1 /* 2 * (MPSAFE) 3 * 4 * Copyright (c) 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * The Mach Operating System project at Carnegie-Mellon University. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 35 * 36 * 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 41 * 42 * Permission to use, copy, modify and distribute this software and 43 * its documentation is hereby granted, provided that both the copyright 44 * notice and this permission notice appear in all copies of the 45 * software, derivative works or modified versions, and any portions 46 * thereof, and that both notices appear in supporting documentation. 47 * 48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 51 * 52 * Carnegie Mellon requests users of this software to return to 53 * 54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55 * School of Computer Science 56 * Carnegie Mellon University 57 * Pittsburgh PA 15213-3890 58 * 59 * any improvements or extensions that they make and grant Carnegie the 60 * rights to redistribute these changes. 61 * 62 * $FreeBSD: src/sys/vm/vm_kern.c,v 1.61.2.2 2002/03/12 18:25:26 tegge Exp $ 63 */ 64 65 /* 66 * Kernel memory management. 67 */ 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/proc.h> 72 #include <sys/malloc.h> 73 #include <sys/kernel.h> 74 #include <sys/sysctl.h> 75 76 #include <vm/vm.h> 77 #include <vm/vm_param.h> 78 #include <sys/lock.h> 79 #include <vm/pmap.h> 80 #include <vm/vm_map.h> 81 #include <vm/vm_object.h> 82 #include <vm/vm_page.h> 83 #include <vm/vm_pageout.h> 84 #include <vm/vm_kern.h> 85 #include <vm/vm_extern.h> 86 87 static struct vm_map kernel_map_store; 88 static struct vm_map clean_map_store; 89 static struct vm_map buffer_map_store; 90 91 struct vm_map *kernel_map = &kernel_map_store; 92 struct vm_map *clean_map = &clean_map_store; 93 struct vm_map *buffer_map = &buffer_map_store; 94 95 static __inline 96 int 97 KMVMCPU(int kmflags) 98 { 99 if ((kmflags & KM_CPU_SPEC) == 0) 100 return 0; 101 return VM_ALLOC_CPU(KM_GETCPU(kmflags)); 102 } 103 104 /* 105 * Allocate pageable swap-backed anonymous memory 106 */ 107 void * 108 kmem_alloc_swapbacked(kmem_anon_desc_t *kp, vm_size_t size, vm_subsys_t id) 109 { 110 int error; 111 vm_pindex_t npages; 112 113 size = round_page(size); 114 npages = size / PAGE_SIZE; 115 116 if (kp->map == NULL) 117 kp->map = kernel_map; 118 kp->data = vm_map_min(kernel_map); 119 kp->size = size; 120 kp->object = vm_object_allocate(OBJT_DEFAULT, npages); 121 122 error = vm_map_find(kp->map, kp->object, NULL, 0, 123 &kp->data, size, 124 PAGE_SIZE, TRUE, 125 VM_MAPTYPE_NORMAL, id, 126 VM_PROT_ALL, VM_PROT_ALL, 0); 127 if (error) { 128 kprintf("kmem_alloc_swapbacked: %zd bytes failed %d\n", 129 size, error); 130 kp->data = (vm_offset_t)0; 131 kmem_free_swapbacked(kp); 132 return NULL; 133 } 134 return ((void *)(intptr_t)kp->data); 135 } 136 137 void 138 kmem_free_swapbacked(kmem_anon_desc_t *kp) 139 { 140 if (kp->data) { 141 /* 142 * The object will be deallocated by kmem_free(). 143 */ 144 kmem_free(kp->map, kp->data, kp->size); 145 kp->data = (vm_offset_t)0; 146 } else { 147 /* 148 * Failure during allocation, object must be deallocated 149 * manually. 150 */ 151 vm_object_deallocate(kp->object); 152 } 153 kp->object = NULL; 154 } 155 156 /* 157 * Allocate pageable memory to the kernel's address map. "map" must 158 * be kernel_map or a submap of kernel_map. Caller must adjust map or 159 * enter VM pages itself. 160 * 161 * No requirements. 162 */ 163 vm_offset_t 164 kmem_alloc_pageable(vm_map_t map, vm_size_t size, vm_subsys_t id) 165 { 166 vm_offset_t addr; 167 int result; 168 169 size = round_page(size); 170 addr = vm_map_min(map); 171 result = vm_map_find(map, NULL, NULL, 172 (vm_offset_t) 0, &addr, size, 173 PAGE_SIZE, TRUE, 174 VM_MAPTYPE_NORMAL, id, 175 VM_PROT_ALL, VM_PROT_ALL, 0); 176 if (result != KERN_SUCCESS) 177 return (0); 178 return (addr); 179 } 180 181 /* 182 * Same as kmem_alloc_pageable, except that it create a nofault entry. 183 * 184 * No requirements. 185 */ 186 vm_offset_t 187 kmem_alloc_nofault(vm_map_t map, vm_size_t size, vm_subsys_t id, 188 vm_size_t align) 189 { 190 vm_offset_t addr; 191 int result; 192 193 size = round_page(size); 194 addr = vm_map_min(map); 195 result = vm_map_find(map, NULL, NULL, 196 (vm_offset_t) 0, &addr, size, 197 align, TRUE, 198 VM_MAPTYPE_NORMAL, id, 199 VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT); 200 if (result != KERN_SUCCESS) 201 return (0); 202 return (addr); 203 } 204 205 /* 206 * Allocate wired-down memory in the kernel's address map or a submap. 207 * 208 * No requirements. 209 */ 210 vm_offset_t 211 kmem_alloc3(vm_map_t map, vm_size_t size, vm_subsys_t id, int kmflags) 212 { 213 vm_offset_t addr; 214 vm_offset_t gstart; 215 vm_offset_t i; 216 int count; 217 int cow; 218 219 size = round_page(size); 220 221 if (kmflags & KM_KRESERVE) 222 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT); 223 else 224 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 225 226 if (kmflags & KM_STACK) { 227 cow = MAP_IS_KSTACK; 228 gstart = PAGE_SIZE; 229 } else { 230 cow = 0; 231 gstart = 0; 232 } 233 234 /* 235 * Use the kernel object for wired-down kernel pages. Assume that no 236 * region of the kernel object is referenced more than once. 237 * 238 * Locate sufficient space in the map. This will give us the final 239 * virtual address for the new memory, and thus will tell us the 240 * offset within the kernel map. 241 */ 242 vm_map_lock(map); 243 if (vm_map_findspace(map, vm_map_min(map), size, PAGE_SIZE, 0, &addr)) { 244 vm_map_unlock(map); 245 if (kmflags & KM_KRESERVE) 246 vm_map_entry_krelease(count); 247 else 248 vm_map_entry_release(count); 249 return (0); 250 } 251 vm_object_hold(kernel_object); 252 vm_object_reference_locked(kernel_object); 253 vm_map_insert(map, &count, 254 kernel_object, NULL, 255 addr, NULL, 256 addr, addr + size, 257 VM_MAPTYPE_NORMAL, id, 258 VM_PROT_ALL, VM_PROT_ALL, cow); 259 vm_object_drop(kernel_object); 260 261 vm_map_unlock(map); 262 if (kmflags & KM_KRESERVE) 263 vm_map_entry_krelease(count); 264 else 265 vm_map_entry_release(count); 266 267 /* 268 * Guarantee that there are pages already in this object before 269 * calling vm_map_wire. This is to prevent the following 270 * scenario: 271 * 272 * 1) Threads have swapped out, so that there is a pager for the 273 * kernel_object. 2) The kmsg zone is empty, and so we are 274 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault; 275 * there is no page, but there is a pager, so we call 276 * pager_data_request. But the kmsg zone is empty, so we must 277 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 278 * we get the data back from the pager, it will be (very stale) 279 * non-zero data. kmem_alloc is defined to return zero-filled memory. 280 * 281 * We're intentionally not activating the pages we allocate to prevent a 282 * race with page-out. vm_map_wire will wire the pages. 283 */ 284 vm_object_hold(kernel_object); 285 for (i = gstart; i < size; i += PAGE_SIZE) { 286 vm_page_t mem; 287 288 mem = vm_page_grab(kernel_object, OFF_TO_IDX(addr + i), 289 VM_ALLOC_FORCE_ZERO | VM_ALLOC_NORMAL | 290 VM_ALLOC_RETRY | KMVMCPU(kmflags)); 291 vm_page_unqueue_nowakeup(mem); 292 vm_page_wakeup(mem); 293 } 294 vm_object_drop(kernel_object); 295 296 /* 297 * And finally, mark the data as non-pageable. 298 * 299 * NOTE: vm_map_wire() handles any kstack guard. 300 */ 301 vm_map_wire(map, addr, addr + size, kmflags); 302 303 return (addr); 304 } 305 306 /* 307 * Release a region of kernel virtual memory allocated with kmem_alloc, 308 * and return the physical pages associated with that region. 309 * 310 * WARNING! If the caller entered pages into the region using pmap_kenter() 311 * it must remove the pages using pmap_kremove[_quick]() before freeing the 312 * underlying kmem, otherwise resident_count will be mistabulated. 313 * 314 * No requirements. 315 */ 316 void 317 kmem_free(vm_map_t map, vm_offset_t addr, vm_size_t size) 318 { 319 vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 320 } 321 322 /* 323 * Used to break a system map into smaller maps, usually to reduce 324 * contention and to provide large KVA spaces for subsystems like the 325 * buffer cache. 326 * 327 * parent Map to take range from 328 * result 329 * size Size of range to find 330 * min, max Returned endpoints of map 331 * pageable Can the region be paged 332 * 333 * No requirements. 334 */ 335 void 336 kmem_suballoc(vm_map_t parent, vm_map_t result, 337 vm_offset_t *min, vm_offset_t *max, vm_size_t size) 338 { 339 int ret; 340 341 size = round_page(size); 342 343 *min = (vm_offset_t) vm_map_min(parent); 344 ret = vm_map_find(parent, NULL, NULL, 345 (vm_offset_t) 0, min, size, 346 PAGE_SIZE, TRUE, 347 VM_MAPTYPE_UNSPECIFIED, VM_SUBSYS_SYSMAP, 348 VM_PROT_ALL, VM_PROT_ALL, 0); 349 if (ret != KERN_SUCCESS) { 350 kprintf("kmem_suballoc: bad status return of %d.\n", ret); 351 panic("kmem_suballoc"); 352 } 353 *max = *min + size; 354 pmap_reference(vm_map_pmap(parent)); 355 vm_map_init(result, *min, *max, vm_map_pmap(parent)); 356 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 357 panic("kmem_suballoc: unable to change range to submap"); 358 } 359 360 /* 361 * Allocates pageable memory from a sub-map of the kernel. If the submap 362 * has no room, the caller sleeps waiting for more memory in the submap. 363 * 364 * No requirements. 365 */ 366 vm_offset_t 367 kmem_alloc_wait(vm_map_t map, vm_size_t size, vm_subsys_t id) 368 { 369 vm_offset_t addr; 370 int count; 371 372 size = round_page(size); 373 374 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 375 376 for (;;) { 377 /* 378 * To make this work for more than one map, use the map's lock 379 * to lock out sleepers/wakers. 380 */ 381 vm_map_lock(map); 382 if (vm_map_findspace(map, vm_map_min(map), 383 size, PAGE_SIZE, 0, &addr) == 0) { 384 break; 385 } 386 /* no space now; see if we can ever get space */ 387 if (vm_map_max(map) - vm_map_min(map) < size) { 388 vm_map_entry_release(count); 389 vm_map_unlock(map); 390 return (0); 391 } 392 vm_map_unlock(map); 393 tsleep(map, 0, "kmaw", 0); 394 } 395 vm_map_insert(map, &count, 396 NULL, NULL, 397 (vm_offset_t)0, NULL, 398 addr, addr + size, 399 VM_MAPTYPE_NORMAL, id, 400 VM_PROT_ALL, VM_PROT_ALL, 0); 401 vm_map_unlock(map); 402 vm_map_entry_release(count); 403 404 return (addr); 405 } 406 407 /* 408 * Allocates a region from the kernel address map and physical pages 409 * within the specified address range to the kernel object. Creates a 410 * wired mapping from this region to these pages, and returns the 411 * region's starting virtual address. The allocated pages are not 412 * necessarily physically contiguous. If M_ZERO is specified through the 413 * given flags, then the pages are zeroed before they are mapped. 414 */ 415 vm_offset_t 416 kmem_alloc_attr(vm_map_t map, vm_size_t size, vm_subsys_t id, 417 int flags, vm_paddr_t low, 418 vm_paddr_t high, vm_memattr_t memattr) 419 { 420 vm_offset_t addr, i, offset; 421 vm_page_t m; 422 int count; 423 424 size = round_page(size); 425 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 426 vm_map_lock(map); 427 if (vm_map_findspace(map, vm_map_min(map), size, PAGE_SIZE, 428 flags, &addr)) { 429 vm_map_unlock(map); 430 vm_map_entry_release(count); 431 return (0); 432 } 433 offset = addr - vm_map_min(kernel_map); 434 vm_object_hold(kernel_object); 435 vm_object_reference_locked(kernel_object); 436 vm_map_insert(map, &count, 437 kernel_object, NULL, 438 offset, NULL, 439 addr, addr + size, 440 VM_MAPTYPE_NORMAL, id, 441 VM_PROT_ALL, VM_PROT_ALL, 0); 442 vm_map_unlock(map); 443 vm_map_entry_release(count); 444 vm_object_drop(kernel_object); 445 for (i = 0; i < size; i += PAGE_SIZE) { 446 m = vm_page_alloc_contig(low, high, PAGE_SIZE, 0, 447 PAGE_SIZE, memattr); 448 if (!m) { 449 return (0); 450 } 451 vm_object_hold(kernel_object); 452 vm_page_insert(m, kernel_object, OFF_TO_IDX(offset + i)); 453 vm_object_drop(kernel_object); 454 if (flags & M_ZERO) 455 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 456 m->valid = VM_PAGE_BITS_ALL; 457 } 458 vm_map_wire(map, addr, addr + size, 0); 459 return (addr); 460 } 461 462 463 /* 464 * Returns memory to a submap of the kernel, and wakes up any processes 465 * waiting for memory in that map. 466 * 467 * No requirements. 468 */ 469 void 470 kmem_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size) 471 { 472 int count; 473 474 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 475 vm_map_lock(map); 476 vm_map_delete(map, trunc_page(addr), round_page(addr + size), &count); 477 wakeup(map); 478 vm_map_unlock(map); 479 vm_map_entry_release(count); 480 } 481 482 /* 483 * Create the kernel_ma for (KvaStart,KvaEnd) and insert mappings to 484 * cover areas already allocated or reserved thus far. 485 * 486 * The areas (virtual_start, virtual_end) and (virtual2_start, virtual2_end) 487 * are available so the cutouts are the areas around these ranges between 488 * KvaStart and KvaEnd. 489 * 490 * Depend on the zalloc bootstrap cache to get our vm_map_entry_t. 491 * Called from the low level boot code only. 492 */ 493 void 494 kmem_init(void) 495 { 496 vm_offset_t addr; 497 vm_map_t m; 498 int count; 499 500 m = kernel_map; 501 vm_map_init(m, KvaStart, KvaEnd, kernel_pmap); 502 vm_map_lock(m); 503 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 504 m->system_map = 1; 505 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 506 addr = KvaStart; 507 if (virtual2_start) { 508 if (addr < virtual2_start) { 509 vm_map_insert(m, &count, 510 NULL, NULL, 511 (vm_offset_t) 0, NULL, 512 addr, virtual2_start, 513 VM_MAPTYPE_NORMAL, VM_SUBSYS_RESERVED, 514 VM_PROT_ALL, VM_PROT_ALL, 0); 515 } 516 addr = virtual2_end; 517 } 518 if (addr < virtual_start) { 519 vm_map_insert(m, &count, 520 NULL, NULL, 521 (vm_offset_t) 0, NULL, 522 addr, virtual_start, 523 VM_MAPTYPE_NORMAL, VM_SUBSYS_RESERVED, 524 VM_PROT_ALL, VM_PROT_ALL, 0); 525 } 526 addr = virtual_end; 527 if (addr < KvaEnd) { 528 vm_map_insert(m, &count, 529 NULL, NULL, 530 (vm_offset_t) 0, NULL, 531 addr, KvaEnd, 532 VM_MAPTYPE_NORMAL, VM_SUBSYS_RESERVED, 533 VM_PROT_ALL, VM_PROT_ALL, 0); 534 } 535 /* ... and ending with the completion of the above `insert' */ 536 vm_map_unlock(m); 537 vm_map_entry_release(count); 538 } 539 540 /* 541 * No requirements. 542 */ 543 static int 544 kvm_size(SYSCTL_HANDLER_ARGS) 545 { 546 unsigned long ksize = KvaSize; 547 548 return sysctl_handle_long(oidp, &ksize, 0, req); 549 } 550 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_ULONG|CTLFLAG_RD, 551 0, 0, kvm_size, "LU", "Size of KVM"); 552 553 /* 554 * No requirements. 555 */ 556 static int 557 kvm_free(SYSCTL_HANDLER_ARGS) 558 { 559 unsigned long kfree = virtual_end - kernel_vm_end; 560 561 return sysctl_handle_long(oidp, &kfree, 0, req); 562 } 563 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_ULONG|CTLFLAG_RD, 564 0, 0, kvm_free, "LU", "Amount of KVM free"); 565 566