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. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 39 * 40 * 41 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 42 * All rights reserved. 43 * 44 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 45 * 46 * Permission to use, copy, modify and distribute this software and 47 * its documentation is hereby granted, provided that both the copyright 48 * notice and this permission notice appear in all copies of the 49 * software, derivative works or modified versions, and any portions 50 * thereof, and that both notices appear in supporting documentation. 51 * 52 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 53 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 54 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 55 * 56 * Carnegie Mellon requests users of this software to return to 57 * 58 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 59 * School of Computer Science 60 * Carnegie Mellon University 61 * Pittsburgh PA 15213-3890 62 * 63 * any improvements or extensions that they make and grant Carnegie the 64 * rights to redistribute these changes. 65 * 66 * $FreeBSD: src/sys/vm/vm_kern.c,v 1.61.2.2 2002/03/12 18:25:26 tegge Exp $ 67 * $DragonFly: src/sys/vm/vm_kern.c,v 1.29 2007/06/07 23:14:29 dillon Exp $ 68 */ 69 70 /* 71 * Kernel memory management. 72 */ 73 74 #include <sys/param.h> 75 #include <sys/systm.h> 76 #include <sys/proc.h> 77 #include <sys/malloc.h> 78 #include <sys/kernel.h> 79 #include <sys/sysctl.h> 80 81 #include <vm/vm.h> 82 #include <vm/vm_param.h> 83 #include <sys/lock.h> 84 #include <vm/pmap.h> 85 #include <vm/vm_map.h> 86 #include <vm/vm_object.h> 87 #include <vm/vm_page.h> 88 #include <vm/vm_pageout.h> 89 #include <vm/vm_kern.h> 90 #include <vm/vm_extern.h> 91 92 struct vm_map kernel_map; 93 struct vm_map clean_map; 94 struct vm_map buffer_map; 95 96 /* 97 * Allocate pageable memory to the kernel's address map. "map" must 98 * be kernel_map or a submap of kernel_map. 99 * 100 * No requirements. 101 */ 102 vm_offset_t 103 kmem_alloc_pageable(vm_map_t map, vm_size_t size) 104 { 105 vm_offset_t addr; 106 int result; 107 108 size = round_page(size); 109 addr = vm_map_min(map); 110 result = vm_map_find(map, NULL, (vm_offset_t) 0, 111 &addr, size, PAGE_SIZE, 112 TRUE, VM_MAPTYPE_NORMAL, 113 VM_PROT_ALL, VM_PROT_ALL, 114 0); 115 if (result != KERN_SUCCESS) { 116 return (0); 117 } 118 return (addr); 119 } 120 121 /* 122 * Same as kmem_alloc_pageable, except that it create a nofault entry. 123 * 124 * No requirements. 125 */ 126 vm_offset_t 127 kmem_alloc_nofault(vm_map_t map, vm_size_t size, vm_size_t align) 128 { 129 vm_offset_t addr; 130 int result; 131 132 size = round_page(size); 133 addr = vm_map_min(map); 134 result = vm_map_find(map, NULL, (vm_offset_t) 0, 135 &addr, size, align, 136 TRUE, VM_MAPTYPE_NORMAL, 137 VM_PROT_ALL, VM_PROT_ALL, 138 MAP_NOFAULT); 139 if (result != KERN_SUCCESS) { 140 return (0); 141 } 142 return (addr); 143 } 144 145 /* 146 * Allocate wired-down memory in the kernel's address map or a submap. 147 * 148 * No requirements. 149 */ 150 vm_offset_t 151 kmem_alloc3(vm_map_t map, vm_size_t size, int kmflags) 152 { 153 vm_offset_t addr; 154 vm_offset_t i; 155 int count; 156 157 size = round_page(size); 158 159 if (kmflags & KM_KRESERVE) 160 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT); 161 else 162 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 163 164 /* 165 * Use the kernel object for wired-down kernel pages. Assume that no 166 * region of the kernel object is referenced more than once. 167 * 168 * Locate sufficient space in the map. This will give us the final 169 * virtual address for the new memory, and thus will tell us the 170 * offset within the kernel map. 171 */ 172 vm_map_lock(map); 173 if (vm_map_findspace(map, vm_map_min(map), size, PAGE_SIZE, 0, &addr)) { 174 vm_map_unlock(map); 175 if (kmflags & KM_KRESERVE) 176 vm_map_entry_krelease(count); 177 else 178 vm_map_entry_release(count); 179 return (0); 180 } 181 vm_object_reference(&kernel_object); 182 vm_map_insert(map, &count, 183 &kernel_object, addr, addr, addr + size, 184 VM_MAPTYPE_NORMAL, 185 VM_PROT_ALL, VM_PROT_ALL, 186 0); 187 vm_map_unlock(map); 188 if (kmflags & KM_KRESERVE) 189 vm_map_entry_krelease(count); 190 else 191 vm_map_entry_release(count); 192 193 /* 194 * Guarantee that there are pages already in this object before 195 * calling vm_map_wire. This is to prevent the following 196 * scenario: 197 * 198 * 1) Threads have swapped out, so that there is a pager for the 199 * kernel_object. 2) The kmsg zone is empty, and so we are 200 * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault; 201 * there is no page, but there is a pager, so we call 202 * pager_data_request. But the kmsg zone is empty, so we must 203 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 204 * we get the data back from the pager, it will be (very stale) 205 * non-zero data. kmem_alloc is defined to return zero-filled memory. 206 * 207 * We're intentionally not activating the pages we allocate to prevent a 208 * race with page-out. vm_map_wire will wire the pages. 209 */ 210 211 lwkt_gettoken(&vm_token); 212 for (i = 0; i < size; i += PAGE_SIZE) { 213 vm_page_t mem; 214 215 mem = vm_page_grab(&kernel_object, OFF_TO_IDX(addr + i), 216 VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 217 if ((mem->flags & PG_ZERO) == 0) 218 vm_page_zero_fill(mem); 219 mem->valid = VM_PAGE_BITS_ALL; 220 vm_page_flag_clear(mem, PG_ZERO); 221 vm_page_wakeup(mem); 222 } 223 lwkt_reltoken(&vm_token); 224 225 /* 226 * And finally, mark the data as non-pageable. 227 */ 228 vm_map_wire(map, (vm_offset_t)addr, addr + size, kmflags); 229 230 return (addr); 231 } 232 233 /* 234 * Release a region of kernel virtual memory allocated with kmem_alloc, 235 * and return the physical pages associated with that region. 236 * 237 * WARNING! If the caller entered pages into the region using pmap_kenter() 238 * it must remove the pages using pmap_kremove[_quick]() before freeing the 239 * underlying kmem, otherwise resident_count will be mistabulated. 240 * 241 * No requirements. 242 */ 243 void 244 kmem_free(vm_map_t map, vm_offset_t addr, vm_size_t size) 245 { 246 vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 247 } 248 249 /* 250 * Used to break a system map into smaller maps, usually to reduce 251 * contention and to provide large KVA spaces for subsystems like the 252 * buffer cache. 253 * 254 * parent Map to take range from 255 * result 256 * size Size of range to find 257 * min, max Returned endpoints of map 258 * pageable Can the region be paged 259 * 260 * No requirements. 261 */ 262 void 263 kmem_suballoc(vm_map_t parent, vm_map_t result, 264 vm_offset_t *min, vm_offset_t *max, vm_size_t size) 265 { 266 int ret; 267 268 size = round_page(size); 269 270 lwkt_gettoken(&vm_token); 271 *min = (vm_offset_t) vm_map_min(parent); 272 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 273 min, size, PAGE_SIZE, 274 TRUE, VM_MAPTYPE_UNSPECIFIED, 275 VM_PROT_ALL, VM_PROT_ALL, 276 0); 277 if (ret != KERN_SUCCESS) { 278 kprintf("kmem_suballoc: bad status return of %d.\n", ret); 279 panic("kmem_suballoc"); 280 } 281 *max = *min + size; 282 pmap_reference(vm_map_pmap(parent)); 283 vm_map_init(result, *min, *max, vm_map_pmap(parent)); 284 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 285 panic("kmem_suballoc: unable to change range to submap"); 286 lwkt_reltoken(&vm_token); 287 } 288 289 /* 290 * Allocates pageable memory from a sub-map of the kernel. If the submap 291 * has no room, the caller sleeps waiting for more memory in the submap. 292 * 293 * No requirements. 294 */ 295 vm_offset_t 296 kmem_alloc_wait(vm_map_t map, vm_size_t size) 297 { 298 vm_offset_t addr; 299 int count; 300 301 size = round_page(size); 302 303 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 304 305 for (;;) { 306 /* 307 * To make this work for more than one map, use the map's lock 308 * to lock out sleepers/wakers. 309 */ 310 vm_map_lock(map); 311 if (vm_map_findspace(map, vm_map_min(map), 312 size, PAGE_SIZE, 0, &addr) == 0) { 313 break; 314 } 315 /* no space now; see if we can ever get space */ 316 if (vm_map_max(map) - vm_map_min(map) < size) { 317 vm_map_entry_release(count); 318 vm_map_unlock(map); 319 return (0); 320 } 321 vm_map_unlock(map); 322 tsleep(map, 0, "kmaw", 0); 323 } 324 vm_map_insert(map, &count, 325 NULL, (vm_offset_t) 0, 326 addr, addr + size, 327 VM_MAPTYPE_NORMAL, 328 VM_PROT_ALL, VM_PROT_ALL, 329 0); 330 vm_map_unlock(map); 331 vm_map_entry_release(count); 332 333 return (addr); 334 } 335 336 /* 337 * Returns memory to a submap of the kernel, and wakes up any processes 338 * waiting for memory in that map. 339 * 340 * No requirements. 341 */ 342 void 343 kmem_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size) 344 { 345 int count; 346 347 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 348 vm_map_lock(map); 349 vm_map_delete(map, trunc_page(addr), round_page(addr + size), &count); 350 wakeup(map); 351 vm_map_unlock(map); 352 vm_map_entry_release(count); 353 } 354 355 /* 356 * Create the kernel_map and insert mappings to cover areas already 357 * allocated or reserved thus far. That is, the area (KvaStart,start) 358 * and (end,KvaEnd) must be marked as allocated. 359 * 360 * virtual2_start/end is a cutout Between KvaStart and start, 361 * for x86_64 due to the location of KERNBASE (at -2G). 362 * 363 * We could use a min_offset of 0 instead of KvaStart, but since the 364 * min_offset is not used for any calculations other then a bounds check 365 * it does not effect readability. KvaStart is more appropriate. 366 * 367 * Depend on the zalloc bootstrap cache to get our vm_map_entry_t. 368 * Called from the low level boot code only. 369 */ 370 void 371 kmem_init(vm_offset_t start, vm_offset_t end) 372 { 373 vm_offset_t addr; 374 vm_map_t m; 375 int count; 376 377 m = vm_map_create(&kernel_map, &kernel_pmap, KvaStart, KvaEnd); 378 vm_map_lock(m); 379 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 380 m->system_map = 1; 381 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 382 addr = KvaStart; 383 if (virtual2_start) { 384 if (addr < virtual2_start) { 385 vm_map_insert(m, &count, NULL, (vm_offset_t) 0, 386 addr, virtual2_start, 387 VM_MAPTYPE_NORMAL, 388 VM_PROT_ALL, VM_PROT_ALL, 389 0); 390 } 391 addr = virtual2_end; 392 } 393 if (addr < start) { 394 vm_map_insert(m, &count, NULL, (vm_offset_t) 0, 395 addr, start, 396 VM_MAPTYPE_NORMAL, 397 VM_PROT_ALL, VM_PROT_ALL, 398 0); 399 } 400 addr = end; 401 if (addr < KvaEnd) { 402 vm_map_insert(m, &count, NULL, (vm_offset_t) 0, 403 addr, KvaEnd, 404 VM_MAPTYPE_NORMAL, 405 VM_PROT_ALL, VM_PROT_ALL, 406 0); 407 } 408 /* ... and ending with the completion of the above `insert' */ 409 vm_map_unlock(m); 410 vm_map_entry_release(count); 411 } 412 413 /* 414 * No requirements. 415 */ 416 static int 417 kvm_size(SYSCTL_HANDLER_ARGS) 418 { 419 unsigned long ksize = KvaSize; 420 421 return sysctl_handle_long(oidp, &ksize, 0, req); 422 } 423 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD, 424 0, 0, kvm_size, "IU", "Size of KVM"); 425 426 /* 427 * No requirements. 428 */ 429 static int 430 kvm_free(SYSCTL_HANDLER_ARGS) 431 { 432 unsigned long kfree = virtual_end - kernel_vm_end; 433 434 return sysctl_handle_long(oidp, &kfree, 0, req); 435 } 436 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD, 437 0, 0, kvm_free, "IU", "Amount of KVM free"); 438 439