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 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 33 * 34 * 35 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 36 * All rights reserved. 37 * 38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 39 * 40 * Permission to use, copy, modify and distribute this software and 41 * its documentation is hereby granted, provided that both the copyright 42 * notice and this permission notice appear in all copies of the 43 * software, derivative works or modified versions, and any portions 44 * thereof, and that both notices appear in supporting documentation. 45 * 46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 49 * 50 * Carnegie Mellon requests users of this software to return to 51 * 52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 53 * School of Computer Science 54 * Carnegie Mellon University 55 * Pittsburgh PA 15213-3890 56 * 57 * any improvements or extensions that they make and grant Carnegie the 58 * rights to redistribute these changes. 59 */ 60 61/* 62 * Kernel memory management. 63 */ 64 65#include <sys/cdefs.h> 66__FBSDID("$FreeBSD$"); 67 68#include <sys/param.h> 69#include <sys/systm.h> 70#include <sys/kernel.h> /* for ticks and hz */ 71#include <sys/eventhandler.h> 72#include <sys/lock.h> 73#include <sys/mutex.h> 74#include <sys/proc.h> 75#include <sys/malloc.h> 76#include <sys/sysctl.h> 77 78#include <vm/vm.h> 79#include <vm/vm_param.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#include <vm/uma.h> 87 88vm_map_t kernel_map; 89vm_map_t kmem_map; 90vm_map_t exec_map; 91vm_map_t pipe_map; 92vm_map_t buffer_map; 93vm_map_t bio_transient_map; 94 95const void *zero_region; 96CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0); 97 98/* 99 * kmem_alloc_nofault: 100 * 101 * Allocate a virtual address range with no underlying object and 102 * no initial mapping to physical memory. Any mapping from this 103 * range to physical memory must be explicitly created prior to 104 * its use, typically with pmap_qenter(). Any attempt to create 105 * a mapping on demand through vm_fault() will result in a panic. 106 */ 107vm_offset_t 108kmem_alloc_nofault(map, size) 109 vm_map_t map; 110 vm_size_t size; 111{ 112 vm_offset_t addr; 113 int result; 114 115 size = round_page(size); 116 addr = vm_map_min(map); 117 result = vm_map_find(map, NULL, 0, &addr, size, VMFS_ANY_SPACE, 118 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT); 119 if (result != KERN_SUCCESS) { 120 return (0); 121 } 122 return (addr); 123} 124 125/* 126 * kmem_alloc_nofault_space: 127 * 128 * Allocate a virtual address range with no underlying object and 129 * no initial mapping to physical memory within the specified 130 * address space. Any mapping from this range to physical memory 131 * must be explicitly created prior to its use, typically with 132 * pmap_qenter(). Any attempt to create a mapping on demand 133 * through vm_fault() will result in a panic. 134 */ 135vm_offset_t 136kmem_alloc_nofault_space(map, size, find_space) 137 vm_map_t map; 138 vm_size_t size; 139 int find_space; 140{ 141 vm_offset_t addr; 142 int result; 143 144 size = round_page(size); 145 addr = vm_map_min(map); 146 result = vm_map_find(map, NULL, 0, &addr, size, find_space, 147 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT); 148 if (result != KERN_SUCCESS) { 149 return (0); 150 } 151 return (addr); 152} 153 154/* 155 * Allocate wired-down memory in the kernel's address map 156 * or a submap. 157 */ 158vm_offset_t 159kmem_alloc(map, size) 160 vm_map_t map; 161 vm_size_t size; 162{ 163 vm_offset_t addr; 164 vm_offset_t offset; 165 166 size = round_page(size); 167 168 /* 169 * Use the kernel object for wired-down kernel pages. Assume that no 170 * region of the kernel object is referenced more than once. 171 */ 172 173 /* 174 * Locate sufficient space in the map. This will give us the final 175 * virtual address for the new memory, and thus will tell us the 176 * offset within the kernel map. 177 */ 178 vm_map_lock(map); 179 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) { 180 vm_map_unlock(map); 181 return (0); 182 } 183 offset = addr - VM_MIN_KERNEL_ADDRESS; 184 vm_object_reference(kernel_object); 185 vm_map_insert(map, kernel_object, offset, addr, addr + size, 186 VM_PROT_ALL, VM_PROT_ALL, 0); 187 vm_map_unlock(map); 188 189 /* 190 * And finally, mark the data as non-pageable. 191 */ 192 (void) vm_map_wire(map, addr, addr + size, 193 VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES); 194 195 return (addr); 196} 197 198/* 199 * kmem_free: 200 * 201 * Release a region of kernel virtual memory allocated 202 * with kmem_alloc, and return the physical pages 203 * associated with that region. 204 * 205 * This routine may not block on kernel maps. 206 */ 207void 208kmem_free(map, addr, size) 209 vm_map_t map; 210 vm_offset_t addr; 211 vm_size_t size; 212{ 213 214 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 215} 216 217/* 218 * kmem_suballoc: 219 * 220 * Allocates a map to manage a subrange 221 * of the kernel virtual address space. 222 * 223 * Arguments are as follows: 224 * 225 * parent Map to take range from 226 * min, max Returned endpoints of map 227 * size Size of range to find 228 * superpage_align Request that min is superpage aligned 229 */ 230vm_map_t 231kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max, 232 vm_size_t size, boolean_t superpage_align) 233{ 234 int ret; 235 vm_map_t result; 236 237 size = round_page(size); 238 239 *min = vm_map_min(parent); 240 ret = vm_map_find(parent, NULL, 0, min, size, superpage_align ? 241 VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL, 242 MAP_ACC_NO_CHARGE); 243 if (ret != KERN_SUCCESS) 244 panic("kmem_suballoc: bad status return of %d", ret); 245 *max = *min + size; 246 result = vm_map_create(vm_map_pmap(parent), *min, *max); 247 if (result == NULL) 248 panic("kmem_suballoc: cannot create submap"); 249 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS) 250 panic("kmem_suballoc: unable to change range to submap"); 251 return (result); 252} 253 254/* 255 * kmem_malloc: 256 * 257 * Allocate wired-down memory in the kernel's address map for the higher 258 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 259 * kmem_alloc() because we may need to allocate memory at interrupt 260 * level where we cannot block (canwait == FALSE). 261 * 262 * This routine has its own private kernel submap (kmem_map) and object 263 * (kmem_object). This, combined with the fact that only malloc uses 264 * this routine, ensures that we will never block in map or object waits. 265 * 266 * We don't worry about expanding the map (adding entries) since entries 267 * for wired maps are statically allocated. 268 * 269 * `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to 270 * which we never free. 271 */ 272vm_offset_t 273kmem_malloc(map, size, flags) 274 vm_map_t map; 275 vm_size_t size; 276 int flags; 277{ 278 vm_offset_t addr; 279 int i, rv; 280 281 size = round_page(size); 282 addr = vm_map_min(map); 283 284 /* 285 * Locate sufficient space in the map. This will give us the final 286 * virtual address for the new memory, and thus will tell us the 287 * offset within the kernel map. 288 */ 289 vm_map_lock(map); 290 if (vm_map_findspace(map, vm_map_min(map), size, &addr)) { 291 vm_map_unlock(map); 292 if ((flags & M_NOWAIT) == 0) { 293 for (i = 0; i < 8; i++) { 294 EVENTHANDLER_INVOKE(vm_lowmem, 0); 295 uma_reclaim(); 296 vm_map_lock(map); 297 if (vm_map_findspace(map, vm_map_min(map), 298 size, &addr) == 0) { 299 break; 300 } 301 vm_map_unlock(map); 302 tsleep(&i, 0, "nokva", (hz / 4) * (i + 1)); 303 } 304 if (i == 8) { 305 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated", 306 (long)size, (long)map->size); 307 } 308 } else { 309 return (0); 310 } 311 } 312 313 rv = kmem_back(map, addr, size, flags); 314 vm_map_unlock(map); 315 return (rv == KERN_SUCCESS ? addr : 0); 316} 317 318/* 319 * kmem_back: 320 * 321 * Allocate physical pages for the specified virtual address range. 322 */ 323int 324kmem_back(vm_map_t map, vm_offset_t addr, vm_size_t size, int flags) 325{ 326 vm_offset_t offset, i; 327 vm_map_entry_t entry; 328 vm_page_t m; 329 int pflags; 330 boolean_t found; 331 332 KASSERT(vm_map_locked(map), ("kmem_back: map %p is not locked", map)); 333 offset = addr - VM_MIN_KERNEL_ADDRESS; 334 vm_object_reference(kmem_object); 335 vm_map_insert(map, kmem_object, offset, addr, addr + size, 336 VM_PROT_ALL, VM_PROT_ALL, 0); 337 338 /* 339 * Assert: vm_map_insert() will never be able to extend the 340 * previous entry so vm_map_lookup_entry() will find a new 341 * entry exactly corresponding to this address range and it 342 * will have wired_count == 0. 343 */ 344 found = vm_map_lookup_entry(map, addr, &entry); 345 KASSERT(found && entry->start == addr && entry->end == addr + size && 346 entry->wired_count == 0 && (entry->eflags & MAP_ENTRY_IN_TRANSITION) 347 == 0, ("kmem_back: entry not found or misaligned")); 348 349 if ((flags & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT) 350 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED; 351 else 352 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED; 353 354 if (flags & M_ZERO) 355 pflags |= VM_ALLOC_ZERO; 356 if (flags & M_NODUMP) 357 pflags |= VM_ALLOC_NODUMP; 358 359 VM_OBJECT_LOCK(kmem_object); 360 for (i = 0; i < size; i += PAGE_SIZE) { 361retry: 362 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags); 363 364 /* 365 * Ran out of space, free everything up and return. Don't need 366 * to lock page queues here as we know that the pages we got 367 * aren't on any queues. 368 */ 369 if (m == NULL) { 370 if ((flags & M_NOWAIT) == 0) { 371 VM_OBJECT_UNLOCK(kmem_object); 372 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 373 vm_map_unlock(map); 374 VM_WAIT; 375 vm_map_lock(map); 376 KASSERT( 377(entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_NEEDS_WAKEUP)) == 378 MAP_ENTRY_IN_TRANSITION, 379 ("kmem_back: volatile entry")); 380 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 381 VM_OBJECT_LOCK(kmem_object); 382 goto retry; 383 } 384 /* 385 * Free the pages before removing the map entry. 386 * They are already marked busy. Calling 387 * vm_map_delete before the pages has been freed or 388 * unbusied will cause a deadlock. 389 */ 390 while (i != 0) { 391 i -= PAGE_SIZE; 392 m = vm_page_lookup(kmem_object, 393 OFF_TO_IDX(offset + i)); 394 vm_page_unwire(m, 0); 395 vm_page_free(m); 396 } 397 VM_OBJECT_UNLOCK(kmem_object); 398 vm_map_delete(map, addr, addr + size); 399 return (KERN_NO_SPACE); 400 } 401 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0) 402 pmap_zero_page(m); 403 m->valid = VM_PAGE_BITS_ALL; 404 KASSERT((m->oflags & VPO_UNMANAGED) != 0, 405 ("kmem_malloc: page %p is managed", m)); 406 } 407 VM_OBJECT_UNLOCK(kmem_object); 408 409 /* 410 * Mark map entry as non-pageable. Repeat the assert. 411 */ 412 KASSERT(entry->start == addr && entry->end == addr + size && 413 entry->wired_count == 0, 414 ("kmem_back: entry not found or misaligned after allocation")); 415 entry->wired_count = 1; 416 417 /* 418 * At this point, the kmem_object must be unlocked because 419 * vm_map_simplify_entry() calls vm_object_deallocate(), which 420 * locks the kmem_object. 421 */ 422 vm_map_simplify_entry(map, entry); 423 424 /* 425 * Loop thru pages, entering them in the pmap. 426 */ 427 VM_OBJECT_LOCK(kmem_object); 428 for (i = 0; i < size; i += PAGE_SIZE) { 429 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i)); 430 /* 431 * Because this is kernel_pmap, this call will not block. 432 */ 433 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL, 434 TRUE); 435 vm_page_wakeup(m); 436 } 437 VM_OBJECT_UNLOCK(kmem_object); 438 439 return (KERN_SUCCESS); 440} 441 442/* 443 * kmem_alloc_wait: 444 * 445 * Allocates pageable memory from a sub-map of the kernel. If the submap 446 * has no room, the caller sleeps waiting for more memory in the submap. 447 * 448 * This routine may block. 449 */ 450vm_offset_t 451kmem_alloc_wait(map, size) 452 vm_map_t map; 453 vm_size_t size; 454{ 455 vm_offset_t addr; 456 457 size = round_page(size); 458 if (!swap_reserve(size)) 459 return (0); 460 461 for (;;) { 462 /* 463 * To make this work for more than one map, use the map's lock 464 * to lock out sleepers/wakers. 465 */ 466 vm_map_lock(map); 467 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0) 468 break; 469 /* no space now; see if we can ever get space */ 470 if (vm_map_max(map) - vm_map_min(map) < size) { 471 vm_map_unlock(map); 472 swap_release(size); 473 return (0); 474 } 475 map->needs_wakeup = TRUE; 476 vm_map_unlock_and_wait(map, 0); 477 } 478 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL, 479 VM_PROT_ALL, MAP_ACC_CHARGED); 480 vm_map_unlock(map); 481 return (addr); 482} 483 484/* 485 * kmem_free_wakeup: 486 * 487 * Returns memory to a submap of the kernel, and wakes up any processes 488 * waiting for memory in that map. 489 */ 490void 491kmem_free_wakeup(map, addr, size) 492 vm_map_t map; 493 vm_offset_t addr; 494 vm_size_t size; 495{ 496 497 vm_map_lock(map); 498 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 499 if (map->needs_wakeup) { 500 map->needs_wakeup = FALSE; 501 vm_map_wakeup(map); 502 } 503 vm_map_unlock(map); 504} 505 506static void 507kmem_init_zero_region(void) 508{ 509 vm_offset_t addr, i; 510 vm_page_t m; 511 int error; 512 513 /* 514 * Map a single physical page of zeros to a larger virtual range. 515 * This requires less looping in places that want large amounts of 516 * zeros, while not using much more physical resources. 517 */ 518 addr = kmem_alloc_nofault(kernel_map, ZERO_REGION_SIZE); 519 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL | 520 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO); 521 if ((m->flags & PG_ZERO) == 0) 522 pmap_zero_page(m); 523 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE) 524 pmap_qenter(addr + i, &m, 1); 525 error = vm_map_protect(kernel_map, addr, addr + ZERO_REGION_SIZE, 526 VM_PROT_READ, TRUE); 527 KASSERT(error == 0, ("error=%d", error)); 528 529 zero_region = (const void *)addr; 530} 531 532/* 533 * kmem_init: 534 * 535 * Create the kernel map; insert a mapping covering kernel text, 536 * data, bss, and all space allocated thus far (`boostrap' data). The 537 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and 538 * `start' as allocated, and the range between `start' and `end' as free. 539 */ 540void 541kmem_init(start, end) 542 vm_offset_t start, end; 543{ 544 vm_map_t m; 545 546 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end); 547 m->system_map = 1; 548 vm_map_lock(m); 549 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 550 kernel_map = m; 551 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0, 552#ifdef __amd64__ 553 KERNBASE, 554#else 555 VM_MIN_KERNEL_ADDRESS, 556#endif 557 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT); 558 /* ... and ending with the completion of the above `insert' */ 559 vm_map_unlock(m); 560 561 kmem_init_zero_region(); 562} 563 564#ifdef DIAGNOSTIC 565/* 566 * Allow userspace to directly trigger the VM drain routine for testing 567 * purposes. 568 */ 569static int 570debug_vm_lowmem(SYSCTL_HANDLER_ARGS) 571{ 572 int error, i; 573 574 i = 0; 575 error = sysctl_handle_int(oidp, &i, 0, req); 576 if (error) 577 return (error); 578 if (i) 579 EVENTHANDLER_INVOKE(vm_lowmem, 0); 580 return (0); 581} 582 583SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0, 584 debug_vm_lowmem, "I", "set to trigger vm_lowmem event"); 585#endif 586