Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | /* |
| 3 | * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk). |
| 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/gfp.h> |
| 16 | #include <linux/overflow.h> |
| 17 | #include <linux/types.h> |
| 18 | #include <linux/workqueue.h> |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 19 | #include <linux/percpu-refcount.h> |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 20 | |
| 21 | |
| 22 | /* |
| 23 | * Flags to pass to kmem_cache_create(). |
| 24 | * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set. |
| 25 | */ |
| 26 | /* DEBUG: Perform (expensive) checks on alloc/free */ |
| 27 | #define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100U) |
| 28 | /* DEBUG: Red zone objs in a cache */ |
| 29 | #define SLAB_RED_ZONE ((slab_flags_t __force)0x00000400U) |
| 30 | /* DEBUG: Poison objects */ |
| 31 | #define SLAB_POISON ((slab_flags_t __force)0x00000800U) |
| 32 | /* Align objs on cache lines */ |
| 33 | #define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000U) |
| 34 | /* Use GFP_DMA memory */ |
| 35 | #define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000U) |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 36 | /* Use GFP_DMA32 memory */ |
| 37 | #define SLAB_CACHE_DMA32 ((slab_flags_t __force)0x00008000U) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 38 | /* DEBUG: Store the last owner for bug hunting */ |
| 39 | #define SLAB_STORE_USER ((slab_flags_t __force)0x00010000U) |
| 40 | /* Panic if kmem_cache_create() fails */ |
| 41 | #define SLAB_PANIC ((slab_flags_t __force)0x00040000U) |
| 42 | /* |
| 43 | * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS! |
| 44 | * |
| 45 | * This delays freeing the SLAB page by a grace period, it does _NOT_ |
| 46 | * delay object freeing. This means that if you do kmem_cache_free() |
| 47 | * that memory location is free to be reused at any time. Thus it may |
| 48 | * be possible to see another object there in the same RCU grace period. |
| 49 | * |
| 50 | * This feature only ensures the memory location backing the object |
| 51 | * stays valid, the trick to using this is relying on an independent |
| 52 | * object validation pass. Something like: |
| 53 | * |
| 54 | * rcu_read_lock() |
| 55 | * again: |
| 56 | * obj = lockless_lookup(key); |
| 57 | * if (obj) { |
| 58 | * if (!try_get_ref(obj)) // might fail for free objects |
| 59 | * goto again; |
| 60 | * |
| 61 | * if (obj->key != key) { // not the object we expected |
| 62 | * put_ref(obj); |
| 63 | * goto again; |
| 64 | * } |
| 65 | * } |
| 66 | * rcu_read_unlock(); |
| 67 | * |
| 68 | * This is useful if we need to approach a kernel structure obliquely, |
| 69 | * from its address obtained without the usual locking. We can lock |
| 70 | * the structure to stabilize it and check it's still at the given address, |
| 71 | * only if we can be sure that the memory has not been meanwhile reused |
| 72 | * for some other kind of object (which our subsystem's lock might corrupt). |
| 73 | * |
| 74 | * rcu_read_lock before reading the address, then rcu_read_unlock after |
| 75 | * taking the spinlock within the structure expected at that address. |
| 76 | * |
| 77 | * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU. |
| 78 | */ |
| 79 | /* Defer freeing slabs to RCU */ |
| 80 | #define SLAB_TYPESAFE_BY_RCU ((slab_flags_t __force)0x00080000U) |
| 81 | /* Spread some memory over cpuset */ |
| 82 | #define SLAB_MEM_SPREAD ((slab_flags_t __force)0x00100000U) |
| 83 | /* Trace allocations and frees */ |
| 84 | #define SLAB_TRACE ((slab_flags_t __force)0x00200000U) |
| 85 | |
| 86 | /* Flag to prevent checks on free */ |
| 87 | #ifdef CONFIG_DEBUG_OBJECTS |
| 88 | # define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00400000U) |
| 89 | #else |
| 90 | # define SLAB_DEBUG_OBJECTS 0 |
| 91 | #endif |
| 92 | |
| 93 | /* Avoid kmemleak tracing */ |
| 94 | #define SLAB_NOLEAKTRACE ((slab_flags_t __force)0x00800000U) |
| 95 | |
| 96 | /* Fault injection mark */ |
| 97 | #ifdef CONFIG_FAILSLAB |
| 98 | # define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000U) |
| 99 | #else |
| 100 | # define SLAB_FAILSLAB 0 |
| 101 | #endif |
| 102 | /* Account to memcg */ |
| 103 | #ifdef CONFIG_MEMCG_KMEM |
| 104 | # define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000U) |
| 105 | #else |
| 106 | # define SLAB_ACCOUNT 0 |
| 107 | #endif |
| 108 | |
| 109 | #ifdef CONFIG_KASAN |
| 110 | #define SLAB_KASAN ((slab_flags_t __force)0x08000000U) |
| 111 | #else |
| 112 | #define SLAB_KASAN 0 |
| 113 | #endif |
| 114 | |
| 115 | /* The following flags affect the page allocator grouping pages by mobility */ |
| 116 | /* Objects are reclaimable */ |
| 117 | #define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000U) |
| 118 | #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 119 | |
| 120 | /* Slab deactivation flag */ |
| 121 | #define SLAB_DEACTIVATED ((slab_flags_t __force)0x10000000U) |
| 122 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 123 | /* |
| 124 | * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests. |
| 125 | * |
| 126 | * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault. |
| 127 | * |
| 128 | * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can. |
| 129 | * Both make kfree a no-op. |
| 130 | */ |
| 131 | #define ZERO_SIZE_PTR ((void *)16) |
| 132 | |
| 133 | #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \ |
| 134 | (unsigned long)ZERO_SIZE_PTR) |
| 135 | |
| 136 | #include <linux/kasan.h> |
| 137 | |
| 138 | struct mem_cgroup; |
| 139 | /* |
| 140 | * struct kmem_cache related prototypes |
| 141 | */ |
| 142 | void __init kmem_cache_init(void); |
| 143 | bool slab_is_available(void); |
| 144 | |
| 145 | extern bool usercopy_fallback; |
| 146 | |
| 147 | struct kmem_cache *kmem_cache_create(const char *name, unsigned int size, |
| 148 | unsigned int align, slab_flags_t flags, |
| 149 | void (*ctor)(void *)); |
| 150 | struct kmem_cache *kmem_cache_create_usercopy(const char *name, |
| 151 | unsigned int size, unsigned int align, |
| 152 | slab_flags_t flags, |
| 153 | unsigned int useroffset, unsigned int usersize, |
| 154 | void (*ctor)(void *)); |
| 155 | void kmem_cache_destroy(struct kmem_cache *); |
| 156 | int kmem_cache_shrink(struct kmem_cache *); |
| 157 | |
| 158 | void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 159 | void memcg_deactivate_kmem_caches(struct mem_cgroup *, struct mem_cgroup *); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 160 | |
| 161 | /* |
| 162 | * Please use this macro to create slab caches. Simply specify the |
| 163 | * name of the structure and maybe some flags that are listed above. |
| 164 | * |
| 165 | * The alignment of the struct determines object alignment. If you |
| 166 | * f.e. add ____cacheline_aligned_in_smp to the struct declaration |
| 167 | * then the objects will be properly aligned in SMP configurations. |
| 168 | */ |
| 169 | #define KMEM_CACHE(__struct, __flags) \ |
| 170 | kmem_cache_create(#__struct, sizeof(struct __struct), \ |
| 171 | __alignof__(struct __struct), (__flags), NULL) |
| 172 | |
| 173 | /* |
| 174 | * To whitelist a single field for copying to/from usercopy, use this |
| 175 | * macro instead for KMEM_CACHE() above. |
| 176 | */ |
| 177 | #define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \ |
| 178 | kmem_cache_create_usercopy(#__struct, \ |
| 179 | sizeof(struct __struct), \ |
| 180 | __alignof__(struct __struct), (__flags), \ |
| 181 | offsetof(struct __struct, __field), \ |
| 182 | sizeof_field(struct __struct, __field), NULL) |
| 183 | |
| 184 | /* |
| 185 | * Common kmalloc functions provided by all allocators |
| 186 | */ |
| 187 | void * __must_check __krealloc(const void *, size_t, gfp_t); |
| 188 | void * __must_check krealloc(const void *, size_t, gfp_t); |
| 189 | void kfree(const void *); |
| 190 | void kzfree(const void *); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 191 | size_t __ksize(const void *); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 192 | size_t ksize(const void *); |
| 193 | |
| 194 | #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR |
| 195 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
| 196 | bool to_user); |
| 197 | #else |
| 198 | static inline void __check_heap_object(const void *ptr, unsigned long n, |
| 199 | struct page *page, bool to_user) { } |
| 200 | #endif |
| 201 | |
| 202 | /* |
| 203 | * Some archs want to perform DMA into kmalloc caches and need a guaranteed |
| 204 | * alignment larger than the alignment of a 64-bit integer. |
| 205 | * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that. |
| 206 | */ |
| 207 | #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8 |
| 208 | #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN |
| 209 | #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN |
| 210 | #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN) |
| 211 | #else |
| 212 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
| 213 | #endif |
| 214 | |
| 215 | /* |
| 216 | * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment. |
| 217 | * Intended for arches that get misalignment faults even for 64 bit integer |
| 218 | * aligned buffers. |
| 219 | */ |
| 220 | #ifndef ARCH_SLAB_MINALIGN |
| 221 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
| 222 | #endif |
| 223 | |
| 224 | /* |
| 225 | * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned |
| 226 | * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN |
| 227 | * aligned pointers. |
| 228 | */ |
| 229 | #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN) |
| 230 | #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN) |
| 231 | #define __assume_page_alignment __assume_aligned(PAGE_SIZE) |
| 232 | |
| 233 | /* |
| 234 | * Kmalloc array related definitions |
| 235 | */ |
| 236 | |
| 237 | #ifdef CONFIG_SLAB |
| 238 | /* |
| 239 | * The largest kmalloc size supported by the SLAB allocators is |
| 240 | * 32 megabyte (2^25) or the maximum allocatable page order if that is |
| 241 | * less than 32 MB. |
| 242 | * |
| 243 | * WARNING: Its not easy to increase this value since the allocators have |
| 244 | * to do various tricks to work around compiler limitations in order to |
| 245 | * ensure proper constant folding. |
| 246 | */ |
| 247 | #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \ |
| 248 | (MAX_ORDER + PAGE_SHIFT - 1) : 25) |
| 249 | #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH |
| 250 | #ifndef KMALLOC_SHIFT_LOW |
| 251 | #define KMALLOC_SHIFT_LOW 5 |
| 252 | #endif |
| 253 | #endif |
| 254 | |
| 255 | #ifdef CONFIG_SLUB |
| 256 | /* |
| 257 | * SLUB directly allocates requests fitting in to an order-1 page |
| 258 | * (PAGE_SIZE*2). Larger requests are passed to the page allocator. |
| 259 | */ |
| 260 | #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1) |
| 261 | #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1) |
| 262 | #ifndef KMALLOC_SHIFT_LOW |
| 263 | #define KMALLOC_SHIFT_LOW 3 |
| 264 | #endif |
| 265 | #endif |
| 266 | |
| 267 | #ifdef CONFIG_SLOB |
| 268 | /* |
| 269 | * SLOB passes all requests larger than one page to the page allocator. |
| 270 | * No kmalloc array is necessary since objects of different sizes can |
| 271 | * be allocated from the same page. |
| 272 | */ |
| 273 | #define KMALLOC_SHIFT_HIGH PAGE_SHIFT |
| 274 | #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1) |
| 275 | #ifndef KMALLOC_SHIFT_LOW |
| 276 | #define KMALLOC_SHIFT_LOW 3 |
| 277 | #endif |
| 278 | #endif |
| 279 | |
| 280 | /* Maximum allocatable size */ |
| 281 | #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX) |
| 282 | /* Maximum size for which we actually use a slab cache */ |
| 283 | #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH) |
| 284 | /* Maximum order allocatable via the slab allocagtor */ |
| 285 | #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT) |
| 286 | |
| 287 | /* |
| 288 | * Kmalloc subsystem. |
| 289 | */ |
| 290 | #ifndef KMALLOC_MIN_SIZE |
| 291 | #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW) |
| 292 | #endif |
| 293 | |
| 294 | /* |
| 295 | * This restriction comes from byte sized index implementation. |
| 296 | * Page size is normally 2^12 bytes and, in this case, if we want to use |
| 297 | * byte sized index which can represent 2^8 entries, the size of the object |
| 298 | * should be equal or greater to 2^12 / 2^8 = 2^4 = 16. |
| 299 | * If minimum size of kmalloc is less than 16, we use it as minimum object |
| 300 | * size and give up to use byte sized index. |
| 301 | */ |
| 302 | #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \ |
| 303 | (KMALLOC_MIN_SIZE) : 16) |
| 304 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 305 | /* |
| 306 | * Whenever changing this, take care of that kmalloc_type() and |
| 307 | * create_kmalloc_caches() still work as intended. |
| 308 | */ |
| 309 | enum kmalloc_cache_type { |
| 310 | KMALLOC_NORMAL = 0, |
| 311 | KMALLOC_RECLAIM, |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 312 | #ifdef CONFIG_ZONE_DMA |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 313 | KMALLOC_DMA, |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 314 | #endif |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 315 | NR_KMALLOC_TYPES |
| 316 | }; |
| 317 | |
| 318 | #ifndef CONFIG_SLOB |
| 319 | extern struct kmem_cache * |
| 320 | kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1]; |
| 321 | |
| 322 | static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags) |
| 323 | { |
| 324 | #ifdef CONFIG_ZONE_DMA |
| 325 | /* |
| 326 | * The most common case is KMALLOC_NORMAL, so test for it |
| 327 | * with a single branch for both flags. |
| 328 | */ |
| 329 | if (likely((flags & (__GFP_DMA | __GFP_RECLAIMABLE)) == 0)) |
| 330 | return KMALLOC_NORMAL; |
| 331 | |
| 332 | /* |
| 333 | * At least one of the flags has to be set. If both are, __GFP_DMA |
| 334 | * is more important. |
| 335 | */ |
| 336 | return flags & __GFP_DMA ? KMALLOC_DMA : KMALLOC_RECLAIM; |
| 337 | #else |
| 338 | return flags & __GFP_RECLAIMABLE ? KMALLOC_RECLAIM : KMALLOC_NORMAL; |
| 339 | #endif |
| 340 | } |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 341 | |
| 342 | /* |
| 343 | * Figure out which kmalloc slab an allocation of a certain size |
| 344 | * belongs to. |
| 345 | * 0 = zero alloc |
| 346 | * 1 = 65 .. 96 bytes |
| 347 | * 2 = 129 .. 192 bytes |
| 348 | * n = 2^(n-1)+1 .. 2^n |
| 349 | */ |
| 350 | static __always_inline unsigned int kmalloc_index(size_t size) |
| 351 | { |
| 352 | if (!size) |
| 353 | return 0; |
| 354 | |
| 355 | if (size <= KMALLOC_MIN_SIZE) |
| 356 | return KMALLOC_SHIFT_LOW; |
| 357 | |
| 358 | if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96) |
| 359 | return 1; |
| 360 | if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192) |
| 361 | return 2; |
| 362 | if (size <= 8) return 3; |
| 363 | if (size <= 16) return 4; |
| 364 | if (size <= 32) return 5; |
| 365 | if (size <= 64) return 6; |
| 366 | if (size <= 128) return 7; |
| 367 | if (size <= 256) return 8; |
| 368 | if (size <= 512) return 9; |
| 369 | if (size <= 1024) return 10; |
| 370 | if (size <= 2 * 1024) return 11; |
| 371 | if (size <= 4 * 1024) return 12; |
| 372 | if (size <= 8 * 1024) return 13; |
| 373 | if (size <= 16 * 1024) return 14; |
| 374 | if (size <= 32 * 1024) return 15; |
| 375 | if (size <= 64 * 1024) return 16; |
| 376 | if (size <= 128 * 1024) return 17; |
| 377 | if (size <= 256 * 1024) return 18; |
| 378 | if (size <= 512 * 1024) return 19; |
| 379 | if (size <= 1024 * 1024) return 20; |
| 380 | if (size <= 2 * 1024 * 1024) return 21; |
| 381 | if (size <= 4 * 1024 * 1024) return 22; |
| 382 | if (size <= 8 * 1024 * 1024) return 23; |
| 383 | if (size <= 16 * 1024 * 1024) return 24; |
| 384 | if (size <= 32 * 1024 * 1024) return 25; |
| 385 | if (size <= 64 * 1024 * 1024) return 26; |
| 386 | BUG(); |
| 387 | |
| 388 | /* Will never be reached. Needed because the compiler may complain */ |
| 389 | return -1; |
| 390 | } |
| 391 | #endif /* !CONFIG_SLOB */ |
| 392 | |
| 393 | void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc; |
| 394 | void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc; |
| 395 | void kmem_cache_free(struct kmem_cache *, void *); |
| 396 | |
| 397 | /* |
| 398 | * Bulk allocation and freeing operations. These are accelerated in an |
| 399 | * allocator specific way to avoid taking locks repeatedly or building |
| 400 | * metadata structures unnecessarily. |
| 401 | * |
| 402 | * Note that interrupts must be enabled when calling these functions. |
| 403 | */ |
| 404 | void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **); |
| 405 | int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **); |
| 406 | |
| 407 | /* |
| 408 | * Caller must not use kfree_bulk() on memory not originally allocated |
| 409 | * by kmalloc(), because the SLOB allocator cannot handle this. |
| 410 | */ |
| 411 | static __always_inline void kfree_bulk(size_t size, void **p) |
| 412 | { |
| 413 | kmem_cache_free_bulk(NULL, size, p); |
| 414 | } |
| 415 | |
| 416 | #ifdef CONFIG_NUMA |
| 417 | void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc; |
| 418 | void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc; |
| 419 | #else |
| 420 | static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node) |
| 421 | { |
| 422 | return __kmalloc(size, flags); |
| 423 | } |
| 424 | |
| 425 | static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node) |
| 426 | { |
| 427 | return kmem_cache_alloc(s, flags); |
| 428 | } |
| 429 | #endif |
| 430 | |
| 431 | #ifdef CONFIG_TRACING |
| 432 | extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc; |
| 433 | |
| 434 | #ifdef CONFIG_NUMA |
| 435 | extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
| 436 | gfp_t gfpflags, |
| 437 | int node, size_t size) __assume_slab_alignment __malloc; |
| 438 | #else |
| 439 | static __always_inline void * |
| 440 | kmem_cache_alloc_node_trace(struct kmem_cache *s, |
| 441 | gfp_t gfpflags, |
| 442 | int node, size_t size) |
| 443 | { |
| 444 | return kmem_cache_alloc_trace(s, gfpflags, size); |
| 445 | } |
| 446 | #endif /* CONFIG_NUMA */ |
| 447 | |
| 448 | #else /* CONFIG_TRACING */ |
| 449 | static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s, |
| 450 | gfp_t flags, size_t size) |
| 451 | { |
| 452 | void *ret = kmem_cache_alloc(s, flags); |
| 453 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 454 | ret = kasan_kmalloc(s, ret, size, flags); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 455 | return ret; |
| 456 | } |
| 457 | |
| 458 | static __always_inline void * |
| 459 | kmem_cache_alloc_node_trace(struct kmem_cache *s, |
| 460 | gfp_t gfpflags, |
| 461 | int node, size_t size) |
| 462 | { |
| 463 | void *ret = kmem_cache_alloc_node(s, gfpflags, node); |
| 464 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 465 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 466 | return ret; |
| 467 | } |
| 468 | #endif /* CONFIG_TRACING */ |
| 469 | |
| 470 | extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc; |
| 471 | |
| 472 | #ifdef CONFIG_TRACING |
| 473 | extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc; |
| 474 | #else |
| 475 | static __always_inline void * |
| 476 | kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) |
| 477 | { |
| 478 | return kmalloc_order(size, flags, order); |
| 479 | } |
| 480 | #endif |
| 481 | |
| 482 | static __always_inline void *kmalloc_large(size_t size, gfp_t flags) |
| 483 | { |
| 484 | unsigned int order = get_order(size); |
| 485 | return kmalloc_order_trace(size, flags, order); |
| 486 | } |
| 487 | |
| 488 | /** |
| 489 | * kmalloc - allocate memory |
| 490 | * @size: how many bytes of memory are required. |
| 491 | * @flags: the type of memory to allocate. |
| 492 | * |
| 493 | * kmalloc is the normal method of allocating memory |
| 494 | * for objects smaller than page size in the kernel. |
| 495 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 496 | * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN |
| 497 | * bytes. For @size of power of two bytes, the alignment is also guaranteed |
| 498 | * to be at least to the size. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 499 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 500 | * The @flags argument may be one of the GFP flags defined at |
| 501 | * include/linux/gfp.h and described at |
| 502 | * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>` |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 503 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 504 | * The recommended usage of the @flags is described at |
| 505 | * :ref:`Documentation/core-api/memory-allocation.rst <memory-allocation>` |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 506 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 507 | * Below is a brief outline of the most useful GFP flags |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 508 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 509 | * %GFP_KERNEL |
| 510 | * Allocate normal kernel ram. May sleep. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 511 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 512 | * %GFP_NOWAIT |
| 513 | * Allocation will not sleep. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 514 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 515 | * %GFP_ATOMIC |
| 516 | * Allocation will not sleep. May use emergency pools. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 517 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 518 | * %GFP_HIGHUSER |
| 519 | * Allocate memory from high memory on behalf of user. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 520 | * |
| 521 | * Also it is possible to set different flags by OR'ing |
| 522 | * in one or more of the following additional @flags: |
| 523 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 524 | * %__GFP_HIGH |
| 525 | * This allocation has high priority and may use emergency pools. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 526 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 527 | * %__GFP_NOFAIL |
| 528 | * Indicate that this allocation is in no way allowed to fail |
| 529 | * (think twice before using). |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 530 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 531 | * %__GFP_NORETRY |
| 532 | * If memory is not immediately available, |
| 533 | * then give up at once. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 534 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 535 | * %__GFP_NOWARN |
| 536 | * If allocation fails, don't issue any warnings. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 537 | * |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 538 | * %__GFP_RETRY_MAYFAIL |
| 539 | * Try really hard to succeed the allocation but fail |
| 540 | * eventually. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 541 | */ |
| 542 | static __always_inline void *kmalloc(size_t size, gfp_t flags) |
| 543 | { |
| 544 | if (__builtin_constant_p(size)) { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 545 | #ifndef CONFIG_SLOB |
| 546 | unsigned int index; |
| 547 | #endif |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 548 | if (size > KMALLOC_MAX_CACHE_SIZE) |
| 549 | return kmalloc_large(size, flags); |
| 550 | #ifndef CONFIG_SLOB |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 551 | index = kmalloc_index(size); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 552 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 553 | if (!index) |
| 554 | return ZERO_SIZE_PTR; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 555 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 556 | return kmem_cache_alloc_trace( |
| 557 | kmalloc_caches[kmalloc_type(flags)][index], |
| 558 | flags, size); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 559 | #endif |
| 560 | } |
| 561 | return __kmalloc(size, flags); |
| 562 | } |
| 563 | |
| 564 | /* |
| 565 | * Determine size used for the nth kmalloc cache. |
| 566 | * return size or 0 if a kmalloc cache for that |
| 567 | * size does not exist |
| 568 | */ |
| 569 | static __always_inline unsigned int kmalloc_size(unsigned int n) |
| 570 | { |
| 571 | #ifndef CONFIG_SLOB |
| 572 | if (n > 2) |
| 573 | return 1U << n; |
| 574 | |
| 575 | if (n == 1 && KMALLOC_MIN_SIZE <= 32) |
| 576 | return 96; |
| 577 | |
| 578 | if (n == 2 && KMALLOC_MIN_SIZE <= 64) |
| 579 | return 192; |
| 580 | #endif |
| 581 | return 0; |
| 582 | } |
| 583 | |
| 584 | static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node) |
| 585 | { |
| 586 | #ifndef CONFIG_SLOB |
| 587 | if (__builtin_constant_p(size) && |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 588 | size <= KMALLOC_MAX_CACHE_SIZE) { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 589 | unsigned int i = kmalloc_index(size); |
| 590 | |
| 591 | if (!i) |
| 592 | return ZERO_SIZE_PTR; |
| 593 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 594 | return kmem_cache_alloc_node_trace( |
| 595 | kmalloc_caches[kmalloc_type(flags)][i], |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 596 | flags, node, size); |
| 597 | } |
| 598 | #endif |
| 599 | return __kmalloc_node(size, flags, node); |
| 600 | } |
| 601 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 602 | int memcg_update_all_caches(int num_memcgs); |
| 603 | |
| 604 | /** |
| 605 | * kmalloc_array - allocate memory for an array. |
| 606 | * @n: number of elements. |
| 607 | * @size: element size. |
| 608 | * @flags: the type of memory to allocate (see kmalloc). |
| 609 | */ |
| 610 | static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags) |
| 611 | { |
| 612 | size_t bytes; |
| 613 | |
| 614 | if (unlikely(check_mul_overflow(n, size, &bytes))) |
| 615 | return NULL; |
| 616 | if (__builtin_constant_p(n) && __builtin_constant_p(size)) |
| 617 | return kmalloc(bytes, flags); |
| 618 | return __kmalloc(bytes, flags); |
| 619 | } |
| 620 | |
| 621 | /** |
| 622 | * kcalloc - allocate memory for an array. The memory is set to zero. |
| 623 | * @n: number of elements. |
| 624 | * @size: element size. |
| 625 | * @flags: the type of memory to allocate (see kmalloc). |
| 626 | */ |
| 627 | static inline void *kcalloc(size_t n, size_t size, gfp_t flags) |
| 628 | { |
| 629 | return kmalloc_array(n, size, flags | __GFP_ZERO); |
| 630 | } |
| 631 | |
| 632 | /* |
| 633 | * kmalloc_track_caller is a special version of kmalloc that records the |
| 634 | * calling function of the routine calling it for slab leak tracking instead |
| 635 | * of just the calling function (confusing, eh?). |
| 636 | * It's useful when the call to kmalloc comes from a widely-used standard |
| 637 | * allocator where we care about the real place the memory allocation |
| 638 | * request comes from. |
| 639 | */ |
| 640 | extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long); |
| 641 | #define kmalloc_track_caller(size, flags) \ |
| 642 | __kmalloc_track_caller(size, flags, _RET_IP_) |
| 643 | |
| 644 | static inline void *kmalloc_array_node(size_t n, size_t size, gfp_t flags, |
| 645 | int node) |
| 646 | { |
| 647 | size_t bytes; |
| 648 | |
| 649 | if (unlikely(check_mul_overflow(n, size, &bytes))) |
| 650 | return NULL; |
| 651 | if (__builtin_constant_p(n) && __builtin_constant_p(size)) |
| 652 | return kmalloc_node(bytes, flags, node); |
| 653 | return __kmalloc_node(bytes, flags, node); |
| 654 | } |
| 655 | |
| 656 | static inline void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node) |
| 657 | { |
| 658 | return kmalloc_array_node(n, size, flags | __GFP_ZERO, node); |
| 659 | } |
| 660 | |
| 661 | |
| 662 | #ifdef CONFIG_NUMA |
| 663 | extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long); |
| 664 | #define kmalloc_node_track_caller(size, flags, node) \ |
| 665 | __kmalloc_node_track_caller(size, flags, node, \ |
| 666 | _RET_IP_) |
| 667 | |
| 668 | #else /* CONFIG_NUMA */ |
| 669 | |
| 670 | #define kmalloc_node_track_caller(size, flags, node) \ |
| 671 | kmalloc_track_caller(size, flags) |
| 672 | |
| 673 | #endif /* CONFIG_NUMA */ |
| 674 | |
| 675 | /* |
| 676 | * Shortcuts |
| 677 | */ |
| 678 | static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags) |
| 679 | { |
| 680 | return kmem_cache_alloc(k, flags | __GFP_ZERO); |
| 681 | } |
| 682 | |
| 683 | /** |
| 684 | * kzalloc - allocate memory. The memory is set to zero. |
| 685 | * @size: how many bytes of memory are required. |
| 686 | * @flags: the type of memory to allocate (see kmalloc). |
| 687 | */ |
| 688 | static inline void *kzalloc(size_t size, gfp_t flags) |
| 689 | { |
| 690 | return kmalloc(size, flags | __GFP_ZERO); |
| 691 | } |
| 692 | |
| 693 | /** |
| 694 | * kzalloc_node - allocate zeroed memory from a particular memory node. |
| 695 | * @size: how many bytes of memory are required. |
| 696 | * @flags: the type of memory to allocate (see kmalloc). |
| 697 | * @node: memory node from which to allocate |
| 698 | */ |
| 699 | static inline void *kzalloc_node(size_t size, gfp_t flags, int node) |
| 700 | { |
| 701 | return kmalloc_node(size, flags | __GFP_ZERO, node); |
| 702 | } |
| 703 | |
| 704 | unsigned int kmem_cache_size(struct kmem_cache *s); |
| 705 | void __init kmem_cache_init_late(void); |
| 706 | |
| 707 | #if defined(CONFIG_SMP) && defined(CONFIG_SLAB) |
| 708 | int slab_prepare_cpu(unsigned int cpu); |
| 709 | int slab_dead_cpu(unsigned int cpu); |
| 710 | #else |
| 711 | #define slab_prepare_cpu NULL |
| 712 | #define slab_dead_cpu NULL |
| 713 | #endif |
| 714 | |
| 715 | #endif /* _LINUX_SLAB_H */ |