Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
1da177e4 | 2 | /* |
2e892f43 CL |
3 | * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk). |
4 | * | |
cde53535 | 5 | * (C) SGI 2006, Christoph Lameter |
2e892f43 CL |
6 | * Cleaned up and restructured to ease the addition of alternative |
7 | * implementations of SLAB allocators. | |
f1b6eb6e CL |
8 | * (C) Linux Foundation 2008-2013 |
9 | * Unified interface for all slab allocators | |
1da177e4 LT |
10 | */ |
11 | ||
12 | #ifndef _LINUX_SLAB_H | |
13 | #define _LINUX_SLAB_H | |
14 | ||
1b1cec4b | 15 | #include <linux/gfp.h> |
49b7f898 | 16 | #include <linux/overflow.h> |
1b1cec4b | 17 | #include <linux/types.h> |
1f458cbf | 18 | #include <linux/workqueue.h> |
f0a3a24b | 19 | #include <linux/percpu-refcount.h> |
1f458cbf | 20 | |
1da177e4 | 21 | |
2e892f43 CL |
22 | /* |
23 | * Flags to pass to kmem_cache_create(). | |
124dee09 | 24 | * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set. |
1da177e4 | 25 | */ |
d50112ed | 26 | /* DEBUG: Perform (expensive) checks on alloc/free */ |
4fd0b46e | 27 | #define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100U) |
d50112ed | 28 | /* DEBUG: Red zone objs in a cache */ |
4fd0b46e | 29 | #define SLAB_RED_ZONE ((slab_flags_t __force)0x00000400U) |
d50112ed | 30 | /* DEBUG: Poison objects */ |
4fd0b46e | 31 | #define SLAB_POISON ((slab_flags_t __force)0x00000800U) |
6edf2576 FT |
32 | /* Indicate a kmalloc slab */ |
33 | #define SLAB_KMALLOC ((slab_flags_t __force)0x00001000U) | |
d50112ed | 34 | /* Align objs on cache lines */ |
4fd0b46e | 35 | #define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000U) |
d50112ed | 36 | /* Use GFP_DMA memory */ |
4fd0b46e | 37 | #define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000U) |
6d6ea1e9 NB |
38 | /* Use GFP_DMA32 memory */ |
39 | #define SLAB_CACHE_DMA32 ((slab_flags_t __force)0x00008000U) | |
d50112ed | 40 | /* DEBUG: Store the last owner for bug hunting */ |
4fd0b46e | 41 | #define SLAB_STORE_USER ((slab_flags_t __force)0x00010000U) |
d50112ed | 42 | /* Panic if kmem_cache_create() fails */ |
4fd0b46e | 43 | #define SLAB_PANIC ((slab_flags_t __force)0x00040000U) |
d7de4c1d | 44 | /* |
5f0d5a3a | 45 | * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS! |
d7de4c1d PZ |
46 | * |
47 | * This delays freeing the SLAB page by a grace period, it does _NOT_ | |
48 | * delay object freeing. This means that if you do kmem_cache_free() | |
49 | * that memory location is free to be reused at any time. Thus it may | |
50 | * be possible to see another object there in the same RCU grace period. | |
51 | * | |
52 | * This feature only ensures the memory location backing the object | |
53 | * stays valid, the trick to using this is relying on an independent | |
54 | * object validation pass. Something like: | |
55 | * | |
56 | * rcu_read_lock() | |
57 | * again: | |
58 | * obj = lockless_lookup(key); | |
59 | * if (obj) { | |
60 | * if (!try_get_ref(obj)) // might fail for free objects | |
61 | * goto again; | |
62 | * | |
63 | * if (obj->key != key) { // not the object we expected | |
64 | * put_ref(obj); | |
65 | * goto again; | |
66 | * } | |
67 | * } | |
68 | * rcu_read_unlock(); | |
69 | * | |
68126702 JK |
70 | * This is useful if we need to approach a kernel structure obliquely, |
71 | * from its address obtained without the usual locking. We can lock | |
72 | * the structure to stabilize it and check it's still at the given address, | |
73 | * only if we can be sure that the memory has not been meanwhile reused | |
74 | * for some other kind of object (which our subsystem's lock might corrupt). | |
75 | * | |
76 | * rcu_read_lock before reading the address, then rcu_read_unlock after | |
77 | * taking the spinlock within the structure expected at that address. | |
5f0d5a3a | 78 | * |
e9f8a790 PM |
79 | * Note that it is not possible to acquire a lock within a structure |
80 | * allocated with SLAB_TYPESAFE_BY_RCU without first acquiring a reference | |
81 | * as described above. The reason is that SLAB_TYPESAFE_BY_RCU pages | |
82 | * are not zeroed before being given to the slab, which means that any | |
83 | * locks must be initialized after each and every kmem_struct_alloc(). | |
84 | * Alternatively, make the ctor passed to kmem_cache_create() initialize | |
85 | * the locks at page-allocation time, as is done in __i915_request_ctor(), | |
86 | * sighand_ctor(), and anon_vma_ctor(). Such a ctor permits readers | |
87 | * to safely acquire those ctor-initialized locks under rcu_read_lock() | |
88 | * protection. | |
89 | * | |
5f0d5a3a | 90 | * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU. |
d7de4c1d | 91 | */ |
d50112ed | 92 | /* Defer freeing slabs to RCU */ |
4fd0b46e | 93 | #define SLAB_TYPESAFE_BY_RCU ((slab_flags_t __force)0x00080000U) |
d50112ed | 94 | /* Spread some memory over cpuset */ |
4fd0b46e | 95 | #define SLAB_MEM_SPREAD ((slab_flags_t __force)0x00100000U) |
d50112ed | 96 | /* Trace allocations and frees */ |
4fd0b46e | 97 | #define SLAB_TRACE ((slab_flags_t __force)0x00200000U) |
1da177e4 | 98 | |
30327acf TG |
99 | /* Flag to prevent checks on free */ |
100 | #ifdef CONFIG_DEBUG_OBJECTS | |
4fd0b46e | 101 | # define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00400000U) |
30327acf | 102 | #else |
4fd0b46e | 103 | # define SLAB_DEBUG_OBJECTS 0 |
30327acf TG |
104 | #endif |
105 | ||
d50112ed | 106 | /* Avoid kmemleak tracing */ |
4fd0b46e | 107 | #define SLAB_NOLEAKTRACE ((slab_flags_t __force)0x00800000U) |
d5cff635 | 108 | |
d50112ed | 109 | /* Fault injection mark */ |
4c13dd3b | 110 | #ifdef CONFIG_FAILSLAB |
4fd0b46e | 111 | # define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000U) |
4c13dd3b | 112 | #else |
4fd0b46e | 113 | # define SLAB_FAILSLAB 0 |
4c13dd3b | 114 | #endif |
d50112ed | 115 | /* Account to memcg */ |
84c07d11 | 116 | #ifdef CONFIG_MEMCG_KMEM |
4fd0b46e | 117 | # define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000U) |
230e9fc2 | 118 | #else |
4fd0b46e | 119 | # define SLAB_ACCOUNT 0 |
230e9fc2 | 120 | #endif |
2dff4405 | 121 | |
682ed089 | 122 | #ifdef CONFIG_KASAN_GENERIC |
4fd0b46e | 123 | #define SLAB_KASAN ((slab_flags_t __force)0x08000000U) |
7ed2f9e6 | 124 | #else |
4fd0b46e | 125 | #define SLAB_KASAN 0 |
7ed2f9e6 AP |
126 | #endif |
127 | ||
a285909f HY |
128 | /* |
129 | * Ignore user specified debugging flags. | |
130 | * Intended for caches created for self-tests so they have only flags | |
131 | * specified in the code and other flags are ignored. | |
132 | */ | |
133 | #define SLAB_NO_USER_FLAGS ((slab_flags_t __force)0x10000000U) | |
134 | ||
b84e04f1 IK |
135 | #ifdef CONFIG_KFENCE |
136 | #define SLAB_SKIP_KFENCE ((slab_flags_t __force)0x20000000U) | |
137 | #else | |
138 | #define SLAB_SKIP_KFENCE 0 | |
139 | #endif | |
140 | ||
e12ba74d | 141 | /* The following flags affect the page allocator grouping pages by mobility */ |
d50112ed | 142 | /* Objects are reclaimable */ |
3d97d976 | 143 | #ifndef CONFIG_SLUB_TINY |
4fd0b46e | 144 | #define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000U) |
3d97d976 VB |
145 | #else |
146 | #define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0) | |
147 | #endif | |
e12ba74d | 148 | #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */ |
fcf8a1e4 | 149 | |
6cb8f913 CL |
150 | /* |
151 | * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests. | |
152 | * | |
153 | * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault. | |
154 | * | |
155 | * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can. | |
156 | * Both make kfree a no-op. | |
157 | */ | |
158 | #define ZERO_SIZE_PTR ((void *)16) | |
159 | ||
1d4ec7b1 | 160 | #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \ |
6cb8f913 CL |
161 | (unsigned long)ZERO_SIZE_PTR) |
162 | ||
0316bec2 | 163 | #include <linux/kasan.h> |
3b0efdfa | 164 | |
88f2ef73 | 165 | struct list_lru; |
2633d7a0 | 166 | struct mem_cgroup; |
2e892f43 CL |
167 | /* |
168 | * struct kmem_cache related prototypes | |
169 | */ | |
170 | void __init kmem_cache_init(void); | |
fda90124 | 171 | bool slab_is_available(void); |
1da177e4 | 172 | |
f4957d5b AD |
173 | struct kmem_cache *kmem_cache_create(const char *name, unsigned int size, |
174 | unsigned int align, slab_flags_t flags, | |
8eb8284b DW |
175 | void (*ctor)(void *)); |
176 | struct kmem_cache *kmem_cache_create_usercopy(const char *name, | |
f4957d5b AD |
177 | unsigned int size, unsigned int align, |
178 | slab_flags_t flags, | |
7bbdb81e | 179 | unsigned int useroffset, unsigned int usersize, |
8eb8284b | 180 | void (*ctor)(void *)); |
72d67229 KC |
181 | void kmem_cache_destroy(struct kmem_cache *s); |
182 | int kmem_cache_shrink(struct kmem_cache *s); | |
2a4db7eb | 183 | |
0a31bd5f CL |
184 | /* |
185 | * Please use this macro to create slab caches. Simply specify the | |
186 | * name of the structure and maybe some flags that are listed above. | |
187 | * | |
188 | * The alignment of the struct determines object alignment. If you | |
189 | * f.e. add ____cacheline_aligned_in_smp to the struct declaration | |
190 | * then the objects will be properly aligned in SMP configurations. | |
191 | */ | |
8eb8284b DW |
192 | #define KMEM_CACHE(__struct, __flags) \ |
193 | kmem_cache_create(#__struct, sizeof(struct __struct), \ | |
194 | __alignof__(struct __struct), (__flags), NULL) | |
195 | ||
196 | /* | |
197 | * To whitelist a single field for copying to/from usercopy, use this | |
198 | * macro instead for KMEM_CACHE() above. | |
199 | */ | |
200 | #define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \ | |
201 | kmem_cache_create_usercopy(#__struct, \ | |
202 | sizeof(struct __struct), \ | |
203 | __alignof__(struct __struct), (__flags), \ | |
204 | offsetof(struct __struct, __field), \ | |
205 | sizeof_field(struct __struct, __field), NULL) | |
0a31bd5f | 206 | |
34504667 CL |
207 | /* |
208 | * Common kmalloc functions provided by all allocators | |
209 | */ | |
9ed9cac1 | 210 | void * __must_check krealloc(const void *objp, size_t new_size, gfp_t flags) __realloc_size(2); |
72d67229 KC |
211 | void kfree(const void *objp); |
212 | void kfree_sensitive(const void *objp); | |
213 | size_t __ksize(const void *objp); | |
05a94065 KC |
214 | |
215 | /** | |
216 | * ksize - Report actual allocation size of associated object | |
217 | * | |
218 | * @objp: Pointer returned from a prior kmalloc()-family allocation. | |
219 | * | |
220 | * This should not be used for writing beyond the originally requested | |
221 | * allocation size. Either use krealloc() or round up the allocation size | |
222 | * with kmalloc_size_roundup() prior to allocation. If this is used to | |
223 | * access beyond the originally requested allocation size, UBSAN_BOUNDS | |
224 | * and/or FORTIFY_SOURCE may trip, since they only know about the | |
225 | * originally allocated size via the __alloc_size attribute. | |
226 | */ | |
72d67229 | 227 | size_t ksize(const void *objp); |
05a94065 | 228 | |
5bb1bb35 | 229 | #ifdef CONFIG_PRINTK |
8e7f37f2 PM |
230 | bool kmem_valid_obj(void *object); |
231 | void kmem_dump_obj(void *object); | |
5bb1bb35 | 232 | #endif |
34504667 | 233 | |
c601fd69 CL |
234 | /* |
235 | * Some archs want to perform DMA into kmalloc caches and need a guaranteed | |
236 | * alignment larger than the alignment of a 64-bit integer. | |
8cf9e121 | 237 | * Setting ARCH_DMA_MINALIGN in arch headers allows that. |
c601fd69 CL |
238 | */ |
239 | #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8 | |
240 | #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN | |
241 | #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN | |
242 | #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN) | |
243 | #else | |
244 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) | |
245 | #endif | |
246 | ||
94a58c36 RV |
247 | /* |
248 | * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment. | |
249 | * Intended for arches that get misalignment faults even for 64 bit integer | |
250 | * aligned buffers. | |
251 | */ | |
252 | #ifndef ARCH_SLAB_MINALIGN | |
253 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) | |
254 | #endif | |
255 | ||
d949a815 PC |
256 | /* |
257 | * Arches can define this function if they want to decide the minimum slab | |
258 | * alignment at runtime. The value returned by the function must be a power | |
259 | * of two and >= ARCH_SLAB_MINALIGN. | |
260 | */ | |
261 | #ifndef arch_slab_minalign | |
262 | static inline unsigned int arch_slab_minalign(void) | |
263 | { | |
264 | return ARCH_SLAB_MINALIGN; | |
265 | } | |
266 | #endif | |
267 | ||
94a58c36 | 268 | /* |
154036a3 AK |
269 | * kmem_cache_alloc and friends return pointers aligned to ARCH_SLAB_MINALIGN. |
270 | * kmalloc and friends return pointers aligned to both ARCH_KMALLOC_MINALIGN | |
271 | * and ARCH_SLAB_MINALIGN, but here we only assume the former alignment. | |
94a58c36 RV |
272 | */ |
273 | #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN) | |
274 | #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN) | |
275 | #define __assume_page_alignment __assume_aligned(PAGE_SIZE) | |
276 | ||
0aa817f0 | 277 | /* |
95a05b42 CL |
278 | * Kmalloc array related definitions |
279 | */ | |
280 | ||
281 | #ifdef CONFIG_SLAB | |
282 | /* | |
d6a71648 HY |
283 | * SLAB and SLUB directly allocates requests fitting in to an order-1 page |
284 | * (PAGE_SIZE*2). Larger requests are passed to the page allocator. | |
0aa817f0 | 285 | */ |
d6a71648 | 286 | #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1) |
23baf831 | 287 | #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT) |
c601fd69 | 288 | #ifndef KMALLOC_SHIFT_LOW |
95a05b42 | 289 | #define KMALLOC_SHIFT_LOW 5 |
c601fd69 | 290 | #endif |
069e2b35 CL |
291 | #endif |
292 | ||
293 | #ifdef CONFIG_SLUB | |
95a05b42 | 294 | #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1) |
23baf831 | 295 | #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT) |
c601fd69 | 296 | #ifndef KMALLOC_SHIFT_LOW |
95a05b42 CL |
297 | #define KMALLOC_SHIFT_LOW 3 |
298 | #endif | |
c601fd69 | 299 | #endif |
0aa817f0 | 300 | |
069e2b35 CL |
301 | #ifdef CONFIG_SLOB |
302 | /* | |
433a91ff | 303 | * SLOB passes all requests larger than one page to the page allocator. |
069e2b35 CL |
304 | * No kmalloc array is necessary since objects of different sizes can |
305 | * be allocated from the same page. | |
306 | */ | |
069e2b35 | 307 | #define KMALLOC_SHIFT_HIGH PAGE_SHIFT |
23baf831 | 308 | #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT) |
069e2b35 CL |
309 | #ifndef KMALLOC_SHIFT_LOW |
310 | #define KMALLOC_SHIFT_LOW 3 | |
311 | #endif | |
312 | #endif | |
313 | ||
95a05b42 CL |
314 | /* Maximum allocatable size */ |
315 | #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX) | |
316 | /* Maximum size for which we actually use a slab cache */ | |
317 | #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH) | |
d7cff4de | 318 | /* Maximum order allocatable via the slab allocator */ |
95a05b42 | 319 | #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT) |
0aa817f0 | 320 | |
ce6a5026 CL |
321 | /* |
322 | * Kmalloc subsystem. | |
323 | */ | |
c601fd69 | 324 | #ifndef KMALLOC_MIN_SIZE |
95a05b42 | 325 | #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW) |
ce6a5026 CL |
326 | #endif |
327 | ||
24f870d8 JK |
328 | /* |
329 | * This restriction comes from byte sized index implementation. | |
330 | * Page size is normally 2^12 bytes and, in this case, if we want to use | |
331 | * byte sized index which can represent 2^8 entries, the size of the object | |
332 | * should be equal or greater to 2^12 / 2^8 = 2^4 = 16. | |
333 | * If minimum size of kmalloc is less than 16, we use it as minimum object | |
334 | * size and give up to use byte sized index. | |
335 | */ | |
336 | #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \ | |
337 | (KMALLOC_MIN_SIZE) : 16) | |
338 | ||
1291523f VB |
339 | /* |
340 | * Whenever changing this, take care of that kmalloc_type() and | |
341 | * create_kmalloc_caches() still work as intended. | |
494c1dfe WL |
342 | * |
343 | * KMALLOC_NORMAL can contain only unaccounted objects whereas KMALLOC_CGROUP | |
344 | * is for accounted but unreclaimable and non-dma objects. All the other | |
345 | * kmem caches can have both accounted and unaccounted objects. | |
1291523f | 346 | */ |
cc252eae VB |
347 | enum kmalloc_cache_type { |
348 | KMALLOC_NORMAL = 0, | |
494c1dfe WL |
349 | #ifndef CONFIG_ZONE_DMA |
350 | KMALLOC_DMA = KMALLOC_NORMAL, | |
351 | #endif | |
352 | #ifndef CONFIG_MEMCG_KMEM | |
353 | KMALLOC_CGROUP = KMALLOC_NORMAL, | |
494c1dfe | 354 | #endif |
2f7c1c13 VB |
355 | #ifdef CONFIG_SLUB_TINY |
356 | KMALLOC_RECLAIM = KMALLOC_NORMAL, | |
357 | #else | |
1291523f | 358 | KMALLOC_RECLAIM, |
2f7c1c13 | 359 | #endif |
cc252eae VB |
360 | #ifdef CONFIG_ZONE_DMA |
361 | KMALLOC_DMA, | |
2f7c1c13 VB |
362 | #endif |
363 | #ifdef CONFIG_MEMCG_KMEM | |
364 | KMALLOC_CGROUP, | |
cc252eae VB |
365 | #endif |
366 | NR_KMALLOC_TYPES | |
367 | }; | |
368 | ||
069e2b35 | 369 | #ifndef CONFIG_SLOB |
cc252eae VB |
370 | extern struct kmem_cache * |
371 | kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1]; | |
372 | ||
494c1dfe WL |
373 | /* |
374 | * Define gfp bits that should not be set for KMALLOC_NORMAL. | |
375 | */ | |
376 | #define KMALLOC_NOT_NORMAL_BITS \ | |
377 | (__GFP_RECLAIMABLE | \ | |
378 | (IS_ENABLED(CONFIG_ZONE_DMA) ? __GFP_DMA : 0) | \ | |
379 | (IS_ENABLED(CONFIG_MEMCG_KMEM) ? __GFP_ACCOUNT : 0)) | |
380 | ||
cc252eae VB |
381 | static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags) |
382 | { | |
4e45f712 VB |
383 | /* |
384 | * The most common case is KMALLOC_NORMAL, so test for it | |
494c1dfe | 385 | * with a single branch for all the relevant flags. |
4e45f712 | 386 | */ |
494c1dfe | 387 | if (likely((flags & KMALLOC_NOT_NORMAL_BITS) == 0)) |
4e45f712 | 388 | return KMALLOC_NORMAL; |
1291523f VB |
389 | |
390 | /* | |
494c1dfe WL |
391 | * At least one of the flags has to be set. Their priorities in |
392 | * decreasing order are: | |
393 | * 1) __GFP_DMA | |
394 | * 2) __GFP_RECLAIMABLE | |
395 | * 3) __GFP_ACCOUNT | |
1291523f | 396 | */ |
494c1dfe WL |
397 | if (IS_ENABLED(CONFIG_ZONE_DMA) && (flags & __GFP_DMA)) |
398 | return KMALLOC_DMA; | |
399 | if (!IS_ENABLED(CONFIG_MEMCG_KMEM) || (flags & __GFP_RECLAIMABLE)) | |
400 | return KMALLOC_RECLAIM; | |
401 | else | |
402 | return KMALLOC_CGROUP; | |
cc252eae VB |
403 | } |
404 | ||
ce6a5026 CL |
405 | /* |
406 | * Figure out which kmalloc slab an allocation of a certain size | |
407 | * belongs to. | |
408 | * 0 = zero alloc | |
409 | * 1 = 65 .. 96 bytes | |
1ed58b60 RV |
410 | * 2 = 129 .. 192 bytes |
411 | * n = 2^(n-1)+1 .. 2^n | |
588c7fa0 HY |
412 | * |
413 | * Note: __kmalloc_index() is compile-time optimized, and not runtime optimized; | |
414 | * typical usage is via kmalloc_index() and therefore evaluated at compile-time. | |
415 | * Callers where !size_is_constant should only be test modules, where runtime | |
416 | * overheads of __kmalloc_index() can be tolerated. Also see kmalloc_slab(). | |
ce6a5026 | 417 | */ |
588c7fa0 HY |
418 | static __always_inline unsigned int __kmalloc_index(size_t size, |
419 | bool size_is_constant) | |
ce6a5026 CL |
420 | { |
421 | if (!size) | |
422 | return 0; | |
423 | ||
424 | if (size <= KMALLOC_MIN_SIZE) | |
425 | return KMALLOC_SHIFT_LOW; | |
426 | ||
427 | if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96) | |
428 | return 1; | |
429 | if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192) | |
430 | return 2; | |
431 | if (size <= 8) return 3; | |
432 | if (size <= 16) return 4; | |
433 | if (size <= 32) return 5; | |
434 | if (size <= 64) return 6; | |
435 | if (size <= 128) return 7; | |
436 | if (size <= 256) return 8; | |
437 | if (size <= 512) return 9; | |
438 | if (size <= 1024) return 10; | |
439 | if (size <= 2 * 1024) return 11; | |
440 | if (size <= 4 * 1024) return 12; | |
441 | if (size <= 8 * 1024) return 13; | |
442 | if (size <= 16 * 1024) return 14; | |
443 | if (size <= 32 * 1024) return 15; | |
444 | if (size <= 64 * 1024) return 16; | |
445 | if (size <= 128 * 1024) return 17; | |
446 | if (size <= 256 * 1024) return 18; | |
447 | if (size <= 512 * 1024) return 19; | |
448 | if (size <= 1024 * 1024) return 20; | |
449 | if (size <= 2 * 1024 * 1024) return 21; | |
588c7fa0 | 450 | |
57b2b72a | 451 | if (!IS_ENABLED(CONFIG_PROFILE_ALL_BRANCHES) && size_is_constant) |
588c7fa0 HY |
452 | BUILD_BUG_ON_MSG(1, "unexpected size in kmalloc_index()"); |
453 | else | |
454 | BUG(); | |
ce6a5026 CL |
455 | |
456 | /* Will never be reached. Needed because the compiler may complain */ | |
457 | return -1; | |
458 | } | |
d6a71648 | 459 | static_assert(PAGE_SHIFT <= 20); |
588c7fa0 | 460 | #define kmalloc_index(s) __kmalloc_index(s, true) |
069e2b35 | 461 | #endif /* !CONFIG_SLOB */ |
ce6a5026 | 462 | |
c37495d6 | 463 | void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __alloc_size(1); |
838de63b VB |
464 | |
465 | /** | |
466 | * kmem_cache_alloc - Allocate an object | |
467 | * @cachep: The cache to allocate from. | |
468 | * @flags: See kmalloc(). | |
469 | * | |
470 | * Allocate an object from this cache. | |
471 | * See kmem_cache_zalloc() for a shortcut of adding __GFP_ZERO to flags. | |
472 | * | |
473 | * Return: pointer to the new object or %NULL in case of error | |
474 | */ | |
475 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) __assume_slab_alignment __malloc; | |
88f2ef73 MS |
476 | void *kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru, |
477 | gfp_t gfpflags) __assume_slab_alignment __malloc; | |
72d67229 | 478 | void kmem_cache_free(struct kmem_cache *s, void *objp); |
f1b6eb6e | 479 | |
484748f0 | 480 | /* |
9f706d68 | 481 | * Bulk allocation and freeing operations. These are accelerated in an |
484748f0 CL |
482 | * allocator specific way to avoid taking locks repeatedly or building |
483 | * metadata structures unnecessarily. | |
484 | * | |
485 | * Note that interrupts must be enabled when calling these functions. | |
486 | */ | |
72d67229 KC |
487 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p); |
488 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, void **p); | |
484748f0 | 489 | |
ca257195 JDB |
490 | /* |
491 | * Caller must not use kfree_bulk() on memory not originally allocated | |
492 | * by kmalloc(), because the SLOB allocator cannot handle this. | |
493 | */ | |
494 | static __always_inline void kfree_bulk(size_t size, void **p) | |
495 | { | |
496 | kmem_cache_free_bulk(NULL, size, p); | |
497 | } | |
498 | ||
c37495d6 KC |
499 | void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment |
500 | __alloc_size(1); | |
72d67229 KC |
501 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node) __assume_slab_alignment |
502 | __malloc; | |
f1b6eb6e | 503 | |
26a40990 HY |
504 | void *kmalloc_trace(struct kmem_cache *s, gfp_t flags, size_t size) |
505 | __assume_kmalloc_alignment __alloc_size(3); | |
f1b6eb6e | 506 | |
26a40990 HY |
507 | void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags, |
508 | int node, size_t size) __assume_kmalloc_alignment | |
509 | __alloc_size(4); | |
e4c98d68 HY |
510 | void *kmalloc_large(size_t size, gfp_t flags) __assume_page_alignment |
511 | __alloc_size(1); | |
a0c3b940 HY |
512 | |
513 | void *kmalloc_large_node(size_t size, gfp_t flags, int node) __assume_page_alignment | |
514 | __alloc_size(1); | |
515 | ||
f1b6eb6e | 516 | /** |
838de63b | 517 | * kmalloc - allocate kernel memory |
f1b6eb6e | 518 | * @size: how many bytes of memory are required. |
838de63b | 519 | * @flags: describe the allocation context |
f1b6eb6e CL |
520 | * |
521 | * kmalloc is the normal method of allocating memory | |
522 | * for objects smaller than page size in the kernel. | |
7e3528c3 | 523 | * |
59bb4798 VB |
524 | * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN |
525 | * bytes. For @size of power of two bytes, the alignment is also guaranteed | |
526 | * to be at least to the size. | |
527 | * | |
01598ba6 MR |
528 | * The @flags argument may be one of the GFP flags defined at |
529 | * include/linux/gfp.h and described at | |
530 | * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>` | |
7e3528c3 | 531 | * |
01598ba6 | 532 | * The recommended usage of the @flags is described at |
2370ae4b | 533 | * :ref:`Documentation/core-api/memory-allocation.rst <memory_allocation>` |
7e3528c3 | 534 | * |
01598ba6 | 535 | * Below is a brief outline of the most useful GFP flags |
7e3528c3 | 536 | * |
01598ba6 MR |
537 | * %GFP_KERNEL |
538 | * Allocate normal kernel ram. May sleep. | |
7e3528c3 | 539 | * |
01598ba6 MR |
540 | * %GFP_NOWAIT |
541 | * Allocation will not sleep. | |
7e3528c3 | 542 | * |
01598ba6 MR |
543 | * %GFP_ATOMIC |
544 | * Allocation will not sleep. May use emergency pools. | |
7e3528c3 | 545 | * |
7e3528c3 RD |
546 | * Also it is possible to set different flags by OR'ing |
547 | * in one or more of the following additional @flags: | |
548 | * | |
838de63b VB |
549 | * %__GFP_ZERO |
550 | * Zero the allocated memory before returning. Also see kzalloc(). | |
551 | * | |
01598ba6 MR |
552 | * %__GFP_HIGH |
553 | * This allocation has high priority and may use emergency pools. | |
7e3528c3 | 554 | * |
01598ba6 MR |
555 | * %__GFP_NOFAIL |
556 | * Indicate that this allocation is in no way allowed to fail | |
557 | * (think twice before using). | |
7e3528c3 | 558 | * |
01598ba6 MR |
559 | * %__GFP_NORETRY |
560 | * If memory is not immediately available, | |
561 | * then give up at once. | |
7e3528c3 | 562 | * |
01598ba6 MR |
563 | * %__GFP_NOWARN |
564 | * If allocation fails, don't issue any warnings. | |
7e3528c3 | 565 | * |
01598ba6 MR |
566 | * %__GFP_RETRY_MAYFAIL |
567 | * Try really hard to succeed the allocation but fail | |
568 | * eventually. | |
f1b6eb6e | 569 | */ |
3bf01933 | 570 | #ifndef CONFIG_SLOB |
c37495d6 | 571 | static __always_inline __alloc_size(1) void *kmalloc(size_t size, gfp_t flags) |
f1b6eb6e | 572 | { |
6fa57d78 | 573 | if (__builtin_constant_p(size) && size) { |
cc252eae | 574 | unsigned int index; |
3bf01933 | 575 | |
f1b6eb6e CL |
576 | if (size > KMALLOC_MAX_CACHE_SIZE) |
577 | return kmalloc_large(size, flags); | |
f1b6eb6e | 578 | |
cc252eae | 579 | index = kmalloc_index(size); |
26a40990 | 580 | return kmalloc_trace( |
cc252eae VB |
581 | kmalloc_caches[kmalloc_type(flags)][index], |
582 | flags, size); | |
f1b6eb6e CL |
583 | } |
584 | return __kmalloc(size, flags); | |
585 | } | |
3bf01933 KC |
586 | #else |
587 | static __always_inline __alloc_size(1) void *kmalloc(size_t size, gfp_t flags) | |
588 | { | |
589 | if (__builtin_constant_p(size) && size > KMALLOC_MAX_CACHE_SIZE) | |
590 | return kmalloc_large(size, flags); | |
591 | ||
592 | return __kmalloc(size, flags); | |
593 | } | |
594 | #endif | |
f1b6eb6e | 595 | |
bf37d791 | 596 | #ifndef CONFIG_SLOB |
c37495d6 | 597 | static __always_inline __alloc_size(1) void *kmalloc_node(size_t size, gfp_t flags, int node) |
f1b6eb6e | 598 | { |
6fa57d78 | 599 | if (__builtin_constant_p(size) && size) { |
bf37d791 | 600 | unsigned int index; |
f1b6eb6e | 601 | |
bf37d791 HY |
602 | if (size > KMALLOC_MAX_CACHE_SIZE) |
603 | return kmalloc_large_node(size, flags, node); | |
604 | ||
605 | index = kmalloc_index(size); | |
26a40990 | 606 | return kmalloc_node_trace( |
bf37d791 | 607 | kmalloc_caches[kmalloc_type(flags)][index], |
26a40990 | 608 | flags, node, size); |
f1b6eb6e | 609 | } |
f1b6eb6e CL |
610 | return __kmalloc_node(size, flags, node); |
611 | } | |
bf37d791 HY |
612 | #else |
613 | static __always_inline __alloc_size(1) void *kmalloc_node(size_t size, gfp_t flags, int node) | |
614 | { | |
615 | if (__builtin_constant_p(size) && size > KMALLOC_MAX_CACHE_SIZE) | |
616 | return kmalloc_large_node(size, flags, node); | |
617 | ||
618 | return __kmalloc_node(size, flags, node); | |
619 | } | |
620 | #endif | |
f1b6eb6e | 621 | |
e7efa615 MO |
622 | /** |
623 | * kmalloc_array - allocate memory for an array. | |
624 | * @n: number of elements. | |
625 | * @size: element size. | |
626 | * @flags: the type of memory to allocate (see kmalloc). | |
800590f5 | 627 | */ |
c37495d6 | 628 | static inline __alloc_size(1, 2) void *kmalloc_array(size_t n, size_t size, gfp_t flags) |
1da177e4 | 629 | { |
49b7f898 KC |
630 | size_t bytes; |
631 | ||
632 | if (unlikely(check_mul_overflow(n, size, &bytes))) | |
6193a2ff | 633 | return NULL; |
91c6a05f | 634 | if (__builtin_constant_p(n) && __builtin_constant_p(size)) |
49b7f898 KC |
635 | return kmalloc(bytes, flags); |
636 | return __kmalloc(bytes, flags); | |
a8203725 XW |
637 | } |
638 | ||
f0dbd2bd BG |
639 | /** |
640 | * krealloc_array - reallocate memory for an array. | |
641 | * @p: pointer to the memory chunk to reallocate | |
642 | * @new_n: new number of elements to alloc | |
643 | * @new_size: new size of a single member of the array | |
644 | * @flags: the type of memory to allocate (see kmalloc) | |
645 | */ | |
9ed9cac1 KC |
646 | static inline __realloc_size(2, 3) void * __must_check krealloc_array(void *p, |
647 | size_t new_n, | |
648 | size_t new_size, | |
649 | gfp_t flags) | |
f0dbd2bd BG |
650 | { |
651 | size_t bytes; | |
652 | ||
653 | if (unlikely(check_mul_overflow(new_n, new_size, &bytes))) | |
654 | return NULL; | |
655 | ||
656 | return krealloc(p, bytes, flags); | |
657 | } | |
658 | ||
a8203725 XW |
659 | /** |
660 | * kcalloc - allocate memory for an array. The memory is set to zero. | |
661 | * @n: number of elements. | |
662 | * @size: element size. | |
663 | * @flags: the type of memory to allocate (see kmalloc). | |
664 | */ | |
c37495d6 | 665 | static inline __alloc_size(1, 2) void *kcalloc(size_t n, size_t size, gfp_t flags) |
a8203725 XW |
666 | { |
667 | return kmalloc_array(n, size, flags | __GFP_ZERO); | |
1da177e4 LT |
668 | } |
669 | ||
c45248db HY |
670 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, int node, |
671 | unsigned long caller) __alloc_size(1); | |
672 | #define kmalloc_node_track_caller(size, flags, node) \ | |
673 | __kmalloc_node_track_caller(size, flags, node, \ | |
674 | _RET_IP_) | |
675 | ||
1d2c8eea CH |
676 | /* |
677 | * kmalloc_track_caller is a special version of kmalloc that records the | |
678 | * calling function of the routine calling it for slab leak tracking instead | |
679 | * of just the calling function (confusing, eh?). | |
680 | * It's useful when the call to kmalloc comes from a widely-used standard | |
681 | * allocator where we care about the real place the memory allocation | |
682 | * request comes from. | |
683 | */ | |
1d2c8eea | 684 | #define kmalloc_track_caller(size, flags) \ |
c45248db HY |
685 | __kmalloc_node_track_caller(size, flags, \ |
686 | NUMA_NO_NODE, _RET_IP_) | |
1da177e4 | 687 | |
c37495d6 KC |
688 | static inline __alloc_size(1, 2) void *kmalloc_array_node(size_t n, size_t size, gfp_t flags, |
689 | int node) | |
5799b255 | 690 | { |
49b7f898 KC |
691 | size_t bytes; |
692 | ||
693 | if (unlikely(check_mul_overflow(n, size, &bytes))) | |
5799b255 JT |
694 | return NULL; |
695 | if (__builtin_constant_p(n) && __builtin_constant_p(size)) | |
49b7f898 KC |
696 | return kmalloc_node(bytes, flags, node); |
697 | return __kmalloc_node(bytes, flags, node); | |
5799b255 JT |
698 | } |
699 | ||
c37495d6 | 700 | static inline __alloc_size(1, 2) void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node) |
5799b255 JT |
701 | { |
702 | return kmalloc_array_node(n, size, flags | __GFP_ZERO, node); | |
703 | } | |
704 | ||
81cda662 CL |
705 | /* |
706 | * Shortcuts | |
707 | */ | |
708 | static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags) | |
709 | { | |
710 | return kmem_cache_alloc(k, flags | __GFP_ZERO); | |
711 | } | |
712 | ||
713 | /** | |
714 | * kzalloc - allocate memory. The memory is set to zero. | |
715 | * @size: how many bytes of memory are required. | |
716 | * @flags: the type of memory to allocate (see kmalloc). | |
717 | */ | |
c37495d6 | 718 | static inline __alloc_size(1) void *kzalloc(size_t size, gfp_t flags) |
81cda662 CL |
719 | { |
720 | return kmalloc(size, flags | __GFP_ZERO); | |
721 | } | |
722 | ||
979b0fea JL |
723 | /** |
724 | * kzalloc_node - allocate zeroed memory from a particular memory node. | |
725 | * @size: how many bytes of memory are required. | |
726 | * @flags: the type of memory to allocate (see kmalloc). | |
727 | * @node: memory node from which to allocate | |
728 | */ | |
c37495d6 | 729 | static inline __alloc_size(1) void *kzalloc_node(size_t size, gfp_t flags, int node) |
979b0fea JL |
730 | { |
731 | return kmalloc_node(size, flags | __GFP_ZERO, node); | |
732 | } | |
733 | ||
56bcf40f KC |
734 | extern void *kvmalloc_node(size_t size, gfp_t flags, int node) __alloc_size(1); |
735 | static inline __alloc_size(1) void *kvmalloc(size_t size, gfp_t flags) | |
8587ca6f MWO |
736 | { |
737 | return kvmalloc_node(size, flags, NUMA_NO_NODE); | |
738 | } | |
56bcf40f | 739 | static inline __alloc_size(1) void *kvzalloc_node(size_t size, gfp_t flags, int node) |
8587ca6f MWO |
740 | { |
741 | return kvmalloc_node(size, flags | __GFP_ZERO, node); | |
742 | } | |
56bcf40f | 743 | static inline __alloc_size(1) void *kvzalloc(size_t size, gfp_t flags) |
8587ca6f MWO |
744 | { |
745 | return kvmalloc(size, flags | __GFP_ZERO); | |
746 | } | |
747 | ||
56bcf40f | 748 | static inline __alloc_size(1, 2) void *kvmalloc_array(size_t n, size_t size, gfp_t flags) |
8587ca6f MWO |
749 | { |
750 | size_t bytes; | |
751 | ||
752 | if (unlikely(check_mul_overflow(n, size, &bytes))) | |
753 | return NULL; | |
754 | ||
755 | return kvmalloc(bytes, flags); | |
756 | } | |
757 | ||
56bcf40f | 758 | static inline __alloc_size(1, 2) void *kvcalloc(size_t n, size_t size, gfp_t flags) |
8587ca6f MWO |
759 | { |
760 | return kvmalloc_array(n, size, flags | __GFP_ZERO); | |
761 | } | |
762 | ||
56bcf40f | 763 | extern void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags) |
9ed9cac1 | 764 | __realloc_size(3); |
8587ca6f MWO |
765 | extern void kvfree(const void *addr); |
766 | extern void kvfree_sensitive(const void *addr, size_t len); | |
767 | ||
07f361b2 | 768 | unsigned int kmem_cache_size(struct kmem_cache *s); |
05a94065 KC |
769 | |
770 | /** | |
771 | * kmalloc_size_roundup - Report allocation bucket size for the given size | |
772 | * | |
773 | * @size: Number of bytes to round up from. | |
774 | * | |
775 | * This returns the number of bytes that would be available in a kmalloc() | |
776 | * allocation of @size bytes. For example, a 126 byte request would be | |
777 | * rounded up to the next sized kmalloc bucket, 128 bytes. (This is strictly | |
778 | * for the general-purpose kmalloc()-based allocations, and is not for the | |
779 | * pre-sized kmem_cache_alloc()-based allocations.) | |
780 | * | |
781 | * Use this to kmalloc() the full bucket size ahead of time instead of using | |
782 | * ksize() to query the size after an allocation. | |
783 | */ | |
784 | size_t kmalloc_size_roundup(size_t size); | |
785 | ||
7e85ee0c PE |
786 | void __init kmem_cache_init_late(void); |
787 | ||
6731d4f1 SAS |
788 | #if defined(CONFIG_SMP) && defined(CONFIG_SLAB) |
789 | int slab_prepare_cpu(unsigned int cpu); | |
790 | int slab_dead_cpu(unsigned int cpu); | |
791 | #else | |
792 | #define slab_prepare_cpu NULL | |
793 | #define slab_dead_cpu NULL | |
794 | #endif | |
795 | ||
1da177e4 | 796 | #endif /* _LINUX_SLAB_H */ |