Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 1 | /* |
| 2 | * Slab allocator functions that are independent of the allocator strategy |
| 3 | * |
| 4 | * (C) 2012 Christoph Lameter <cl@linux.com> |
| 5 | */ |
| 6 | #include <linux/slab.h> |
| 7 | |
| 8 | #include <linux/mm.h> |
| 9 | #include <linux/poison.h> |
| 10 | #include <linux/interrupt.h> |
| 11 | #include <linux/memory.h> |
| 12 | #include <linux/compiler.h> |
| 13 | #include <linux/module.h> |
Christoph Lameter | 20cea96 | 2012-07-06 15:25:13 -0500 | [diff] [blame] | 14 | #include <linux/cpu.h> |
| 15 | #include <linux/uaccess.h> |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 16 | #include <linux/seq_file.h> |
| 17 | #include <linux/proc_fs.h> |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 18 | #include <asm/cacheflush.h> |
| 19 | #include <asm/tlbflush.h> |
| 20 | #include <asm/page.h> |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 21 | #include <linux/memcontrol.h> |
Andrey Ryabinin | 928cec9 | 2014-08-06 16:04:44 -0700 | [diff] [blame] | 22 | |
| 23 | #define CREATE_TRACE_POINTS |
Christoph Lameter | f1b6eb6 | 2013-09-04 16:35:34 +0000 | [diff] [blame] | 24 | #include <trace/events/kmem.h> |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 25 | |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 26 | #include "slab.h" |
| 27 | |
| 28 | enum slab_state slab_state; |
Christoph Lameter | 18004c5 | 2012-07-06 15:25:12 -0500 | [diff] [blame] | 29 | LIST_HEAD(slab_caches); |
| 30 | DEFINE_MUTEX(slab_mutex); |
Christoph Lameter | 9b030cb | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 31 | struct kmem_cache *kmem_cache; |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 32 | |
Joonsoo Kim | 07f361b | 2014-10-09 15:26:00 -0700 | [diff] [blame] | 33 | /* |
Joonsoo Kim | 423c929 | 2014-10-09 15:26:22 -0700 | [diff] [blame] | 34 | * Set of flags that will prevent slab merging |
| 35 | */ |
| 36 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
| 37 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
| 38 | SLAB_FAILSLAB) |
| 39 | |
| 40 | #define SLAB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ |
| 41 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
| 42 | |
| 43 | /* |
| 44 | * Merge control. If this is set then no merging of slab caches will occur. |
| 45 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
| 46 | */ |
| 47 | static int slab_nomerge; |
| 48 | |
| 49 | static int __init setup_slab_nomerge(char *str) |
| 50 | { |
| 51 | slab_nomerge = 1; |
| 52 | return 1; |
| 53 | } |
| 54 | |
| 55 | #ifdef CONFIG_SLUB |
| 56 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); |
| 57 | #endif |
| 58 | |
| 59 | __setup("slab_nomerge", setup_slab_nomerge); |
| 60 | |
| 61 | /* |
Joonsoo Kim | 07f361b | 2014-10-09 15:26:00 -0700 | [diff] [blame] | 62 | * Determine the size of a slab object |
| 63 | */ |
| 64 | unsigned int kmem_cache_size(struct kmem_cache *s) |
| 65 | { |
| 66 | return s->object_size; |
| 67 | } |
| 68 | EXPORT_SYMBOL(kmem_cache_size); |
| 69 | |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 70 | #ifdef CONFIG_DEBUG_VM |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 71 | static int kmem_cache_sanity_check(const char *name, size_t size) |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 72 | { |
| 73 | struct kmem_cache *s = NULL; |
| 74 | |
| 75 | if (!name || in_interrupt() || size < sizeof(void *) || |
| 76 | size > KMALLOC_MAX_SIZE) { |
| 77 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
| 78 | return -EINVAL; |
| 79 | } |
| 80 | |
| 81 | list_for_each_entry(s, &slab_caches, list) { |
| 82 | char tmp; |
| 83 | int res; |
| 84 | |
| 85 | /* |
| 86 | * This happens when the module gets unloaded and doesn't |
| 87 | * destroy its slab cache and no-one else reuses the vmalloc |
| 88 | * area of the module. Print a warning. |
| 89 | */ |
| 90 | res = probe_kernel_address(s->name, tmp); |
| 91 | if (res) { |
| 92 | pr_err("Slab cache with size %d has lost its name\n", |
| 93 | s->object_size); |
| 94 | continue; |
| 95 | } |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 96 | } |
| 97 | |
| 98 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
| 99 | return 0; |
| 100 | } |
| 101 | #else |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 102 | static inline int kmem_cache_sanity_check(const char *name, size_t size) |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 103 | { |
| 104 | return 0; |
| 105 | } |
| 106 | #endif |
| 107 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 108 | #ifdef CONFIG_MEMCG_KMEM |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 109 | static int memcg_alloc_cache_params(struct mem_cgroup *memcg, |
| 110 | struct kmem_cache *s, struct kmem_cache *root_cache) |
| 111 | { |
| 112 | size_t size; |
| 113 | |
| 114 | if (!memcg_kmem_enabled()) |
| 115 | return 0; |
| 116 | |
| 117 | if (!memcg) { |
| 118 | size = offsetof(struct memcg_cache_params, memcg_caches); |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame^] | 119 | size += memcg_nr_cache_ids * sizeof(void *); |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 120 | } else |
| 121 | size = sizeof(struct memcg_cache_params); |
| 122 | |
| 123 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
| 124 | if (!s->memcg_params) |
| 125 | return -ENOMEM; |
| 126 | |
| 127 | if (memcg) { |
| 128 | s->memcg_params->memcg = memcg; |
| 129 | s->memcg_params->root_cache = root_cache; |
| 130 | } else |
| 131 | s->memcg_params->is_root_cache = true; |
| 132 | |
| 133 | return 0; |
| 134 | } |
| 135 | |
| 136 | static void memcg_free_cache_params(struct kmem_cache *s) |
| 137 | { |
| 138 | kfree(s->memcg_params); |
| 139 | } |
| 140 | |
Vladimir Davydov | 6f817f4 | 2014-10-09 15:28:47 -0700 | [diff] [blame] | 141 | static int memcg_update_cache_params(struct kmem_cache *s, int num_memcgs) |
| 142 | { |
| 143 | int size; |
| 144 | struct memcg_cache_params *new_params, *cur_params; |
| 145 | |
| 146 | BUG_ON(!is_root_cache(s)); |
| 147 | |
| 148 | size = offsetof(struct memcg_cache_params, memcg_caches); |
| 149 | size += num_memcgs * sizeof(void *); |
| 150 | |
| 151 | new_params = kzalloc(size, GFP_KERNEL); |
| 152 | if (!new_params) |
| 153 | return -ENOMEM; |
| 154 | |
| 155 | cur_params = s->memcg_params; |
| 156 | memcpy(new_params->memcg_caches, cur_params->memcg_caches, |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame^] | 157 | memcg_nr_cache_ids * sizeof(void *)); |
Vladimir Davydov | 6f817f4 | 2014-10-09 15:28:47 -0700 | [diff] [blame] | 158 | |
| 159 | new_params->is_root_cache = true; |
| 160 | |
| 161 | rcu_assign_pointer(s->memcg_params, new_params); |
| 162 | if (cur_params) |
| 163 | kfree_rcu(cur_params, rcu_head); |
| 164 | |
| 165 | return 0; |
| 166 | } |
| 167 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 168 | int memcg_update_all_caches(int num_memcgs) |
| 169 | { |
| 170 | struct kmem_cache *s; |
| 171 | int ret = 0; |
| 172 | mutex_lock(&slab_mutex); |
| 173 | |
| 174 | list_for_each_entry(s, &slab_caches, list) { |
| 175 | if (!is_root_cache(s)) |
| 176 | continue; |
| 177 | |
Vladimir Davydov | 6f817f4 | 2014-10-09 15:28:47 -0700 | [diff] [blame] | 178 | ret = memcg_update_cache_params(s, num_memcgs); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 179 | /* |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 180 | * Instead of freeing the memory, we'll just leave the caches |
| 181 | * up to this point in an updated state. |
| 182 | */ |
| 183 | if (ret) |
| 184 | goto out; |
| 185 | } |
| 186 | |
| 187 | memcg_update_array_size(num_memcgs); |
| 188 | out: |
| 189 | mutex_unlock(&slab_mutex); |
| 190 | return ret; |
| 191 | } |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 192 | #else |
| 193 | static inline int memcg_alloc_cache_params(struct mem_cgroup *memcg, |
| 194 | struct kmem_cache *s, struct kmem_cache *root_cache) |
| 195 | { |
| 196 | return 0; |
| 197 | } |
| 198 | |
| 199 | static inline void memcg_free_cache_params(struct kmem_cache *s) |
| 200 | { |
| 201 | } |
| 202 | #endif /* CONFIG_MEMCG_KMEM */ |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 203 | |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 204 | /* |
Joonsoo Kim | 423c929 | 2014-10-09 15:26:22 -0700 | [diff] [blame] | 205 | * Find a mergeable slab cache |
| 206 | */ |
| 207 | int slab_unmergeable(struct kmem_cache *s) |
| 208 | { |
| 209 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) |
| 210 | return 1; |
| 211 | |
| 212 | if (!is_root_cache(s)) |
| 213 | return 1; |
| 214 | |
| 215 | if (s->ctor) |
| 216 | return 1; |
| 217 | |
| 218 | /* |
| 219 | * We may have set a slab to be unmergeable during bootstrap. |
| 220 | */ |
| 221 | if (s->refcount < 0) |
| 222 | return 1; |
| 223 | |
| 224 | return 0; |
| 225 | } |
| 226 | |
| 227 | struct kmem_cache *find_mergeable(size_t size, size_t align, |
| 228 | unsigned long flags, const char *name, void (*ctor)(void *)) |
| 229 | { |
| 230 | struct kmem_cache *s; |
| 231 | |
| 232 | if (slab_nomerge || (flags & SLAB_NEVER_MERGE)) |
| 233 | return NULL; |
| 234 | |
| 235 | if (ctor) |
| 236 | return NULL; |
| 237 | |
| 238 | size = ALIGN(size, sizeof(void *)); |
| 239 | align = calculate_alignment(flags, align, size); |
| 240 | size = ALIGN(size, align); |
| 241 | flags = kmem_cache_flags(size, flags, name, NULL); |
| 242 | |
Joonsoo Kim | 5436205 | 2014-12-10 15:42:18 -0800 | [diff] [blame] | 243 | list_for_each_entry_reverse(s, &slab_caches, list) { |
Joonsoo Kim | 423c929 | 2014-10-09 15:26:22 -0700 | [diff] [blame] | 244 | if (slab_unmergeable(s)) |
| 245 | continue; |
| 246 | |
| 247 | if (size > s->size) |
| 248 | continue; |
| 249 | |
| 250 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) |
| 251 | continue; |
| 252 | /* |
| 253 | * Check if alignment is compatible. |
| 254 | * Courtesy of Adrian Drzewiecki |
| 255 | */ |
| 256 | if ((s->size & ~(align - 1)) != s->size) |
| 257 | continue; |
| 258 | |
| 259 | if (s->size - size >= sizeof(void *)) |
| 260 | continue; |
| 261 | |
Joonsoo Kim | 95069ac8 | 2014-11-13 15:19:25 -0800 | [diff] [blame] | 262 | if (IS_ENABLED(CONFIG_SLAB) && align && |
| 263 | (align > s->align || s->align % align)) |
| 264 | continue; |
| 265 | |
Joonsoo Kim | 423c929 | 2014-10-09 15:26:22 -0700 | [diff] [blame] | 266 | return s; |
| 267 | } |
| 268 | return NULL; |
| 269 | } |
| 270 | |
| 271 | /* |
Christoph Lameter | 4590685 | 2012-11-28 16:23:16 +0000 | [diff] [blame] | 272 | * Figure out what the alignment of the objects will be given a set of |
| 273 | * flags, a user specified alignment and the size of the objects. |
| 274 | */ |
| 275 | unsigned long calculate_alignment(unsigned long flags, |
| 276 | unsigned long align, unsigned long size) |
| 277 | { |
| 278 | /* |
| 279 | * If the user wants hardware cache aligned objects then follow that |
| 280 | * suggestion if the object is sufficiently large. |
| 281 | * |
| 282 | * The hardware cache alignment cannot override the specified |
| 283 | * alignment though. If that is greater then use it. |
| 284 | */ |
| 285 | if (flags & SLAB_HWCACHE_ALIGN) { |
| 286 | unsigned long ralign = cache_line_size(); |
| 287 | while (size <= ralign / 2) |
| 288 | ralign /= 2; |
| 289 | align = max(align, ralign); |
| 290 | } |
| 291 | |
| 292 | if (align < ARCH_SLAB_MINALIGN) |
| 293 | align = ARCH_SLAB_MINALIGN; |
| 294 | |
| 295 | return ALIGN(align, sizeof(void *)); |
| 296 | } |
| 297 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 298 | static struct kmem_cache * |
| 299 | do_kmem_cache_create(char *name, size_t object_size, size_t size, size_t align, |
| 300 | unsigned long flags, void (*ctor)(void *), |
| 301 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) |
| 302 | { |
| 303 | struct kmem_cache *s; |
| 304 | int err; |
| 305 | |
| 306 | err = -ENOMEM; |
| 307 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); |
| 308 | if (!s) |
| 309 | goto out; |
| 310 | |
| 311 | s->name = name; |
| 312 | s->object_size = object_size; |
| 313 | s->size = size; |
| 314 | s->align = align; |
| 315 | s->ctor = ctor; |
| 316 | |
| 317 | err = memcg_alloc_cache_params(memcg, s, root_cache); |
| 318 | if (err) |
| 319 | goto out_free_cache; |
| 320 | |
| 321 | err = __kmem_cache_create(s, flags); |
| 322 | if (err) |
| 323 | goto out_free_cache; |
| 324 | |
| 325 | s->refcount = 1; |
| 326 | list_add(&s->list, &slab_caches); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 327 | out: |
| 328 | if (err) |
| 329 | return ERR_PTR(err); |
| 330 | return s; |
| 331 | |
| 332 | out_free_cache: |
| 333 | memcg_free_cache_params(s); |
Vaishali Thakkar | 7c4da06 | 2015-02-10 14:09:40 -0800 | [diff] [blame] | 334 | kmem_cache_free(kmem_cache, s); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 335 | goto out; |
| 336 | } |
Christoph Lameter | 4590685 | 2012-11-28 16:23:16 +0000 | [diff] [blame] | 337 | |
| 338 | /* |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 339 | * kmem_cache_create - Create a cache. |
| 340 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
| 341 | * @size: The size of objects to be created in this cache. |
| 342 | * @align: The required alignment for the objects. |
| 343 | * @flags: SLAB flags |
| 344 | * @ctor: A constructor for the objects. |
| 345 | * |
| 346 | * Returns a ptr to the cache on success, NULL on failure. |
| 347 | * Cannot be called within a interrupt, but can be interrupted. |
| 348 | * The @ctor is run when new pages are allocated by the cache. |
| 349 | * |
| 350 | * The flags are |
| 351 | * |
| 352 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) |
| 353 | * to catch references to uninitialised memory. |
| 354 | * |
| 355 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check |
| 356 | * for buffer overruns. |
| 357 | * |
| 358 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
| 359 | * cacheline. This can be beneficial if you're counting cycles as closely |
| 360 | * as davem. |
| 361 | */ |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 362 | struct kmem_cache * |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 363 | kmem_cache_create(const char *name, size_t size, size_t align, |
| 364 | unsigned long flags, void (*ctor)(void *)) |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 365 | { |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 366 | struct kmem_cache *s; |
| 367 | char *cache_name; |
Vladimir Davydov | 3965fc3 | 2014-01-23 15:52:55 -0800 | [diff] [blame] | 368 | int err; |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 369 | |
Pekka Enberg | b920536 | 2012-08-16 10:12:18 +0300 | [diff] [blame] | 370 | get_online_cpus(); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 371 | get_online_mems(); |
| 372 | |
Pekka Enberg | b920536 | 2012-08-16 10:12:18 +0300 | [diff] [blame] | 373 | mutex_lock(&slab_mutex); |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 374 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 375 | err = kmem_cache_sanity_check(name, size); |
Andrew Morton | 3aa24f5 | 2014-10-09 15:25:58 -0700 | [diff] [blame] | 376 | if (err) { |
| 377 | s = NULL; /* suppress uninit var warning */ |
Vladimir Davydov | 3965fc3 | 2014-01-23 15:52:55 -0800 | [diff] [blame] | 378 | goto out_unlock; |
Andrew Morton | 3aa24f5 | 2014-10-09 15:25:58 -0700 | [diff] [blame] | 379 | } |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 380 | |
Glauber Costa | d884392 | 2012-10-17 15:36:51 +0400 | [diff] [blame] | 381 | /* |
| 382 | * Some allocators will constraint the set of valid flags to a subset |
| 383 | * of all flags. We expect them to define CACHE_CREATE_MASK in this |
| 384 | * case, and we'll just provide them with a sanitized version of the |
| 385 | * passed flags. |
| 386 | */ |
| 387 | flags &= CACHE_CREATE_MASK; |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 388 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 389 | s = __kmem_cache_alias(name, size, align, flags, ctor); |
| 390 | if (s) |
Vladimir Davydov | 3965fc3 | 2014-01-23 15:52:55 -0800 | [diff] [blame] | 391 | goto out_unlock; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 392 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 393 | cache_name = kstrdup(name, GFP_KERNEL); |
| 394 | if (!cache_name) { |
| 395 | err = -ENOMEM; |
| 396 | goto out_unlock; |
| 397 | } |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 398 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 399 | s = do_kmem_cache_create(cache_name, size, size, |
| 400 | calculate_alignment(flags, align, size), |
| 401 | flags, ctor, NULL, NULL); |
| 402 | if (IS_ERR(s)) { |
| 403 | err = PTR_ERR(s); |
| 404 | kfree(cache_name); |
| 405 | } |
Vladimir Davydov | 3965fc3 | 2014-01-23 15:52:55 -0800 | [diff] [blame] | 406 | |
| 407 | out_unlock: |
Christoph Lameter | 20cea96 | 2012-07-06 15:25:13 -0500 | [diff] [blame] | 408 | mutex_unlock(&slab_mutex); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 409 | |
| 410 | put_online_mems(); |
Christoph Lameter | 20cea96 | 2012-07-06 15:25:13 -0500 | [diff] [blame] | 411 | put_online_cpus(); |
| 412 | |
Dave Jones | ba3253c7 | 2014-01-29 14:05:48 -0800 | [diff] [blame] | 413 | if (err) { |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 414 | if (flags & SLAB_PANIC) |
| 415 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", |
| 416 | name, err); |
| 417 | else { |
| 418 | printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d", |
| 419 | name, err); |
| 420 | dump_stack(); |
| 421 | } |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 422 | return NULL; |
| 423 | } |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 424 | return s; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 425 | } |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 426 | EXPORT_SYMBOL(kmem_cache_create); |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 427 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 428 | static int do_kmem_cache_shutdown(struct kmem_cache *s, |
| 429 | struct list_head *release, bool *need_rcu_barrier) |
| 430 | { |
| 431 | if (__kmem_cache_shutdown(s) != 0) { |
| 432 | printk(KERN_ERR "kmem_cache_destroy %s: " |
| 433 | "Slab cache still has objects\n", s->name); |
| 434 | dump_stack(); |
| 435 | return -EBUSY; |
| 436 | } |
| 437 | |
| 438 | if (s->flags & SLAB_DESTROY_BY_RCU) |
| 439 | *need_rcu_barrier = true; |
| 440 | |
| 441 | #ifdef CONFIG_MEMCG_KMEM |
| 442 | if (!is_root_cache(s)) { |
| 443 | struct kmem_cache *root_cache = s->memcg_params->root_cache; |
| 444 | int memcg_id = memcg_cache_id(s->memcg_params->memcg); |
| 445 | |
| 446 | BUG_ON(root_cache->memcg_params->memcg_caches[memcg_id] != s); |
| 447 | root_cache->memcg_params->memcg_caches[memcg_id] = NULL; |
| 448 | } |
| 449 | #endif |
| 450 | list_move(&s->list, release); |
| 451 | return 0; |
| 452 | } |
| 453 | |
| 454 | static void do_kmem_cache_release(struct list_head *release, |
| 455 | bool need_rcu_barrier) |
| 456 | { |
| 457 | struct kmem_cache *s, *s2; |
| 458 | |
| 459 | if (need_rcu_barrier) |
| 460 | rcu_barrier(); |
| 461 | |
| 462 | list_for_each_entry_safe(s, s2, release, list) { |
| 463 | #ifdef SLAB_SUPPORTS_SYSFS |
| 464 | sysfs_slab_remove(s); |
| 465 | #else |
| 466 | slab_kmem_cache_release(s); |
| 467 | #endif |
| 468 | } |
| 469 | } |
| 470 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 471 | #ifdef CONFIG_MEMCG_KMEM |
| 472 | /* |
Vladimir Davydov | 776ed0f | 2014-06-04 16:10:02 -0700 | [diff] [blame] | 473 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 474 | * @memcg: The memory cgroup the new cache is for. |
| 475 | * @root_cache: The parent of the new cache. |
| 476 | * |
| 477 | * This function attempts to create a kmem cache that will serve allocation |
| 478 | * requests going from @memcg to @root_cache. The new cache inherits properties |
| 479 | * from its parent. |
| 480 | */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 481 | void memcg_create_kmem_cache(struct mem_cgroup *memcg, |
| 482 | struct kmem_cache *root_cache) |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 483 | { |
Vladimir Davydov | 3e0350a | 2015-02-10 14:11:44 -0800 | [diff] [blame] | 484 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 485 | int memcg_id = memcg_cache_id(memcg); |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 486 | struct kmem_cache *s = NULL; |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 487 | char *cache_name; |
| 488 | |
| 489 | get_online_cpus(); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 490 | get_online_mems(); |
| 491 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 492 | mutex_lock(&slab_mutex); |
| 493 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 494 | /* |
| 495 | * Since per-memcg caches are created asynchronously on first |
| 496 | * allocation (see memcg_kmem_get_cache()), several threads can try to |
| 497 | * create the same cache, but only one of them may succeed. |
| 498 | */ |
| 499 | if (cache_from_memcg_idx(root_cache, memcg_id)) |
| 500 | goto out_unlock; |
| 501 | |
Vladimir Davydov | 3e0350a | 2015-02-10 14:11:44 -0800 | [diff] [blame] | 502 | cgroup_name(mem_cgroup_css(memcg)->cgroup, |
| 503 | memcg_name_buf, sizeof(memcg_name_buf)); |
Vladimir Davydov | 073ee1c | 2014-06-04 16:08:23 -0700 | [diff] [blame] | 504 | cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name, |
Vladimir Davydov | 3e0350a | 2015-02-10 14:11:44 -0800 | [diff] [blame] | 505 | memcg_cache_id(memcg), memcg_name_buf); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 506 | if (!cache_name) |
| 507 | goto out_unlock; |
| 508 | |
| 509 | s = do_kmem_cache_create(cache_name, root_cache->object_size, |
| 510 | root_cache->size, root_cache->align, |
| 511 | root_cache->flags, root_cache->ctor, |
| 512 | memcg, root_cache); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 513 | /* |
| 514 | * If we could not create a memcg cache, do not complain, because |
| 515 | * that's not critical at all as we can always proceed with the root |
| 516 | * cache. |
| 517 | */ |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 518 | if (IS_ERR(s)) { |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 519 | kfree(cache_name); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 520 | goto out_unlock; |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 521 | } |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 522 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 523 | /* |
| 524 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
| 525 | * barrier here to ensure nobody will see the kmem_cache partially |
| 526 | * initialized. |
| 527 | */ |
| 528 | smp_wmb(); |
| 529 | root_cache->memcg_params->memcg_caches[memcg_id] = s; |
| 530 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 531 | out_unlock: |
| 532 | mutex_unlock(&slab_mutex); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 533 | |
| 534 | put_online_mems(); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 535 | put_online_cpus(); |
| 536 | } |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 537 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 538 | void memcg_destroy_kmem_caches(struct mem_cgroup *memcg) |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 539 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 540 | LIST_HEAD(release); |
| 541 | bool need_rcu_barrier = false; |
| 542 | struct kmem_cache *s, *s2; |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 543 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 544 | get_online_cpus(); |
| 545 | get_online_mems(); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 546 | |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 547 | mutex_lock(&slab_mutex); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 548 | list_for_each_entry_safe(s, s2, &slab_caches, list) { |
| 549 | if (is_root_cache(s) || s->memcg_params->memcg != memcg) |
| 550 | continue; |
| 551 | /* |
| 552 | * The cgroup is about to be freed and therefore has no charges |
| 553 | * left. Hence, all its caches must be empty by now. |
| 554 | */ |
| 555 | BUG_ON(do_kmem_cache_shutdown(s, &release, &need_rcu_barrier)); |
| 556 | } |
| 557 | mutex_unlock(&slab_mutex); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 558 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 559 | put_online_mems(); |
| 560 | put_online_cpus(); |
| 561 | |
| 562 | do_kmem_cache_release(&release, need_rcu_barrier); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 563 | } |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 564 | #endif /* CONFIG_MEMCG_KMEM */ |
| 565 | |
Christoph Lameter | 41a2128 | 2014-05-06 12:50:08 -0700 | [diff] [blame] | 566 | void slab_kmem_cache_release(struct kmem_cache *s) |
| 567 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 568 | memcg_free_cache_params(s); |
Christoph Lameter | 41a2128 | 2014-05-06 12:50:08 -0700 | [diff] [blame] | 569 | kfree(s->name); |
| 570 | kmem_cache_free(kmem_cache, s); |
| 571 | } |
| 572 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 573 | void kmem_cache_destroy(struct kmem_cache *s) |
| 574 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 575 | int i; |
| 576 | LIST_HEAD(release); |
| 577 | bool need_rcu_barrier = false; |
| 578 | bool busy = false; |
| 579 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 580 | get_online_cpus(); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 581 | get_online_mems(); |
| 582 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 583 | mutex_lock(&slab_mutex); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 584 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 585 | s->refcount--; |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 586 | if (s->refcount) |
| 587 | goto out_unlock; |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 588 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 589 | for_each_memcg_cache_index(i) { |
| 590 | struct kmem_cache *c = cache_from_memcg_idx(s, i); |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 591 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 592 | if (c && do_kmem_cache_shutdown(c, &release, &need_rcu_barrier)) |
| 593 | busy = true; |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 594 | } |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 595 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 596 | if (!busy) |
| 597 | do_kmem_cache_shutdown(s, &release, &need_rcu_barrier); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 598 | |
| 599 | out_unlock: |
| 600 | mutex_unlock(&slab_mutex); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 601 | |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 602 | put_online_mems(); |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 603 | put_online_cpus(); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 604 | |
| 605 | do_kmem_cache_release(&release, need_rcu_barrier); |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 606 | } |
| 607 | EXPORT_SYMBOL(kmem_cache_destroy); |
| 608 | |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 609 | /** |
| 610 | * kmem_cache_shrink - Shrink a cache. |
| 611 | * @cachep: The cache to shrink. |
| 612 | * |
| 613 | * Releases as many slabs as possible for a cache. |
| 614 | * To help debugging, a zero exit status indicates all slabs were released. |
| 615 | */ |
| 616 | int kmem_cache_shrink(struct kmem_cache *cachep) |
| 617 | { |
| 618 | int ret; |
| 619 | |
| 620 | get_online_cpus(); |
| 621 | get_online_mems(); |
| 622 | ret = __kmem_cache_shrink(cachep); |
| 623 | put_online_mems(); |
| 624 | put_online_cpus(); |
| 625 | return ret; |
| 626 | } |
| 627 | EXPORT_SYMBOL(kmem_cache_shrink); |
| 628 | |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 629 | int slab_is_available(void) |
| 630 | { |
| 631 | return slab_state >= UP; |
| 632 | } |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 633 | |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 634 | #ifndef CONFIG_SLOB |
| 635 | /* Create a cache during boot when no slab services are available yet */ |
| 636 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, |
| 637 | unsigned long flags) |
| 638 | { |
| 639 | int err; |
| 640 | |
| 641 | s->name = name; |
| 642 | s->size = s->object_size = size; |
Christoph Lameter | 4590685 | 2012-11-28 16:23:16 +0000 | [diff] [blame] | 643 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 644 | err = __kmem_cache_create(s, flags); |
| 645 | |
| 646 | if (err) |
Christoph Lameter | 31ba734 | 2013-01-10 19:00:53 +0000 | [diff] [blame] | 647 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 648 | name, size, err); |
| 649 | |
| 650 | s->refcount = -1; /* Exempt from merging for now */ |
| 651 | } |
| 652 | |
| 653 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, |
| 654 | unsigned long flags) |
| 655 | { |
| 656 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
| 657 | |
| 658 | if (!s) |
| 659 | panic("Out of memory when creating slab %s\n", name); |
| 660 | |
| 661 | create_boot_cache(s, name, size, flags); |
| 662 | list_add(&s->list, &slab_caches); |
| 663 | s->refcount = 1; |
| 664 | return s; |
| 665 | } |
| 666 | |
Christoph Lameter | 9425c58 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 667 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; |
| 668 | EXPORT_SYMBOL(kmalloc_caches); |
| 669 | |
| 670 | #ifdef CONFIG_ZONE_DMA |
| 671 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; |
| 672 | EXPORT_SYMBOL(kmalloc_dma_caches); |
| 673 | #endif |
| 674 | |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 675 | /* |
Christoph Lameter | 2c59dd6 | 2013-01-10 19:14:19 +0000 | [diff] [blame] | 676 | * Conversion table for small slabs sizes / 8 to the index in the |
| 677 | * kmalloc array. This is necessary for slabs < 192 since we have non power |
| 678 | * of two cache sizes there. The size of larger slabs can be determined using |
| 679 | * fls. |
| 680 | */ |
| 681 | static s8 size_index[24] = { |
| 682 | 3, /* 8 */ |
| 683 | 4, /* 16 */ |
| 684 | 5, /* 24 */ |
| 685 | 5, /* 32 */ |
| 686 | 6, /* 40 */ |
| 687 | 6, /* 48 */ |
| 688 | 6, /* 56 */ |
| 689 | 6, /* 64 */ |
| 690 | 1, /* 72 */ |
| 691 | 1, /* 80 */ |
| 692 | 1, /* 88 */ |
| 693 | 1, /* 96 */ |
| 694 | 7, /* 104 */ |
| 695 | 7, /* 112 */ |
| 696 | 7, /* 120 */ |
| 697 | 7, /* 128 */ |
| 698 | 2, /* 136 */ |
| 699 | 2, /* 144 */ |
| 700 | 2, /* 152 */ |
| 701 | 2, /* 160 */ |
| 702 | 2, /* 168 */ |
| 703 | 2, /* 176 */ |
| 704 | 2, /* 184 */ |
| 705 | 2 /* 192 */ |
| 706 | }; |
| 707 | |
| 708 | static inline int size_index_elem(size_t bytes) |
| 709 | { |
| 710 | return (bytes - 1) / 8; |
| 711 | } |
| 712 | |
| 713 | /* |
| 714 | * Find the kmem_cache structure that serves a given size of |
| 715 | * allocation |
| 716 | */ |
| 717 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) |
| 718 | { |
| 719 | int index; |
| 720 | |
Joonsoo Kim | 9de1bc8 | 2013-08-02 11:02:42 +0900 | [diff] [blame] | 721 | if (unlikely(size > KMALLOC_MAX_SIZE)) { |
Sasha Levin | 907985f | 2013-06-10 15:18:00 -0400 | [diff] [blame] | 722 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); |
Christoph Lameter | 6286ae9 | 2013-05-03 15:43:18 +0000 | [diff] [blame] | 723 | return NULL; |
Sasha Levin | 907985f | 2013-06-10 15:18:00 -0400 | [diff] [blame] | 724 | } |
Christoph Lameter | 6286ae9 | 2013-05-03 15:43:18 +0000 | [diff] [blame] | 725 | |
Christoph Lameter | 2c59dd6 | 2013-01-10 19:14:19 +0000 | [diff] [blame] | 726 | if (size <= 192) { |
| 727 | if (!size) |
| 728 | return ZERO_SIZE_PTR; |
| 729 | |
| 730 | index = size_index[size_index_elem(size)]; |
| 731 | } else |
| 732 | index = fls(size - 1); |
| 733 | |
| 734 | #ifdef CONFIG_ZONE_DMA |
Joonsoo Kim | b1e0541 | 2013-02-04 23:46:46 +0900 | [diff] [blame] | 735 | if (unlikely((flags & GFP_DMA))) |
Christoph Lameter | 2c59dd6 | 2013-01-10 19:14:19 +0000 | [diff] [blame] | 736 | return kmalloc_dma_caches[index]; |
| 737 | |
| 738 | #endif |
| 739 | return kmalloc_caches[index]; |
| 740 | } |
| 741 | |
| 742 | /* |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 743 | * Create the kmalloc array. Some of the regular kmalloc arrays |
| 744 | * may already have been created because they were needed to |
| 745 | * enable allocations for slab creation. |
| 746 | */ |
| 747 | void __init create_kmalloc_caches(unsigned long flags) |
| 748 | { |
| 749 | int i; |
| 750 | |
Christoph Lameter | 2c59dd6 | 2013-01-10 19:14:19 +0000 | [diff] [blame] | 751 | /* |
| 752 | * Patch up the size_index table if we have strange large alignment |
| 753 | * requirements for the kmalloc array. This is only the case for |
| 754 | * MIPS it seems. The standard arches will not generate any code here. |
| 755 | * |
| 756 | * Largest permitted alignment is 256 bytes due to the way we |
| 757 | * handle the index determination for the smaller caches. |
| 758 | * |
| 759 | * Make sure that nothing crazy happens if someone starts tinkering |
| 760 | * around with ARCH_KMALLOC_MINALIGN |
| 761 | */ |
| 762 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
| 763 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); |
| 764 | |
| 765 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
| 766 | int elem = size_index_elem(i); |
| 767 | |
| 768 | if (elem >= ARRAY_SIZE(size_index)) |
| 769 | break; |
| 770 | size_index[elem] = KMALLOC_SHIFT_LOW; |
| 771 | } |
| 772 | |
| 773 | if (KMALLOC_MIN_SIZE >= 64) { |
| 774 | /* |
| 775 | * The 96 byte size cache is not used if the alignment |
| 776 | * is 64 byte. |
| 777 | */ |
| 778 | for (i = 64 + 8; i <= 96; i += 8) |
| 779 | size_index[size_index_elem(i)] = 7; |
| 780 | |
| 781 | } |
| 782 | |
| 783 | if (KMALLOC_MIN_SIZE >= 128) { |
| 784 | /* |
| 785 | * The 192 byte sized cache is not used if the alignment |
| 786 | * is 128 byte. Redirect kmalloc to use the 256 byte cache |
| 787 | * instead. |
| 788 | */ |
| 789 | for (i = 128 + 8; i <= 192; i += 8) |
| 790 | size_index[size_index_elem(i)] = 8; |
| 791 | } |
Christoph Lameter | 8a965b3 | 2013-05-03 18:04:18 +0000 | [diff] [blame] | 792 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { |
| 793 | if (!kmalloc_caches[i]) { |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 794 | kmalloc_caches[i] = create_kmalloc_cache(NULL, |
| 795 | 1 << i, flags); |
Christoph Lameter | 8a965b3 | 2013-05-03 18:04:18 +0000 | [diff] [blame] | 796 | } |
Chris Mason | 956e46e | 2013-05-08 15:56:28 -0400 | [diff] [blame] | 797 | |
| 798 | /* |
| 799 | * Caches that are not of the two-to-the-power-of size. |
| 800 | * These have to be created immediately after the |
| 801 | * earlier power of two caches |
| 802 | */ |
| 803 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6) |
| 804 | kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags); |
| 805 | |
| 806 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7) |
| 807 | kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags); |
Christoph Lameter | 8a965b3 | 2013-05-03 18:04:18 +0000 | [diff] [blame] | 808 | } |
| 809 | |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 810 | /* Kmalloc array is now usable */ |
| 811 | slab_state = UP; |
| 812 | |
| 813 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { |
| 814 | struct kmem_cache *s = kmalloc_caches[i]; |
| 815 | char *n; |
| 816 | |
| 817 | if (s) { |
| 818 | n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i)); |
| 819 | |
| 820 | BUG_ON(!n); |
| 821 | s->name = n; |
| 822 | } |
| 823 | } |
| 824 | |
| 825 | #ifdef CONFIG_ZONE_DMA |
| 826 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { |
| 827 | struct kmem_cache *s = kmalloc_caches[i]; |
| 828 | |
| 829 | if (s) { |
| 830 | int size = kmalloc_size(i); |
| 831 | char *n = kasprintf(GFP_NOWAIT, |
| 832 | "dma-kmalloc-%d", size); |
| 833 | |
| 834 | BUG_ON(!n); |
| 835 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, |
| 836 | size, SLAB_CACHE_DMA | flags); |
| 837 | } |
| 838 | } |
| 839 | #endif |
| 840 | } |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 841 | #endif /* !CONFIG_SLOB */ |
| 842 | |
Vladimir Davydov | cea371f | 2014-06-04 16:07:04 -0700 | [diff] [blame] | 843 | /* |
| 844 | * To avoid unnecessary overhead, we pass through large allocation requests |
| 845 | * directly to the page allocator. We use __GFP_COMP, because we will need to |
| 846 | * know the allocation order to free the pages properly in kfree. |
| 847 | */ |
Vladimir Davydov | 5238343 | 2014-06-04 16:06:39 -0700 | [diff] [blame] | 848 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
| 849 | { |
| 850 | void *ret; |
| 851 | struct page *page; |
| 852 | |
| 853 | flags |= __GFP_COMP; |
| 854 | page = alloc_kmem_pages(flags, order); |
| 855 | ret = page ? page_address(page) : NULL; |
| 856 | kmemleak_alloc(ret, size, 1, flags); |
| 857 | return ret; |
| 858 | } |
| 859 | EXPORT_SYMBOL(kmalloc_order); |
| 860 | |
Christoph Lameter | f1b6eb6 | 2013-09-04 16:35:34 +0000 | [diff] [blame] | 861 | #ifdef CONFIG_TRACING |
| 862 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) |
| 863 | { |
| 864 | void *ret = kmalloc_order(size, flags, order); |
| 865 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); |
| 866 | return ret; |
| 867 | } |
| 868 | EXPORT_SYMBOL(kmalloc_order_trace); |
| 869 | #endif |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 870 | |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 871 | #ifdef CONFIG_SLABINFO |
Wanpeng Li | e9b4db2 | 2013-07-04 08:33:24 +0800 | [diff] [blame] | 872 | |
| 873 | #ifdef CONFIG_SLAB |
| 874 | #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR) |
| 875 | #else |
| 876 | #define SLABINFO_RIGHTS S_IRUSR |
| 877 | #endif |
| 878 | |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 879 | static void print_slabinfo_header(struct seq_file *m) |
Glauber Costa | bcee6e2 | 2012-10-19 18:20:26 +0400 | [diff] [blame] | 880 | { |
| 881 | /* |
| 882 | * Output format version, so at least we can change it |
| 883 | * without _too_ many complaints. |
| 884 | */ |
| 885 | #ifdef CONFIG_DEBUG_SLAB |
| 886 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); |
| 887 | #else |
| 888 | seq_puts(m, "slabinfo - version: 2.1\n"); |
| 889 | #endif |
| 890 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " |
| 891 | "<objperslab> <pagesperslab>"); |
| 892 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
| 893 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); |
| 894 | #ifdef CONFIG_DEBUG_SLAB |
| 895 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " |
| 896 | "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
| 897 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
| 898 | #endif |
| 899 | seq_putc(m, '\n'); |
| 900 | } |
| 901 | |
Vladimir Davydov | 1df3b26 | 2014-12-10 15:42:16 -0800 | [diff] [blame] | 902 | void *slab_start(struct seq_file *m, loff_t *pos) |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 903 | { |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 904 | mutex_lock(&slab_mutex); |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 905 | return seq_list_start(&slab_caches, *pos); |
| 906 | } |
| 907 | |
Wanpeng Li | 276a243 | 2013-07-08 08:08:28 +0800 | [diff] [blame] | 908 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 909 | { |
| 910 | return seq_list_next(p, &slab_caches, pos); |
| 911 | } |
| 912 | |
Wanpeng Li | 276a243 | 2013-07-08 08:08:28 +0800 | [diff] [blame] | 913 | void slab_stop(struct seq_file *m, void *p) |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 914 | { |
| 915 | mutex_unlock(&slab_mutex); |
| 916 | } |
| 917 | |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 918 | static void |
| 919 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 920 | { |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 921 | struct kmem_cache *c; |
| 922 | struct slabinfo sinfo; |
| 923 | int i; |
| 924 | |
| 925 | if (!is_root_cache(s)) |
| 926 | return; |
| 927 | |
| 928 | for_each_memcg_cache_index(i) { |
Qiang Huang | 2ade4de | 2013-11-12 15:08:23 -0800 | [diff] [blame] | 929 | c = cache_from_memcg_idx(s, i); |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 930 | if (!c) |
| 931 | continue; |
| 932 | |
| 933 | memset(&sinfo, 0, sizeof(sinfo)); |
| 934 | get_slabinfo(c, &sinfo); |
| 935 | |
| 936 | info->active_slabs += sinfo.active_slabs; |
| 937 | info->num_slabs += sinfo.num_slabs; |
| 938 | info->shared_avail += sinfo.shared_avail; |
| 939 | info->active_objs += sinfo.active_objs; |
| 940 | info->num_objs += sinfo.num_objs; |
| 941 | } |
| 942 | } |
| 943 | |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 944 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 945 | { |
Glauber Costa | 0d7561c | 2012-10-19 18:20:27 +0400 | [diff] [blame] | 946 | struct slabinfo sinfo; |
| 947 | |
| 948 | memset(&sinfo, 0, sizeof(sinfo)); |
| 949 | get_slabinfo(s, &sinfo); |
| 950 | |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 951 | memcg_accumulate_slabinfo(s, &sinfo); |
| 952 | |
Glauber Costa | 0d7561c | 2012-10-19 18:20:27 +0400 | [diff] [blame] | 953 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 954 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
Glauber Costa | 0d7561c | 2012-10-19 18:20:27 +0400 | [diff] [blame] | 955 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
| 956 | |
| 957 | seq_printf(m, " : tunables %4u %4u %4u", |
| 958 | sinfo.limit, sinfo.batchcount, sinfo.shared); |
| 959 | seq_printf(m, " : slabdata %6lu %6lu %6lu", |
| 960 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); |
| 961 | slabinfo_show_stats(m, s); |
| 962 | seq_putc(m, '\n'); |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 963 | } |
| 964 | |
Vladimir Davydov | 1df3b26 | 2014-12-10 15:42:16 -0800 | [diff] [blame] | 965 | static int slab_show(struct seq_file *m, void *p) |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 966 | { |
| 967 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); |
| 968 | |
Vladimir Davydov | 1df3b26 | 2014-12-10 15:42:16 -0800 | [diff] [blame] | 969 | if (p == slab_caches.next) |
| 970 | print_slabinfo_header(m); |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 971 | if (is_root_cache(s)) |
| 972 | cache_show(s, m); |
| 973 | return 0; |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 974 | } |
| 975 | |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 976 | #ifdef CONFIG_MEMCG_KMEM |
| 977 | int memcg_slab_show(struct seq_file *m, void *p) |
| 978 | { |
| 979 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); |
| 980 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 981 | |
| 982 | if (p == slab_caches.next) |
| 983 | print_slabinfo_header(m); |
| 984 | if (!is_root_cache(s) && s->memcg_params->memcg == memcg) |
| 985 | cache_show(s, m); |
| 986 | return 0; |
| 987 | } |
| 988 | #endif |
| 989 | |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 990 | /* |
| 991 | * slabinfo_op - iterator that generates /proc/slabinfo |
| 992 | * |
| 993 | * Output layout: |
| 994 | * cache-name |
| 995 | * num-active-objs |
| 996 | * total-objs |
| 997 | * object size |
| 998 | * num-active-slabs |
| 999 | * total-slabs |
| 1000 | * num-pages-per-slab |
| 1001 | * + further values on SMP and with statistics enabled |
| 1002 | */ |
| 1003 | static const struct seq_operations slabinfo_op = { |
Vladimir Davydov | 1df3b26 | 2014-12-10 15:42:16 -0800 | [diff] [blame] | 1004 | .start = slab_start, |
Wanpeng Li | 276a243 | 2013-07-08 08:08:28 +0800 | [diff] [blame] | 1005 | .next = slab_next, |
| 1006 | .stop = slab_stop, |
Vladimir Davydov | 1df3b26 | 2014-12-10 15:42:16 -0800 | [diff] [blame] | 1007 | .show = slab_show, |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1008 | }; |
| 1009 | |
| 1010 | static int slabinfo_open(struct inode *inode, struct file *file) |
| 1011 | { |
| 1012 | return seq_open(file, &slabinfo_op); |
| 1013 | } |
| 1014 | |
| 1015 | static const struct file_operations proc_slabinfo_operations = { |
| 1016 | .open = slabinfo_open, |
| 1017 | .read = seq_read, |
| 1018 | .write = slabinfo_write, |
| 1019 | .llseek = seq_lseek, |
| 1020 | .release = seq_release, |
| 1021 | }; |
| 1022 | |
| 1023 | static int __init slab_proc_init(void) |
| 1024 | { |
Wanpeng Li | e9b4db2 | 2013-07-04 08:33:24 +0800 | [diff] [blame] | 1025 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, |
| 1026 | &proc_slabinfo_operations); |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1027 | return 0; |
| 1028 | } |
| 1029 | module_init(slab_proc_init); |
| 1030 | #endif /* CONFIG_SLABINFO */ |
Andrey Ryabinin | 928cec9 | 2014-08-06 16:04:44 -0700 | [diff] [blame] | 1031 | |
| 1032 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, |
| 1033 | gfp_t flags) |
| 1034 | { |
| 1035 | void *ret; |
| 1036 | size_t ks = 0; |
| 1037 | |
| 1038 | if (p) |
| 1039 | ks = ksize(p); |
| 1040 | |
| 1041 | if (ks >= new_size) |
| 1042 | return (void *)p; |
| 1043 | |
| 1044 | ret = kmalloc_track_caller(new_size, flags); |
| 1045 | if (ret && p) |
| 1046 | memcpy(ret, p, ks); |
| 1047 | |
| 1048 | return ret; |
| 1049 | } |
| 1050 | |
| 1051 | /** |
| 1052 | * __krealloc - like krealloc() but don't free @p. |
| 1053 | * @p: object to reallocate memory for. |
| 1054 | * @new_size: how many bytes of memory are required. |
| 1055 | * @flags: the type of memory to allocate. |
| 1056 | * |
| 1057 | * This function is like krealloc() except it never frees the originally |
| 1058 | * allocated buffer. Use this if you don't want to free the buffer immediately |
| 1059 | * like, for example, with RCU. |
| 1060 | */ |
| 1061 | void *__krealloc(const void *p, size_t new_size, gfp_t flags) |
| 1062 | { |
| 1063 | if (unlikely(!new_size)) |
| 1064 | return ZERO_SIZE_PTR; |
| 1065 | |
| 1066 | return __do_krealloc(p, new_size, flags); |
| 1067 | |
| 1068 | } |
| 1069 | EXPORT_SYMBOL(__krealloc); |
| 1070 | |
| 1071 | /** |
| 1072 | * krealloc - reallocate memory. The contents will remain unchanged. |
| 1073 | * @p: object to reallocate memory for. |
| 1074 | * @new_size: how many bytes of memory are required. |
| 1075 | * @flags: the type of memory to allocate. |
| 1076 | * |
| 1077 | * The contents of the object pointed to are preserved up to the |
| 1078 | * lesser of the new and old sizes. If @p is %NULL, krealloc() |
| 1079 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a |
| 1080 | * %NULL pointer, the object pointed to is freed. |
| 1081 | */ |
| 1082 | void *krealloc(const void *p, size_t new_size, gfp_t flags) |
| 1083 | { |
| 1084 | void *ret; |
| 1085 | |
| 1086 | if (unlikely(!new_size)) { |
| 1087 | kfree(p); |
| 1088 | return ZERO_SIZE_PTR; |
| 1089 | } |
| 1090 | |
| 1091 | ret = __do_krealloc(p, new_size, flags); |
| 1092 | if (ret && p != ret) |
| 1093 | kfree(p); |
| 1094 | |
| 1095 | return ret; |
| 1096 | } |
| 1097 | EXPORT_SYMBOL(krealloc); |
| 1098 | |
| 1099 | /** |
| 1100 | * kzfree - like kfree but zero memory |
| 1101 | * @p: object to free memory of |
| 1102 | * |
| 1103 | * The memory of the object @p points to is zeroed before freed. |
| 1104 | * If @p is %NULL, kzfree() does nothing. |
| 1105 | * |
| 1106 | * Note: this function zeroes the whole allocated buffer which can be a good |
| 1107 | * deal bigger than the requested buffer size passed to kmalloc(). So be |
| 1108 | * careful when using this function in performance sensitive code. |
| 1109 | */ |
| 1110 | void kzfree(const void *p) |
| 1111 | { |
| 1112 | size_t ks; |
| 1113 | void *mem = (void *)p; |
| 1114 | |
| 1115 | if (unlikely(ZERO_OR_NULL_PTR(mem))) |
| 1116 | return; |
| 1117 | ks = ksize(mem); |
| 1118 | memset(mem, 0, ks); |
| 1119 | kfree(mem); |
| 1120 | } |
| 1121 | EXPORT_SYMBOL(kzfree); |
| 1122 | |
| 1123 | /* Tracepoints definitions. */ |
| 1124 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); |
| 1125 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); |
| 1126 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); |
| 1127 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); |
| 1128 | EXPORT_TRACEPOINT_SYMBOL(kfree); |
| 1129 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); |