vm_kern.c revision 10548
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 1.15 1995/07/29 11:44:20 bde 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" below. 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#if 0 110 if (map != kernel_map) 111 panic("kmem_alloc_pageable: not called with kernel_map"); 112#endif 113 114 size = round_page(size); 115 116 addr = vm_map_min(map); 117 result = vm_map_find(map, NULL, (vm_offset_t) 0, 118 &addr, size, TRUE); 119 if (result != KERN_SUCCESS) { 120 return (0); 121 } 122 return (addr); 123} 124 125/* 126 * Allocate wired-down memory in the kernel's address map 127 * or a submap. 128 */ 129vm_offset_t 130kmem_alloc(map, size) 131 register vm_map_t map; 132 register vm_size_t size; 133{ 134 vm_offset_t addr; 135 register vm_offset_t offset; 136 vm_offset_t i; 137 138 size = round_page(size); 139 140 /* 141 * Use the kernel object for wired-down kernel pages. Assume that no 142 * region of the kernel object is referenced more than once. 143 */ 144 145 /* 146 * Locate sufficient space in the map. This will give us the final 147 * virtual address for the new memory, and thus will tell us the 148 * offset within the kernel map. 149 */ 150 vm_map_lock(map); 151 if (vm_map_findspace(map, 0, size, &addr)) { 152 vm_map_unlock(map); 153 return (0); 154 } 155 offset = addr - VM_MIN_KERNEL_ADDRESS; 156 vm_object_reference(kernel_object); 157 vm_map_insert(map, kernel_object, offset, addr, addr + size); 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, offset + i, (VM_ALLOC_NORMAL|VM_ALLOC_ZERO))) == NULL) { 182 VM_WAIT; 183 } 184 if ((mem->flags & PG_ZERO) == 0) 185 vm_page_zero_fill(mem); 186 mem->flags &= ~(PG_BUSY|PG_ZERO); 187 mem->valid = VM_PAGE_BITS_ALL; 188 } 189 190 /* 191 * And finally, mark the data as non-pageable. 192 */ 193 194 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 195 196 /* 197 * Try to coalesce the map 198 */ 199 vm_map_simplify(map, addr); 200 201 return (addr); 202} 203 204/* 205 * kmem_free: 206 * 207 * Release a region of kernel virtual memory allocated 208 * with kmem_alloc, and return the physical pages 209 * associated with that region. 210 */ 211void 212kmem_free(map, addr, size) 213 vm_map_t map; 214 register vm_offset_t addr; 215 vm_size_t size; 216{ 217 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 218} 219 220/* 221 * kmem_suballoc: 222 * 223 * Allocates a map to manage a subrange 224 * of the kernel virtual address space. 225 * 226 * Arguments are as follows: 227 * 228 * parent Map to take range from 229 * size Size of range to find 230 * min, max Returned endpoints of map 231 * pageable Can the region be paged 232 */ 233vm_map_t 234kmem_suballoc(parent, min, max, size, pageable) 235 register vm_map_t parent; 236 vm_offset_t *min, *max; 237 register vm_size_t size; 238 boolean_t pageable; 239{ 240 register int ret; 241 vm_map_t result; 242 243 size = round_page(size); 244 245 *min = (vm_offset_t) vm_map_min(parent); 246 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 247 min, size, TRUE); 248 if (ret != KERN_SUCCESS) { 249 printf("kmem_suballoc: bad status return of %d.\n", ret); 250 panic("kmem_suballoc"); 251 } 252 *max = *min + size; 253 pmap_reference(vm_map_pmap(parent)); 254 result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 255 if (result == NULL) 256 panic("kmem_suballoc: cannot create submap"); 257 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 258 panic("kmem_suballoc: unable to change range to submap"); 259 return (result); 260} 261 262/* 263 * Allocate wired-down memory in the kernel's address map for the higher 264 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 265 * kmem_alloc() because we may need to allocate memory at interrupt 266 * level where we cannot block (canwait == FALSE). 267 * 268 * This routine has its own private kernel submap (kmem_map) and object 269 * (kmem_object). This, combined with the fact that only malloc uses 270 * this routine, ensures that we will never block in map or object waits. 271 * 272 * Note that this still only works in a uni-processor environment and 273 * when called at splhigh(). 274 * 275 * We don't worry about expanding the map (adding entries) since entries 276 * for wired maps are statically allocated. 277 */ 278vm_offset_t 279kmem_malloc(map, size, waitflag) 280 register vm_map_t map; 281 register vm_size_t size; 282 boolean_t waitflag; 283{ 284 register vm_offset_t offset, i; 285 vm_map_entry_t entry; 286 vm_offset_t addr; 287 vm_page_t m; 288 289 if (map != kmem_map && map != mb_map) 290 panic("kmem_malloc: map != {kmem,mb}_map"); 291 292 size = round_page(size); 293 addr = vm_map_min(map); 294 295 /* 296 * Locate sufficient space in the map. This will give us the final 297 * virtual address for the new memory, and thus will tell us the 298 * offset within the kernel map. 299 */ 300 vm_map_lock(map); 301 if (vm_map_findspace(map, 0, size, &addr)) { 302 vm_map_unlock(map); 303 if (map == mb_map) { 304 mb_map_full = TRUE; 305 log(LOG_ERR, "mb_map full\n"); 306 return (0); 307 } 308 if (waitflag == M_WAITOK) 309 panic("kmem_malloc: kmem_map too small"); 310 return (0); 311 } 312 offset = addr - vm_map_min(kmem_map); 313 vm_object_reference(kmem_object); 314 vm_map_insert(map, kmem_object, offset, addr, addr + size); 315 316 /* 317 * If we can wait, just mark the range as wired (will fault pages as 318 * necessary). 319 */ 320 if (waitflag == M_WAITOK) { 321 vm_map_unlock(map); 322 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, 323 FALSE); 324 vm_map_simplify(map, addr); 325 return (addr); 326 } 327 /* 328 * If we cannot wait then we must allocate all memory up front, 329 * pulling it off the active queue to prevent pageout. 330 */ 331 for (i = 0; i < size; i += PAGE_SIZE) { 332 m = vm_page_alloc(kmem_object, offset + i, 333 (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 334 335 /* 336 * Ran out of space, free everything up and return. Don't need 337 * to lock page queues here as we know that the pages we got 338 * aren't on any queues. 339 */ 340 if (m == NULL) { 341 while (i != 0) { 342 i -= PAGE_SIZE; 343 m = vm_page_lookup(kmem_object, offset + i); 344 vm_page_free(m); 345 } 346 vm_map_delete(map, addr, addr + size); 347 vm_map_unlock(map); 348 return (0); 349 } 350 m->flags &= ~(PG_BUSY|PG_ZERO); 351 m->valid = VM_PAGE_BITS_ALL; 352 } 353 354 /* 355 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 356 * be able to extend the previous entry so there will be a new entry 357 * exactly corresponding to this address range and it will have 358 * wired_count == 0. 359 */ 360 if (!vm_map_lookup_entry(map, addr, &entry) || 361 entry->start != addr || entry->end != addr + size || 362 entry->wired_count) 363 panic("kmem_malloc: entry not found or misaligned"); 364 entry->wired_count++; 365 366 /* 367 * Loop thru pages, entering them in the pmap. (We cannot add them to 368 * the wired count without wrapping the vm_page_queue_lock in 369 * splimp...) 370 */ 371 for (i = 0; i < size; i += PAGE_SIZE) { 372 m = vm_page_lookup(kmem_object, offset + i); 373 pmap_kenter(addr + i, VM_PAGE_TO_PHYS(m)); 374 } 375 vm_map_unlock(map); 376 377 vm_map_simplify(map, addr); 378 return (addr); 379} 380 381/* 382 * kmem_alloc_wait 383 * 384 * Allocates pageable memory from a sub-map of the kernel. If the submap 385 * has no room, the caller sleeps waiting for more memory in the submap. 386 * 387 */ 388vm_offset_t 389kmem_alloc_wait(map, size) 390 vm_map_t map; 391 vm_size_t size; 392{ 393 vm_offset_t addr; 394 395 size = round_page(size); 396 397 for (;;) { 398 /* 399 * To make this work for more than one map, use the map's lock 400 * to lock out sleepers/wakers. 401 */ 402 vm_map_lock(map); 403 if (vm_map_findspace(map, 0, size, &addr) == 0) 404 break; 405 /* no space now; see if we can ever get space */ 406 if (vm_map_max(map) - vm_map_min(map) < size) { 407 vm_map_unlock(map); 408 return (0); 409 } 410 vm_map_unlock(map); 411 tsleep(map, PVM, "kmaw", 0); 412 } 413 vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size); 414 vm_map_unlock(map); 415 return (addr); 416} 417 418/* 419 * kmem_free_wakeup 420 * 421 * Returns memory to a submap of the kernel, and wakes up any processes 422 * waiting for memory in that map. 423 */ 424void 425kmem_free_wakeup(map, addr, size) 426 vm_map_t map; 427 vm_offset_t addr; 428 vm_size_t size; 429{ 430 vm_map_lock(map); 431 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 432 wakeup(map); 433 vm_map_unlock(map); 434} 435 436/* 437 * Create the kernel map; insert a mapping covering kernel text, data, bss, 438 * and all space allocated thus far (`boostrap' data). The new map will thus 439 * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 440 * the range between `start' and `end' as free. 441 */ 442void 443kmem_init(start, end) 444 vm_offset_t start, end; 445{ 446 register vm_map_t m; 447 448 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 449 vm_map_lock(m); 450 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 451 kernel_map = m; 452 (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 453 VM_MIN_KERNEL_ADDRESS, start); 454 /* ... and ending with the completion of the above `insert' */ 455 vm_map_unlock(m); 456} 457