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 1.18 2004/07/31 07:52:51 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_kern.h> 86 #include <vm/vm_extern.h> 87 88 vm_map_t kernel_map=0; 89 vm_map_t exec_map=0; 90 vm_map_t clean_map=0; 91 vm_map_t buffer_map=0; 92 93 /* 94 * kmem_alloc_pageable: 95 * 96 * Allocate pageable memory to the kernel's address map. 97 * "map" must be kernel_map or a submap of kernel_map. 98 */ 99 vm_offset_t 100 kmem_alloc_pageable(vm_map_t map, vm_size_t size) 101 { 102 vm_offset_t addr; 103 int result; 104 105 size = round_page(size); 106 addr = vm_map_min(map); 107 result = vm_map_find(map, NULL, (vm_offset_t) 0, 108 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 109 if (result != KERN_SUCCESS) { 110 return (0); 111 } 112 return (addr); 113 } 114 115 /* 116 * kmem_alloc_nofault: 117 * 118 * Same as kmem_alloc_pageable, except that it create a nofault entry. 119 */ 120 vm_offset_t 121 kmem_alloc_nofault(vm_map_t map, vm_size_t size) 122 { 123 vm_offset_t addr; 124 int result; 125 126 size = round_page(size); 127 addr = vm_map_min(map); 128 result = vm_map_find(map, NULL, (vm_offset_t) 0, 129 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT); 130 if (result != KERN_SUCCESS) { 131 return (0); 132 } 133 return (addr); 134 } 135 136 /* 137 * Allocate wired-down memory in the kernel's address map 138 * or a submap. 139 */ 140 vm_offset_t 141 kmem_alloc3(vm_map_t map, vm_size_t size, int kmflags) 142 { 143 vm_offset_t addr; 144 vm_offset_t offset; 145 vm_offset_t i; 146 int count; 147 148 size = round_page(size); 149 150 if (kmflags & KM_KRESERVE) 151 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT); 152 else 153 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 154 155 /* 156 * Use the kernel object for wired-down kernel pages. Assume that no 157 * region of the kernel object is referenced more than once. 158 * 159 * Locate sufficient space in the map. This will give us the final 160 * virtual address for the new memory, and thus will tell us the 161 * offset within the kernel map. 162 */ 163 vm_map_lock(map); 164 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) { 165 vm_map_unlock(map); 166 if (kmflags & KM_KRESERVE) 167 vm_map_entry_krelease(count); 168 else 169 vm_map_entry_release(count); 170 return (0); 171 } 172 offset = addr - VM_MIN_KERNEL_ADDRESS; 173 vm_object_reference(kernel_object); 174 vm_map_insert(map, &count, 175 kernel_object, offset, addr, addr + size, 176 VM_PROT_ALL, VM_PROT_ALL, 0); 177 vm_map_unlock(map); 178 if (kmflags & KM_KRESERVE) 179 vm_map_entry_krelease(count); 180 else 181 vm_map_entry_release(count); 182 183 /* 184 * Guarantee that there are pages already in this object before 185 * calling vm_map_wire. This is to prevent the following 186 * scenario: 187 * 188 * 1) Threads have swapped out, so that there is a pager for the 189 * kernel_object. 2) The kmsg zone is empty, and so we are 190 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault; 191 * there is no page, but there is a pager, so we call 192 * pager_data_request. But the kmsg zone is empty, so we must 193 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 194 * we get the data back from the pager, it will be (very stale) 195 * non-zero data. kmem_alloc is defined to return zero-filled memory. 196 * 197 * We're intentionally not activating the pages we allocate to prevent a 198 * race with page-out. vm_map_wire will wire the pages. 199 */ 200 201 for (i = 0; i < size; i += PAGE_SIZE) { 202 vm_page_t mem; 203 204 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i), 205 VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 206 if ((mem->flags & PG_ZERO) == 0) 207 vm_page_zero_fill(mem); 208 mem->valid = VM_PAGE_BITS_ALL; 209 vm_page_flag_clear(mem, PG_ZERO); 210 vm_page_wakeup(mem); 211 } 212 213 /* 214 * And finally, mark the data as non-pageable. 215 */ 216 217 (void) vm_map_wire(map, (vm_offset_t) addr, addr + size, kmflags); 218 219 return (addr); 220 } 221 222 /* 223 * kmem_free: 224 * 225 * Release a region of kernel virtual memory allocated 226 * with kmem_alloc, and return the physical pages 227 * associated with that region. 228 * 229 * This routine may not block on kernel maps. 230 */ 231 void 232 kmem_free(vm_map_t map, vm_offset_t addr, vm_size_t size) 233 { 234 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 235 } 236 237 /* 238 * kmem_suballoc: 239 * 240 * Allocates a map to manage a subrange 241 * of the kernel virtual address space. 242 * 243 * Arguments are as follows: 244 * 245 * parent Map to take range from 246 * size Size of range to find 247 * min, max Returned endpoints of map 248 * pageable Can the region be paged 249 */ 250 vm_map_t 251 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max, 252 vm_size_t size) 253 { 254 int ret; 255 vm_map_t result; 256 257 size = round_page(size); 258 259 *min = (vm_offset_t) vm_map_min(parent); 260 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 261 min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 262 if (ret != KERN_SUCCESS) { 263 printf("kmem_suballoc: bad status return of %d.\n", ret); 264 panic("kmem_suballoc"); 265 } 266 *max = *min + size; 267 pmap_reference(vm_map_pmap(parent)); 268 result = vm_map_create(vm_map_pmap(parent), *min, *max); 269 if (result == NULL) 270 panic("kmem_suballoc: cannot create submap"); 271 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 272 panic("kmem_suballoc: unable to change range to submap"); 273 return (result); 274 } 275 276 /* 277 * kmem_malloc: 278 * 279 * Allocate wired-down memory in the kernel's address map for the higher 280 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 281 * kmem_alloc() because we may need to allocate memory at interrupt 282 * level where we cannot block (canwait == FALSE). 283 * 284 * We don't worry about expanding the map (adding entries) since entries 285 * for wired maps are statically allocated. 286 * 287 * NOTE: Please see kmem_slab_alloc() for a better explanation of the 288 * M_* flags. 289 */ 290 vm_offset_t 291 kmem_malloc(vm_map_t map, vm_size_t size, int flags) 292 { 293 vm_offset_t offset, i; 294 vm_map_entry_t entry; 295 vm_offset_t addr; 296 vm_page_t m; 297 int count; 298 thread_t td; 299 int wanted_reserve; 300 301 if (map != kernel_map) 302 panic("kmem_malloc: map != kernel_map"); 303 304 size = round_page(size); 305 addr = vm_map_min(map); 306 307 /* 308 * Locate sufficient space in the map. This will give us the final 309 * virtual address for the new memory, and thus will tell us the 310 * offset within the kernel map. 311 */ 312 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 313 vm_map_lock(map); 314 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr)) { 315 vm_map_unlock(map); 316 vm_map_entry_release(count); 317 if ((flags & (M_RNOWAIT|M_NULLOK)) == 0 || 318 (flags & (M_FAILSAFE|M_NULLOK)) == M_FAILSAFE 319 ) { 320 panic("kmem_malloc(%ld): kernel_map too small: " 321 "%ld total allocated", 322 (long)size, (long)map->size); 323 } 324 return (0); 325 } 326 offset = addr - VM_MIN_KERNEL_ADDRESS; 327 vm_object_reference(kmem_object); 328 vm_map_insert(map, &count, 329 kmem_object, offset, addr, addr + size, 330 VM_PROT_ALL, VM_PROT_ALL, 0); 331 332 td = curthread; 333 wanted_reserve = 0; 334 335 for (i = 0; i < size; i += PAGE_SIZE) { 336 int vmflags; 337 338 vmflags = VM_ALLOC_SYSTEM; /* XXX M_USE_RESERVE? */ 339 if ((flags & (M_WAITOK|M_RNOWAIT)) == 0) 340 printf("kmem_malloc: bad flags %08x (%p)\n", flags, ((int **)&map)[-1]); 341 if (flags & M_USE_INTERRUPT_RESERVE) 342 vmflags |= VM_ALLOC_INTERRUPT; 343 if (flags & (M_FAILSAFE|M_WAITOK)) { 344 if (td->td_preempted) { 345 wanted_reserve = 1; 346 } else { 347 vmflags |= VM_ALLOC_NORMAL; 348 wanted_reserve = 0; 349 } 350 } 351 352 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), vmflags); 353 354 /* 355 * Ran out of space, free everything up and return. Don't need 356 * to lock page queues here as we know that the pages we got 357 * aren't on any queues. 358 * 359 * If M_WAITOK or M_FAILSAFE is set we can yield or block. 360 */ 361 if (m == NULL) { 362 if (flags & (M_FAILSAFE|M_WAITOK)) { 363 if (wanted_reserve) { 364 if (flags & M_FAILSAFE) 365 printf("kmem_malloc: no memory, try failsafe\n"); 366 vm_map_unlock(map); 367 lwkt_yield(); 368 vm_map_lock(map); 369 } else { 370 if (flags & M_FAILSAFE) 371 printf("kmem_malloc: no memory, block even though we shouldn't\n"); 372 vm_map_unlock(map); 373 vm_wait(); 374 vm_map_lock(map); 375 } 376 i -= PAGE_SIZE; /* retry */ 377 continue; 378 } 379 /* 380 * Free the pages before removing the map entry. 381 * They are already marked busy. Calling 382 * vm_map_delete before the pages has been freed or 383 * unbusied will cause a deadlock. 384 */ 385 while (i != 0) { 386 i -= PAGE_SIZE; 387 m = vm_page_lookup(kmem_object, 388 OFF_TO_IDX(offset + i)); 389 vm_page_free(m); 390 } 391 vm_map_delete(map, addr, addr + size, &count); 392 vm_map_unlock(map); 393 vm_map_entry_release(count); 394 return (0); 395 } 396 vm_page_flag_clear(m, PG_ZERO); 397 m->valid = VM_PAGE_BITS_ALL; 398 } 399 400 /* 401 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 402 * be able to extend the previous entry so there will be a new entry 403 * exactly corresponding to this address range and it will have 404 * wired_count == 0. 405 */ 406 if (!vm_map_lookup_entry(map, addr, &entry) || 407 entry->start != addr || entry->end != addr + size || 408 entry->wired_count != 0) 409 panic("kmem_malloc: entry not found or misaligned"); 410 entry->wired_count = 1; 411 412 vm_map_simplify_entry(map, entry, &count); 413 414 /* 415 * Loop thru pages, entering them in the pmap. (We cannot add them to 416 * the wired count without wrapping the vm_page_queue_lock in 417 * splimp...) 418 */ 419 for (i = 0; i < size; i += PAGE_SIZE) { 420 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i)); 421 vm_page_wire(m); 422 vm_page_wakeup(m); 423 /* 424 * Because this is kernel_pmap, this call will not block. 425 */ 426 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1); 427 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE | PG_REFERENCED); 428 } 429 vm_map_unlock(map); 430 vm_map_entry_release(count); 431 432 return (addr); 433 } 434 435 /* 436 * kmem_alloc_wait: 437 * 438 * Allocates pageable memory from a sub-map of the kernel. If the submap 439 * has no room, the caller sleeps waiting for more memory in the submap. 440 * 441 * This routine may block. 442 */ 443 444 vm_offset_t 445 kmem_alloc_wait(vm_map_t map, vm_size_t size) 446 { 447 vm_offset_t addr; 448 int count; 449 450 size = round_page(size); 451 452 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 453 454 for (;;) { 455 /* 456 * To make this work for more than one map, use the map's lock 457 * to lock out sleepers/wakers. 458 */ 459 vm_map_lock(map); 460 if (vm_map_findspace(map, vm_map_min(map), size, 1, &addr) == 0) 461 break; 462 /* no space now; see if we can ever get space */ 463 if (vm_map_max(map) - vm_map_min(map) < size) { 464 vm_map_entry_release(count); 465 vm_map_unlock(map); 466 return (0); 467 } 468 vm_map_unlock(map); 469 tsleep(map, 0, "kmaw", 0); 470 } 471 vm_map_insert(map, &count, 472 NULL, (vm_offset_t) 0, 473 addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0); 474 vm_map_unlock(map); 475 vm_map_entry_release(count); 476 return (addr); 477 } 478 479 /* 480 * kmem_free_wakeup: 481 * 482 * Returns memory to a submap of the kernel, and wakes up any processes 483 * waiting for memory in that map. 484 */ 485 void 486 kmem_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size) 487 { 488 int count; 489 490 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 491 vm_map_lock(map); 492 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size), &count); 493 wakeup(map); 494 vm_map_unlock(map); 495 vm_map_entry_release(count); 496 } 497 498 /* 499 * kmem_init: 500 * 501 * Create the kernel map; insert a mapping covering kernel text, 502 * data, bss, and all space allocated thus far (`boostrap' data). The 503 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and 504 * `start' as allocated, and the range between `start' and `end' as free. 505 * 506 * Depend on the zalloc bootstrap cache to get our vm_map_entry_t. 507 */ 508 void 509 kmem_init(vm_offset_t start, vm_offset_t end) 510 { 511 vm_map_t m; 512 int count; 513 514 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end); 515 vm_map_lock(m); 516 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 517 kernel_map = m; 518 kernel_map->system_map = 1; 519 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 520 (void) vm_map_insert(m, &count, NULL, (vm_offset_t) 0, 521 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0); 522 /* ... and ending with the completion of the above `insert' */ 523 vm_map_unlock(m); 524 vm_map_entry_release(count); 525 } 526