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/proc.h> 74 #include <sys/malloc.h> 75 #include <sys/rwlock.h> 76 #include <sys/sysctl.h> 77 #include <sys/vmem.h> 78 79 #include <vm/vm.h> 80 #include <vm/vm_param.h> 81 #include <vm/vm_kern.h> 82 #include <vm/pmap.h> 83 #include <vm/vm_map.h> 84 #include <vm/vm_object.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_pageout.h> 87 #include <vm/vm_extern.h> 88 #include <vm/uma.h> 89 90 vm_map_t kernel_map; 91 vm_map_t exec_map; 92 vm_map_t pipe_map; 93 94 const void *zero_region; 95 CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0); 96 97 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD, 98 NULL, VM_MIN_KERNEL_ADDRESS, "Min kernel address"); 99 100 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD, 101 #if defined(__arm__) || defined(__sparc64__) 102 &vm_max_kernel_address, 0, 103 #else 104 NULL, VM_MAX_KERNEL_ADDRESS, 105 #endif 106 "Max kernel address"); 107 108 /* 109 * kva_alloc: 110 * 111 * Allocate a virtual address range with no underlying object and 112 * no initial mapping to physical memory. Any mapping from this 113 * range to physical memory must be explicitly created prior to 114 * its use, typically with pmap_qenter(). Any attempt to create 115 * a mapping on demand through vm_fault() will result in a panic. 116 */ 117 vm_offset_t 118 kva_alloc(size) 119 vm_size_t size; 120 { 121 vm_offset_t addr; 122 123 size = round_page(size); 124 if (vmem_alloc(kernel_arena, size, M_BESTFIT | M_NOWAIT, &addr)) 125 return (0); 126 127 return (addr); 128 } 129 130 /* 131 * kva_free: 132 * 133 * Release a region of kernel virtual memory allocated 134 * with kva_alloc, and return the physical pages 135 * associated with that region. 136 * 137 * This routine may not block on kernel maps. 138 */ 139 void 140 kva_free(addr, size) 141 vm_offset_t addr; 142 vm_size_t size; 143 { 144 145 size = round_page(size); 146 vmem_free(kernel_arena, addr, size); 147 } 148 149 /* 150 * Allocates a region from the kernel address map and physical pages 151 * within the specified address range to the kernel object. Creates a 152 * wired mapping from this region to these pages, and returns the 153 * region's starting virtual address. The allocated pages are not 154 * necessarily physically contiguous. If M_ZERO is specified through the 155 * given flags, then the pages are zeroed before they are mapped. 156 */ 157 vm_offset_t 158 kmem_alloc_attr(vmem_t *vmem, vm_size_t size, int flags, vm_paddr_t low, 159 vm_paddr_t high, vm_memattr_t memattr) 160 { 161 vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object; 162 vm_offset_t addr; 163 vm_ooffset_t offset; 164 vm_page_t m; 165 int pflags, tries; 166 int i; 167 168 size = round_page(size); 169 if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr)) 170 return (0); 171 offset = addr - VM_MIN_KERNEL_ADDRESS; 172 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED; 173 VM_OBJECT_WLOCK(object); 174 for (i = 0; i < size; i += PAGE_SIZE) { 175 tries = 0; 176 retry: 177 m = vm_page_alloc_contig(object, OFF_TO_IDX(offset + i), 178 pflags, 1, low, high, PAGE_SIZE, 0, memattr); 179 if (m == NULL) { 180 VM_OBJECT_WUNLOCK(object); 181 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) { 182 vm_pageout_grow_cache(tries, low, high); 183 VM_OBJECT_WLOCK(object); 184 tries++; 185 goto retry; 186 } 187 /* 188 * Unmap and free the pages. 189 */ 190 if (i != 0) 191 pmap_remove(kernel_pmap, addr, addr + i); 192 while (i != 0) { 193 i -= PAGE_SIZE; 194 m = vm_page_lookup(object, 195 OFF_TO_IDX(offset + i)); 196 vm_page_unwire(m, PQ_INACTIVE); 197 vm_page_free(m); 198 } 199 vmem_free(vmem, addr, size); 200 return (0); 201 } 202 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0) 203 pmap_zero_page(m); 204 m->valid = VM_PAGE_BITS_ALL; 205 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL, 206 TRUE); 207 } 208 VM_OBJECT_WUNLOCK(object); 209 return (addr); 210 } 211 212 /* 213 * Allocates a region from the kernel address map and physically 214 * contiguous pages within the specified address range to the kernel 215 * object. Creates a wired mapping from this region to these pages, and 216 * returns the region's starting virtual address. If M_ZERO is specified 217 * through the given flags, then the pages are zeroed before they are 218 * mapped. 219 */ 220 vm_offset_t 221 kmem_alloc_contig(struct vmem *vmem, vm_size_t size, int flags, vm_paddr_t low, 222 vm_paddr_t high, u_long alignment, vm_paddr_t boundary, 223 vm_memattr_t memattr) 224 { 225 vm_object_t object = vmem == kmem_arena ? kmem_object : kernel_object; 226 vm_offset_t addr, tmp; 227 vm_ooffset_t offset; 228 vm_page_t end_m, m; 229 int pflags, tries; 230 231 size = round_page(size); 232 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr)) 233 return (0); 234 offset = addr - VM_MIN_KERNEL_ADDRESS; 235 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED; 236 VM_OBJECT_WLOCK(object); 237 tries = 0; 238 retry: 239 m = vm_page_alloc_contig(object, OFF_TO_IDX(offset), pflags, 240 atop(size), low, high, alignment, boundary, memattr); 241 if (m == NULL) { 242 VM_OBJECT_WUNLOCK(object); 243 if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) { 244 vm_pageout_grow_cache(tries, low, high); 245 VM_OBJECT_WLOCK(object); 246 tries++; 247 goto retry; 248 } 249 vmem_free(vmem, addr, size); 250 return (0); 251 } 252 end_m = m + atop(size); 253 tmp = addr; 254 for (; m < end_m; m++) { 255 if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0) 256 pmap_zero_page(m); 257 m->valid = VM_PAGE_BITS_ALL; 258 pmap_enter(kernel_pmap, tmp, VM_PROT_ALL, m, VM_PROT_ALL, true); 259 tmp += PAGE_SIZE; 260 } 261 VM_OBJECT_WUNLOCK(object); 262 return (addr); 263 } 264 265 /* 266 * kmem_suballoc: 267 * 268 * Allocates a map to manage a subrange 269 * of the kernel virtual address space. 270 * 271 * Arguments are as follows: 272 * 273 * parent Map to take range from 274 * min, max Returned endpoints of map 275 * size Size of range to find 276 * superpage_align Request that min is superpage aligned 277 */ 278 vm_map_t 279 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max, 280 vm_size_t size, boolean_t superpage_align) 281 { 282 int ret; 283 vm_map_t result; 284 285 size = round_page(size); 286 287 *min = vm_map_min(parent); 288 ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ? 289 VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL, 290 MAP_ACC_NO_CHARGE); 291 if (ret != KERN_SUCCESS) 292 panic("kmem_suballoc: bad status return of %d", ret); 293 *max = *min + size; 294 result = vm_map_create(vm_map_pmap(parent), *min, *max); 295 if (result == NULL) 296 panic("kmem_suballoc: cannot create submap"); 297 if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS) 298 panic("kmem_suballoc: unable to change range to submap"); 299 return (result); 300 } 301 302 /* 303 * kmem_malloc: 304 * 305 * Allocate wired-down pages in the kernel's address space. 306 */ 307 vm_offset_t 308 kmem_malloc(struct vmem *vmem, vm_size_t size, int flags) 309 { 310 vm_offset_t addr; 311 int rv; 312 313 size = round_page(size); 314 if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr)) 315 return (0); 316 317 rv = kmem_back((vmem == kmem_arena) ? kmem_object : kernel_object, 318 addr, size, flags); 319 if (rv != KERN_SUCCESS) { 320 vmem_free(vmem, addr, size); 321 return (0); 322 } 323 return (addr); 324 } 325 326 /* 327 * kmem_back: 328 * 329 * Allocate physical pages for the specified virtual address range. 330 */ 331 int 332 kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags) 333 { 334 vm_offset_t offset, i; 335 vm_page_t m; 336 int pflags; 337 338 KASSERT(object == kmem_object || object == kernel_object, 339 ("kmem_back: only supports kernel objects.")); 340 341 offset = addr - VM_MIN_KERNEL_ADDRESS; 342 pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED; 343 344 VM_OBJECT_WLOCK(object); 345 for (i = 0; i < size; i += PAGE_SIZE) { 346 retry: 347 m = vm_page_alloc(object, OFF_TO_IDX(offset + i), pflags); 348 349 /* 350 * Ran out of space, free everything up and return. Don't need 351 * to lock page queues here as we know that the pages we got 352 * aren't on any queues. 353 */ 354 if (m == NULL) { 355 if ((flags & M_NOWAIT) == 0) { 356 VM_OBJECT_WUNLOCK(object); 357 VM_WAIT; 358 VM_OBJECT_WLOCK(object); 359 goto retry; 360 } 361 /* 362 * Unmap and free the pages. 363 */ 364 if (i != 0) 365 pmap_remove(kernel_pmap, addr, addr + i); 366 while (i != 0) { 367 i -= PAGE_SIZE; 368 m = vm_page_lookup(object, 369 OFF_TO_IDX(offset + i)); 370 vm_page_unwire(m, PQ_INACTIVE); 371 vm_page_free(m); 372 } 373 VM_OBJECT_WUNLOCK(object); 374 return (KERN_NO_SPACE); 375 } 376 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0) 377 pmap_zero_page(m); 378 KASSERT((m->oflags & VPO_UNMANAGED) != 0, 379 ("kmem_malloc: page %p is managed", m)); 380 m->valid = VM_PAGE_BITS_ALL; 381 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL, 382 TRUE); 383 } 384 VM_OBJECT_WUNLOCK(object); 385 386 return (KERN_SUCCESS); 387 } 388 389 void 390 kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size) 391 { 392 vm_page_t m; 393 vm_offset_t offset; 394 int i; 395 396 KASSERT(object == kmem_object || object == kernel_object, 397 ("kmem_unback: only supports kernel objects.")); 398 399 pmap_remove(kernel_pmap, addr, addr + size); 400 offset = addr - VM_MIN_KERNEL_ADDRESS; 401 VM_OBJECT_WLOCK(object); 402 for (i = 0; i < size; i += PAGE_SIZE) { 403 m = vm_page_lookup(object, OFF_TO_IDX(offset + i)); 404 vm_page_unwire(m, PQ_INACTIVE); 405 vm_page_free(m); 406 } 407 VM_OBJECT_WUNLOCK(object); 408 } 409 410 /* 411 * kmem_free: 412 * 413 * Free memory allocated with kmem_malloc. The size must match the 414 * original allocation. 415 */ 416 void 417 kmem_free(struct vmem *vmem, vm_offset_t addr, vm_size_t size) 418 { 419 420 size = round_page(size); 421 kmem_unback((vmem == kmem_arena) ? kmem_object : kernel_object, 422 addr, size); 423 vmem_free(vmem, addr, size); 424 } 425 426 /* 427 * kmap_alloc_wait: 428 * 429 * Allocates pageable memory from a sub-map of the kernel. If the submap 430 * has no room, the caller sleeps waiting for more memory in the submap. 431 * 432 * This routine may block. 433 */ 434 vm_offset_t 435 kmap_alloc_wait(map, size) 436 vm_map_t map; 437 vm_size_t size; 438 { 439 vm_offset_t addr; 440 441 size = round_page(size); 442 if (!swap_reserve(size)) 443 return (0); 444 445 for (;;) { 446 /* 447 * To make this work for more than one map, use the map's lock 448 * to lock out sleepers/wakers. 449 */ 450 vm_map_lock(map); 451 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0) 452 break; 453 /* no space now; see if we can ever get space */ 454 if (vm_map_max(map) - vm_map_min(map) < size) { 455 vm_map_unlock(map); 456 swap_release(size); 457 return (0); 458 } 459 map->needs_wakeup = TRUE; 460 vm_map_unlock_and_wait(map, 0); 461 } 462 vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL, 463 VM_PROT_ALL, MAP_ACC_CHARGED); 464 vm_map_unlock(map); 465 return (addr); 466 } 467 468 /* 469 * kmap_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 kmap_free_wakeup(map, addr, size) 476 vm_map_t map; 477 vm_offset_t addr; 478 vm_size_t size; 479 { 480 481 vm_map_lock(map); 482 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 483 if (map->needs_wakeup) { 484 map->needs_wakeup = FALSE; 485 vm_map_wakeup(map); 486 } 487 vm_map_unlock(map); 488 } 489 490 void 491 kmem_init_zero_region(void) 492 { 493 vm_offset_t addr, i; 494 vm_page_t m; 495 496 /* 497 * Map a single physical page of zeros to a larger virtual range. 498 * This requires less looping in places that want large amounts of 499 * zeros, while not using much more physical resources. 500 */ 501 addr = kva_alloc(ZERO_REGION_SIZE); 502 m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL | 503 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO); 504 if ((m->flags & PG_ZERO) == 0) 505 pmap_zero_page(m); 506 for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE) 507 pmap_qenter(addr + i, &m, 1); 508 pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ); 509 510 zero_region = (const void *)addr; 511 } 512 513 /* 514 * kmem_init: 515 * 516 * Create the kernel map; insert a mapping covering kernel text, 517 * data, bss, and all space allocated thus far (`boostrap' data). The 518 * new map will thus map the range between VM_MIN_KERNEL_ADDRESS and 519 * `start' as allocated, and the range between `start' and `end' as free. 520 */ 521 void 522 kmem_init(start, end) 523 vm_offset_t start, end; 524 { 525 vm_map_t m; 526 527 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end); 528 m->system_map = 1; 529 vm_map_lock(m); 530 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 531 kernel_map = m; 532 (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0, 533 #ifdef __amd64__ 534 KERNBASE, 535 #else 536 VM_MIN_KERNEL_ADDRESS, 537 #endif 538 start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT); 539 /* ... and ending with the completion of the above `insert' */ 540 vm_map_unlock(m); 541 } 542 543 #ifdef DIAGNOSTIC 544 /* 545 * Allow userspace to directly trigger the VM drain routine for testing 546 * purposes. 547 */ 548 static int 549 debug_vm_lowmem(SYSCTL_HANDLER_ARGS) 550 { 551 int error, i; 552 553 i = 0; 554 error = sysctl_handle_int(oidp, &i, 0, req); 555 if (error) 556 return (error); 557 if (i) 558 EVENTHANDLER_INVOKE(vm_lowmem, 0); 559 return (0); 560 } 561 562 SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem, CTLTYPE_INT | CTLFLAG_RW, 0, 0, 563 debug_vm_lowmem, "I", "set to trigger vm_lowmem event"); 564 #endif 565