vm_kern.c revision 5455
1108983Simp/* 2108983Simp * Copyright (c) 1991, 1993 3108983Simp * The Regents of the University of California. All rights reserved. 4108983Simp * 5108983Simp * This code is derived from software contributed to Berkeley by 6108983Simp * The Mach Operating System project at Carnegie-Mellon University. 7108983Simp * 8108983Simp * Redistribution and use in source and binary forms, with or without 9108983Simp * modification, are permitted provided that the following conditions 10108983Simp * are met: 11108983Simp * 1. Redistributions of source code must retain the above copyright 12108983Simp * notice, this list of conditions and the following disclaimer. 13108983Simp * 2. Redistributions in binary form must reproduce the above copyright 14108983Simp * notice, this list of conditions and the following disclaimer in the 15108983Simp * documentation and/or other materials provided with the distribution. 16108983Simp * 3. All advertising materials mentioning features or use of this software 17108983Simp * must display the following acknowledgement: 18108983Simp * This product includes software developed by the University of 19108983Simp * California, Berkeley and its contributors. 20148471Simp * 4. Neither the name of the University nor the names of its contributors 21108983Simp * may be used to endorse or promote products derived from this software 22139027Sbrueffer * without specific prior written permission. 23146969Smarius * 24139027Sbrueffer * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25108983Simp * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26108983Simp * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27108983Simp * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28108983Simp * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29108983Simp * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30108983Simp * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31159126Sthompsa * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32159126Sthompsa * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33108983Simp * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34159126Sthompsa * SUCH DAMAGE. 35159126Sthompsa * 36159126Sthompsa * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 37159126Sthompsa * 38108983Simp * 39108983Simp * Copyright (c) 1987, 1990 Carnegie-Mellon University. 40159126Sthompsa * All rights reserved. 41159126Sthompsa * 42159126Sthompsa * Authors: Avadis Tevanian, Jr., Michael Wayne Young 43159126Sthompsa * 44108983Simp * Permission to use, copy, modify and distribute this software and 45108983Simp * its documentation is hereby granted, provided that both the copyright 46147088Sbrooks * notice and this permission notice appear in all copies of the 47147088Sbrooks * software, derivative works or modified versions, and any portions 48147088Sbrooks * thereof, and that both notices appear in supporting documentation. 49147088Sbrooks * 50147088Sbrooks * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 51147088Sbrooks * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 52147088Sbrooks * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 53147088Sbrooks * 54147088Sbrooks * Carnegie Mellon requests users of this software to return to 55148642Ssam * 56147088Sbrooks * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 57147088Sbrooks * School of Computer Science 58147088Sbrooks * Carnegie Mellon University 59148642Ssam * Pittsburgh PA 15213-3890 60148642Ssam * 61148642Ssam * any improvements or extensions that they make and grant Carnegie the 62148642Ssam * rights to redistribute these changes. 63148642Ssam * 64148642Ssam * $Id: vm_kern.c,v 1.7 1994/08/18 22:36:02 wollman Exp $ 65148642Ssam */ 66148642Ssam 67148642Ssam/* 68148642Ssam * Kernel memory management. 69148642Ssam */ 70148642Ssam 71148642Ssam#include <sys/param.h> 72148642Ssam#include <sys/systm.h> 73148642Ssam#include <sys/kernel.h> 74148642Ssam#include <sys/proc.h> 75148642Ssam 76148642Ssam#include <vm/vm.h> 77148642Ssam#include <vm/vm_page.h> 78148642Ssam#include <vm/vm_pageout.h> 79108983Simp#include <vm/vm_kern.h> 80108983Simp 81108983Simpvm_map_t buffer_map; 82139281Sbrueffervm_map_t kernel_map; 83108983Simpvm_map_t kmem_map; 84108983Simpvm_map_t mb_map; 85108983Simpvm_map_t io_map; 86108983Simpvm_map_t clean_map; 87108983Simpvm_map_t pager_map; 88108983Simpvm_map_t phys_map; 89108983Simpvm_map_t exec_map; 90108983Simpvm_map_t u_map; 91152326Semax 92152326Semax/* 93152326Semax * kmem_alloc_pageable: 94152326Semax * 95152326Semax * Allocate pageable memory to the kernel's address map. 96152326Semax * map must be "kernel_map" below. 97152326Semax */ 98152326Semax 99152326Semaxvm_offset_t 100152326Semaxkmem_alloc_pageable(map, size) 101139281Sbrueffer vm_map_t map; 102134584Sbrooks register vm_size_t size; 103134584Sbrooks{ 104156782Semax vm_offset_t addr; 105134584Sbrooks register int result; 106134584Sbrooks 107134584Sbrooks#if 0 108156331Semax if (map != kernel_map) 109134584Sbrooks panic("kmem_alloc_pageable: not called with kernel_map"); 110134584Sbrooks#endif 111138175Siedowse 112138175Siedowse size = round_page(size); 113138175Siedowse 114138175Siedowse addr = vm_map_min(map); 115138175Siedowse result = vm_map_find(map, NULL, (vm_offset_t) 0, 116138175Siedowse &addr, size, TRUE); 117138175Siedowse if (result != KERN_SUCCESS) { 118153300Siedowse return (0); 119153300Siedowse } 120153300Siedowse return (addr); 121153300Siedowse} 122153300Siedowse 123153300Siedowse/* 124153300Siedowse * Allocate wired-down memory in the kernel's address map 125153300Siedowse * or a submap. 126153300Siedowse */ 127153300Siedowsevm_offset_t 128153300Siedowsekmem_alloc(map, size) 129153300Siedowse register vm_map_t map; 130153300Siedowse register vm_size_t size; 131153300Siedowse{ 132153300Siedowse vm_offset_t addr; 133153300Siedowse register vm_offset_t offset; 134153300Siedowse vm_offset_t i; 135153300Siedowse 136153300Siedowse size = round_page(size); 137153300Siedowse 138153300Siedowse /* 139153300Siedowse * Use the kernel object for wired-down kernel pages. Assume that no 140153300Siedowse * region of the kernel object is referenced more than once. 141153300Siedowse */ 142153300Siedowse 143153300Siedowse /* 144153300Siedowse * Locate sufficient space in the map. This will give us the final 145153300Siedowse * virtual address for the new memory, and thus will tell us the 146153300Siedowse * offset within the kernel map. 147108983Simp */ 148148471Simp vm_map_lock(map); 149148471Simp if (vm_map_findspace(map, 0, size, &addr)) { 150108983Simp vm_map_unlock(map); 151108983Simp return (0); 152108983Simp } 153131646Simp offset = addr - VM_MIN_KERNEL_ADDRESS; 154108983Simp vm_object_reference(kernel_object); 155108983Simp vm_map_insert(map, kernel_object, offset, addr, addr + size); 156108983Simp vm_map_unlock(map); 157114799Simp 158139281Sbrueffer /* 159119254Simp * Guarantee that there are pages already in this object before 160114852Simp * calling vm_map_pageable. This is to prevent the following 161119254Simp * scenario: 162108983Simp * 163166701Sjoerg * 1) Threads have swapped out, so that there is a pager for the 164166701Sjoerg * kernel_object. 2) The kmsg zone is empty, and so we are 165166701Sjoerg * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 166166701Sjoerg * there is no page, but there is a pager, so we call 167166701Sjoerg * pager_data_request. But the kmsg zone is empty, so we must 168166701Sjoerg * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 169166701Sjoerg * we get the data back from the pager, it will be (very stale) 170166701Sjoerg * non-zero data. kmem_alloc is defined to return zero-filled memory. 171166701Sjoerg * 172166701Sjoerg * We're intentionally not activating the pages we allocate to prevent a 173166701Sjoerg * race with page-out. vm_map_pageable will wire the pages. 174166701Sjoerg */ 175166701Sjoerg 176166701Sjoerg vm_object_lock(kernel_object); 177166701Sjoerg for (i = 0; i < size; i += PAGE_SIZE) { 178166701Sjoerg vm_page_t mem; 179166701Sjoerg 180166701Sjoerg while ((mem = vm_page_alloc(kernel_object, offset + i, 0)) == NULL) { 181166701Sjoerg vm_object_unlock(kernel_object); 182166701Sjoerg VM_WAIT; 183166701Sjoerg vm_object_lock(kernel_object); 184166701Sjoerg } 185166701Sjoerg vm_page_zero_fill(mem); 186166701Sjoerg mem->flags &= ~PG_BUSY; 187166701Sjoerg mem->valid |= VM_PAGE_BITS_ALL; 188166701Sjoerg } 189166701Sjoerg vm_object_unlock(kernel_object); 190166701Sjoerg 191166701Sjoerg /* 192166701Sjoerg * And finally, mark the data as non-pageable. 193139281Sbrueffer */ 194123626Snjl 195123626Snjl (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 196123626Snjl 197125366Snjl /* 198123626Snjl * Try to coalesce the map 199123626Snjl */ 200125366Snjl vm_map_simplify(map, addr); 201125366Snjl 202125366Snjl return (addr); 203125366Snjl} 204125366Snjl 205125366Snjl/* 206125366Snjl * kmem_free: 207125366Snjl * 208125366Snjl * Release a region of kernel virtual memory allocated 209125366Snjl * with kmem_alloc, and return the physical pages 210108983Simp * associated with that region. 211108983Simp */ 212108983Simpvoid 213108983Simpkmem_free(map, addr, size) 214108983Simp vm_map_t map; 215108983Simp register vm_offset_t addr; 216108983Simp vm_size_t size; 217108983Simp{ 218108983Simp (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 219108983Simp} 220108983Simp 221108983Simp/* 222108983Simp * kmem_suballoc: 223108983Simp * 224108983Simp * Allocates a map to manage a subrange 225108983Simp * of the kernel virtual address space. 226108983Simp * 227108983Simp * Arguments are as follows: 228108983Simp * 229108983Simp * parent Map to take range from 230108983Simp * size Size of range to find 231108983Simp * min, max Returned endpoints of map 232108983Simp * pageable Can the region be paged 233121493Snjl */ 234121493Snjlvm_map_t 235121493Snjlkmem_suballoc(parent, min, max, size, pageable) 236121493Snjl register vm_map_t parent; 237121493Snjl vm_offset_t *min, *max; 238121493Snjl register vm_size_t size; 239121493Snjl boolean_t pageable; 240121493Snjl{ 241121493Snjl register int ret; 242121493Snjl vm_map_t result; 243121493Snjl 244121493Snjl size = round_page(size); 245121493Snjl 246121493Snjl *min = (vm_offset_t) vm_map_min(parent); 247121493Snjl ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 248121493Snjl min, size, TRUE); 249121493Snjl if (ret != KERN_SUCCESS) { 250121493Snjl printf("kmem_suballoc: bad status return of %d.\n", ret); 251121493Snjl panic("kmem_suballoc"); 252121493Snjl } 253121493Snjl *max = *min + size; 254121493Snjl pmap_reference(vm_map_pmap(parent)); 255121493Snjl result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 256121493Snjl if (result == NULL) 257121493Snjl panic("kmem_suballoc: cannot create submap"); 258121493Snjl if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 259108983Simp panic("kmem_suballoc: unable to change range to submap"); 260 return (result); 261} 262 263/* 264 * Allocate wired-down memory in the kernel's address map for the higher 265 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 266 * kmem_alloc() because we may need to allocate memory at interrupt 267 * level where we cannot block (canwait == FALSE). 268 * 269 * This routine has its own private kernel submap (kmem_map) and object 270 * (kmem_object). This, combined with the fact that only malloc uses 271 * this routine, ensures that we will never block in map or object waits. 272 * 273 * Note that this still only works in a uni-processor environment and 274 * when called at splhigh(). 275 * 276 * We don't worry about expanding the map (adding entries) since entries 277 * for wired maps are statically allocated. 278 */ 279vm_offset_t 280kmem_malloc(map, size, canwait) 281 register vm_map_t map; 282 register vm_size_t size; 283 boolean_t canwait; 284{ 285 register vm_offset_t offset, i; 286 vm_map_entry_t entry; 287 vm_offset_t addr; 288 vm_page_t m; 289 290 if (map != kmem_map && map != mb_map) 291 panic("kern_malloc_alloc: map != {kmem,mb}_map"); 292 293 size = round_page(size); 294 addr = vm_map_min(map); 295 296 /* 297 * Locate sufficient space in the map. This will give us the final 298 * virtual address for the new memory, and thus will tell us the 299 * offset within the kernel map. 300 */ 301 vm_map_lock(map); 302 if (vm_map_findspace(map, 0, size, &addr)) { 303 vm_map_unlock(map); 304#if 0 305 if (canwait) /* XXX should wait */ 306 panic("kmem_malloc: %s too small", 307 map == kmem_map ? "kmem_map" : "mb_map"); 308#endif 309 if (canwait) 310 panic("kmem_malloc: map too small"); 311 return (0); 312 } 313 offset = addr - vm_map_min(kmem_map); 314 vm_object_reference(kmem_object); 315 vm_map_insert(map, kmem_object, offset, addr, addr + size); 316 317 /* 318 * If we can wait, just mark the range as wired (will fault pages as 319 * necessary). 320 */ 321 if (canwait) { 322 vm_map_unlock(map); 323 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, 324 FALSE); 325 vm_map_simplify(map, addr); 326 return (addr); 327 } 328 /* 329 * If we cannot wait then we must allocate all memory up front, 330 * pulling it off the active queue to prevent pageout. 331 */ 332 vm_object_lock(kmem_object); 333 for (i = 0; i < size; i += PAGE_SIZE) { 334 m = vm_page_alloc(kmem_object, offset + i, 1); 335 336 /* 337 * Ran out of space, free everything up and return. Don't need 338 * to lock page queues here as we know that the pages we got 339 * aren't on any queues. 340 */ 341 if (m == NULL) { 342 while (i != 0) { 343 i -= PAGE_SIZE; 344 m = vm_page_lookup(kmem_object, offset + i); 345 vm_page_free(m); 346 } 347 vm_object_unlock(kmem_object); 348 vm_map_delete(map, addr, addr + size); 349 vm_map_unlock(map); 350 return (0); 351 } 352#if 0 353 vm_page_zero_fill(m); 354#endif 355 m->flags &= ~PG_BUSY; 356 m->valid |= VM_PAGE_BITS_ALL; 357 } 358 vm_object_unlock(kmem_object); 359 360 /* 361 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 362 * be able to extend the previous entry so there will be a new entry 363 * exactly corresponding to this address range and it will have 364 * wired_count == 0. 365 */ 366 if (!vm_map_lookup_entry(map, addr, &entry) || 367 entry->start != addr || entry->end != addr + size || 368 entry->wired_count) 369 panic("kmem_malloc: entry not found or misaligned"); 370 entry->wired_count++; 371 372 /* 373 * Loop thru pages, entering them in the pmap. (We cannot add them to 374 * the wired count without wrapping the vm_page_queue_lock in 375 * splimp...) 376 */ 377 for (i = 0; i < size; i += PAGE_SIZE) { 378 vm_object_lock(kmem_object); 379 m = vm_page_lookup(kmem_object, offset + i); 380 vm_object_unlock(kmem_object); 381 pmap_kenter(addr + i, VM_PAGE_TO_PHYS(m)); 382 } 383 vm_map_unlock(map); 384 385 vm_map_simplify(map, addr); 386 return (addr); 387} 388 389/* 390 * kmem_alloc_wait 391 * 392 * Allocates pageable memory from a sub-map of the kernel. If the submap 393 * has no room, the caller sleeps waiting for more memory in the submap. 394 * 395 */ 396vm_offset_t 397kmem_alloc_wait(map, size) 398 vm_map_t map; 399 vm_size_t size; 400{ 401 vm_offset_t addr; 402 403 size = round_page(size); 404 405 for (;;) { 406 /* 407 * To make this work for more than one map, use the map's lock 408 * to lock out sleepers/wakers. 409 */ 410 vm_map_lock(map); 411 if (vm_map_findspace(map, 0, size, &addr) == 0) 412 break; 413 /* no space now; see if we can ever get space */ 414 if (vm_map_max(map) - vm_map_min(map) < size) { 415 vm_map_unlock(map); 416 return (0); 417 } 418 assert_wait((int) map, TRUE); 419 vm_map_unlock(map); 420 thread_block("kmaw"); 421 } 422 vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size); 423 vm_map_unlock(map); 424 return (addr); 425} 426 427/* 428 * kmem_free_wakeup 429 * 430 * Returns memory to a submap of the kernel, and wakes up any threads 431 * waiting for memory in that map. 432 */ 433void 434kmem_free_wakeup(map, addr, size) 435 vm_map_t map; 436 vm_offset_t addr; 437 vm_size_t size; 438{ 439 vm_map_lock(map); 440 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 441 thread_wakeup((int) map); 442 vm_map_unlock(map); 443} 444 445/* 446 * Create the kernel map; insert a mapping covering kernel text, data, bss, 447 * and all space allocated thus far (`boostrap' data). The new map will thus 448 * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 449 * the range between `start' and `end' as free. 450 */ 451void 452kmem_init(start, end) 453 vm_offset_t start, end; 454{ 455 register vm_map_t m; 456 457 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 458 vm_map_lock(m); 459 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 460 kernel_map = m; 461 (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 462 VM_MIN_KERNEL_ADDRESS, start); 463 /* ... and ending with the completion of the above `insert' */ 464 vm_map_unlock(m); 465} 466