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 * $FreeBSD: src/sys/vm/vm_kern.c,v 1.61.2.2 2002/03/12 18:25:26 tegge Exp $ 65 * $DragonFly: src/sys/vm/vm_kern.c,v 1.11 2003/10/19 00:23:30 dillon Exp $ 66 */ 67 68 /* 69 * Kernel memory management. 70 */ 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/proc.h> 75 #include <sys/malloc.h> 76 77 #include <vm/vm.h> 78 #include <vm/vm_param.h> 79 #include <sys/lock.h> 80 #include <vm/pmap.h> 81 #include <vm/vm_map.h> 82 #include <vm/vm_object.h> 83 #include <vm/vm_page.h> 84 #include <vm/vm_pageout.h> 85 #include <vm/vm_extern.h> 86 87 vm_map_t kernel_map=0; 88 vm_map_t exec_map=0; 89 vm_map_t clean_map=0; 90 vm_map_t buffer_map=0; 91 vm_map_t mb_map=0; 92 int mb_map_full=0; 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 */ 100 101 vm_offset_t 102 kmem_alloc_pageable(map, size) 103 vm_map_t map; 104 vm_size_t size; 105 { 106 vm_offset_t addr; 107 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, VM_PROT_ALL, VM_PROT_ALL, 0); 113 if (result != KERN_SUCCESS) { 114 return (0); 115 } 116 return (addr); 117 } 118 119 /* 120 * kmem_alloc_nofault: 121 * 122 * Same as kmem_alloc_pageable, except that it create a nofault entry. 123 */ 124 125 vm_offset_t 126 kmem_alloc_nofault(map, size) 127 vm_map_t map; 128 vm_size_t size; 129 { 130 vm_offset_t addr; 131 int result; 132 133 size = round_page(size); 134 addr = vm_map_min(map); 135 result = vm_map_find(map, NULL, (vm_offset_t) 0, 136 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT); 137 if (result != KERN_SUCCESS) { 138 return (0); 139 } 140 return (addr); 141 } 142 143 /* 144 * Allocate wired-down memory in the kernel's address map 145 * or a submap. 146 */ 147 vm_offset_t 148 kmem_alloc(vm_map_t map, vm_size_t size) 149 { 150 vm_offset_t addr; 151 vm_offset_t offset; 152 vm_offset_t i; 153 int count; 154 155 size = round_page(size); 156 157 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT); 158 159 /* 160 * Use the kernel object for wired-down kernel pages. Assume that no 161 * region of the kernel object is referenced more than once. 162 * 163 * Locate sufficient space in the map. This will give us the final 164 * virtual address for the new memory, and thus will tell us the 165 * offset within the kernel map. 166 */ 167 vm_map_lock(map); 168 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) { 169 vm_map_unlock(map); 170 vm_map_entry_krelease(count); 171 return (0); 172 } 173 offset = addr - VM_MIN_KERNEL_ADDRESS; 174 vm_object_reference(kernel_object); 175 vm_map_insert(map, &count, 176 kernel_object, offset, addr, addr + size, 177 VM_PROT_ALL, VM_PROT_ALL, 0); 178 vm_map_unlock(map); 179 vm_map_entry_krelease(count); 180 181 /* 182 * Guarantee that there are pages already in this object before 183 * calling vm_map_wire. This is to prevent the following 184 * scenario: 185 * 186 * 1) Threads have swapped out, so that there is a pager for the 187 * kernel_object. 2) The kmsg zone is empty, and so we are 188 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault; 189 * there is no page, but there is a pager, so we call 190 * pager_data_request. But the kmsg zone is empty, so we must 191 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 192 * we get the data back from the pager, it will be (very stale) 193 * non-zero data. kmem_alloc is defined to return zero-filled memory. 194 * 195 * We're intentionally not activating the pages we allocate to prevent a 196 * race with page-out. vm_map_wire will wire the pages. 197 */ 198 199 for (i = 0; i < size; i += PAGE_SIZE) { 200 vm_page_t mem; 201 202 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i), 203 VM_ALLOC_ZERO | VM_ALLOC_RETRY); 204 if ((mem->flags & PG_ZERO) == 0) 205 vm_page_zero_fill(mem); 206 mem->valid = VM_PAGE_BITS_ALL; 207 vm_page_flag_clear(mem, PG_ZERO); 208 vm_page_wakeup(mem); 209 } 210 211 /* 212 * And finally, mark the data as non-pageable. 213 */ 214 215 (void) vm_map_wire(map, (vm_offset_t) addr, addr + size, FALSE); 216 217 return (addr); 218 } 219 220 /* 221 * kmem_free: 222 * 223 * Release a region of kernel virtual memory allocated 224 * with kmem_alloc, and return the physical pages 225 * associated with that region. 226 * 227 * This routine may not block on kernel maps. 228 */ 229 void 230 kmem_free(map, addr, size) 231 vm_map_t map; 232 vm_offset_t addr; 233 vm_size_t size; 234 { 235 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 236 } 237 238 /* 239 * kmem_suballoc: 240 * 241 * Allocates a map to manage a subrange 242 * of the kernel virtual address space. 243 * 244 * Arguments are as follows: 245 * 246 * parent Map to take range from 247 * size Size of range to find 248 * min, max Returned endpoints of map 249 * pageable Can the region be paged 250 */ 251 vm_map_t 252 kmem_suballoc(parent, min, max, size) 253 vm_map_t parent; 254 vm_offset_t *min, *max; 255 vm_size_t size; 256 { 257 int ret; 258 vm_map_t result; 259 260 size = round_page(size); 261 262 *min = (vm_offset_t) vm_map_min(parent); 263 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 264 min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 265 if (ret != KERN_SUCCESS) { 266 printf("kmem_suballoc: bad status return of %d.\n", ret); 267 panic("kmem_suballoc"); 268 } 269 *max = *min + size; 270 pmap_reference(vm_map_pmap(parent)); 271 result = vm_map_create(vm_map_pmap(parent), *min, *max); 272 if (result == NULL) 273 panic("kmem_suballoc: cannot create submap"); 274 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 275 panic("kmem_suballoc: unable to change range to submap"); 276 return (result); 277 } 278 279 /* 280 * kmem_malloc: 281 * 282 * Allocate wired-down memory in the kernel's address map for the higher 283 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 284 * kmem_alloc() because we may need to allocate memory at interrupt 285 * level where we cannot block (canwait == FALSE). 286 * 287 * This routine has its own private kernel submap (kmem_map) and object 288 * (kmem_object). This, combined with the fact that only malloc uses 289 * this routine, ensures that we will never block in map or object waits. 290 * 291 * Note that this still only works in a uni-processor environment and 292 * when called at splhigh(). 293 * 294 * We don't worry about expanding the map (adding entries) since entries 295 * for wired maps are statically allocated. 296 * 297 * NOTE: This routine is not supposed to block if M_NOWAIT is set, but 298 * I have not verified that it actually does not block. 299 */ 300 vm_offset_t 301 kmem_malloc(vm_map_t map, vm_size_t size, int flags) 302 { 303 vm_offset_t offset, i; 304 vm_map_entry_t entry; 305 vm_offset_t addr; 306 vm_page_t m; 307 int count; 308 309 if (map != kernel_map && map != mb_map) 310 panic("kmem_malloc: map != {kmem,mb}_map"); 311 312 size = round_page(size); 313 addr = vm_map_min(map); 314 315 /* 316 * Locate sufficient space in the map. This will give us the final 317 * virtual address for the new memory, and thus will tell us the 318 * offset within the kernel map. 319 */ 320 vm_map_lock(map); 321 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT); 322 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) { 323 vm_map_unlock(map); 324 vm_map_entry_krelease(count); 325 if (map == mb_map) { 326 mb_map_full = TRUE; 327 printf("Out of mbuf clusters - adjust NMBCLUSTERS or increase maxusers!\n"); 328 return (0); 329 } 330 if ((flags & M_NOWAIT) == 0) 331 panic("kmem_malloc(%ld): kernel_map too small: %ld total allocated", 332 (long)size, (long)map->size); 333 return (0); 334 } 335 offset = addr - VM_MIN_KERNEL_ADDRESS; 336 vm_object_reference(kmem_object); 337 vm_map_insert(map, &count, 338 kmem_object, offset, addr, addr + size, 339 VM_PROT_ALL, VM_PROT_ALL, 0); 340 341 for (i = 0; i < size; i += PAGE_SIZE) { 342 /* 343 * Note: if M_NOWAIT specified alone, allocate from 344 * interrupt-safe queues only (just the free list). If 345 * M_USE_RESERVE is also specified, we can also 346 * allocate from the cache. Neither of the latter two 347 * flags may be specified from an interrupt since interrupts 348 * are not allowed to mess with the cache queue. 349 */ 350 retry: 351 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), 352 ((flags & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT) ? 353 VM_ALLOC_INTERRUPT : 354 VM_ALLOC_SYSTEM); 355 356 /* 357 * Ran out of space, free everything up and return. Don't need 358 * to lock page queues here as we know that the pages we got 359 * aren't on any queues. 360 */ 361 if (m == NULL) { 362 if ((flags & M_NOWAIT) == 0) { 363 vm_map_unlock(map); 364 VM_WAIT; 365 vm_map_lock(map); 366 goto retry; 367 } 368 /* 369 * Free the pages before removing the map entry. 370 * They are already marked busy. Calling 371 * vm_map_delete before the pages has been freed or 372 * unbusied will cause a deadlock. 373 */ 374 while (i != 0) { 375 i -= PAGE_SIZE; 376 m = vm_page_lookup(kmem_object, 377 OFF_TO_IDX(offset + i)); 378 vm_page_free(m); 379 } 380 vm_map_delete(map, addr, addr + size, &count); 381 vm_map_unlock(map); 382 vm_map_entry_krelease(count); 383 return (0); 384 } 385 vm_page_flag_clear(m, PG_ZERO); 386 m->valid = VM_PAGE_BITS_ALL; 387 } 388 389 /* 390 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 391 * be able to extend the previous entry so there will be a new entry 392 * exactly corresponding to this address range and it will have 393 * wired_count == 0. 394 */ 395 if (!vm_map_lookup_entry(map, addr, &entry) || 396 entry->start != addr || entry->end != addr + size || 397 entry->wired_count != 0) 398 panic("kmem_malloc: entry not found or misaligned"); 399 entry->wired_count = 1; 400 401 vm_map_simplify_entry(map, entry, &count); 402 403 /* 404 * Loop thru pages, entering them in the pmap. (We cannot add them to 405 * the wired count without wrapping the vm_page_queue_lock in 406 * splimp...) 407 */ 408 for (i = 0; i < size; i += PAGE_SIZE) { 409 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i)); 410 vm_page_wire(m); 411 vm_page_wakeup(m); 412 /* 413 * Because this is kernel_pmap, this call will not block. 414 */ 415 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1); 416 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE | PG_REFERENCED); 417 } 418 vm_map_unlock(map); 419 vm_map_entry_krelease(count); 420 421 return (addr); 422 } 423 424 /* 425 * kmem_alloc_wait: 426 * 427 * Allocates pageable memory from a sub-map of the kernel. If the submap 428 * has no room, the caller sleeps waiting for more memory in the submap. 429 * 430 * This routine may block. 431 */ 432 433 vm_offset_t 434 kmem_alloc_wait(vm_map_t map, vm_size_t size) 435 { 436 vm_offset_t addr; 437 int count; 438 439 size = round_page(size); 440 441 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT); 442 443 for (;;) { 444 /* 445 * To make this work for more than one map, use the map's lock 446 * to lock out sleepers/wakers. 447 */ 448 vm_map_lock(map); 449 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr) == 0) 450 break; 451 /* no space now; see if we can ever get space */ 452 if (vm_map_max(map) - vm_map_min(map) < size) { 453 vm_map_entry_krelease(count); 454 vm_map_unlock(map); 455 return (0); 456 } 457 vm_map_unlock(map); 458 tsleep(map, 0, "kmaw", 0); 459 } 460 vm_map_insert(map, &count, 461 NULL, (vm_offset_t) 0, 462 addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0); 463 vm_map_unlock(map); 464 vm_map_entry_krelease(count); 465 return (addr); 466 } 467 468 /* 469 * kmem_free_wakeup: 470 * 471 * Returns memory to a submap of the kernel, and wakes up any processes 472 * waiting for memory in that map. 473 */ 474 void 475 kmem_free_wakeup(map, addr, size) 476 vm_map_t map; 477 vm_offset_t addr; 478 vm_size_t size; 479 { 480 int count; 481 482 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT); 483 vm_map_lock(map); 484 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size), &count); 485 wakeup(map); 486 vm_map_unlock(map); 487 vm_map_entry_krelease(count); 488 } 489 490 /* 491 * kmem_init: 492 * 493 * Create the kernel map; insert a mapping covering kernel text, 494 * data, bss, and all space allocated thus far (`boostrap' data). The 495 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and 496 * `start' as allocated, and the range between `start' and `end' as free. 497 * 498 * Depend on the zalloc bootstrap cache to get our vm_map_entry_t. 499 */ 500 void 501 kmem_init(vm_offset_t start, vm_offset_t end) 502 { 503 vm_map_t m; 504 int count; 505 506 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end); 507 vm_map_lock(m); 508 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 509 kernel_map = m; 510 kernel_map->system_map = 1; 511 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 512 (void) vm_map_insert(m, &count, NULL, (vm_offset_t) 0, 513 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0); 514 /* ... and ending with the completion of the above `insert' */ 515 vm_map_unlock(m); 516 vm_map_entry_release(count); 517 } 518 519 /* 520 * kmem_cpu_init: 521 * 522 * Load up extra vm_map_entry structures in each cpu's globaldata 523 * cache. These allow us to expand the mapent zone for kernel_map. 524 * Without them we would get into a recursion deadlock trying to 525 * reserve map entries (reserve->zalloc->kmem_alloc->reserve->...) 526 */ 527 void 528 kmem_cpu_init(void) 529 { 530 vm_map_entry_reserve(MAP_RESERVE_COUNT * 2); 531 } 532 533