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 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 37 * 38 * 39 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 40 * All rights reserved. 41 * 42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 43 * 44 * Permission to use, copy, modify and distribute this software and 45 * its documentation is hereby granted, provided that both the copyright 46 * notice and this permission notice appear in all copies of the 47 * software, derivative works or modified versions, and any portions 48 * thereof, and that both notices appear in supporting documentation. 49 * 50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 53 * 54 * Carnegie Mellon requests users of this software to return to 55 * 56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 57 * School of Computer Science 58 * Carnegie Mellon University 59 * Pittsburgh PA 15213-3890 60 * 61 * any improvements or extensions that they make and grant Carnegie the 62 * rights to redistribute these changes. 63 * 64 * $Id: vm_kern.c,v 1.28 1996/11/17 02:38:29 dyson Exp $ 65 */ 66 67 /* 68 * Kernel memory management. 69 */ 70 71 #include <sys/param.h> 72 #include <sys/systm.h> 73 #include <sys/kernel.h> 74 #include <sys/proc.h> 75 #include <sys/malloc.h> 76 #include <sys/syslog.h> 77 #include <sys/queue.h> 78 #include <sys/vmmeter.h> 79 80 #include <vm/vm.h> 81 #include <vm/vm_param.h> 82 #include <vm/vm_prot.h> 83 #include <vm/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 vm_map_t kernel_map=0; 93 vm_map_t kmem_map=0; 94 vm_map_t exec_map=0; 95 vm_map_t exech_map=0; 96 vm_map_t clean_map=0; 97 vm_map_t u_map=0; 98 vm_map_t buffer_map=0; 99 vm_map_t mb_map=0; 100 int mb_map_full=0; 101 vm_map_t mcl_map=0; 102 int mcl_map_full=0; 103 vm_map_t io_map=0; 104 vm_map_t phys_map=0; 105 106 /* 107 * kmem_alloc_pageable: 108 * 109 * Allocate pageable memory to the kernel's address map. 110 * "map" must be kernel_map or a submap of kernel_map. 111 */ 112 113 vm_offset_t 114 kmem_alloc_pageable(map, size) 115 vm_map_t map; 116 register vm_size_t size; 117 { 118 vm_offset_t addr; 119 register int result; 120 121 size = round_page(size); 122 addr = vm_map_min(map); 123 result = vm_map_find(map, NULL, (vm_offset_t) 0, 124 &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 125 if (result != KERN_SUCCESS) { 126 return (0); 127 } 128 return (addr); 129 } 130 131 /* 132 * Allocate wired-down memory in the kernel's address map 133 * or a submap. 134 */ 135 vm_offset_t 136 kmem_alloc(map, size) 137 register vm_map_t map; 138 register vm_size_t size; 139 { 140 vm_offset_t addr; 141 register vm_offset_t offset; 142 vm_offset_t i; 143 144 size = round_page(size); 145 146 /* 147 * Use the kernel object for wired-down kernel pages. Assume that no 148 * region of the kernel object is referenced more than once. 149 */ 150 151 /* 152 * Locate sufficient space in the map. This will give us the final 153 * virtual address for the new memory, and thus will tell us the 154 * offset within the kernel map. 155 */ 156 vm_map_lock(map); 157 if (vm_map_findspace(map, 0, size, &addr)) { 158 vm_map_unlock(map); 159 return (0); 160 } 161 offset = addr - VM_MIN_KERNEL_ADDRESS; 162 vm_object_reference(kernel_object); 163 vm_map_insert(map, kernel_object, offset, addr, addr + size, 164 VM_PROT_ALL, VM_PROT_ALL, 0); 165 vm_map_unlock(map); 166 167 /* 168 * Guarantee that there are pages already in this object before 169 * calling vm_map_pageable. This is to prevent the following 170 * scenario: 171 * 172 * 1) Threads have swapped out, so that there is a pager for the 173 * kernel_object. 2) The kmsg zone is empty, and so we are 174 * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 175 * there is no page, but there is a pager, so we call 176 * pager_data_request. But the kmsg zone is empty, so we must 177 * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 178 * we get the data back from the pager, it will be (very stale) 179 * non-zero data. kmem_alloc is defined to return zero-filled memory. 180 * 181 * We're intentionally not activating the pages we allocate to prevent a 182 * race with page-out. vm_map_pageable will wire the pages. 183 */ 184 185 for (i = 0; i < size; i += PAGE_SIZE) { 186 vm_page_t mem; 187 188 while ((mem = vm_page_alloc(kernel_object, 189 OFF_TO_IDX(offset + i), VM_ALLOC_ZERO)) == NULL) { 190 VM_WAIT; 191 } 192 if ((mem->flags & PG_ZERO) == 0) 193 vm_page_zero_fill(mem); 194 mem->flags &= ~(PG_BUSY|PG_ZERO); 195 mem->valid = VM_PAGE_BITS_ALL; 196 } 197 198 /* 199 * And finally, mark the data as non-pageable. 200 */ 201 202 (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 203 204 return (addr); 205 } 206 207 /* 208 * kmem_free: 209 * 210 * Release a region of kernel virtual memory allocated 211 * with kmem_alloc, and return the physical pages 212 * associated with that region. 213 */ 214 void 215 kmem_free(map, addr, size) 216 vm_map_t map; 217 register vm_offset_t addr; 218 vm_size_t size; 219 { 220 (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 221 } 222 223 /* 224 * kmem_suballoc: 225 * 226 * Allocates a map to manage a subrange 227 * of the kernel virtual address space. 228 * 229 * Arguments are as follows: 230 * 231 * parent Map to take range from 232 * size Size of range to find 233 * min, max Returned endpoints of map 234 * pageable Can the region be paged 235 */ 236 vm_map_t 237 kmem_suballoc(parent, min, max, size, pageable) 238 register vm_map_t parent; 239 vm_offset_t *min, *max; 240 register vm_size_t size; 241 boolean_t pageable; 242 { 243 register int ret; 244 vm_map_t result; 245 246 size = round_page(size); 247 248 *min = (vm_offset_t) vm_map_min(parent); 249 ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 250 min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 251 if (ret != KERN_SUCCESS) { 252 printf("kmem_suballoc: bad status return of %d.\n", ret); 253 panic("kmem_suballoc"); 254 } 255 *max = *min + size; 256 pmap_reference(vm_map_pmap(parent)); 257 result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 258 if (result == NULL) 259 panic("kmem_suballoc: cannot create submap"); 260 if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 261 panic("kmem_suballoc: unable to change range to submap"); 262 return (result); 263 } 264 265 /* 266 * Allocate wired-down memory in the kernel's address map for the higher 267 * level kernel memory allocator (kern/kern_malloc.c). We cannot use 268 * kmem_alloc() because we may need to allocate memory at interrupt 269 * level where we cannot block (canwait == FALSE). 270 * 271 * This routine has its own private kernel submap (kmem_map) and object 272 * (kmem_object). This, combined with the fact that only malloc uses 273 * this routine, ensures that we will never block in map or object waits. 274 * 275 * Note that this still only works in a uni-processor environment and 276 * when called at splhigh(). 277 * 278 * We don't worry about expanding the map (adding entries) since entries 279 * for wired maps are statically allocated. 280 */ 281 vm_offset_t 282 kmem_malloc(map, size, waitflag) 283 register vm_map_t map; 284 register vm_size_t size; 285 boolean_t waitflag; 286 { 287 register vm_offset_t offset, i; 288 vm_map_entry_t entry; 289 vm_offset_t addr; 290 vm_page_t m; 291 292 if (map != kmem_map && map != mb_map && map != mcl_map) 293 panic("kmem_malloc: map != {kmem,mb,mcl}_map"); 294 295 size = round_page(size); 296 addr = vm_map_min(map); 297 298 /* 299 * Locate sufficient space in the map. This will give us the final 300 * virtual address for the new memory, and thus will tell us the 301 * offset within the kernel map. 302 */ 303 vm_map_lock(map); 304 if (vm_map_findspace(map, 0, size, &addr)) { 305 vm_map_unlock(map); 306 if (map == mb_map) { 307 mb_map_full = TRUE; 308 log(LOG_ERR, "Out of mbufs - increase maxusers!\n"); 309 return (0); 310 } 311 if (map == mcl_map) { 312 mcl_map_full = TRUE; 313 log(LOG_ERR, 314 "Out of mbuf clusters - increase maxusers!\n"); 315 return (0); 316 } 317 if (waitflag == M_WAITOK) 318 panic("kmem_malloc: kmem_map too small"); 319 return (0); 320 } 321 offset = addr - VM_MIN_KERNEL_ADDRESS; 322 vm_object_reference(kmem_object); 323 vm_map_insert(map, kmem_object, offset, addr, addr + size, 324 VM_PROT_ALL, VM_PROT_ALL, 0); 325 326 for (i = 0; i < size; i += PAGE_SIZE) { 327 retry: 328 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), 329 (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 330 331 /* 332 * Ran out of space, free everything up and return. Don't need 333 * to lock page queues here as we know that the pages we got 334 * aren't on any queues. 335 */ 336 if (m == NULL) { 337 if (waitflag == M_WAITOK) { 338 VM_WAIT; 339 goto retry; 340 } 341 while (i != 0) { 342 i -= PAGE_SIZE; 343 m = vm_page_lookup(kmem_object, 344 OFF_TO_IDX(offset + i)); 345 PAGE_WAKEUP(m); 346 vm_page_free(m); 347 } 348 vm_map_delete(map, addr, addr + size); 349 vm_map_unlock(map); 350 return (0); 351 } 352 m->flags &= ~PG_ZERO; 353 m->valid = VM_PAGE_BITS_ALL; 354 } 355 356 /* 357 * Mark map entry as non-pageable. Assert: vm_map_insert() will never 358 * be able to extend the previous entry so there will be a new entry 359 * exactly corresponding to this address range and it will have 360 * wired_count == 0. 361 */ 362 if (!vm_map_lookup_entry(map, addr, &entry) || 363 entry->start != addr || entry->end != addr + size || 364 entry->wired_count) 365 panic("kmem_malloc: entry not found or misaligned"); 366 entry->wired_count++; 367 368 vm_map_simplify_entry(map, entry); 369 370 /* 371 * Loop thru pages, entering them in the pmap. (We cannot add them to 372 * the wired count without wrapping the vm_page_queue_lock in 373 * splimp...) 374 */ 375 for (i = 0; i < size; i += PAGE_SIZE) { 376 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i)); 377 vm_page_wire(m); 378 PAGE_WAKEUP(m); 379 pmap_enter(kernel_pmap, addr + i, VM_PAGE_TO_PHYS(m), 380 VM_PROT_ALL, 1); 381 m->flags |= PG_MAPPED|PG_WRITEABLE; 382 } 383 vm_map_unlock(map); 384 385 return (addr); 386 } 387 388 /* 389 * kmem_alloc_wait 390 * 391 * Allocates pageable memory from a sub-map of the kernel. If the submap 392 * has no room, the caller sleeps waiting for more memory in the submap. 393 * 394 */ 395 vm_offset_t 396 kmem_alloc_wait(map, size) 397 vm_map_t map; 398 vm_size_t size; 399 { 400 vm_offset_t addr; 401 402 size = round_page(size); 403 404 for (;;) { 405 /* 406 * To make this work for more than one map, use the map's lock 407 * to lock out sleepers/wakers. 408 */ 409 vm_map_lock(map); 410 if (vm_map_findspace(map, 0, size, &addr) == 0) 411 break; 412 /* no space now; see if we can ever get space */ 413 if (vm_map_max(map) - vm_map_min(map) < size) { 414 vm_map_unlock(map); 415 return (0); 416 } 417 vm_map_unlock(map); 418 tsleep(map, PVM, "kmaw", 0); 419 } 420 vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0); 421 vm_map_unlock(map); 422 return (addr); 423 } 424 425 /* 426 * kmem_free_wakeup 427 * 428 * Returns memory to a submap of the kernel, and wakes up any processes 429 * waiting for memory in that map. 430 */ 431 void 432 kmem_free_wakeup(map, addr, size) 433 vm_map_t map; 434 vm_offset_t addr; 435 vm_size_t size; 436 { 437 vm_map_lock(map); 438 (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 439 wakeup(map); 440 vm_map_unlock(map); 441 } 442 443 /* 444 * Create the kernel map; insert a mapping covering kernel text, data, bss, 445 * and all space allocated thus far (`boostrap' data). The new map will thus 446 * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 447 * the range between `start' and `end' as free. 448 */ 449 void 450 kmem_init(start, end) 451 vm_offset_t start, end; 452 { 453 register vm_map_t m; 454 455 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 456 vm_map_lock(m); 457 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 458 kernel_map = m; 459 (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 460 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0); 461 /* ... and ending with the completion of the above `insert' */ 462 vm_map_unlock(m); 463 } 464