subr_kmem.c revision 1.59
1 /* $NetBSD: subr_kmem.c,v 1.59 2014/07/03 08:43:49 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. This allocator has some debug features 60 * enabled with "option DIAGNOSTIC" and "option DEBUG". 61 */ 62 63 /* 64 * KMEM_SIZE: detect alloc/free size mismatch bugs. 65 * Prefix each allocations with a fixed-sized, aligned header and record 66 * the exact user-requested allocation size in it. When freeing, compare 67 * it with kmem_free's "size" argument. 68 */ 69 70 /* 71 * KMEM_REDZONE: detect overrun bugs. 72 * Add a 2-byte pattern (allocate one more memory chunk if needed) at the 73 * end of each allocated buffer. Check this pattern on kmem_free. 74 * 75 * KMEM_POISON: detect modify-after-free bugs. 76 * Fill freed (in the sense of kmem_free) memory with a garbage pattern. 77 * Check the pattern on allocation. 78 * 79 * KMEM_GUARD 80 * A kernel with "option DEBUG" has "kmguard" debugging feature compiled 81 * in. See the comment in uvm/uvm_kmguard.c for what kind of bugs it tries 82 * to detect. Even if compiled in, it's disabled by default because it's 83 * very expensive. You can enable it on boot by: 84 * boot -d 85 * db> w kmem_guard_depth 0t30000 86 * db> c 87 * 88 * The default value of kmem_guard_depth is 0, which means disabled. 89 * It can be changed by KMEM_GUARD_DEPTH kernel config option. 90 */ 91 92 #include <sys/cdefs.h> 93 __KERNEL_RCSID(0, "$NetBSD: subr_kmem.c,v 1.59 2014/07/03 08:43:49 maxv Exp $"); 94 95 #include <sys/param.h> 96 #include <sys/callback.h> 97 #include <sys/kmem.h> 98 #include <sys/pool.h> 99 #include <sys/debug.h> 100 #include <sys/lockdebug.h> 101 #include <sys/cpu.h> 102 103 #include <uvm/uvm_extern.h> 104 #include <uvm/uvm_map.h> 105 #include <uvm/uvm_kmguard.h> 106 107 #include <lib/libkern/libkern.h> 108 109 struct kmem_cache_info { 110 size_t kc_size; 111 const char * kc_name; 112 }; 113 114 static const struct kmem_cache_info kmem_cache_sizes[] = { 115 { 8, "kmem-8" }, 116 { 16, "kmem-16" }, 117 { 24, "kmem-24" }, 118 { 32, "kmem-32" }, 119 { 40, "kmem-40" }, 120 { 48, "kmem-48" }, 121 { 56, "kmem-56" }, 122 { 64, "kmem-64" }, 123 { 80, "kmem-80" }, 124 { 96, "kmem-96" }, 125 { 112, "kmem-112" }, 126 { 128, "kmem-128" }, 127 { 160, "kmem-160" }, 128 { 192, "kmem-192" }, 129 { 224, "kmem-224" }, 130 { 256, "kmem-256" }, 131 { 320, "kmem-320" }, 132 { 384, "kmem-384" }, 133 { 448, "kmem-448" }, 134 { 512, "kmem-512" }, 135 { 768, "kmem-768" }, 136 { 1024, "kmem-1024" }, 137 { 0, NULL } 138 }; 139 140 static const struct kmem_cache_info kmem_cache_big_sizes[] = { 141 { 2048, "kmem-2048" }, 142 { 4096, "kmem-4096" }, 143 { 8192, "kmem-8192" }, 144 { 16384, "kmem-16384" }, 145 { 0, NULL } 146 }; 147 148 /* 149 * KMEM_ALIGN is the smallest guaranteed alignment and also the 150 * smallest allocateable quantum. 151 * Every cache size >= CACHE_LINE_SIZE gets CACHE_LINE_SIZE alignment. 152 */ 153 #define KMEM_ALIGN 8 154 #define KMEM_SHIFT 3 155 #define KMEM_MAXSIZE 1024 156 #define KMEM_CACHE_COUNT (KMEM_MAXSIZE >> KMEM_SHIFT) 157 158 static pool_cache_t kmem_cache[KMEM_CACHE_COUNT] __cacheline_aligned; 159 static size_t kmem_cache_maxidx __read_mostly; 160 161 #define KMEM_BIG_ALIGN 2048 162 #define KMEM_BIG_SHIFT 11 163 #define KMEM_BIG_MAXSIZE 16384 164 #define KMEM_CACHE_BIG_COUNT (KMEM_BIG_MAXSIZE >> KMEM_BIG_SHIFT) 165 166 static pool_cache_t kmem_cache_big[KMEM_CACHE_BIG_COUNT] __cacheline_aligned; 167 static size_t kmem_cache_big_maxidx __read_mostly; 168 169 #if defined(DIAGNOSTIC) && defined(_HARDKERNEL) 170 #define KMEM_SIZE 171 #endif /* defined(DIAGNOSTIC) */ 172 173 #if defined(DEBUG) && defined(_HARDKERNEL) 174 #define KMEM_POISON 175 #define KMEM_REDZONE 176 #define KMEM_GUARD 177 #endif /* defined(DEBUG) */ 178 179 #if defined(KMEM_POISON) 180 static int kmem_poison_ctor(void *, void *, int); 181 static void kmem_poison_fill(void *, size_t); 182 static void kmem_poison_check(void *, size_t); 183 #else /* defined(KMEM_POISON) */ 184 #define kmem_poison_fill(p, sz) /* nothing */ 185 #define kmem_poison_check(p, sz) /* nothing */ 186 #endif /* defined(KMEM_POISON) */ 187 188 #if defined(KMEM_REDZONE) 189 #define REDZONE_SIZE 2 190 static void kmem_redzone_fill(void *, size_t); 191 static void kmem_redzone_check(void *, size_t); 192 #else /* defined(KMEM_REDZONE) */ 193 #define REDZONE_SIZE 0 194 #define kmem_redzone_fill(p, sz) /* nothing */ 195 #define kmem_redzone_check(p, sz) /* nothing */ 196 #endif /* defined(KMEM_REDZONE) */ 197 198 #if defined(KMEM_SIZE) 199 struct kmem_header { 200 size_t size; 201 } __aligned(KMEM_ALIGN); 202 #define SIZE_SIZE sizeof(struct kmem_header) 203 static void kmem_size_set(void *, size_t); 204 static void kmem_size_check(void *, size_t); 205 #else 206 #define SIZE_SIZE 0 207 #define kmem_size_set(p, sz) /* nothing */ 208 #define kmem_size_check(p, sz) /* nothing */ 209 #endif 210 211 #if defined(KMEM_GUARD) 212 #ifndef KMEM_GUARD_DEPTH 213 #define KMEM_GUARD_DEPTH 0 214 #endif 215 int kmem_guard_depth = KMEM_GUARD_DEPTH; 216 size_t kmem_guard_size; 217 static struct uvm_kmguard kmem_guard; 218 static void *kmem_freecheck; 219 #endif /* defined(KMEM_GUARD) */ 220 221 CTASSERT(KM_SLEEP == PR_WAITOK); 222 CTASSERT(KM_NOSLEEP == PR_NOWAIT); 223 224 /* 225 * kmem_intr_alloc: allocate wired memory. 226 */ 227 228 void * 229 kmem_intr_alloc(size_t requested_size, km_flag_t kmflags) 230 { 231 size_t allocsz, index; 232 size_t size; 233 pool_cache_t pc; 234 uint8_t *p; 235 236 KASSERT(requested_size > 0); 237 238 #ifdef KMEM_GUARD 239 if (requested_size <= kmem_guard_size) { 240 return uvm_kmguard_alloc(&kmem_guard, requested_size, 241 (kmflags & KM_SLEEP) != 0); 242 } 243 #endif 244 size = kmem_roundup_size(requested_size); 245 allocsz = size + SIZE_SIZE; 246 247 #ifdef KMEM_REDZONE 248 if (size - requested_size < REDZONE_SIZE) { 249 /* If there isn't enough space in the padding, allocate 250 * one more memory chunk for the red zone. */ 251 allocsz += kmem_roundup_size(REDZONE_SIZE); 252 } 253 #endif 254 255 if ((index = ((allocsz -1) >> KMEM_SHIFT)) 256 < kmem_cache_maxidx) { 257 pc = kmem_cache[index]; 258 } else if ((index = ((allocsz - 1) >> KMEM_BIG_SHIFT)) 259 < kmem_cache_big_maxidx) { 260 pc = kmem_cache_big[index]; 261 } else { 262 int ret = uvm_km_kmem_alloc(kmem_va_arena, 263 (vsize_t)round_page(size), 264 ((kmflags & KM_SLEEP) ? VM_SLEEP : VM_NOSLEEP) 265 | VM_INSTANTFIT, (vmem_addr_t *)&p); 266 if (ret) { 267 return NULL; 268 } 269 FREECHECK_OUT(&kmem_freecheck, p); 270 return p; 271 } 272 273 p = pool_cache_get(pc, kmflags); 274 275 if (__predict_true(p != NULL)) { 276 kmem_poison_check(p, allocsz); 277 FREECHECK_OUT(&kmem_freecheck, p); 278 kmem_size_set(p, requested_size); 279 kmem_redzone_fill(p, requested_size + SIZE_SIZE); 280 281 return p + SIZE_SIZE; 282 } 283 return p; 284 } 285 286 /* 287 * kmem_intr_zalloc: allocate zeroed wired memory. 288 */ 289 290 void * 291 kmem_intr_zalloc(size_t size, km_flag_t kmflags) 292 { 293 void *p; 294 295 p = kmem_intr_alloc(size, kmflags); 296 if (p != NULL) { 297 memset(p, 0, size); 298 } 299 return p; 300 } 301 302 /* 303 * kmem_intr_free: free wired memory allocated by kmem_alloc. 304 */ 305 306 void 307 kmem_intr_free(void *p, size_t requested_size) 308 { 309 size_t allocsz, index; 310 size_t size; 311 pool_cache_t pc; 312 313 KASSERT(p != NULL); 314 KASSERT(requested_size > 0); 315 316 #ifdef KMEM_GUARD 317 if (requested_size <= kmem_guard_size) { 318 uvm_kmguard_free(&kmem_guard, requested_size, p); 319 return; 320 } 321 #endif 322 323 size = kmem_roundup_size(requested_size); 324 allocsz = size + SIZE_SIZE; 325 326 #ifdef KMEM_REDZONE 327 if (size - requested_size < REDZONE_SIZE) { 328 allocsz += kmem_roundup_size(REDZONE_SIZE); 329 } 330 #endif 331 332 if ((index = ((allocsz -1) >> KMEM_SHIFT)) 333 < kmem_cache_maxidx) { 334 pc = kmem_cache[index]; 335 } else if ((index = ((allocsz - 1) >> KMEM_BIG_SHIFT)) 336 < kmem_cache_big_maxidx) { 337 pc = kmem_cache_big[index]; 338 } else { 339 FREECHECK_IN(&kmem_freecheck, p); 340 uvm_km_kmem_free(kmem_va_arena, (vaddr_t)p, 341 round_page(size)); 342 return; 343 } 344 345 p = (uint8_t *)p - SIZE_SIZE; 346 kmem_size_check(p, requested_size); 347 kmem_redzone_check(p, requested_size + SIZE_SIZE); 348 FREECHECK_IN(&kmem_freecheck, p); 349 LOCKDEBUG_MEM_CHECK(p, size); 350 kmem_poison_fill(p, allocsz); 351 352 pool_cache_put(pc, p); 353 } 354 355 /* ---- kmem API */ 356 357 /* 358 * kmem_alloc: allocate wired memory. 359 * => must not be called from interrupt context. 360 */ 361 362 void * 363 kmem_alloc(size_t size, km_flag_t kmflags) 364 { 365 366 KASSERTMSG((!cpu_intr_p() && !cpu_softintr_p()), 367 "kmem(9) should not be used from the interrupt context"); 368 return kmem_intr_alloc(size, kmflags); 369 } 370 371 /* 372 * kmem_zalloc: allocate zeroed wired memory. 373 * => must not be called from interrupt context. 374 */ 375 376 void * 377 kmem_zalloc(size_t size, km_flag_t kmflags) 378 { 379 380 KASSERTMSG((!cpu_intr_p() && !cpu_softintr_p()), 381 "kmem(9) should not be used from the interrupt context"); 382 return kmem_intr_zalloc(size, kmflags); 383 } 384 385 /* 386 * kmem_free: free wired memory allocated by kmem_alloc. 387 * => must not be called from interrupt context. 388 */ 389 390 void 391 kmem_free(void *p, size_t size) 392 { 393 394 KASSERT(!cpu_intr_p()); 395 KASSERT(!cpu_softintr_p()); 396 kmem_intr_free(p, size); 397 } 398 399 static size_t 400 kmem_create_caches(const struct kmem_cache_info *array, 401 pool_cache_t alloc_table[], size_t maxsize, int shift, int ipl) 402 { 403 size_t maxidx = 0; 404 size_t table_unit = (1 << shift); 405 size_t size = table_unit; 406 int i; 407 408 for (i = 0; array[i].kc_size != 0 ; i++) { 409 const char *name = array[i].kc_name; 410 size_t cache_size = array[i].kc_size; 411 struct pool_allocator *pa; 412 int flags = PR_NOALIGN; 413 pool_cache_t pc; 414 size_t align; 415 416 if ((cache_size & (CACHE_LINE_SIZE - 1)) == 0) 417 align = CACHE_LINE_SIZE; 418 else if ((cache_size & (PAGE_SIZE - 1)) == 0) 419 align = PAGE_SIZE; 420 else 421 align = KMEM_ALIGN; 422 423 if (cache_size < CACHE_LINE_SIZE) 424 flags |= PR_NOTOUCH; 425 426 /* check if we reached the requested size */ 427 if (cache_size > maxsize || cache_size > PAGE_SIZE) { 428 break; 429 } 430 if ((cache_size >> shift) > maxidx) { 431 maxidx = cache_size >> shift; 432 } 433 434 if ((cache_size >> shift) > maxidx) { 435 maxidx = cache_size >> shift; 436 } 437 438 pa = &pool_allocator_kmem; 439 #if defined(KMEM_POISON) 440 pc = pool_cache_init(cache_size, align, 0, flags, 441 name, pa, ipl, kmem_poison_ctor, 442 NULL, (void *)cache_size); 443 #else /* defined(KMEM_POISON) */ 444 pc = pool_cache_init(cache_size, align, 0, flags, 445 name, pa, ipl, NULL, NULL, NULL); 446 #endif /* defined(KMEM_POISON) */ 447 448 while (size <= cache_size) { 449 alloc_table[(size - 1) >> shift] = pc; 450 size += table_unit; 451 } 452 } 453 return maxidx; 454 } 455 456 void 457 kmem_init(void) 458 { 459 460 #ifdef KMEM_GUARD 461 uvm_kmguard_init(&kmem_guard, &kmem_guard_depth, &kmem_guard_size, 462 kmem_va_arena); 463 #endif 464 kmem_cache_maxidx = kmem_create_caches(kmem_cache_sizes, 465 kmem_cache, KMEM_MAXSIZE, KMEM_SHIFT, IPL_VM); 466 kmem_cache_big_maxidx = kmem_create_caches(kmem_cache_big_sizes, 467 kmem_cache_big, PAGE_SIZE, KMEM_BIG_SHIFT, IPL_VM); 468 } 469 470 size_t 471 kmem_roundup_size(size_t size) 472 { 473 474 return (size + (KMEM_ALIGN - 1)) & ~(KMEM_ALIGN - 1); 475 } 476 477 /* ------------------ DEBUG / DIAGNOSTIC ------------------ */ 478 479 #if defined(KMEM_POISON) || defined(KMEM_REDZONE) 480 #if defined(_LP64) 481 #define PRIME 0x9e37fffffffc0000UL 482 #else /* defined(_LP64) */ 483 #define PRIME 0x9e3779b1 484 #endif /* defined(_LP64) */ 485 486 static inline uint8_t 487 kmem_pattern_generate(const void *p) 488 { 489 return (uint8_t)(((uintptr_t)p) * PRIME 490 >> ((sizeof(uintptr_t) - sizeof(uint8_t))) * CHAR_BIT); 491 } 492 #endif /* defined(KMEM_POISON) || defined(KMEM_REDZONE) */ 493 494 #if defined(KMEM_POISON) 495 static int 496 kmem_poison_ctor(void *arg, void *obj, int flag) 497 { 498 size_t sz = (size_t)arg; 499 500 kmem_poison_fill(obj, sz); 501 502 return 0; 503 } 504 505 static void 506 kmem_poison_fill(void *p, size_t sz) 507 { 508 uint8_t *cp; 509 const uint8_t *ep; 510 511 cp = p; 512 ep = cp + sz; 513 while (cp < ep) { 514 *cp = kmem_pattern_generate(cp); 515 cp++; 516 } 517 } 518 519 static void 520 kmem_poison_check(void *p, size_t sz) 521 { 522 uint8_t *cp; 523 const uint8_t *ep; 524 525 cp = p; 526 ep = cp + sz; 527 while (cp < ep) { 528 const uint8_t expected = kmem_pattern_generate(cp); 529 530 if (*cp != expected) { 531 panic("%s: %p: 0x%02x != 0x%02x\n", 532 __func__, cp, *cp, expected); 533 } 534 cp++; 535 } 536 } 537 #endif /* defined(KMEM_POISON) */ 538 539 #if defined(KMEM_SIZE) 540 static void 541 kmem_size_set(void *p, size_t sz) 542 { 543 struct kmem_header *hd; 544 hd = (struct kmem_header *)p; 545 hd->size = sz; 546 } 547 548 static void 549 kmem_size_check(void *p, size_t sz) 550 { 551 struct kmem_header *hd; 552 size_t hsz; 553 554 hd = (struct kmem_header *)p; 555 hsz = hd->size; 556 557 if (hsz != sz) { 558 panic("kmem_free(%p, %zu) != allocated size %zu", 559 (const uint8_t *)p + SIZE_SIZE, sz, hsz); 560 } 561 } 562 #endif /* defined(KMEM_SIZE) */ 563 564 #if defined(KMEM_REDZONE) 565 #define STATIC_BYTE 0xFE 566 CTASSERT(REDZONE_SIZE > 1); 567 static void 568 kmem_redzone_fill(void *p, size_t sz) 569 { 570 uint8_t *cp, pat; 571 const uint8_t *ep; 572 573 cp = (uint8_t *)p + sz; 574 ep = cp + REDZONE_SIZE; 575 576 /* 577 * We really don't want the first byte of the red zone to be '\0'; 578 * an off-by-one in a string may not be properly detected. 579 */ 580 pat = kmem_pattern_generate(cp); 581 *cp = (pat == '\0') ? STATIC_BYTE: pat; 582 cp++; 583 584 while (cp < ep) { 585 *cp = kmem_pattern_generate(cp); 586 cp++; 587 } 588 } 589 590 static void 591 kmem_redzone_check(void *p, size_t sz) 592 { 593 uint8_t *cp, pat, expected; 594 const uint8_t *ep; 595 596 cp = (uint8_t *)p + sz; 597 ep = cp + REDZONE_SIZE; 598 599 pat = kmem_pattern_generate(cp); 600 expected = (pat == '\0') ? STATIC_BYTE: pat; 601 if (expected != *cp) { 602 panic("%s: %p: 0x%02x != 0x%02x\n", 603 __func__, cp, *cp, expected); 604 } 605 cp++; 606 607 while (cp < ep) { 608 expected = kmem_pattern_generate(cp); 609 if (*cp != expected) { 610 panic("%s: %p: 0x%02x != 0x%02x\n", 611 __func__, cp, *cp, expected); 612 } 613 cp++; 614 } 615 } 616 #endif /* defined(KMEM_REDZONE) */ 617 618 619 /* 620 * Used to dynamically allocate string with kmem accordingly to format. 621 */ 622 char * 623 kmem_asprintf(const char *fmt, ...) 624 { 625 int size __diagused, len; 626 va_list va; 627 char *str; 628 629 va_start(va, fmt); 630 len = vsnprintf(NULL, 0, fmt, va); 631 va_end(va); 632 633 str = kmem_alloc(len + 1, KM_SLEEP); 634 635 va_start(va, fmt); 636 size = vsnprintf(str, len + 1, fmt, va); 637 va_end(va); 638 639 KASSERT(size == len); 640 641 return str; 642 } 643
Indexes created Sat Sep 20 22:09:52 GMT 2025