subr_kmem.c revision 1.54
1 /* $NetBSD: subr_kmem.c,v 1.54 2014/06/24 07:28:23 maxv Exp $ */ 2 3 /*- 4 * Copyright (c) 2009 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Andrew Doran. 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c)2006 YAMAMOTO Takashi, 34 * All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 45 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 48 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 55 * SUCH DAMAGE. 56 */ 57 58 /* 59 * Allocator of kernel wired memory. 60 */ 61 62 /* 63 * This allocator has some debug features enabled with "option DEBUG" and 64 * "option DIAGNOSTIC". 65 * 66 * KMEM_POISON 67 * Try to detect modify-after-free bugs. 68 * 69 * Fill freed (in the sense of kmem_free) memory with a garbage pattern. 70 * Check the pattern on allocation. 71 * 72 * KMEM_REDZONE 73 * Try to detect overrun bugs. 74 * 75 * Add a 2-byte pattern (allocate some more bytes if needed) at the end 76 * of each allocated buffer. Check this pattern on kmem_free. 77 * 78 * KMEM_SIZE 79 * Try to detect alloc/free size mismatch bugs. 80 * 81 * Prefix each allocations with a fixed-sized header and record 82 * the exact user-requested allocation size in it. 83 * When freeing, compare it with kmem_free's "size" argument. 84 * 85 * KMEM_GUARD 86 * See the below "kmguard" section. 87 */ 88 89 /* 90 * kmguard 91 * 92 * A kernel with "option DEBUG" has "kmguard" debugging feature compiled in. 93 * See the comment in uvm/uvm_kmguard.c for what kind of bugs it tries to 94 * detect. Even if compiled in, it's disabled by default because it's very 95 * expensive. You can enable it on boot by: 96 * 97 * boot -d 98 * db> w kmem_guard_depth 0t30000 99 * db> c 100 * 101 * The default value of kmem_guard_depth is 0, which means disabled. 102 * It can be changed by KMEM_GUARD_DEPTH kernel config option. 103 */ 104 105 #include <sys/cdefs.h> 106 __KERNEL_RCSID(0, "$NetBSD: subr_kmem.c,v 1.54 2014/06/24 07:28:23 maxv Exp $"); 107 108 #include <sys/param.h> 109 #include <sys/callback.h> 110 #include <sys/kmem.h> 111 #include <sys/pool.h> 112 #include <sys/debug.h> 113 #include <sys/lockdebug.h> 114 #include <sys/cpu.h> 115 116 #include <uvm/uvm_extern.h> 117 #include <uvm/uvm_map.h> 118 #include <uvm/uvm_kmguard.h> 119 120 #include <lib/libkern/libkern.h> 121 122 struct kmem_cache_info { 123 size_t kc_size; 124 const char * kc_name; 125 }; 126 127 static const struct kmem_cache_info kmem_cache_sizes[] = { 128 { 8, "kmem-8" }, 129 { 16, "kmem-16" }, 130 { 24, "kmem-24" }, 131 { 32, "kmem-32" }, 132 { 40, "kmem-40" }, 133 { 48, "kmem-48" }, 134 { 56, "kmem-56" }, 135 { 64, "kmem-64" }, 136 { 80, "kmem-80" }, 137 { 96, "kmem-96" }, 138 { 112, "kmem-112" }, 139 { 128, "kmem-128" }, 140 { 160, "kmem-160" }, 141 { 192, "kmem-192" }, 142 { 224, "kmem-224" }, 143 { 256, "kmem-256" }, 144 { 320, "kmem-320" }, 145 { 384, "kmem-384" }, 146 { 448, "kmem-448" }, 147 { 512, "kmem-512" }, 148 { 768, "kmem-768" }, 149 { 1024, "kmem-1024" }, 150 { 0, NULL } 151 }; 152 153 static const struct kmem_cache_info kmem_cache_big_sizes[] = { 154 { 2048, "kmem-2048" }, 155 { 4096, "kmem-4096" }, 156 { 8192, "kmem-8192" }, 157 { 16384, "kmem-16384" }, 158 { 0, NULL } 159 }; 160 161 /* 162 * KMEM_ALIGN is the smallest guaranteed alignment and also the 163 * smallest allocateable quantum. 164 * Every cache size >= CACHE_LINE_SIZE gets CACHE_LINE_SIZE alignment. 165 */ 166 #define KMEM_ALIGN 8 167 #define KMEM_SHIFT 3 168 #define KMEM_MAXSIZE 1024 169 #define KMEM_CACHE_COUNT (KMEM_MAXSIZE >> KMEM_SHIFT) 170 171 static pool_cache_t kmem_cache[KMEM_CACHE_COUNT] __cacheline_aligned; 172 static size_t kmem_cache_maxidx __read_mostly; 173 174 #define KMEM_BIG_ALIGN 2048 175 #define KMEM_BIG_SHIFT 11 176 #define KMEM_BIG_MAXSIZE 16384 177 #define KMEM_CACHE_BIG_COUNT (KMEM_BIG_MAXSIZE >> KMEM_BIG_SHIFT) 178 179 static pool_cache_t kmem_cache_big[KMEM_CACHE_BIG_COUNT] __cacheline_aligned; 180 static size_t kmem_cache_big_maxidx __read_mostly; 181 182 #if defined(DIAGNOSTIC) && defined(_HARDKERNEL) 183 #define KMEM_SIZE 184 #endif /* defined(DIAGNOSTIC) */ 185 186 #if defined(DEBUG) && defined(_HARDKERNEL) 187 #define KMEM_POISON 188 #define KMEM_REDZONE 189 #define KMEM_GUARD 190 #endif /* defined(DEBUG) */ 191 192 #if defined(KMEM_POISON) 193 static int kmem_poison_ctor(void *, void *, int); 194 static void kmem_poison_fill(void *, size_t); 195 static void kmem_poison_check(void *, size_t); 196 #else /* defined(KMEM_POISON) */ 197 #define kmem_poison_fill(p, sz) /* nothing */ 198 #define kmem_poison_check(p, sz) /* nothing */ 199 #endif /* defined(KMEM_POISON) */ 200 201 #if defined(KMEM_REDZONE) 202 #define REDZONE_SIZE 2 203 static void kmem_redzone_fill(void *p, size_t sz); 204 static void kmem_redzone_check(void *p, size_t sz); 205 #else /* defined(KMEM_REDZONE) */ 206 #define REDZONE_SIZE 0 207 #define kmem_redzone_fill(p, sz) /* nothing */ 208 #define kmem_redzone_check(p, sz) /* nothing */ 209 #endif /* defined(KMEM_REDZONE) */ 210 211 #if defined(KMEM_SIZE) 212 #define SIZE_SIZE (MAX(KMEM_ALIGN, sizeof(size_t))) 213 static void kmem_size_set(void *, size_t); 214 static void kmem_size_check(void *, size_t); 215 #else 216 #define SIZE_SIZE 0 217 #define kmem_size_set(p, sz) /* nothing */ 218 #define kmem_size_check(p, sz) /* nothing */ 219 #endif 220 221 #if defined(KMEM_GUARD) 222 #ifndef KMEM_GUARD_DEPTH 223 #define KMEM_GUARD_DEPTH 0 224 #endif 225 int kmem_guard_depth = KMEM_GUARD_DEPTH; 226 size_t kmem_guard_size; 227 static struct uvm_kmguard kmem_guard; 228 static void *kmem_freecheck; 229 #endif /* defined(KMEM_GUARD) */ 230 231 CTASSERT(KM_SLEEP == PR_WAITOK); 232 CTASSERT(KM_NOSLEEP == PR_NOWAIT); 233 234 /* 235 * kmem_intr_alloc: allocate wired memory. 236 */ 237 238 void * 239 kmem_intr_alloc(size_t requested_size, km_flag_t kmflags) 240 { 241 size_t allocsz, index; 242 size_t size; 243 pool_cache_t pc; 244 uint8_t *p; 245 246 KASSERT(requested_size > 0); 247 248 #ifdef KMEM_GUARD 249 if (requested_size <= kmem_guard_size) { 250 return uvm_kmguard_alloc(&kmem_guard, requested_size, 251 (kmflags & KM_SLEEP) != 0); 252 } 253 #endif 254 size = kmem_roundup_size(requested_size); 255 allocsz = size + SIZE_SIZE; 256 257 #ifdef KMEM_REDZONE 258 if (size - requested_size < REDZONE_SIZE) { 259 /* If there isn't enough space in the page padding, 260 * allocate two more bytes for the red zone. */ 261 allocsz += REDZONE_SIZE; 262 } 263 #endif 264 265 if ((index = ((allocsz -1) >> KMEM_SHIFT)) 266 < kmem_cache_maxidx) { 267 pc = kmem_cache[index]; 268 } else if ((index = ((allocsz - 1) >> KMEM_BIG_SHIFT)) 269 < kmem_cache_big_maxidx) { 270 pc = kmem_cache_big[index]; 271 } else { 272 int ret = uvm_km_kmem_alloc(kmem_va_arena, 273 (vsize_t)round_page(size), 274 ((kmflags & KM_SLEEP) ? VM_SLEEP : VM_NOSLEEP) 275 | VM_INSTANTFIT, (vmem_addr_t *)&p); 276 if (ret) { 277 return NULL; 278 } 279 FREECHECK_OUT(&kmem_freecheck, p); 280 return p; 281 } 282 283 p = pool_cache_get(pc, kmflags); 284 285 if (__predict_true(p != NULL)) { 286 kmem_poison_check(p, size); 287 FREECHECK_OUT(&kmem_freecheck, p); 288 kmem_size_set(p, requested_size); 289 kmem_redzone_fill(p, requested_size + SIZE_SIZE); 290 291 return p + SIZE_SIZE; 292 } 293 return p; 294 } 295 296 /* 297 * kmem_intr_zalloc: allocate zeroed wired memory. 298 */ 299 300 void * 301 kmem_intr_zalloc(size_t size, km_flag_t kmflags) 302 { 303 void *p; 304 305 p = kmem_intr_alloc(size, kmflags); 306 if (p != NULL) { 307 memset(p, 0, size); 308 } 309 return p; 310 } 311 312 /* 313 * kmem_intr_free: free wired memory allocated by kmem_alloc. 314 */ 315 316 void 317 kmem_intr_free(void *p, size_t requested_size) 318 { 319 size_t allocsz, index; 320 size_t size; 321 pool_cache_t pc; 322 323 KASSERT(p != NULL); 324 KASSERT(requested_size > 0); 325 326 #ifdef KMEM_GUARD 327 if (requested_size <= kmem_guard_size) { 328 uvm_kmguard_free(&kmem_guard, requested_size, p); 329 return; 330 } 331 #endif 332 333 size = kmem_roundup_size(requested_size); 334 allocsz = size + SIZE_SIZE; 335 336 #ifdef KMEM_REDZONE 337 if (size - requested_size < REDZONE_SIZE) { 338 allocsz += REDZONE_SIZE; 339 } 340 #endif 341 342 if ((index = ((allocsz -1) >> KMEM_SHIFT)) 343 < kmem_cache_maxidx) { 344 pc = kmem_cache[index]; 345 } else if ((index = ((allocsz - 1) >> KMEM_BIG_SHIFT)) 346 < kmem_cache_big_maxidx) { 347 pc = kmem_cache_big[index]; 348 } else { 349 FREECHECK_IN(&kmem_freecheck, p); 350 uvm_km_kmem_free(kmem_va_arena, (vaddr_t)p, 351 round_page(size)); 352 return; 353 } 354 355 p = (uint8_t *)p - SIZE_SIZE; 356 kmem_size_check(p, requested_size); 357 kmem_redzone_check(p, requested_size + SIZE_SIZE); 358 FREECHECK_IN(&kmem_freecheck, p); 359 LOCKDEBUG_MEM_CHECK(p, size); 360 kmem_poison_fill(p, allocsz); 361 362 pool_cache_put(pc, p); 363 } 364 365 /* ---- kmem API */ 366 367 /* 368 * kmem_alloc: allocate wired memory. 369 * => must not be called from interrupt context. 370 */ 371 372 void * 373 kmem_alloc(size_t size, km_flag_t kmflags) 374 { 375 376 KASSERTMSG((!cpu_intr_p() && !cpu_softintr_p()), 377 "kmem(9) should not be used from the interrupt context"); 378 return kmem_intr_alloc(size, kmflags); 379 } 380 381 /* 382 * kmem_zalloc: allocate zeroed wired memory. 383 * => must not be called from interrupt context. 384 */ 385 386 void * 387 kmem_zalloc(size_t size, km_flag_t kmflags) 388 { 389 390 KASSERTMSG((!cpu_intr_p() && !cpu_softintr_p()), 391 "kmem(9) should not be used from the interrupt context"); 392 return kmem_intr_zalloc(size, kmflags); 393 } 394 395 /* 396 * kmem_free: free wired memory allocated by kmem_alloc. 397 * => must not be called from interrupt context. 398 */ 399 400 void 401 kmem_free(void *p, size_t size) 402 { 403 404 KASSERT(!cpu_intr_p()); 405 KASSERT(!cpu_softintr_p()); 406 kmem_intr_free(p, size); 407 } 408 409 static size_t 410 kmem_create_caches(const struct kmem_cache_info *array, 411 pool_cache_t alloc_table[], size_t maxsize, int shift, int ipl) 412 { 413 size_t maxidx = 0; 414 size_t table_unit = (1 << shift); 415 size_t size = table_unit; 416 int i; 417 418 for (i = 0; array[i].kc_size != 0 ; i++) { 419 const char *name = array[i].kc_name; 420 size_t cache_size = array[i].kc_size; 421 struct pool_allocator *pa; 422 int flags = PR_NOALIGN; 423 pool_cache_t pc; 424 size_t align; 425 426 if ((cache_size & (CACHE_LINE_SIZE - 1)) == 0) 427 align = CACHE_LINE_SIZE; 428 else if ((cache_size & (PAGE_SIZE - 1)) == 0) 429 align = PAGE_SIZE; 430 else 431 align = KMEM_ALIGN; 432 433 if (cache_size < CACHE_LINE_SIZE) 434 flags |= PR_NOTOUCH; 435 436 /* check if we reached the requested size */ 437 if (cache_size > maxsize || cache_size > PAGE_SIZE) { 438 break; 439 } 440 if ((cache_size >> shift) > maxidx) { 441 maxidx = cache_size >> shift; 442 } 443 444 if ((cache_size >> shift) > maxidx) { 445 maxidx = cache_size >> shift; 446 } 447 448 pa = &pool_allocator_kmem; 449 #if defined(KMEM_POISON) 450 pc = pool_cache_init(cache_size, align, 0, flags, 451 name, pa, ipl, kmem_poison_ctor, 452 NULL, (void *)cache_size); 453 #else /* defined(KMEM_POISON) */ 454 pc = pool_cache_init(cache_size, align, 0, flags, 455 name, pa, ipl, NULL, NULL, NULL); 456 #endif /* defined(KMEM_POISON) */ 457 458 while (size <= cache_size) { 459 alloc_table[(size - 1) >> shift] = pc; 460 size += table_unit; 461 } 462 } 463 return maxidx; 464 } 465 466 void 467 kmem_init(void) 468 { 469 470 #ifdef KMEM_GUARD 471 uvm_kmguard_init(&kmem_guard, &kmem_guard_depth, &kmem_guard_size, 472 kmem_va_arena); 473 #endif 474 kmem_cache_maxidx = kmem_create_caches(kmem_cache_sizes, 475 kmem_cache, KMEM_MAXSIZE, KMEM_SHIFT, IPL_VM); 476 kmem_cache_big_maxidx = kmem_create_caches(kmem_cache_big_sizes, 477 kmem_cache_big, PAGE_SIZE, KMEM_BIG_SHIFT, IPL_VM); 478 } 479 480 size_t 481 kmem_roundup_size(size_t size) 482 { 483 484 return (size + (KMEM_ALIGN - 1)) & ~(KMEM_ALIGN - 1); 485 } 486 487 /* ------------------ DEBUG / DIAGNOSTIC ------------------ */ 488 489 #if defined(KMEM_POISON) || defined(KMEM_REDZONE) 490 #if defined(_LP64) 491 #define PRIME 0x9e37fffffffc0000UL 492 #else /* defined(_LP64) */ 493 #define PRIME 0x9e3779b1 494 #endif /* defined(_LP64) */ 495 #endif /* defined(KMEM_POISON) || defined(KMEM_REDZONE) */ 496 497 #if defined(KMEM_POISON) 498 static inline uint8_t 499 kmem_poison_pattern(const void *p) 500 { 501 return (uint8_t)(((uintptr_t)p) * PRIME 502 >> ((sizeof(uintptr_t) - sizeof(uint8_t))) * CHAR_BIT); 503 } 504 505 static int 506 kmem_poison_ctor(void *arg, void *obj, int flag) 507 { 508 size_t sz = (size_t)arg; 509 510 kmem_poison_fill(obj, sz); 511 512 return 0; 513 } 514 515 static void 516 kmem_poison_fill(void *p, size_t sz) 517 { 518 uint8_t *cp; 519 const uint8_t *ep; 520 521 cp = p; 522 ep = cp + sz; 523 while (cp < ep) { 524 *cp = kmem_poison_pattern(cp); 525 cp++; 526 } 527 } 528 529 static void 530 kmem_poison_check(void *p, size_t sz) 531 { 532 uint8_t *cp; 533 const uint8_t *ep; 534 535 cp = p; 536 ep = cp + sz; 537 while (cp < ep) { 538 const uint8_t expected = kmem_poison_pattern(cp); 539 540 if (*cp != expected) { 541 panic("%s: %p: 0x%02x != 0x%02x\n", 542 __func__, cp, *cp, expected); 543 } 544 cp++; 545 } 546 } 547 #endif /* defined(KMEM_POISON) */ 548 549 #if defined(KMEM_SIZE) 550 static void 551 kmem_size_set(void *p, size_t sz) 552 { 553 memcpy(p, &sz, sizeof(sz)); 554 } 555 556 static void 557 kmem_size_check(void *p, size_t sz) 558 { 559 size_t psz; 560 561 memcpy(&psz, p, sizeof(psz)); 562 if (psz != sz) { 563 panic("kmem_free(%p, %zu) != allocated size %zu", 564 (const uint8_t *)p + SIZE_SIZE, sz, psz); 565 } 566 } 567 #endif /* defined(KMEM_SIZE) */ 568 569 #if defined(KMEM_REDZONE) 570 static inline uint8_t 571 kmem_redzone_pattern(const void *p) 572 { 573 return (uint8_t)(((uintptr_t)p) * PRIME 574 >> ((sizeof(uintptr_t) - sizeof(uint8_t))) * CHAR_BIT); 575 } 576 577 static void 578 kmem_redzone_fill(void *p, size_t sz) 579 { 580 uint8_t *cp; 581 const uint8_t *ep; 582 583 cp = (uint8_t *)p + sz; 584 ep = cp + REDZONE_SIZE; 585 while (cp < ep) { 586 *cp = kmem_redzone_pattern(cp); 587 cp++; 588 } 589 } 590 591 static void 592 kmem_redzone_check(void *p, size_t sz) 593 { 594 uint8_t *cp; 595 const uint8_t *ep; 596 597 cp = (uint8_t *)p + sz; 598 ep = (uint8_t *)p + sz + REDZONE_SIZE; 599 while (cp < ep) { 600 const uint8_t expected = kmem_redzone_pattern(cp); 601 602 if (*cp != expected) { 603 panic("%s: %p: 0x%02x != 0x%02x\n", 604 __func__, cp, *cp, expected); 605 } 606 cp++; 607 } 608 } 609 #endif /* defined(KMEM_REDZONE) */ 610 611 612 /* 613 * Used to dynamically allocate string with kmem accordingly to format. 614 */ 615 char * 616 kmem_asprintf(const char *fmt, ...) 617 { 618 int size __diagused, len; 619 va_list va; 620 char *str; 621 622 va_start(va, fmt); 623 len = vsnprintf(NULL, 0, fmt, va); 624 va_end(va); 625 626 str = kmem_alloc(len + 1, KM_SLEEP); 627 628 va_start(va, fmt); 629 size = vsnprintf(str, len + 1, fmt, va); 630 va_end(va); 631 632 KASSERT(size == len); 633 634 return str; 635 } 636
Indexes created Tue Sep 23 02:09:52 GMT 2025