vm_kern.c revision 26258
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 1.35 1997/04/26 11:46:23 peter Exp $ 65 */ 66 67/* 68 * Kernel memory management. 69 */ 70#include "opt_smp_privpages.h" 71 72#include <sys/param.h> 73#include <sys/systm.h> 74#include <sys/kernel.h> 75#include <sys/proc.h> 76#include <sys/malloc.h> 77#include <sys/syslog.h> 78#include <sys/queue.h> 79#include <sys/vmmeter.h> 80 81#include <vm/vm.h> 82#include <vm/vm_param.h> 83#include <vm/vm_prot.h> 84#include <sys/lock.h> 85#include <vm/pmap.h> 86#include <vm/vm_map.h> 87#include <vm/vm_object.h> 88#include <vm/vm_page.h> 89#include <vm/vm_pageout.h> 90#include <vm/vm_kern.h> 91#include <vm/vm_extern.h> 92 93vm_map_t kernel_map=0; 94vm_map_t kmem_map=0; 95vm_map_t exec_map=0; 96vm_map_t clean_map=0; 97vm_map_t u_map=0; 98vm_map_t buffer_map=0; 99vm_map_t mb_map=0; 100int mb_map_full=0; 101vm_map_t io_map=0; 102vm_map_t phys_map=0; 103#if defined(SMP) && defined(SMP_PRIVPAGES) 104vm_map_t ppage_map=0; 105#endif 106 107 108/* 109 * kmem_alloc_pageable: 110 * 111 * Allocate pageable memory to the kernel's address map. 112 * "map" must be kernel_map or a submap of kernel_map. 113 */ 114 115vm_offset_t 116kmem_alloc_pageable(map, size) 117 vm_map_t map; 118 register vm_size_t size; 119{ 120 vm_offset_t addr; 121 register int result; 122 123 size = round_page(size); 124 addr = vm_map_min(map); 125 result = vm_map_find(map, NULL, (vm_offset_t) 0, 126 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 127 if (result != KERN_SUCCESS) { 128 return (0); 129 } 130 return (addr); 131} 132 133/* 134 * Allocate wired-down memory in the kernel's address map 135 * or a submap. 136 */ 137vm_offset_t 138kmem_alloc(map, size) 139 register vm_map_t map; 140 register vm_size_t size; 141{ 142 vm_offset_t addr; 143 register vm_offset_t offset; 144 vm_offset_t i; 145 146 size = round_page(size); 147 148 /* 149 * Use the kernel object for wired-down kernel pages. Assume that no 150 * region of the kernel object is referenced more than once. 151 */ 152 153 /* 154 * Locate sufficient space in the map. This will give us the final 155 * virtual address for the new memory, and thus will tell us the 156 * offset within the kernel map. 157 */ 158 vm_map_lock(map); 159 if (vm_map_findspace(map, 0, size, &addr)) { 160 vm_map_unlock(map); 161 return (0); 162 } 163 offset = addr - VM_MIN_KERNEL_ADDRESS; 164 vm_object_reference(kernel_object); 165 vm_map_insert(map, kernel_object, offset, addr, addr + size, 166 VM_PROT_ALL, VM_PROT_ALL, 0); 167 vm_map_unlock(map); 168 169 /* 170 * Guarantee that there are pages already in this object before 171 * calling vm_map_pageable. This is to prevent the following 172 * scenario: 173 * 174 * 1) Threads have swapped out, so that there is a pager for the 175 * kernel_object. 2) The kmsg zone is empty, and so we are 176 * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 177 * there is no page, but there is a pager, so we call 178 * pager_data_request. But the kmsg zone is empty, so we must 179 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 180 * we get the data back from the pager, it will be (very stale) 181 * non-zero data. kmem_alloc is defined to return zero-filled memory. 182 * 183 * We're intentionally not activating the pages we allocate to prevent a 184 * race with page-out. vm_map_pageable will wire the pages. 185 */ 186 187 for (i = 0; i < size; i += PAGE_SIZE) { 188 vm_page_t mem; 189 190 while ((mem = vm_page_alloc(kernel_object, 191 OFF_TO_IDX(offset + i), VM_ALLOC_ZERO)) == NULL) { 192 VM_WAIT; 193 } 194 if ((mem->flags & PG_ZERO) == 0) 195 vm_page_zero_fill(mem); 196 mem->flags &= ~(PG_BUSY|PG_ZERO); 197 mem->valid = VM_PAGE_BITS_ALL; 198 } 199 200 /* 201 * And finally, mark the data as non-pageable. 202 */ 203 204 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 205 206 return (addr); 207} 208 209/* 210 * kmem_free: 211 * 212 * Release a region of kernel virtual memory allocated 213 * with kmem_alloc, and return the physical pages 214 * associated with that region. 215 */ 216void 217kmem_free(map, addr, size) 218 vm_map_t map; 219 register vm_offset_t addr; 220 vm_size_t size; 221{ 222 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 223} 224 225/* 226 * kmem_suballoc: 227 * 228 * Allocates a map to manage a subrange 229 * of the kernel virtual address space. 230 * 231 * Arguments are as follows: 232 * 233 * parent Map to take range from 234 * size Size of range to find 235 * min, max Returned endpoints of map 236 * pageable Can the region be paged 237 */ 238vm_map_t 239kmem_suballoc(parent, min, max, size, pageable) 240 register vm_map_t parent; 241 vm_offset_t *min, *max; 242 register vm_size_t size; 243 boolean_t pageable; 244{ 245 register int ret; 246 vm_map_t result; 247 248 size = round_page(size); 249 250 *min = (vm_offset_t) vm_map_min(parent); 251 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 252 min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 253 if (ret != KERN_SUCCESS) { 254 printf("kmem_suballoc: bad status return of %d.\n", ret); 255 panic("kmem_suballoc"); 256 } 257 *max = *min + size; 258 pmap_reference(vm_map_pmap(parent)); 259 result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 260 if (result == NULL) 261 panic("kmem_suballoc: cannot create submap"); 262 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 263 panic("kmem_suballoc: unable to change range to submap"); 264 return (result); 265} 266 267/* 268 * Allocate wired-down memory in the kernel's address map for the higher 269 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 270 * kmem_alloc() because we may need to allocate memory at interrupt 271 * level where we cannot block (canwait == FALSE). 272 * 273 * This routine has its own private kernel submap (kmem_map) and object 274 * (kmem_object). This, combined with the fact that only malloc uses 275 * this routine, ensures that we will never block in map or object waits. 276 * 277 * Note that this still only works in a uni-processor environment and 278 * when called at splhigh(). 279 * 280 * We don't worry about expanding the map (adding entries) since entries 281 * for wired maps are statically allocated. 282 */ 283vm_offset_t 284kmem_malloc(map, size, waitflag) 285 register vm_map_t map; 286 register vm_size_t size; 287 boolean_t waitflag; 288{ 289 register vm_offset_t offset, i; 290 vm_map_entry_t entry; 291 vm_offset_t addr; 292 vm_page_t m; 293 294 if (map != kmem_map && map != mb_map) 295 panic("kmem_malloc: map != {kmem,mb}_map"); 296 297 size = round_page(size); 298 addr = vm_map_min(map); 299 300 /* 301 * Locate sufficient space in the map. This will give us the final 302 * virtual address for the new memory, and thus will tell us the 303 * offset within the kernel map. 304 */ 305 vm_map_lock(map); 306 if (vm_map_findspace(map, 0, size, &addr)) { 307 vm_map_unlock(map); 308 if (map == mb_map) { 309 mb_map_full = TRUE; 310 log(LOG_ERR, "Out of mbuf clusters - increase maxusers!\n"); 311 return (0); 312 } 313 if (waitflag == M_WAITOK) 314 panic("kmem_malloc: kmem_map too small"); 315 return (0); 316 } 317 offset = addr - VM_MIN_KERNEL_ADDRESS; 318 vm_object_reference(kmem_object); 319 vm_map_insert(map, kmem_object, offset, addr, addr + size, 320 VM_PROT_ALL, VM_PROT_ALL, 0); 321 322 for (i = 0; i < size; i += PAGE_SIZE) { 323retry: 324 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), 325 (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 326 327 /* 328 * Ran out of space, free everything up and return. Don't need 329 * to lock page queues here as we know that the pages we got 330 * aren't on any queues. 331 */ 332 if (m == NULL) { 333 if (waitflag == M_WAITOK) { 334 VM_WAIT; 335 goto retry; 336 } 337 while (i != 0) { 338 i -= PAGE_SIZE; 339 m = vm_page_lookup(kmem_object, 340 OFF_TO_IDX(offset + i)); 341 PAGE_WAKEUP(m); 342 vm_page_free(m); 343 } 344 vm_map_delete(map, addr, addr + size); 345 vm_map_unlock(map); 346 return (0); 347 } 348 m->flags &= ~PG_ZERO; 349 m->valid = VM_PAGE_BITS_ALL; 350 } 351 352 /* 353 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 354 * be able to extend the previous entry so there will be a new entry 355 * exactly corresponding to this address range and it will have 356 * wired_count == 0. 357 */ 358 if (!vm_map_lookup_entry(map, addr, &entry) || 359 entry->start != addr || entry->end != addr + size || 360 entry->wired_count) 361 panic("kmem_malloc: entry not found or misaligned"); 362 entry->wired_count++; 363 364 vm_map_simplify_entry(map, entry); 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, OFF_TO_IDX(offset + i)); 373 vm_page_wire(m); 374 PAGE_WAKEUP(m); 375 pmap_enter(kernel_pmap, addr + i, VM_PAGE_TO_PHYS(m), 376 VM_PROT_ALL, 1); 377 m->flags |= PG_MAPPED|PG_WRITEABLE; 378 } 379 vm_map_unlock(map); 380 381 return (addr); 382} 383 384/* 385 * kmem_alloc_wait 386 * 387 * Allocates pageable memory from a sub-map of the kernel. If the submap 388 * has no room, the caller sleeps waiting for more memory in the submap. 389 * 390 */ 391vm_offset_t 392kmem_alloc_wait(map, size) 393 vm_map_t map; 394 vm_size_t size; 395{ 396 vm_offset_t addr; 397 398 size = round_page(size); 399 400 for (;;) { 401 /* 402 * To make this work for more than one map, use the map's lock 403 * to lock out sleepers/wakers. 404 */ 405 vm_map_lock(map); 406 if (vm_map_findspace(map, 0, size, &addr) == 0) 407 break; 408 /* no space now; see if we can ever get space */ 409 if (vm_map_max(map) - vm_map_min(map) < size) { 410 vm_map_unlock(map); 411 return (0); 412 } 413 vm_map_unlock(map); 414 tsleep(map, PVM, "kmaw", 0); 415 } 416 vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0); 417 vm_map_unlock(map); 418 return (addr); 419} 420 421/* 422 * kmem_free_wakeup 423 * 424 * Returns memory to a submap of the kernel, and wakes up any processes 425 * waiting for memory in that map. 426 */ 427void 428kmem_free_wakeup(map, addr, size) 429 vm_map_t map; 430 vm_offset_t addr; 431 vm_size_t size; 432{ 433 vm_map_lock(map); 434 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 435 wakeup(map); 436 vm_map_unlock(map); 437} 438 439/* 440 * Create the kernel map; insert a mapping covering kernel text, data, bss, 441 * and all space allocated thus far (`boostrap' data). The new map will thus 442 * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 443 * the range between `start' and `end' as free. 444 */ 445void 446kmem_init(start, end) 447 vm_offset_t start, end; 448{ 449 register vm_map_t m; 450 451 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 452 vm_map_lock(m); 453 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 454 kernel_map = m; 455 (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 456 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0); 457 /* ... and ending with the completion of the above `insert' */ 458 vm_map_unlock(m); 459} 460