); 444 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 445 kernel_map = m; 446 (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 447 VM_MIN_KERNEL_ADDRESS, start); 448 /* ... and ending with the completion of the above `insert' */ 449 vm_map_unlock(m); 450} 451