1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Written by Mark Hemment, 1996 ([email protected]). 4 * 5 * (C) SGI 2006, Christoph Lameter 6 * Cleaned up and restructured to ease the addition of alternative 7 * implementations of SLAB allocators. 8 * (C) Linux Foundation 2008-2013 9 * Unified interface for all slab allocators 10 */ 11 12 #ifndef _LINUX_SLAB_H 13 #define _LINUX_SLAB_H 14 15 #include <linux/cache.h> 16 #include <linux/gfp.h> 17 #include <linux/overflow.h> 18 #include <linux/types.h> 19 #include <linux/workqueue.h> 20 #include <linux/percpu-refcount.h> 21 #include <linux/cleanup.h> 22 #include <linux/hash.h> 23 24 enum _slab_flag_bits { 25 _SLAB_CONSISTENCY_CHECKS, 26 _SLAB_RED_ZONE, 27 _SLAB_POISON, 28 _SLAB_KMALLOC, 29 _SLAB_HWCACHE_ALIGN, 30 _SLAB_CACHE_DMA, 31 _SLAB_CACHE_DMA32, 32 _SLAB_STORE_USER, 33 _SLAB_PANIC, 34 _SLAB_TYPESAFE_BY_RCU, 35 _SLAB_TRACE, 36 #ifdef CONFIG_DEBUG_OBJECTS 37 _SLAB_DEBUG_OBJECTS, 38 #endif 39 _SLAB_NOLEAKTRACE, 40 _SLAB_NO_MERGE, 41 #ifdef CONFIG_FAILSLAB 42 _SLAB_FAILSLAB, 43 #endif 44 #ifdef CONFIG_MEMCG 45 _SLAB_ACCOUNT, 46 #endif 47 #ifdef CONFIG_KASAN_GENERIC 48 _SLAB_KASAN, 49 #endif 50 _SLAB_NO_USER_FLAGS, 51 #ifdef CONFIG_KFENCE 52 _SLAB_SKIP_KFENCE, 53 #endif 54 #ifndef CONFIG_SLUB_TINY 55 _SLAB_RECLAIM_ACCOUNT, 56 #endif 57 _SLAB_OBJECT_POISON, 58 _SLAB_CMPXCHG_DOUBLE, 59 #ifdef CONFIG_SLAB_OBJ_EXT 60 _SLAB_NO_OBJ_EXT, 61 #endif 62 _SLAB_FLAGS_LAST_BIT 63 }; 64 65 #define __SLAB_FLAG_BIT(nr) ((slab_flags_t __force)(1U << (nr))) 66 #define __SLAB_FLAG_UNUSED ((slab_flags_t __force)(0U)) 67 68 /* 69 * Flags to pass to kmem_cache_create(). 70 * The ones marked DEBUG need CONFIG_SLUB_DEBUG enabled, otherwise are no-op 71 */ 72 /* DEBUG: Perform (expensive) checks on alloc/free */ 73 #define SLAB_CONSISTENCY_CHECKS __SLAB_FLAG_BIT(_SLAB_CONSISTENCY_CHECKS) 74 /* DEBUG: Red zone objs in a cache */ 75 #define SLAB_RED_ZONE __SLAB_FLAG_BIT(_SLAB_RED_ZONE) 76 /* DEBUG: Poison objects */ 77 #define SLAB_POISON __SLAB_FLAG_BIT(_SLAB_POISON) 78 /* Indicate a kmalloc slab */ 79 #define SLAB_KMALLOC __SLAB_FLAG_BIT(_SLAB_KMALLOC) 80 /* Align objs on cache lines */ 81 #define SLAB_HWCACHE_ALIGN __SLAB_FLAG_BIT(_SLAB_HWCACHE_ALIGN) 82 /* Use GFP_DMA memory */ 83 #define SLAB_CACHE_DMA __SLAB_FLAG_BIT(_SLAB_CACHE_DMA) 84 /* Use GFP_DMA32 memory */ 85 #define SLAB_CACHE_DMA32 __SLAB_FLAG_BIT(_SLAB_CACHE_DMA32) 86 /* DEBUG: Store the last owner for bug hunting */ 87 #define SLAB_STORE_USER __SLAB_FLAG_BIT(_SLAB_STORE_USER) 88 /* Panic if kmem_cache_create() fails */ 89 #define SLAB_PANIC __SLAB_FLAG_BIT(_SLAB_PANIC) 90 /* 91 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS! 92 * 93 * This delays freeing the SLAB page by a grace period, it does _NOT_ 94 * delay object freeing. This means that if you do kmem_cache_free() 95 * that memory location is free to be reused at any time. Thus it may 96 * be possible to see another object there in the same RCU grace period. 97 * 98 * This feature only ensures the memory location backing the object 99 * stays valid, the trick to using this is relying on an independent 100 * object validation pass. Something like: 101 * 102 * begin: 103 * rcu_read_lock(); 104 * obj = lockless_lookup(key); 105 * if (obj) { 106 * if (!try_get_ref(obj)) // might fail for free objects 107 * rcu_read_unlock(); 108 * goto begin; 109 * 110 * if (obj->key != key) { // not the object we expected 111 * put_ref(obj); 112 * rcu_read_unlock(); 113 * goto begin; 114 * } 115 * } 116 * rcu_read_unlock(); 117 * 118 * This is useful if we need to approach a kernel structure obliquely, 119 * from its address obtained without the usual locking. We can lock 120 * the structure to stabilize it and check it's still at the given address, 121 * only if we can be sure that the memory has not been meanwhile reused 122 * for some other kind of object (which our subsystem's lock might corrupt). 123 * 124 * rcu_read_lock before reading the address, then rcu_read_unlock after 125 * taking the spinlock within the structure expected at that address. 126 * 127 * Note that it is not possible to acquire a lock within a structure 128 * allocated with SLAB_TYPESAFE_BY_RCU without first acquiring a reference 129 * as described above. The reason is that SLAB_TYPESAFE_BY_RCU pages 130 * are not zeroed before being given to the slab, which means that any 131 * locks must be initialized after each and every kmem_struct_alloc(). 132 * Alternatively, make the ctor passed to kmem_cache_create() initialize 133 * the locks at page-allocation time, as is done in __i915_request_ctor(), 134 * sighand_ctor(), and anon_vma_ctor(). Such a ctor permits readers 135 * to safely acquire those ctor-initialized locks under rcu_read_lock() 136 * protection. 137 * 138 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU. 139 */ 140 /* Defer freeing slabs to RCU */ 141 #define SLAB_TYPESAFE_BY_RCU __SLAB_FLAG_BIT(_SLAB_TYPESAFE_BY_RCU) 142 /* Trace allocations and frees */ 143 #define SLAB_TRACE __SLAB_FLAG_BIT(_SLAB_TRACE) 144 145 /* Flag to prevent checks on free */ 146 #ifdef CONFIG_DEBUG_OBJECTS 147 # define SLAB_DEBUG_OBJECTS __SLAB_FLAG_BIT(_SLAB_DEBUG_OBJECTS) 148 #else 149 # define SLAB_DEBUG_OBJECTS __SLAB_FLAG_UNUSED 150 #endif 151 152 /* Avoid kmemleak tracing */ 153 #define SLAB_NOLEAKTRACE __SLAB_FLAG_BIT(_SLAB_NOLEAKTRACE) 154 155 /* 156 * Prevent merging with compatible kmem caches. This flag should be used 157 * cautiously. Valid use cases: 158 * 159 * - caches created for self-tests (e.g. kunit) 160 * - general caches created and used by a subsystem, only when a 161 * (subsystem-specific) debug option is enabled 162 * - performance critical caches, should be very rare and consulted with slab 163 * maintainers, and not used together with CONFIG_SLUB_TINY 164 */ 165 #define SLAB_NO_MERGE __SLAB_FLAG_BIT(_SLAB_NO_MERGE) 166 167 /* Fault injection mark */ 168 #ifdef CONFIG_FAILSLAB 169 # define SLAB_FAILSLAB __SLAB_FLAG_BIT(_SLAB_FAILSLAB) 170 #else 171 # define SLAB_FAILSLAB __SLAB_FLAG_UNUSED 172 #endif 173 /* Account to memcg */ 174 #ifdef CONFIG_MEMCG 175 # define SLAB_ACCOUNT __SLAB_FLAG_BIT(_SLAB_ACCOUNT) 176 #else 177 # define SLAB_ACCOUNT __SLAB_FLAG_UNUSED 178 #endif 179 180 #ifdef CONFIG_KASAN_GENERIC 181 #define SLAB_KASAN __SLAB_FLAG_BIT(_SLAB_KASAN) 182 #else 183 #define SLAB_KASAN __SLAB_FLAG_UNUSED 184 #endif 185 186 /* 187 * Ignore user specified debugging flags. 188 * Intended for caches created for self-tests so they have only flags 189 * specified in the code and other flags are ignored. 190 */ 191 #define SLAB_NO_USER_FLAGS __SLAB_FLAG_BIT(_SLAB_NO_USER_FLAGS) 192 193 #ifdef CONFIG_KFENCE 194 #define SLAB_SKIP_KFENCE __SLAB_FLAG_BIT(_SLAB_SKIP_KFENCE) 195 #else 196 #define SLAB_SKIP_KFENCE __SLAB_FLAG_UNUSED 197 #endif 198 199 /* The following flags affect the page allocator grouping pages by mobility */ 200 /* Objects are reclaimable */ 201 #ifndef CONFIG_SLUB_TINY 202 #define SLAB_RECLAIM_ACCOUNT __SLAB_FLAG_BIT(_SLAB_RECLAIM_ACCOUNT) 203 #else 204 #define SLAB_RECLAIM_ACCOUNT __SLAB_FLAG_UNUSED 205 #endif 206 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */ 207 208 /* Slab created using create_boot_cache */ 209 #ifdef CONFIG_SLAB_OBJ_EXT 210 #define SLAB_NO_OBJ_EXT __SLAB_FLAG_BIT(_SLAB_NO_OBJ_EXT) 211 #else 212 #define SLAB_NO_OBJ_EXT __SLAB_FLAG_UNUSED 213 #endif 214 215 /* 216 * freeptr_t represents a SLUB freelist pointer, which might be encoded 217 * and not dereferenceable if CONFIG_SLAB_FREELIST_HARDENED is enabled. 218 */ 219 typedef struct { unsigned long v; } freeptr_t; 220 221 /* 222 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests. 223 * 224 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault. 225 * 226 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can. 227 * Both make kfree a no-op. 228 */ 229 #define ZERO_SIZE_PTR ((void *)16) 230 231 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \ 232 (unsigned long)ZERO_SIZE_PTR) 233 234 #include <linux/kasan.h> 235 236 struct list_lru; 237 struct mem_cgroup; 238 /* 239 * struct kmem_cache related prototypes 240 */ 241 bool slab_is_available(void); 242 243 /** 244 * struct kmem_cache_args - Less common arguments for kmem_cache_create() 245 * @align: The required alignment for the objects. 246 * @useroffset: Usercopy region offset 247 * @usersize: Usercopy region size 248 * @freeptr_offset: Custom offset for the free pointer in RCU caches 249 * @use_freeptr_offset: Whether a @freeptr_offset is used 250 * @ctor: A constructor for the objects. 251 */ 252 struct kmem_cache_args { 253 unsigned int align; 254 unsigned int useroffset; 255 unsigned int usersize; 256 unsigned int freeptr_offset; 257 bool use_freeptr_offset; 258 void (*ctor)(void *); 259 }; 260 261 struct kmem_cache *__kmem_cache_create_args(const char *name, 262 unsigned int object_size, 263 struct kmem_cache_args *args, 264 slab_flags_t flags); 265 struct kmem_cache *kmem_cache_create(const char *name, unsigned int size, 266 unsigned int align, slab_flags_t flags, 267 void (*ctor)(void *)); 268 struct kmem_cache *kmem_cache_create_usercopy(const char *name, 269 unsigned int size, unsigned int align, 270 slab_flags_t flags, 271 unsigned int useroffset, unsigned int usersize, 272 void (*ctor)(void *)); 273 struct kmem_cache *kmem_cache_create_rcu(const char *name, unsigned int size, 274 unsigned int freeptr_offset, 275 slab_flags_t flags); 276 void kmem_cache_destroy(struct kmem_cache *s); 277 int kmem_cache_shrink(struct kmem_cache *s); 278 279 /* 280 * Please use this macro to create slab caches. Simply specify the 281 * name of the structure and maybe some flags that are listed above. 282 * 283 * The alignment of the struct determines object alignment. If you 284 * f.e. add ____cacheline_aligned_in_smp to the struct declaration 285 * then the objects will be properly aligned in SMP configurations. 286 */ 287 #define KMEM_CACHE(__struct, __flags) \ 288 __kmem_cache_create_args(#__struct, sizeof(struct __struct), \ 289 &(struct kmem_cache_args) { \ 290 .align = __alignof__(struct __struct), \ 291 }, (__flags)) 292 293 /* 294 * To whitelist a single field for copying to/from usercopy, use this 295 * macro instead for KMEM_CACHE() above. 296 */ 297 #define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \ 298 __kmem_cache_create_args(#__struct, sizeof(struct __struct), \ 299 &(struct kmem_cache_args) { \ 300 .align = __alignof__(struct __struct), \ 301 .useroffset = offsetof(struct __struct, __field), \ 302 .usersize = sizeof_field(struct __struct, __field), \ 303 }, (__flags)) 304 305 /* 306 * Common kmalloc functions provided by all allocators 307 */ 308 void * __must_check krealloc_noprof(const void *objp, size_t new_size, 309 gfp_t flags) __realloc_size(2); 310 #define krealloc(...) alloc_hooks(krealloc_noprof(__VA_ARGS__)) 311 312 void kfree(const void *objp); 313 void kfree_sensitive(const void *objp); 314 size_t __ksize(const void *objp); 315 316 DEFINE_FREE(kfree, void *, if (!IS_ERR_OR_NULL(_T)) kfree(_T)) 317 318 /** 319 * ksize - Report actual allocation size of associated object 320 * 321 * @objp: Pointer returned from a prior kmalloc()-family allocation. 322 * 323 * This should not be used for writing beyond the originally requested 324 * allocation size. Either use krealloc() or round up the allocation size 325 * with kmalloc_size_roundup() prior to allocation. If this is used to 326 * access beyond the originally requested allocation size, UBSAN_BOUNDS 327 * and/or FORTIFY_SOURCE may trip, since they only know about the 328 * originally allocated size via the __alloc_size attribute. 329 */ 330 size_t ksize(const void *objp); 331 332 #ifdef CONFIG_PRINTK 333 bool kmem_dump_obj(void *object); 334 #else 335 static inline bool kmem_dump_obj(void *object) { return false; } 336 #endif 337 338 /* 339 * Some archs want to perform DMA into kmalloc caches and need a guaranteed 340 * alignment larger than the alignment of a 64-bit integer. 341 * Setting ARCH_DMA_MINALIGN in arch headers allows that. 342 */ 343 #ifdef ARCH_HAS_DMA_MINALIGN 344 #if ARCH_DMA_MINALIGN > 8 && !defined(ARCH_KMALLOC_MINALIGN) 345 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN 346 #endif 347 #endif 348 349 #ifndef ARCH_KMALLOC_MINALIGN 350 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) 351 #elif ARCH_KMALLOC_MINALIGN > 8 352 #define KMALLOC_MIN_SIZE ARCH_KMALLOC_MINALIGN 353 #define KMALLOC_SHIFT_LOW ilog2(KMALLOC_MIN_SIZE) 354 #endif 355 356 /* 357 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment. 358 * Intended for arches that get misalignment faults even for 64 bit integer 359 * aligned buffers. 360 */ 361 #ifndef ARCH_SLAB_MINALIGN 362 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) 363 #endif 364 365 /* 366 * Arches can define this function if they want to decide the minimum slab 367 * alignment at runtime. The value returned by the function must be a power 368 * of two and >= ARCH_SLAB_MINALIGN. 369 */ 370 #ifndef arch_slab_minalign 371 static inline unsigned int arch_slab_minalign(void) 372 { 373 return ARCH_SLAB_MINALIGN; 374 } 375 #endif 376 377 /* 378 * kmem_cache_alloc and friends return pointers aligned to ARCH_SLAB_MINALIGN. 379 * kmalloc and friends return pointers aligned to both ARCH_KMALLOC_MINALIGN 380 * and ARCH_SLAB_MINALIGN, but here we only assume the former alignment. 381 */ 382 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN) 383 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN) 384 #define __assume_page_alignment __assume_aligned(PAGE_SIZE) 385 386 /* 387 * Kmalloc array related definitions 388 */ 389 390 /* 391 * SLUB directly allocates requests fitting in to an order-1 page 392 * (PAGE_SIZE*2). Larger requests are passed to the page allocator. 393 */ 394 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1) 395 #define KMALLOC_SHIFT_MAX (MAX_PAGE_ORDER + PAGE_SHIFT) 396 #ifndef KMALLOC_SHIFT_LOW 397 #define KMALLOC_SHIFT_LOW 3 398 #endif 399 400 /* Maximum allocatable size */ 401 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX) 402 /* Maximum size for which we actually use a slab cache */ 403 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH) 404 /* Maximum order allocatable via the slab allocator */ 405 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT) 406 407 /* 408 * Kmalloc subsystem. 409 */ 410 #ifndef KMALLOC_MIN_SIZE 411 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW) 412 #endif 413 414 /* 415 * This restriction comes from byte sized index implementation. 416 * Page size is normally 2^12 bytes and, in this case, if we want to use 417 * byte sized index which can represent 2^8 entries, the size of the object 418 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16. 419 * If minimum size of kmalloc is less than 16, we use it as minimum object 420 * size and give up to use byte sized index. 421 */ 422 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \ 423 (KMALLOC_MIN_SIZE) : 16) 424 425 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 426 #define RANDOM_KMALLOC_CACHES_NR 15 // # of cache copies 427 #else 428 #define RANDOM_KMALLOC_CACHES_NR 0 429 #endif 430 431 /* 432 * Whenever changing this, take care of that kmalloc_type() and 433 * create_kmalloc_caches() still work as intended. 434 * 435 * KMALLOC_NORMAL can contain only unaccounted objects whereas KMALLOC_CGROUP 436 * is for accounted but unreclaimable and non-dma objects. All the other 437 * kmem caches can have both accounted and unaccounted objects. 438 */ 439 enum kmalloc_cache_type { 440 KMALLOC_NORMAL = 0, 441 #ifndef CONFIG_ZONE_DMA 442 KMALLOC_DMA = KMALLOC_NORMAL, 443 #endif 444 #ifndef CONFIG_MEMCG 445 KMALLOC_CGROUP = KMALLOC_NORMAL, 446 #endif 447 KMALLOC_RANDOM_START = KMALLOC_NORMAL, 448 KMALLOC_RANDOM_END = KMALLOC_RANDOM_START + RANDOM_KMALLOC_CACHES_NR, 449 #ifdef CONFIG_SLUB_TINY 450 KMALLOC_RECLAIM = KMALLOC_NORMAL, 451 #else 452 KMALLOC_RECLAIM, 453 #endif 454 #ifdef CONFIG_ZONE_DMA 455 KMALLOC_DMA, 456 #endif 457 #ifdef CONFIG_MEMCG 458 KMALLOC_CGROUP, 459 #endif 460 NR_KMALLOC_TYPES 461 }; 462 463 typedef struct kmem_cache * kmem_buckets[KMALLOC_SHIFT_HIGH + 1]; 464 465 extern kmem_buckets kmalloc_caches[NR_KMALLOC_TYPES]; 466 467 /* 468 * Define gfp bits that should not be set for KMALLOC_NORMAL. 469 */ 470 #define KMALLOC_NOT_NORMAL_BITS \ 471 (__GFP_RECLAIMABLE | \ 472 (IS_ENABLED(CONFIG_ZONE_DMA) ? __GFP_DMA : 0) | \ 473 (IS_ENABLED(CONFIG_MEMCG) ? __GFP_ACCOUNT : 0)) 474 475 extern unsigned long random_kmalloc_seed; 476 477 static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags, unsigned long caller) 478 { 479 /* 480 * The most common case is KMALLOC_NORMAL, so test for it 481 * with a single branch for all the relevant flags. 482 */ 483 if (likely((flags & KMALLOC_NOT_NORMAL_BITS) == 0)) 484 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 485 /* RANDOM_KMALLOC_CACHES_NR (=15) copies + the KMALLOC_NORMAL */ 486 return KMALLOC_RANDOM_START + hash_64(caller ^ random_kmalloc_seed, 487 ilog2(RANDOM_KMALLOC_CACHES_NR + 1)); 488 #else 489 return KMALLOC_NORMAL; 490 #endif 491 492 /* 493 * At least one of the flags has to be set. Their priorities in 494 * decreasing order are: 495 * 1) __GFP_DMA 496 * 2) __GFP_RECLAIMABLE 497 * 3) __GFP_ACCOUNT 498 */ 499 if (IS_ENABLED(CONFIG_ZONE_DMA) && (flags & __GFP_DMA)) 500 return KMALLOC_DMA; 501 if (!IS_ENABLED(CONFIG_MEMCG) || (flags & __GFP_RECLAIMABLE)) 502 return KMALLOC_RECLAIM; 503 else 504 return KMALLOC_CGROUP; 505 } 506 507 /* 508 * Figure out which kmalloc slab an allocation of a certain size 509 * belongs to. 510 * 0 = zero alloc 511 * 1 = 65 .. 96 bytes 512 * 2 = 129 .. 192 bytes 513 * n = 2^(n-1)+1 .. 2^n 514 * 515 * Note: __kmalloc_index() is compile-time optimized, and not runtime optimized; 516 * typical usage is via kmalloc_index() and therefore evaluated at compile-time. 517 * Callers where !size_is_constant should only be test modules, where runtime 518 * overheads of __kmalloc_index() can be tolerated. Also see kmalloc_slab(). 519 */ 520 static __always_inline unsigned int __kmalloc_index(size_t size, 521 bool size_is_constant) 522 { 523 if (!size) 524 return 0; 525 526 if (size <= KMALLOC_MIN_SIZE) 527 return KMALLOC_SHIFT_LOW; 528 529 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96) 530 return 1; 531 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192) 532 return 2; 533 if (size <= 8) return 3; 534 if (size <= 16) return 4; 535 if (size <= 32) return 5; 536 if (size <= 64) return 6; 537 if (size <= 128) return 7; 538 if (size <= 256) return 8; 539 if (size <= 512) return 9; 540 if (size <= 1024) return 10; 541 if (size <= 2 * 1024) return 11; 542 if (size <= 4 * 1024) return 12; 543 if (size <= 8 * 1024) return 13; 544 if (size <= 16 * 1024) return 14; 545 if (size <= 32 * 1024) return 15; 546 if (size <= 64 * 1024) return 16; 547 if (size <= 128 * 1024) return 17; 548 if (size <= 256 * 1024) return 18; 549 if (size <= 512 * 1024) return 19; 550 if (size <= 1024 * 1024) return 20; 551 if (size <= 2 * 1024 * 1024) return 21; 552 553 if (!IS_ENABLED(CONFIG_PROFILE_ALL_BRANCHES) && size_is_constant) 554 BUILD_BUG_ON_MSG(1, "unexpected size in kmalloc_index()"); 555 else 556 BUG(); 557 558 /* Will never be reached. Needed because the compiler may complain */ 559 return -1; 560 } 561 static_assert(PAGE_SHIFT <= 20); 562 #define kmalloc_index(s) __kmalloc_index(s, true) 563 564 #include <linux/alloc_tag.h> 565 566 /** 567 * kmem_cache_alloc - Allocate an object 568 * @cachep: The cache to allocate from. 569 * @flags: See kmalloc(). 570 * 571 * Allocate an object from this cache. 572 * See kmem_cache_zalloc() for a shortcut of adding __GFP_ZERO to flags. 573 * 574 * Return: pointer to the new object or %NULL in case of error 575 */ 576 void *kmem_cache_alloc_noprof(struct kmem_cache *cachep, 577 gfp_t flags) __assume_slab_alignment __malloc; 578 #define kmem_cache_alloc(...) alloc_hooks(kmem_cache_alloc_noprof(__VA_ARGS__)) 579 580 void *kmem_cache_alloc_lru_noprof(struct kmem_cache *s, struct list_lru *lru, 581 gfp_t gfpflags) __assume_slab_alignment __malloc; 582 #define kmem_cache_alloc_lru(...) alloc_hooks(kmem_cache_alloc_lru_noprof(__VA_ARGS__)) 583 584 void kmem_cache_free(struct kmem_cache *s, void *objp); 585 586 kmem_buckets *kmem_buckets_create(const char *name, slab_flags_t flags, 587 unsigned int useroffset, unsigned int usersize, 588 void (*ctor)(void *)); 589 590 /* 591 * Bulk allocation and freeing operations. These are accelerated in an 592 * allocator specific way to avoid taking locks repeatedly or building 593 * metadata structures unnecessarily. 594 * 595 * Note that interrupts must be enabled when calling these functions. 596 */ 597 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p); 598 599 int kmem_cache_alloc_bulk_noprof(struct kmem_cache *s, gfp_t flags, size_t size, void **p); 600 #define kmem_cache_alloc_bulk(...) alloc_hooks(kmem_cache_alloc_bulk_noprof(__VA_ARGS__)) 601 602 static __always_inline void kfree_bulk(size_t size, void **p) 603 { 604 kmem_cache_free_bulk(NULL, size, p); 605 } 606 607 void *kmem_cache_alloc_node_noprof(struct kmem_cache *s, gfp_t flags, 608 int node) __assume_slab_alignment __malloc; 609 #define kmem_cache_alloc_node(...) alloc_hooks(kmem_cache_alloc_node_noprof(__VA_ARGS__)) 610 611 /* 612 * These macros allow declaring a kmem_buckets * parameter alongside size, which 613 * can be compiled out with CONFIG_SLAB_BUCKETS=n so that a large number of call 614 * sites don't have to pass NULL. 615 */ 616 #ifdef CONFIG_SLAB_BUCKETS 617 #define DECL_BUCKET_PARAMS(_size, _b) size_t (_size), kmem_buckets *(_b) 618 #define PASS_BUCKET_PARAMS(_size, _b) (_size), (_b) 619 #define PASS_BUCKET_PARAM(_b) (_b) 620 #else 621 #define DECL_BUCKET_PARAMS(_size, _b) size_t (_size) 622 #define PASS_BUCKET_PARAMS(_size, _b) (_size) 623 #define PASS_BUCKET_PARAM(_b) NULL 624 #endif 625 626 /* 627 * The following functions are not to be used directly and are intended only 628 * for internal use from kmalloc() and kmalloc_node() 629 * with the exception of kunit tests 630 */ 631 632 void *__kmalloc_noprof(size_t size, gfp_t flags) 633 __assume_kmalloc_alignment __alloc_size(1); 634 635 void *__kmalloc_node_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node) 636 __assume_kmalloc_alignment __alloc_size(1); 637 638 void *__kmalloc_cache_noprof(struct kmem_cache *s, gfp_t flags, size_t size) 639 __assume_kmalloc_alignment __alloc_size(3); 640 641 void *__kmalloc_cache_node_noprof(struct kmem_cache *s, gfp_t gfpflags, 642 int node, size_t size) 643 __assume_kmalloc_alignment __alloc_size(4); 644 645 void *__kmalloc_large_noprof(size_t size, gfp_t flags) 646 __assume_page_alignment __alloc_size(1); 647 648 void *__kmalloc_large_node_noprof(size_t size, gfp_t flags, int node) 649 __assume_page_alignment __alloc_size(1); 650 651 /** 652 * kmalloc - allocate kernel memory 653 * @size: how many bytes of memory are required. 654 * @flags: describe the allocation context 655 * 656 * kmalloc is the normal method of allocating memory 657 * for objects smaller than page size in the kernel. 658 * 659 * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN 660 * bytes. For @size of power of two bytes, the alignment is also guaranteed 661 * to be at least to the size. For other sizes, the alignment is guaranteed to 662 * be at least the largest power-of-two divisor of @size. 663 * 664 * The @flags argument may be one of the GFP flags defined at 665 * include/linux/gfp_types.h and described at 666 * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>` 667 * 668 * The recommended usage of the @flags is described at 669 * :ref:`Documentation/core-api/memory-allocation.rst <memory_allocation>` 670 * 671 * Below is a brief outline of the most useful GFP flags 672 * 673 * %GFP_KERNEL 674 * Allocate normal kernel ram. May sleep. 675 * 676 * %GFP_NOWAIT 677 * Allocation will not sleep. 678 * 679 * %GFP_ATOMIC 680 * Allocation will not sleep. May use emergency pools. 681 * 682 * Also it is possible to set different flags by OR'ing 683 * in one or more of the following additional @flags: 684 * 685 * %__GFP_ZERO 686 * Zero the allocated memory before returning. Also see kzalloc(). 687 * 688 * %__GFP_HIGH 689 * This allocation has high priority and may use emergency pools. 690 * 691 * %__GFP_NOFAIL 692 * Indicate that this allocation is in no way allowed to fail 693 * (think twice before using). 694 * 695 * %__GFP_NORETRY 696 * If memory is not immediately available, 697 * then give up at once. 698 * 699 * %__GFP_NOWARN 700 * If allocation fails, don't issue any warnings. 701 * 702 * %__GFP_RETRY_MAYFAIL 703 * Try really hard to succeed the allocation but fail 704 * eventually. 705 */ 706 static __always_inline __alloc_size(1) void *kmalloc_noprof(size_t size, gfp_t flags) 707 { 708 if (__builtin_constant_p(size) && size) { 709 unsigned int index; 710 711 if (size > KMALLOC_MAX_CACHE_SIZE) 712 return __kmalloc_large_noprof(size, flags); 713 714 index = kmalloc_index(size); 715 return __kmalloc_cache_noprof( 716 kmalloc_caches[kmalloc_type(flags, _RET_IP_)][index], 717 flags, size); 718 } 719 return __kmalloc_noprof(size, flags); 720 } 721 #define kmalloc(...) alloc_hooks(kmalloc_noprof(__VA_ARGS__)) 722 723 #define kmem_buckets_alloc(_b, _size, _flags) \ 724 alloc_hooks(__kmalloc_node_noprof(PASS_BUCKET_PARAMS(_size, _b), _flags, NUMA_NO_NODE)) 725 726 #define kmem_buckets_alloc_track_caller(_b, _size, _flags) \ 727 alloc_hooks(__kmalloc_node_track_caller_noprof(PASS_BUCKET_PARAMS(_size, _b), _flags, NUMA_NO_NODE, _RET_IP_)) 728 729 static __always_inline __alloc_size(1) void *kmalloc_node_noprof(size_t size, gfp_t flags, int node) 730 { 731 if (__builtin_constant_p(size) && size) { 732 unsigned int index; 733 734 if (size > KMALLOC_MAX_CACHE_SIZE) 735 return __kmalloc_large_node_noprof(size, flags, node); 736 737 index = kmalloc_index(size); 738 return __kmalloc_cache_node_noprof( 739 kmalloc_caches[kmalloc_type(flags, _RET_IP_)][index], 740 flags, node, size); 741 } 742 return __kmalloc_node_noprof(PASS_BUCKET_PARAMS(size, NULL), flags, node); 743 } 744 #define kmalloc_node(...) alloc_hooks(kmalloc_node_noprof(__VA_ARGS__)) 745 746 /** 747 * kmalloc_array - allocate memory for an array. 748 * @n: number of elements. 749 * @size: element size. 750 * @flags: the type of memory to allocate (see kmalloc). 751 */ 752 static inline __alloc_size(1, 2) void *kmalloc_array_noprof(size_t n, size_t size, gfp_t flags) 753 { 754 size_t bytes; 755 756 if (unlikely(check_mul_overflow(n, size, &bytes))) 757 return NULL; 758 if (__builtin_constant_p(n) && __builtin_constant_p(size)) 759 return kmalloc_noprof(bytes, flags); 760 return kmalloc_noprof(bytes, flags); 761 } 762 #define kmalloc_array(...) alloc_hooks(kmalloc_array_noprof(__VA_ARGS__)) 763 764 /** 765 * krealloc_array - reallocate memory for an array. 766 * @p: pointer to the memory chunk to reallocate 767 * @new_n: new number of elements to alloc 768 * @new_size: new size of a single member of the array 769 * @flags: the type of memory to allocate (see kmalloc) 770 */ 771 static inline __realloc_size(2, 3) void * __must_check krealloc_array_noprof(void *p, 772 size_t new_n, 773 size_t new_size, 774 gfp_t flags) 775 { 776 size_t bytes; 777 778 if (unlikely(check_mul_overflow(new_n, new_size, &bytes))) 779 return NULL; 780 781 return krealloc_noprof(p, bytes, flags); 782 } 783 #define krealloc_array(...) alloc_hooks(krealloc_array_noprof(__VA_ARGS__)) 784 785 /** 786 * kcalloc - allocate memory for an array. The memory is set to zero. 787 * @n: number of elements. 788 * @size: element size. 789 * @flags: the type of memory to allocate (see kmalloc). 790 */ 791 #define kcalloc(n, size, flags) kmalloc_array(n, size, (flags) | __GFP_ZERO) 792 793 void *__kmalloc_node_track_caller_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node, 794 unsigned long caller) __alloc_size(1); 795 #define kmalloc_node_track_caller_noprof(size, flags, node, caller) \ 796 __kmalloc_node_track_caller_noprof(PASS_BUCKET_PARAMS(size, NULL), flags, node, caller) 797 #define kmalloc_node_track_caller(...) \ 798 alloc_hooks(kmalloc_node_track_caller_noprof(__VA_ARGS__, _RET_IP_)) 799 800 /* 801 * kmalloc_track_caller is a special version of kmalloc that records the 802 * calling function of the routine calling it for slab leak tracking instead 803 * of just the calling function (confusing, eh?). 804 * It's useful when the call to kmalloc comes from a widely-used standard 805 * allocator where we care about the real place the memory allocation 806 * request comes from. 807 */ 808 #define kmalloc_track_caller(...) kmalloc_node_track_caller(__VA_ARGS__, NUMA_NO_NODE) 809 810 #define kmalloc_track_caller_noprof(...) \ 811 kmalloc_node_track_caller_noprof(__VA_ARGS__, NUMA_NO_NODE, _RET_IP_) 812 813 static inline __alloc_size(1, 2) void *kmalloc_array_node_noprof(size_t n, size_t size, gfp_t flags, 814 int node) 815 { 816 size_t bytes; 817 818 if (unlikely(check_mul_overflow(n, size, &bytes))) 819 return NULL; 820 if (__builtin_constant_p(n) && __builtin_constant_p(size)) 821 return kmalloc_node_noprof(bytes, flags, node); 822 return __kmalloc_node_noprof(PASS_BUCKET_PARAMS(bytes, NULL), flags, node); 823 } 824 #define kmalloc_array_node(...) alloc_hooks(kmalloc_array_node_noprof(__VA_ARGS__)) 825 826 #define kcalloc_node(_n, _size, _flags, _node) \ 827 kmalloc_array_node(_n, _size, (_flags) | __GFP_ZERO, _node) 828 829 /* 830 * Shortcuts 831 */ 832 #define kmem_cache_zalloc(_k, _flags) kmem_cache_alloc(_k, (_flags)|__GFP_ZERO) 833 834 /** 835 * kzalloc - allocate memory. The memory is set to zero. 836 * @size: how many bytes of memory are required. 837 * @flags: the type of memory to allocate (see kmalloc). 838 */ 839 static inline __alloc_size(1) void *kzalloc_noprof(size_t size, gfp_t flags) 840 { 841 return kmalloc_noprof(size, flags | __GFP_ZERO); 842 } 843 #define kzalloc(...) alloc_hooks(kzalloc_noprof(__VA_ARGS__)) 844 #define kzalloc_node(_size, _flags, _node) kmalloc_node(_size, (_flags)|__GFP_ZERO, _node) 845 846 void *__kvmalloc_node_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node) __alloc_size(1); 847 #define kvmalloc_node_noprof(size, flags, node) \ 848 __kvmalloc_node_noprof(PASS_BUCKET_PARAMS(size, NULL), flags, node) 849 #define kvmalloc_node(...) alloc_hooks(kvmalloc_node_noprof(__VA_ARGS__)) 850 851 #define kvmalloc(_size, _flags) kvmalloc_node(_size, _flags, NUMA_NO_NODE) 852 #define kvmalloc_noprof(_size, _flags) kvmalloc_node_noprof(_size, _flags, NUMA_NO_NODE) 853 #define kvzalloc(_size, _flags) kvmalloc(_size, (_flags)|__GFP_ZERO) 854 855 #define kvzalloc_node(_size, _flags, _node) kvmalloc_node(_size, (_flags)|__GFP_ZERO, _node) 856 #define kmem_buckets_valloc(_b, _size, _flags) \ 857 alloc_hooks(__kvmalloc_node_noprof(PASS_BUCKET_PARAMS(_size, _b), _flags, NUMA_NO_NODE)) 858 859 static inline __alloc_size(1, 2) void * 860 kvmalloc_array_node_noprof(size_t n, size_t size, gfp_t flags, int node) 861 { 862 size_t bytes; 863 864 if (unlikely(check_mul_overflow(n, size, &bytes))) 865 return NULL; 866 867 return kvmalloc_node_noprof(bytes, flags, node); 868 } 869 870 #define kvmalloc_array_noprof(...) kvmalloc_array_node_noprof(__VA_ARGS__, NUMA_NO_NODE) 871 #define kvcalloc_node_noprof(_n,_s,_f,_node) kvmalloc_array_node_noprof(_n,_s,(_f)|__GFP_ZERO,_node) 872 #define kvcalloc_noprof(...) kvcalloc_node_noprof(__VA_ARGS__, NUMA_NO_NODE) 873 874 #define kvmalloc_array(...) alloc_hooks(kvmalloc_array_noprof(__VA_ARGS__)) 875 #define kvcalloc_node(...) alloc_hooks(kvcalloc_node_noprof(__VA_ARGS__)) 876 #define kvcalloc(...) alloc_hooks(kvcalloc_noprof(__VA_ARGS__)) 877 878 extern void *kvrealloc_noprof(const void *p, size_t oldsize, size_t newsize, gfp_t flags) 879 __realloc_size(3); 880 #define kvrealloc(...) alloc_hooks(kvrealloc_noprof(__VA_ARGS__)) 881 882 extern void kvfree(const void *addr); 883 DEFINE_FREE(kvfree, void *, if (!IS_ERR_OR_NULL(_T)) kvfree(_T)) 884 885 extern void kvfree_sensitive(const void *addr, size_t len); 886 887 unsigned int kmem_cache_size(struct kmem_cache *s); 888 889 /** 890 * kmalloc_size_roundup - Report allocation bucket size for the given size 891 * 892 * @size: Number of bytes to round up from. 893 * 894 * This returns the number of bytes that would be available in a kmalloc() 895 * allocation of @size bytes. For example, a 126 byte request would be 896 * rounded up to the next sized kmalloc bucket, 128 bytes. (This is strictly 897 * for the general-purpose kmalloc()-based allocations, and is not for the 898 * pre-sized kmem_cache_alloc()-based allocations.) 899 * 900 * Use this to kmalloc() the full bucket size ahead of time instead of using 901 * ksize() to query the size after an allocation. 902 */ 903 size_t kmalloc_size_roundup(size_t size); 904 905 void __init kmem_cache_init_late(void); 906 907 #endif /* _LINUX_SLAB_H */ 908