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> |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 22 | |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 23 | #include "slab.h" |
| 24 | |
| 25 | enum slab_state slab_state; |
Christoph Lameter | 18004c5 | 2012-07-06 15:25:12 -0500 | [diff] [blame] | 26 | LIST_HEAD(slab_caches); |
| 27 | DEFINE_MUTEX(slab_mutex); |
Christoph Lameter | 9b030cb | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 28 | struct kmem_cache *kmem_cache; |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 29 | |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 30 | #ifdef CONFIG_DEBUG_VM |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 31 | static int kmem_cache_sanity_check(struct mem_cgroup *memcg, const char *name, |
| 32 | size_t size) |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 33 | { |
| 34 | struct kmem_cache *s = NULL; |
| 35 | |
| 36 | if (!name || in_interrupt() || size < sizeof(void *) || |
| 37 | size > KMALLOC_MAX_SIZE) { |
| 38 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
| 39 | return -EINVAL; |
| 40 | } |
| 41 | |
| 42 | list_for_each_entry(s, &slab_caches, list) { |
| 43 | char tmp; |
| 44 | int res; |
| 45 | |
| 46 | /* |
| 47 | * This happens when the module gets unloaded and doesn't |
| 48 | * destroy its slab cache and no-one else reuses the vmalloc |
| 49 | * area of the module. Print a warning. |
| 50 | */ |
| 51 | res = probe_kernel_address(s->name, tmp); |
| 52 | if (res) { |
| 53 | pr_err("Slab cache with size %d has lost its name\n", |
| 54 | s->object_size); |
| 55 | continue; |
| 56 | } |
| 57 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 58 | /* |
| 59 | * For simplicity, we won't check this in the list of memcg |
| 60 | * caches. We have control over memcg naming, and if there |
| 61 | * aren't duplicates in the global list, there won't be any |
| 62 | * duplicates in the memcg lists as well. |
| 63 | */ |
| 64 | if (!memcg && !strcmp(s->name, name)) { |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 65 | pr_err("%s (%s): Cache name already exists.\n", |
| 66 | __func__, name); |
| 67 | dump_stack(); |
| 68 | s = NULL; |
| 69 | return -EINVAL; |
| 70 | } |
| 71 | } |
| 72 | |
| 73 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
| 74 | return 0; |
| 75 | } |
| 76 | #else |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 77 | static inline int kmem_cache_sanity_check(struct mem_cgroup *memcg, |
| 78 | const char *name, size_t size) |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 79 | { |
| 80 | return 0; |
| 81 | } |
| 82 | #endif |
| 83 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 84 | #ifdef CONFIG_MEMCG_KMEM |
| 85 | int memcg_update_all_caches(int num_memcgs) |
| 86 | { |
| 87 | struct kmem_cache *s; |
| 88 | int ret = 0; |
| 89 | mutex_lock(&slab_mutex); |
| 90 | |
| 91 | list_for_each_entry(s, &slab_caches, list) { |
| 92 | if (!is_root_cache(s)) |
| 93 | continue; |
| 94 | |
| 95 | ret = memcg_update_cache_size(s, num_memcgs); |
| 96 | /* |
| 97 | * See comment in memcontrol.c, memcg_update_cache_size: |
| 98 | * Instead of freeing the memory, we'll just leave the caches |
| 99 | * up to this point in an updated state. |
| 100 | */ |
| 101 | if (ret) |
| 102 | goto out; |
| 103 | } |
| 104 | |
| 105 | memcg_update_array_size(num_memcgs); |
| 106 | out: |
| 107 | mutex_unlock(&slab_mutex); |
| 108 | return ret; |
| 109 | } |
| 110 | #endif |
| 111 | |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 112 | /* |
Christoph Lameter | 4590685 | 2012-11-28 16:23:16 +0000 | [diff] [blame] | 113 | * Figure out what the alignment of the objects will be given a set of |
| 114 | * flags, a user specified alignment and the size of the objects. |
| 115 | */ |
| 116 | unsigned long calculate_alignment(unsigned long flags, |
| 117 | unsigned long align, unsigned long size) |
| 118 | { |
| 119 | /* |
| 120 | * If the user wants hardware cache aligned objects then follow that |
| 121 | * suggestion if the object is sufficiently large. |
| 122 | * |
| 123 | * The hardware cache alignment cannot override the specified |
| 124 | * alignment though. If that is greater then use it. |
| 125 | */ |
| 126 | if (flags & SLAB_HWCACHE_ALIGN) { |
| 127 | unsigned long ralign = cache_line_size(); |
| 128 | while (size <= ralign / 2) |
| 129 | ralign /= 2; |
| 130 | align = max(align, ralign); |
| 131 | } |
| 132 | |
| 133 | if (align < ARCH_SLAB_MINALIGN) |
| 134 | align = ARCH_SLAB_MINALIGN; |
| 135 | |
| 136 | return ALIGN(align, sizeof(void *)); |
| 137 | } |
| 138 | |
| 139 | |
| 140 | /* |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 141 | * kmem_cache_create - Create a cache. |
| 142 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
| 143 | * @size: The size of objects to be created in this cache. |
| 144 | * @align: The required alignment for the objects. |
| 145 | * @flags: SLAB flags |
| 146 | * @ctor: A constructor for the objects. |
| 147 | * |
| 148 | * Returns a ptr to the cache on success, NULL on failure. |
| 149 | * Cannot be called within a interrupt, but can be interrupted. |
| 150 | * The @ctor is run when new pages are allocated by the cache. |
| 151 | * |
| 152 | * The flags are |
| 153 | * |
| 154 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) |
| 155 | * to catch references to uninitialised memory. |
| 156 | * |
| 157 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check |
| 158 | * for buffer overruns. |
| 159 | * |
| 160 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
| 161 | * cacheline. This can be beneficial if you're counting cycles as closely |
| 162 | * as davem. |
| 163 | */ |
| 164 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 165 | struct kmem_cache * |
| 166 | kmem_cache_create_memcg(struct mem_cgroup *memcg, const char *name, size_t size, |
Glauber Costa | 943a451 | 2012-12-18 14:23:03 -0800 | [diff] [blame] | 167 | size_t align, unsigned long flags, void (*ctor)(void *), |
| 168 | struct kmem_cache *parent_cache) |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 169 | { |
| 170 | struct kmem_cache *s = NULL; |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 171 | int err = 0; |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 172 | |
Pekka Enberg | b920536 | 2012-08-16 10:12:18 +0300 | [diff] [blame] | 173 | get_online_cpus(); |
| 174 | mutex_lock(&slab_mutex); |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 175 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 176 | if (!kmem_cache_sanity_check(memcg, name, size) == 0) |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 177 | goto out_locked; |
| 178 | |
Glauber Costa | d884392 | 2012-10-17 15:36:51 +0400 | [diff] [blame] | 179 | /* |
| 180 | * Some allocators will constraint the set of valid flags to a subset |
| 181 | * of all flags. We expect them to define CACHE_CREATE_MASK in this |
| 182 | * case, and we'll just provide them with a sanitized version of the |
| 183 | * passed flags. |
| 184 | */ |
| 185 | flags &= CACHE_CREATE_MASK; |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 186 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 187 | s = __kmem_cache_alias(memcg, name, size, align, flags, ctor); |
Christoph Lameter | cbb7969 | 2012-09-05 00:18:32 +0000 | [diff] [blame] | 188 | if (s) |
| 189 | goto out_locked; |
| 190 | |
Christoph Lameter | 278b1bb | 2012-09-05 00:20:34 +0000 | [diff] [blame] | 191 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); |
Christoph Lameter | db265ec | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 192 | if (s) { |
Christoph Lameter | 8a13a4c | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 193 | s->object_size = s->size = size; |
Christoph Lameter | 4590685 | 2012-11-28 16:23:16 +0000 | [diff] [blame] | 194 | s->align = calculate_alignment(flags, align, size); |
Christoph Lameter | 8a13a4c | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 195 | s->ctor = ctor; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 196 | |
Glauber Costa | 943a451 | 2012-12-18 14:23:03 -0800 | [diff] [blame] | 197 | if (memcg_register_cache(memcg, s, parent_cache)) { |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 198 | kmem_cache_free(kmem_cache, s); |
| 199 | err = -ENOMEM; |
| 200 | goto out_locked; |
| 201 | } |
| 202 | |
Christoph Lameter | 8a13a4c | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 203 | s->name = kstrdup(name, GFP_KERNEL); |
| 204 | if (!s->name) { |
| 205 | kmem_cache_free(kmem_cache, s); |
| 206 | err = -ENOMEM; |
| 207 | goto out_locked; |
| 208 | } |
| 209 | |
| 210 | err = __kmem_cache_create(s, flags); |
Christoph Lameter | cce89f4 | 2012-09-04 23:38:33 +0000 | [diff] [blame] | 211 | if (!err) { |
Christoph Lameter | cce89f4 | 2012-09-04 23:38:33 +0000 | [diff] [blame] | 212 | s->refcount = 1; |
Christoph Lameter | db265ec | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 213 | list_add(&s->list, &slab_caches); |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 214 | memcg_cache_list_add(memcg, s); |
Christoph Lameter | cce89f4 | 2012-09-04 23:38:33 +0000 | [diff] [blame] | 215 | } else { |
Christoph Lameter | 8a13a4c | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 216 | kfree(s->name); |
Christoph Lameter | 278b1bb | 2012-09-05 00:20:34 +0000 | [diff] [blame] | 217 | kmem_cache_free(kmem_cache, s); |
| 218 | } |
Christoph Lameter | 8a13a4c | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 219 | } else |
Christoph Lameter | 278b1bb | 2012-09-05 00:20:34 +0000 | [diff] [blame] | 220 | err = -ENOMEM; |
Christoph Lameter | 7c9adf5 | 2012-09-04 23:38:33 +0000 | [diff] [blame] | 221 | |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 222 | out_locked: |
Christoph Lameter | 20cea96 | 2012-07-06 15:25:13 -0500 | [diff] [blame] | 223 | mutex_unlock(&slab_mutex); |
| 224 | put_online_cpus(); |
| 225 | |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 226 | if (err) { |
| 227 | |
| 228 | if (flags & SLAB_PANIC) |
| 229 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", |
| 230 | name, err); |
| 231 | else { |
| 232 | printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d", |
| 233 | name, err); |
| 234 | dump_stack(); |
| 235 | } |
| 236 | |
| 237 | return NULL; |
| 238 | } |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 239 | |
| 240 | return s; |
| 241 | } |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 242 | |
| 243 | struct kmem_cache * |
| 244 | kmem_cache_create(const char *name, size_t size, size_t align, |
| 245 | unsigned long flags, void (*ctor)(void *)) |
| 246 | { |
Glauber Costa | 943a451 | 2012-12-18 14:23:03 -0800 | [diff] [blame] | 247 | return kmem_cache_create_memcg(NULL, name, size, align, flags, ctor, NULL); |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 248 | } |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 249 | EXPORT_SYMBOL(kmem_cache_create); |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 250 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 251 | void kmem_cache_destroy(struct kmem_cache *s) |
| 252 | { |
Glauber Costa | 7cf2798 | 2012-12-18 14:22:55 -0800 | [diff] [blame] | 253 | /* Destroy all the children caches if we aren't a memcg cache */ |
| 254 | kmem_cache_destroy_memcg_children(s); |
| 255 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 256 | get_online_cpus(); |
| 257 | mutex_lock(&slab_mutex); |
| 258 | s->refcount--; |
| 259 | if (!s->refcount) { |
| 260 | list_del(&s->list); |
| 261 | |
| 262 | if (!__kmem_cache_shutdown(s)) { |
Jiri Kosina | 210ed9d | 2012-10-08 09:26:01 +0200 | [diff] [blame] | 263 | mutex_unlock(&slab_mutex); |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 264 | if (s->flags & SLAB_DESTROY_BY_RCU) |
| 265 | rcu_barrier(); |
| 266 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 267 | memcg_release_cache(s); |
Christoph Lameter | db265ec | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 268 | kfree(s->name); |
Christoph Lameter | 8f4c765 | 2012-09-05 00:18:32 +0000 | [diff] [blame] | 269 | kmem_cache_free(kmem_cache, s); |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 270 | } else { |
| 271 | list_add(&s->list, &slab_caches); |
Jiri Kosina | 210ed9d | 2012-10-08 09:26:01 +0200 | [diff] [blame] | 272 | mutex_unlock(&slab_mutex); |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 273 | printk(KERN_ERR "kmem_cache_destroy %s: Slab cache still has objects\n", |
| 274 | s->name); |
| 275 | dump_stack(); |
| 276 | } |
Jiri Kosina | 210ed9d | 2012-10-08 09:26:01 +0200 | [diff] [blame] | 277 | } else { |
| 278 | mutex_unlock(&slab_mutex); |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 279 | } |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 280 | put_online_cpus(); |
| 281 | } |
| 282 | EXPORT_SYMBOL(kmem_cache_destroy); |
| 283 | |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 284 | int slab_is_available(void) |
| 285 | { |
| 286 | return slab_state >= UP; |
| 287 | } |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 288 | |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 289 | #ifndef CONFIG_SLOB |
| 290 | /* Create a cache during boot when no slab services are available yet */ |
| 291 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, |
| 292 | unsigned long flags) |
| 293 | { |
| 294 | int err; |
| 295 | |
| 296 | s->name = name; |
| 297 | s->size = s->object_size = size; |
Christoph Lameter | 4590685 | 2012-11-28 16:23:16 +0000 | [diff] [blame] | 298 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 299 | err = __kmem_cache_create(s, flags); |
| 300 | |
| 301 | if (err) |
Christoph Lameter | 31ba734 | 2013-01-10 19:00:53 +0000 | [diff] [blame] | 302 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 303 | name, size, err); |
| 304 | |
| 305 | s->refcount = -1; /* Exempt from merging for now */ |
| 306 | } |
| 307 | |
| 308 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, |
| 309 | unsigned long flags) |
| 310 | { |
| 311 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
| 312 | |
| 313 | if (!s) |
| 314 | panic("Out of memory when creating slab %s\n", name); |
| 315 | |
| 316 | create_boot_cache(s, name, size, flags); |
| 317 | list_add(&s->list, &slab_caches); |
| 318 | s->refcount = 1; |
| 319 | return s; |
| 320 | } |
| 321 | |
Christoph Lameter | 9425c58 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 322 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; |
| 323 | EXPORT_SYMBOL(kmalloc_caches); |
| 324 | |
| 325 | #ifdef CONFIG_ZONE_DMA |
| 326 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; |
| 327 | EXPORT_SYMBOL(kmalloc_dma_caches); |
| 328 | #endif |
| 329 | |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 330 | /* |
Christoph Lameter | 2c59dd6 | 2013-01-10 19:14:19 +0000 | [diff] [blame] | 331 | * Conversion table for small slabs sizes / 8 to the index in the |
| 332 | * kmalloc array. This is necessary for slabs < 192 since we have non power |
| 333 | * of two cache sizes there. The size of larger slabs can be determined using |
| 334 | * fls. |
| 335 | */ |
| 336 | static s8 size_index[24] = { |
| 337 | 3, /* 8 */ |
| 338 | 4, /* 16 */ |
| 339 | 5, /* 24 */ |
| 340 | 5, /* 32 */ |
| 341 | 6, /* 40 */ |
| 342 | 6, /* 48 */ |
| 343 | 6, /* 56 */ |
| 344 | 6, /* 64 */ |
| 345 | 1, /* 72 */ |
| 346 | 1, /* 80 */ |
| 347 | 1, /* 88 */ |
| 348 | 1, /* 96 */ |
| 349 | 7, /* 104 */ |
| 350 | 7, /* 112 */ |
| 351 | 7, /* 120 */ |
| 352 | 7, /* 128 */ |
| 353 | 2, /* 136 */ |
| 354 | 2, /* 144 */ |
| 355 | 2, /* 152 */ |
| 356 | 2, /* 160 */ |
| 357 | 2, /* 168 */ |
| 358 | 2, /* 176 */ |
| 359 | 2, /* 184 */ |
| 360 | 2 /* 192 */ |
| 361 | }; |
| 362 | |
| 363 | static inline int size_index_elem(size_t bytes) |
| 364 | { |
| 365 | return (bytes - 1) / 8; |
| 366 | } |
| 367 | |
| 368 | /* |
| 369 | * Find the kmem_cache structure that serves a given size of |
| 370 | * allocation |
| 371 | */ |
| 372 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) |
| 373 | { |
| 374 | int index; |
| 375 | |
| 376 | if (size <= 192) { |
| 377 | if (!size) |
| 378 | return ZERO_SIZE_PTR; |
| 379 | |
| 380 | index = size_index[size_index_elem(size)]; |
| 381 | } else |
| 382 | index = fls(size - 1); |
| 383 | |
| 384 | #ifdef CONFIG_ZONE_DMA |
| 385 | if (unlikely((flags & SLAB_CACHE_DMA))) |
| 386 | return kmalloc_dma_caches[index]; |
| 387 | |
| 388 | #endif |
| 389 | return kmalloc_caches[index]; |
| 390 | } |
| 391 | |
| 392 | /* |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 393 | * Create the kmalloc array. Some of the regular kmalloc arrays |
| 394 | * may already have been created because they were needed to |
| 395 | * enable allocations for slab creation. |
| 396 | */ |
| 397 | void __init create_kmalloc_caches(unsigned long flags) |
| 398 | { |
| 399 | int i; |
| 400 | |
Christoph Lameter | 2c59dd6 | 2013-01-10 19:14:19 +0000 | [diff] [blame] | 401 | /* |
| 402 | * Patch up the size_index table if we have strange large alignment |
| 403 | * requirements for the kmalloc array. This is only the case for |
| 404 | * MIPS it seems. The standard arches will not generate any code here. |
| 405 | * |
| 406 | * Largest permitted alignment is 256 bytes due to the way we |
| 407 | * handle the index determination for the smaller caches. |
| 408 | * |
| 409 | * Make sure that nothing crazy happens if someone starts tinkering |
| 410 | * around with ARCH_KMALLOC_MINALIGN |
| 411 | */ |
| 412 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
| 413 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); |
| 414 | |
| 415 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
| 416 | int elem = size_index_elem(i); |
| 417 | |
| 418 | if (elem >= ARRAY_SIZE(size_index)) |
| 419 | break; |
| 420 | size_index[elem] = KMALLOC_SHIFT_LOW; |
| 421 | } |
| 422 | |
| 423 | if (KMALLOC_MIN_SIZE >= 64) { |
| 424 | /* |
| 425 | * The 96 byte size cache is not used if the alignment |
| 426 | * is 64 byte. |
| 427 | */ |
| 428 | for (i = 64 + 8; i <= 96; i += 8) |
| 429 | size_index[size_index_elem(i)] = 7; |
| 430 | |
| 431 | } |
| 432 | |
| 433 | if (KMALLOC_MIN_SIZE >= 128) { |
| 434 | /* |
| 435 | * The 192 byte sized cache is not used if the alignment |
| 436 | * is 128 byte. Redirect kmalloc to use the 256 byte cache |
| 437 | * instead. |
| 438 | */ |
| 439 | for (i = 128 + 8; i <= 192; i += 8) |
| 440 | size_index[size_index_elem(i)] = 8; |
| 441 | } |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 442 | /* Caches that are not of the two-to-the-power-of size */ |
| 443 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1]) |
| 444 | kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags); |
| 445 | |
| 446 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2]) |
| 447 | kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags); |
| 448 | |
| 449 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) |
| 450 | if (!kmalloc_caches[i]) |
| 451 | kmalloc_caches[i] = create_kmalloc_cache(NULL, |
| 452 | 1 << i, flags); |
| 453 | |
| 454 | /* Kmalloc array is now usable */ |
| 455 | slab_state = UP; |
| 456 | |
| 457 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { |
| 458 | struct kmem_cache *s = kmalloc_caches[i]; |
| 459 | char *n; |
| 460 | |
| 461 | if (s) { |
| 462 | n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i)); |
| 463 | |
| 464 | BUG_ON(!n); |
| 465 | s->name = n; |
| 466 | } |
| 467 | } |
| 468 | |
| 469 | #ifdef CONFIG_ZONE_DMA |
| 470 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { |
| 471 | struct kmem_cache *s = kmalloc_caches[i]; |
| 472 | |
| 473 | if (s) { |
| 474 | int size = kmalloc_size(i); |
| 475 | char *n = kasprintf(GFP_NOWAIT, |
| 476 | "dma-kmalloc-%d", size); |
| 477 | |
| 478 | BUG_ON(!n); |
| 479 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, |
| 480 | size, SLAB_CACHE_DMA | flags); |
| 481 | } |
| 482 | } |
| 483 | #endif |
| 484 | } |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 485 | #endif /* !CONFIG_SLOB */ |
| 486 | |
| 487 | |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 488 | #ifdef CONFIG_SLABINFO |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 489 | void print_slabinfo_header(struct seq_file *m) |
Glauber Costa | bcee6e2 | 2012-10-19 18:20:26 +0400 | [diff] [blame] | 490 | { |
| 491 | /* |
| 492 | * Output format version, so at least we can change it |
| 493 | * without _too_ many complaints. |
| 494 | */ |
| 495 | #ifdef CONFIG_DEBUG_SLAB |
| 496 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); |
| 497 | #else |
| 498 | seq_puts(m, "slabinfo - version: 2.1\n"); |
| 499 | #endif |
| 500 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " |
| 501 | "<objperslab> <pagesperslab>"); |
| 502 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
| 503 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); |
| 504 | #ifdef CONFIG_DEBUG_SLAB |
| 505 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " |
| 506 | "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
| 507 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
| 508 | #endif |
| 509 | seq_putc(m, '\n'); |
| 510 | } |
| 511 | |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 512 | static void *s_start(struct seq_file *m, loff_t *pos) |
| 513 | { |
| 514 | loff_t n = *pos; |
| 515 | |
| 516 | mutex_lock(&slab_mutex); |
| 517 | if (!n) |
| 518 | print_slabinfo_header(m); |
| 519 | |
| 520 | return seq_list_start(&slab_caches, *pos); |
| 521 | } |
| 522 | |
| 523 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) |
| 524 | { |
| 525 | return seq_list_next(p, &slab_caches, pos); |
| 526 | } |
| 527 | |
| 528 | static void s_stop(struct seq_file *m, void *p) |
| 529 | { |
| 530 | mutex_unlock(&slab_mutex); |
| 531 | } |
| 532 | |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 533 | static void |
| 534 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 535 | { |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 536 | struct kmem_cache *c; |
| 537 | struct slabinfo sinfo; |
| 538 | int i; |
| 539 | |
| 540 | if (!is_root_cache(s)) |
| 541 | return; |
| 542 | |
| 543 | for_each_memcg_cache_index(i) { |
| 544 | c = cache_from_memcg(s, i); |
| 545 | if (!c) |
| 546 | continue; |
| 547 | |
| 548 | memset(&sinfo, 0, sizeof(sinfo)); |
| 549 | get_slabinfo(c, &sinfo); |
| 550 | |
| 551 | info->active_slabs += sinfo.active_slabs; |
| 552 | info->num_slabs += sinfo.num_slabs; |
| 553 | info->shared_avail += sinfo.shared_avail; |
| 554 | info->active_objs += sinfo.active_objs; |
| 555 | info->num_objs += sinfo.num_objs; |
| 556 | } |
| 557 | } |
| 558 | |
| 559 | int cache_show(struct kmem_cache *s, struct seq_file *m) |
| 560 | { |
Glauber Costa | 0d7561c | 2012-10-19 18:20:27 +0400 | [diff] [blame] | 561 | struct slabinfo sinfo; |
| 562 | |
| 563 | memset(&sinfo, 0, sizeof(sinfo)); |
| 564 | get_slabinfo(s, &sinfo); |
| 565 | |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 566 | memcg_accumulate_slabinfo(s, &sinfo); |
| 567 | |
Glauber Costa | 0d7561c | 2012-10-19 18:20:27 +0400 | [diff] [blame] | 568 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 569 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
Glauber Costa | 0d7561c | 2012-10-19 18:20:27 +0400 | [diff] [blame] | 570 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
| 571 | |
| 572 | seq_printf(m, " : tunables %4u %4u %4u", |
| 573 | sinfo.limit, sinfo.batchcount, sinfo.shared); |
| 574 | seq_printf(m, " : slabdata %6lu %6lu %6lu", |
| 575 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); |
| 576 | slabinfo_show_stats(m, s); |
| 577 | seq_putc(m, '\n'); |
| 578 | return 0; |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 579 | } |
| 580 | |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 581 | static int s_show(struct seq_file *m, void *p) |
| 582 | { |
| 583 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); |
| 584 | |
| 585 | if (!is_root_cache(s)) |
| 586 | return 0; |
| 587 | return cache_show(s, m); |
| 588 | } |
| 589 | |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 590 | /* |
| 591 | * slabinfo_op - iterator that generates /proc/slabinfo |
| 592 | * |
| 593 | * Output layout: |
| 594 | * cache-name |
| 595 | * num-active-objs |
| 596 | * total-objs |
| 597 | * object size |
| 598 | * num-active-slabs |
| 599 | * total-slabs |
| 600 | * num-pages-per-slab |
| 601 | * + further values on SMP and with statistics enabled |
| 602 | */ |
| 603 | static const struct seq_operations slabinfo_op = { |
| 604 | .start = s_start, |
| 605 | .next = s_next, |
| 606 | .stop = s_stop, |
| 607 | .show = s_show, |
| 608 | }; |
| 609 | |
| 610 | static int slabinfo_open(struct inode *inode, struct file *file) |
| 611 | { |
| 612 | return seq_open(file, &slabinfo_op); |
| 613 | } |
| 614 | |
| 615 | static const struct file_operations proc_slabinfo_operations = { |
| 616 | .open = slabinfo_open, |
| 617 | .read = seq_read, |
| 618 | .write = slabinfo_write, |
| 619 | .llseek = seq_lseek, |
| 620 | .release = seq_release, |
| 621 | }; |
| 622 | |
| 623 | static int __init slab_proc_init(void) |
| 624 | { |
| 625 | proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations); |
| 626 | return 0; |
| 627 | } |
| 628 | module_init(slab_proc_init); |
| 629 | #endif /* CONFIG_SLABINFO */ |