| 1 | /* |
| 2 | * linux/mm/slab.c |
| 3 | * Written by Mark Hemment, 1996/97. |
| 4 | * (markhe@nextd.demon.co.uk) |
| 5 | * |
| 6 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli |
| 7 | * |
| 8 | * Major cleanup, different bufctl logic, per-cpu arrays |
| 9 | * (c) 2000 Manfred Spraul |
| 10 | * |
| 11 | * Cleanup, make the head arrays unconditional, preparation for NUMA |
| 12 | * (c) 2002 Manfred Spraul |
| 13 | * |
| 14 | * An implementation of the Slab Allocator as described in outline in; |
| 15 | * UNIX Internals: The New Frontiers by Uresh Vahalia |
| 16 | * Pub: Prentice Hall ISBN 0-13-101908-2 |
| 17 | * or with a little more detail in; |
| 18 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator |
| 19 | * Jeff Bonwick (Sun Microsystems). |
| 20 | * Presented at: USENIX Summer 1994 Technical Conference |
| 21 | * |
| 22 | * The memory is organized in caches, one cache for each object type. |
| 23 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) |
| 24 | * Each cache consists out of many slabs (they are small (usually one |
| 25 | * page long) and always contiguous), and each slab contains multiple |
| 26 | * initialized objects. |
| 27 | * |
| 28 | * This means, that your constructor is used only for newly allocated |
| 29 | * slabs and you must pass objects with the same initializations to |
| 30 | * kmem_cache_free. |
| 31 | * |
| 32 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, |
| 33 | * normal). If you need a special memory type, then must create a new |
| 34 | * cache for that memory type. |
| 35 | * |
| 36 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: |
| 37 | * full slabs with 0 free objects |
| 38 | * partial slabs |
| 39 | * empty slabs with no allocated objects |
| 40 | * |
| 41 | * If partial slabs exist, then new allocations come from these slabs, |
| 42 | * otherwise from empty slabs or new slabs are allocated. |
| 43 | * |
| 44 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache |
| 45 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. |
| 46 | * |
| 47 | * Each cache has a short per-cpu head array, most allocs |
| 48 | * and frees go into that array, and if that array overflows, then 1/2 |
| 49 | * of the entries in the array are given back into the global cache. |
| 50 | * The head array is strictly LIFO and should improve the cache hit rates. |
| 51 | * On SMP, it additionally reduces the spinlock operations. |
| 52 | * |
| 53 | * The c_cpuarray may not be read with enabled local interrupts - |
| 54 | * it's changed with a smp_call_function(). |
| 55 | * |
| 56 | * SMP synchronization: |
| 57 | * constructors and destructors are called without any locking. |
| 58 | * Several members in struct kmem_cache and struct slab never change, they |
| 59 | * are accessed without any locking. |
| 60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, |
| 61 | * and local interrupts are disabled so slab code is preempt-safe. |
| 62 | * The non-constant members are protected with a per-cache irq spinlock. |
| 63 | * |
| 64 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch |
| 65 | * in 2000 - many ideas in the current implementation are derived from |
| 66 | * his patch. |
| 67 | * |
| 68 | * Further notes from the original documentation: |
| 69 | * |
| 70 | * 11 April '97. Started multi-threading - markhe |
| 71 | * The global cache-chain is protected by the mutex 'slab_mutex'. |
| 72 | * The sem is only needed when accessing/extending the cache-chain, which |
| 73 | * can never happen inside an interrupt (kmem_cache_create(), |
| 74 | * kmem_cache_shrink() and kmem_cache_reap()). |
| 75 | * |
| 76 | * At present, each engine can be growing a cache. This should be blocked. |
| 77 | * |
| 78 | * 15 March 2005. NUMA slab allocator. |
| 79 | * Shai Fultheim <shai@scalex86.org>. |
| 80 | * Shobhit Dayal <shobhit@calsoftinc.com> |
| 81 | * Alok N Kataria <alokk@calsoftinc.com> |
| 82 | * Christoph Lameter <christoph@lameter.com> |
| 83 | * |
| 84 | * Modified the slab allocator to be node aware on NUMA systems. |
| 85 | * Each node has its own list of partial, free and full slabs. |
| 86 | * All object allocations for a node occur from node specific slab lists. |
| 87 | */ |
| 88 | |
| 89 | #include <linux/slab.h> |
| 90 | #include <linux/mm.h> |
| 91 | #include <linux/poison.h> |
| 92 | #include <linux/swap.h> |
| 93 | #include <linux/cache.h> |
| 94 | #include <linux/interrupt.h> |
| 95 | #include <linux/init.h> |
| 96 | #include <linux/compiler.h> |
| 97 | #include <linux/cpuset.h> |
| 98 | #include <linux/proc_fs.h> |
| 99 | #include <linux/seq_file.h> |
| 100 | #include <linux/notifier.h> |
| 101 | #include <linux/kallsyms.h> |
| 102 | #include <linux/cpu.h> |
| 103 | #include <linux/sysctl.h> |
| 104 | #include <linux/module.h> |
| 105 | #include <linux/rcupdate.h> |
| 106 | #include <linux/string.h> |
| 107 | #include <linux/uaccess.h> |
| 108 | #include <linux/nodemask.h> |
| 109 | #include <linux/kmemleak.h> |
| 110 | #include <linux/mempolicy.h> |
| 111 | #include <linux/mutex.h> |
| 112 | #include <linux/fault-inject.h> |
| 113 | #include <linux/rtmutex.h> |
| 114 | #include <linux/reciprocal_div.h> |
| 115 | #include <linux/debugobjects.h> |
| 116 | #include <linux/kmemcheck.h> |
| 117 | #include <linux/memory.h> |
| 118 | #include <linux/prefetch.h> |
| 119 | |
| 120 | #include <net/sock.h> |
| 121 | |
| 122 | #include <asm/cacheflush.h> |
| 123 | #include <asm/tlbflush.h> |
| 124 | #include <asm/page.h> |
| 125 | |
| 126 | #include <trace/events/kmem.h> |
| 127 | |
| 128 | #include "internal.h" |
| 129 | |
| 130 | #include "slab.h" |
| 131 | |
| 132 | /* |
| 133 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
| 134 | * 0 for faster, smaller code (especially in the critical paths). |
| 135 | * |
| 136 | * STATS - 1 to collect stats for /proc/slabinfo. |
| 137 | * 0 for faster, smaller code (especially in the critical paths). |
| 138 | * |
| 139 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) |
| 140 | */ |
| 141 | |
| 142 | #ifdef CONFIG_DEBUG_SLAB |
| 143 | #define DEBUG 1 |
| 144 | #define STATS 1 |
| 145 | #define FORCED_DEBUG 1 |
| 146 | #else |
| 147 | #define DEBUG 0 |
| 148 | #define STATS 0 |
| 149 | #define FORCED_DEBUG 0 |
| 150 | #endif |
| 151 | |
| 152 | /* Shouldn't this be in a header file somewhere? */ |
| 153 | #define BYTES_PER_WORD sizeof(void *) |
| 154 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
| 155 | |
| 156 | #ifndef ARCH_KMALLOC_FLAGS |
| 157 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN |
| 158 | #endif |
| 159 | |
| 160 | #define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \ |
| 161 | <= SLAB_OBJ_MIN_SIZE) ? 1 : 0) |
| 162 | |
| 163 | #if FREELIST_BYTE_INDEX |
| 164 | typedef unsigned char freelist_idx_t; |
| 165 | #else |
| 166 | typedef unsigned short freelist_idx_t; |
| 167 | #endif |
| 168 | |
| 169 | #define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1) |
| 170 | |
| 171 | /* |
| 172 | * true if a page was allocated from pfmemalloc reserves for network-based |
| 173 | * swap |
| 174 | */ |
| 175 | static bool pfmemalloc_active __read_mostly; |
| 176 | |
| 177 | /* |
| 178 | * struct array_cache |
| 179 | * |
| 180 | * Purpose: |
| 181 | * - LIFO ordering, to hand out cache-warm objects from _alloc |
| 182 | * - reduce the number of linked list operations |
| 183 | * - reduce spinlock operations |
| 184 | * |
| 185 | * The limit is stored in the per-cpu structure to reduce the data cache |
| 186 | * footprint. |
| 187 | * |
| 188 | */ |
| 189 | struct array_cache { |
| 190 | unsigned int avail; |
| 191 | unsigned int limit; |
| 192 | unsigned int batchcount; |
| 193 | unsigned int touched; |
| 194 | void *entry[]; /* |
| 195 | * Must have this definition in here for the proper |
| 196 | * alignment of array_cache. Also simplifies accessing |
| 197 | * the entries. |
| 198 | * |
| 199 | * Entries should not be directly dereferenced as |
| 200 | * entries belonging to slabs marked pfmemalloc will |
| 201 | * have the lower bits set SLAB_OBJ_PFMEMALLOC |
| 202 | */ |
| 203 | }; |
| 204 | |
| 205 | struct alien_cache { |
| 206 | spinlock_t lock; |
| 207 | struct array_cache ac; |
| 208 | }; |
| 209 | |
| 210 | #define SLAB_OBJ_PFMEMALLOC 1 |
| 211 | static inline bool is_obj_pfmemalloc(void *objp) |
| 212 | { |
| 213 | return (unsigned long)objp & SLAB_OBJ_PFMEMALLOC; |
| 214 | } |
| 215 | |
| 216 | static inline void set_obj_pfmemalloc(void **objp) |
| 217 | { |
| 218 | *objp = (void *)((unsigned long)*objp | SLAB_OBJ_PFMEMALLOC); |
| 219 | return; |
| 220 | } |
| 221 | |
| 222 | static inline void clear_obj_pfmemalloc(void **objp) |
| 223 | { |
| 224 | *objp = (void *)((unsigned long)*objp & ~SLAB_OBJ_PFMEMALLOC); |
| 225 | } |
| 226 | |
| 227 | /* |
| 228 | * bootstrap: The caches do not work without cpuarrays anymore, but the |
| 229 | * cpuarrays are allocated from the generic caches... |
| 230 | */ |
| 231 | #define BOOT_CPUCACHE_ENTRIES 1 |
| 232 | struct arraycache_init { |
| 233 | struct array_cache cache; |
| 234 | void *entries[BOOT_CPUCACHE_ENTRIES]; |
| 235 | }; |
| 236 | |
| 237 | /* |
| 238 | * Need this for bootstrapping a per node allocator. |
| 239 | */ |
| 240 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES) |
| 241 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
| 242 | #define CACHE_CACHE 0 |
| 243 | #define SIZE_NODE (MAX_NUMNODES) |
| 244 | |
| 245 | static int drain_freelist(struct kmem_cache *cache, |
| 246 | struct kmem_cache_node *n, int tofree); |
| 247 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
| 248 | int node, struct list_head *list); |
| 249 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list); |
| 250 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
| 251 | static void cache_reap(struct work_struct *unused); |
| 252 | |
| 253 | static int slab_early_init = 1; |
| 254 | |
| 255 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
| 256 | |
| 257 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
| 258 | { |
| 259 | INIT_LIST_HEAD(&parent->slabs_full); |
| 260 | INIT_LIST_HEAD(&parent->slabs_partial); |
| 261 | INIT_LIST_HEAD(&parent->slabs_free); |
| 262 | parent->shared = NULL; |
| 263 | parent->alien = NULL; |
| 264 | parent->colour_next = 0; |
| 265 | spin_lock_init(&parent->list_lock); |
| 266 | parent->free_objects = 0; |
| 267 | parent->free_touched = 0; |
| 268 | } |
| 269 | |
| 270 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
| 271 | do { \ |
| 272 | INIT_LIST_HEAD(listp); \ |
| 273 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
| 274 | } while (0) |
| 275 | |
| 276 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
| 277 | do { \ |
| 278 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
| 279 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ |
| 280 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ |
| 281 | } while (0) |
| 282 | |
| 283 | #define CFLGS_OFF_SLAB (0x80000000UL) |
| 284 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
| 285 | |
| 286 | #define BATCHREFILL_LIMIT 16 |
| 287 | /* |
| 288 | * Optimization question: fewer reaps means less probability for unnessary |
| 289 | * cpucache drain/refill cycles. |
| 290 | * |
| 291 | * OTOH the cpuarrays can contain lots of objects, |
| 292 | * which could lock up otherwise freeable slabs. |
| 293 | */ |
| 294 | #define REAPTIMEOUT_AC (2*HZ) |
| 295 | #define REAPTIMEOUT_NODE (4*HZ) |
| 296 | |
| 297 | #if STATS |
| 298 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) |
| 299 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) |
| 300 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) |
| 301 | #define STATS_INC_GROWN(x) ((x)->grown++) |
| 302 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
| 303 | #define STATS_SET_HIGH(x) \ |
| 304 | do { \ |
| 305 | if ((x)->num_active > (x)->high_mark) \ |
| 306 | (x)->high_mark = (x)->num_active; \ |
| 307 | } while (0) |
| 308 | #define STATS_INC_ERR(x) ((x)->errors++) |
| 309 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) |
| 310 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
| 311 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
| 312 | #define STATS_SET_FREEABLE(x, i) \ |
| 313 | do { \ |
| 314 | if ((x)->max_freeable < i) \ |
| 315 | (x)->max_freeable = i; \ |
| 316 | } while (0) |
| 317 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
| 318 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) |
| 319 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) |
| 320 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) |
| 321 | #else |
| 322 | #define STATS_INC_ACTIVE(x) do { } while (0) |
| 323 | #define STATS_DEC_ACTIVE(x) do { } while (0) |
| 324 | #define STATS_INC_ALLOCED(x) do { } while (0) |
| 325 | #define STATS_INC_GROWN(x) do { } while (0) |
| 326 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
| 327 | #define STATS_SET_HIGH(x) do { } while (0) |
| 328 | #define STATS_INC_ERR(x) do { } while (0) |
| 329 | #define STATS_INC_NODEALLOCS(x) do { } while (0) |
| 330 | #define STATS_INC_NODEFREES(x) do { } while (0) |
| 331 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
| 332 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
| 333 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
| 334 | #define STATS_INC_ALLOCMISS(x) do { } while (0) |
| 335 | #define STATS_INC_FREEHIT(x) do { } while (0) |
| 336 | #define STATS_INC_FREEMISS(x) do { } while (0) |
| 337 | #endif |
| 338 | |
| 339 | #if DEBUG |
| 340 | |
| 341 | /* |
| 342 | * memory layout of objects: |
| 343 | * 0 : objp |
| 344 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
| 345 | * the end of an object is aligned with the end of the real |
| 346 | * allocation. Catches writes behind the end of the allocation. |
| 347 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
| 348 | * redzone word. |
| 349 | * cachep->obj_offset: The real object. |
| 350 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
| 351 | * cachep->size - 1* BYTES_PER_WORD: last caller address |
| 352 | * [BYTES_PER_WORD long] |
| 353 | */ |
| 354 | static int obj_offset(struct kmem_cache *cachep) |
| 355 | { |
| 356 | return cachep->obj_offset; |
| 357 | } |
| 358 | |
| 359 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
| 360 | { |
| 361 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); |
| 362 | return (unsigned long long*) (objp + obj_offset(cachep) - |
| 363 | sizeof(unsigned long long)); |
| 364 | } |
| 365 | |
| 366 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
| 367 | { |
| 368 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); |
| 369 | if (cachep->flags & SLAB_STORE_USER) |
| 370 | return (unsigned long long *)(objp + cachep->size - |
| 371 | sizeof(unsigned long long) - |
| 372 | REDZONE_ALIGN); |
| 373 | return (unsigned long long *) (objp + cachep->size - |
| 374 | sizeof(unsigned long long)); |
| 375 | } |
| 376 | |
| 377 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
| 378 | { |
| 379 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); |
| 380 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
| 381 | } |
| 382 | |
| 383 | #else |
| 384 | |
| 385 | #define obj_offset(x) 0 |
| 386 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
| 387 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
| 388 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
| 389 | |
| 390 | #endif |
| 391 | |
| 392 | #define OBJECT_FREE (0) |
| 393 | #define OBJECT_ACTIVE (1) |
| 394 | |
| 395 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
| 396 | |
| 397 | static void set_obj_status(struct page *page, int idx, int val) |
| 398 | { |
| 399 | int freelist_size; |
| 400 | char *status; |
| 401 | struct kmem_cache *cachep = page->slab_cache; |
| 402 | |
| 403 | freelist_size = cachep->num * sizeof(freelist_idx_t); |
| 404 | status = (char *)page->freelist + freelist_size; |
| 405 | status[idx] = val; |
| 406 | } |
| 407 | |
| 408 | static inline unsigned int get_obj_status(struct page *page, int idx) |
| 409 | { |
| 410 | int freelist_size; |
| 411 | char *status; |
| 412 | struct kmem_cache *cachep = page->slab_cache; |
| 413 | |
| 414 | freelist_size = cachep->num * sizeof(freelist_idx_t); |
| 415 | status = (char *)page->freelist + freelist_size; |
| 416 | |
| 417 | return status[idx]; |
| 418 | } |
| 419 | |
| 420 | #else |
| 421 | static inline void set_obj_status(struct page *page, int idx, int val) {} |
| 422 | |
| 423 | #endif |
| 424 | |
| 425 | /* |
| 426 | * Do not go above this order unless 0 objects fit into the slab or |
| 427 | * overridden on the command line. |
| 428 | */ |
| 429 | #define SLAB_MAX_ORDER_HI 1 |
| 430 | #define SLAB_MAX_ORDER_LO 0 |
| 431 | static int slab_max_order = SLAB_MAX_ORDER_LO; |
| 432 | static bool slab_max_order_set __initdata; |
| 433 | |
| 434 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
| 435 | { |
| 436 | struct page *page = virt_to_head_page(obj); |
| 437 | return page->slab_cache; |
| 438 | } |
| 439 | |
| 440 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
| 441 | unsigned int idx) |
| 442 | { |
| 443 | return page->s_mem + cache->size * idx; |
| 444 | } |
| 445 | |
| 446 | /* |
| 447 | * We want to avoid an expensive divide : (offset / cache->size) |
| 448 | * Using the fact that size is a constant for a particular cache, |
| 449 | * we can replace (offset / cache->size) by |
| 450 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
| 451 | */ |
| 452 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, |
| 453 | const struct page *page, void *obj) |
| 454 | { |
| 455 | u32 offset = (obj - page->s_mem); |
| 456 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
| 457 | } |
| 458 | |
| 459 | /* internal cache of cache description objs */ |
| 460 | static struct kmem_cache kmem_cache_boot = { |
| 461 | .batchcount = 1, |
| 462 | .limit = BOOT_CPUCACHE_ENTRIES, |
| 463 | .shared = 1, |
| 464 | .size = sizeof(struct kmem_cache), |
| 465 | .name = "kmem_cache", |
| 466 | }; |
| 467 | |
| 468 | #define BAD_ALIEN_MAGIC 0x01020304ul |
| 469 | |
| 470 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
| 471 | |
| 472 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
| 473 | { |
| 474 | return this_cpu_ptr(cachep->cpu_cache); |
| 475 | } |
| 476 | |
| 477 | static size_t calculate_freelist_size(int nr_objs, size_t align) |
| 478 | { |
| 479 | size_t freelist_size; |
| 480 | |
| 481 | freelist_size = nr_objs * sizeof(freelist_idx_t); |
| 482 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
| 483 | freelist_size += nr_objs * sizeof(char); |
| 484 | |
| 485 | if (align) |
| 486 | freelist_size = ALIGN(freelist_size, align); |
| 487 | |
| 488 | return freelist_size; |
| 489 | } |
| 490 | |
| 491 | static int calculate_nr_objs(size_t slab_size, size_t buffer_size, |
| 492 | size_t idx_size, size_t align) |
| 493 | { |
| 494 | int nr_objs; |
| 495 | size_t remained_size; |
| 496 | size_t freelist_size; |
| 497 | int extra_space = 0; |
| 498 | |
| 499 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
| 500 | extra_space = sizeof(char); |
| 501 | /* |
| 502 | * Ignore padding for the initial guess. The padding |
| 503 | * is at most @align-1 bytes, and @buffer_size is at |
| 504 | * least @align. In the worst case, this result will |
| 505 | * be one greater than the number of objects that fit |
| 506 | * into the memory allocation when taking the padding |
| 507 | * into account. |
| 508 | */ |
| 509 | nr_objs = slab_size / (buffer_size + idx_size + extra_space); |
| 510 | |
| 511 | /* |
| 512 | * This calculated number will be either the right |
| 513 | * amount, or one greater than what we want. |
| 514 | */ |
| 515 | remained_size = slab_size - nr_objs * buffer_size; |
| 516 | freelist_size = calculate_freelist_size(nr_objs, align); |
| 517 | if (remained_size < freelist_size) |
| 518 | nr_objs--; |
| 519 | |
| 520 | return nr_objs; |
| 521 | } |
| 522 | |
| 523 | /* |
| 524 | * Calculate the number of objects and left-over bytes for a given buffer size. |
| 525 | */ |
| 526 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
| 527 | size_t align, int flags, size_t *left_over, |
| 528 | unsigned int *num) |
| 529 | { |
| 530 | int nr_objs; |
| 531 | size_t mgmt_size; |
| 532 | size_t slab_size = PAGE_SIZE << gfporder; |
| 533 | |
| 534 | /* |
| 535 | * The slab management structure can be either off the slab or |
| 536 | * on it. For the latter case, the memory allocated for a |
| 537 | * slab is used for: |
| 538 | * |
| 539 | * - One unsigned int for each object |
| 540 | * - Padding to respect alignment of @align |
| 541 | * - @buffer_size bytes for each object |
| 542 | * |
| 543 | * If the slab management structure is off the slab, then the |
| 544 | * alignment will already be calculated into the size. Because |
| 545 | * the slabs are all pages aligned, the objects will be at the |
| 546 | * correct alignment when allocated. |
| 547 | */ |
| 548 | if (flags & CFLGS_OFF_SLAB) { |
| 549 | mgmt_size = 0; |
| 550 | nr_objs = slab_size / buffer_size; |
| 551 | |
| 552 | } else { |
| 553 | nr_objs = calculate_nr_objs(slab_size, buffer_size, |
| 554 | sizeof(freelist_idx_t), align); |
| 555 | mgmt_size = calculate_freelist_size(nr_objs, align); |
| 556 | } |
| 557 | *num = nr_objs; |
| 558 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; |
| 559 | } |
| 560 | |
| 561 | #if DEBUG |
| 562 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
| 563 | |
| 564 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
| 565 | char *msg) |
| 566 | { |
| 567 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", |
| 568 | function, cachep->name, msg); |
| 569 | dump_stack(); |
| 570 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
| 571 | } |
| 572 | #endif |
| 573 | |
| 574 | /* |
| 575 | * By default on NUMA we use alien caches to stage the freeing of |
| 576 | * objects allocated from other nodes. This causes massive memory |
| 577 | * inefficiencies when using fake NUMA setup to split memory into a |
| 578 | * large number of small nodes, so it can be disabled on the command |
| 579 | * line |
| 580 | */ |
| 581 | |
| 582 | static int use_alien_caches __read_mostly = 1; |
| 583 | static int __init noaliencache_setup(char *s) |
| 584 | { |
| 585 | use_alien_caches = 0; |
| 586 | return 1; |
| 587 | } |
| 588 | __setup("noaliencache", noaliencache_setup); |
| 589 | |
| 590 | static int __init slab_max_order_setup(char *str) |
| 591 | { |
| 592 | get_option(&str, &slab_max_order); |
| 593 | slab_max_order = slab_max_order < 0 ? 0 : |
| 594 | min(slab_max_order, MAX_ORDER - 1); |
| 595 | slab_max_order_set = true; |
| 596 | |
| 597 | return 1; |
| 598 | } |
| 599 | __setup("slab_max_order=", slab_max_order_setup); |
| 600 | |
| 601 | #ifdef CONFIG_NUMA |
| 602 | /* |
| 603 | * Special reaping functions for NUMA systems called from cache_reap(). |
| 604 | * These take care of doing round robin flushing of alien caches (containing |
| 605 | * objects freed on different nodes from which they were allocated) and the |
| 606 | * flushing of remote pcps by calling drain_node_pages. |
| 607 | */ |
| 608 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
| 609 | |
| 610 | static void init_reap_node(int cpu) |
| 611 | { |
| 612 | int node; |
| 613 | |
| 614 | node = next_node(cpu_to_mem(cpu), node_online_map); |
| 615 | if (node == MAX_NUMNODES) |
| 616 | node = first_node(node_online_map); |
| 617 | |
| 618 | per_cpu(slab_reap_node, cpu) = node; |
| 619 | } |
| 620 | |
| 621 | static void next_reap_node(void) |
| 622 | { |
| 623 | int node = __this_cpu_read(slab_reap_node); |
| 624 | |
| 625 | node = next_node(node, node_online_map); |
| 626 | if (unlikely(node >= MAX_NUMNODES)) |
| 627 | node = first_node(node_online_map); |
| 628 | __this_cpu_write(slab_reap_node, node); |
| 629 | } |
| 630 | |
| 631 | #else |
| 632 | #define init_reap_node(cpu) do { } while (0) |
| 633 | #define next_reap_node(void) do { } while (0) |
| 634 | #endif |
| 635 | |
| 636 | /* |
| 637 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz |
| 638 | * via the workqueue/eventd. |
| 639 | * Add the CPU number into the expiration time to minimize the possibility of |
| 640 | * the CPUs getting into lockstep and contending for the global cache chain |
| 641 | * lock. |
| 642 | */ |
| 643 | static void start_cpu_timer(int cpu) |
| 644 | { |
| 645 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
| 646 | |
| 647 | /* |
| 648 | * When this gets called from do_initcalls via cpucache_init(), |
| 649 | * init_workqueues() has already run, so keventd will be setup |
| 650 | * at that time. |
| 651 | */ |
| 652 | if (keventd_up() && reap_work->work.func == NULL) { |
| 653 | init_reap_node(cpu); |
| 654 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
| 655 | schedule_delayed_work_on(cpu, reap_work, |
| 656 | __round_jiffies_relative(HZ, cpu)); |
| 657 | } |
| 658 | } |
| 659 | |
| 660 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
| 661 | { |
| 662 | /* |
| 663 | * The array_cache structures contain pointers to free object. |
| 664 | * However, when such objects are allocated or transferred to another |
| 665 | * cache the pointers are not cleared and they could be counted as |
| 666 | * valid references during a kmemleak scan. Therefore, kmemleak must |
| 667 | * not scan such objects. |
| 668 | */ |
| 669 | kmemleak_no_scan(ac); |
| 670 | if (ac) { |
| 671 | ac->avail = 0; |
| 672 | ac->limit = limit; |
| 673 | ac->batchcount = batch; |
| 674 | ac->touched = 0; |
| 675 | } |
| 676 | } |
| 677 | |
| 678 | static struct array_cache *alloc_arraycache(int node, int entries, |
| 679 | int batchcount, gfp_t gfp) |
| 680 | { |
| 681 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
| 682 | struct array_cache *ac = NULL; |
| 683 | |
| 684 | ac = kmalloc_node(memsize, gfp, node); |
| 685 | init_arraycache(ac, entries, batchcount); |
| 686 | return ac; |
| 687 | } |
| 688 | |
| 689 | static inline bool is_slab_pfmemalloc(struct page *page) |
| 690 | { |
| 691 | return PageSlabPfmemalloc(page); |
| 692 | } |
| 693 | |
| 694 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ |
| 695 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, |
| 696 | struct array_cache *ac) |
| 697 | { |
| 698 | struct kmem_cache_node *n = get_node(cachep, numa_mem_id()); |
| 699 | struct page *page; |
| 700 | unsigned long flags; |
| 701 | |
| 702 | if (!pfmemalloc_active) |
| 703 | return; |
| 704 | |
| 705 | spin_lock_irqsave(&n->list_lock, flags); |
| 706 | list_for_each_entry(page, &n->slabs_full, lru) |
| 707 | if (is_slab_pfmemalloc(page)) |
| 708 | goto out; |
| 709 | |
| 710 | list_for_each_entry(page, &n->slabs_partial, lru) |
| 711 | if (is_slab_pfmemalloc(page)) |
| 712 | goto out; |
| 713 | |
| 714 | list_for_each_entry(page, &n->slabs_free, lru) |
| 715 | if (is_slab_pfmemalloc(page)) |
| 716 | goto out; |
| 717 | |
| 718 | pfmemalloc_active = false; |
| 719 | out: |
| 720 | spin_unlock_irqrestore(&n->list_lock, flags); |
| 721 | } |
| 722 | |
| 723 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
| 724 | gfp_t flags, bool force_refill) |
| 725 | { |
| 726 | int i; |
| 727 | void *objp = ac->entry[--ac->avail]; |
| 728 | |
| 729 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ |
| 730 | if (unlikely(is_obj_pfmemalloc(objp))) { |
| 731 | struct kmem_cache_node *n; |
| 732 | |
| 733 | if (gfp_pfmemalloc_allowed(flags)) { |
| 734 | clear_obj_pfmemalloc(&objp); |
| 735 | return objp; |
| 736 | } |
| 737 | |
| 738 | /* The caller cannot use PFMEMALLOC objects, find another one */ |
| 739 | for (i = 0; i < ac->avail; i++) { |
| 740 | /* If a !PFMEMALLOC object is found, swap them */ |
| 741 | if (!is_obj_pfmemalloc(ac->entry[i])) { |
| 742 | objp = ac->entry[i]; |
| 743 | ac->entry[i] = ac->entry[ac->avail]; |
| 744 | ac->entry[ac->avail] = objp; |
| 745 | return objp; |
| 746 | } |
| 747 | } |
| 748 | |
| 749 | /* |
| 750 | * If there are empty slabs on the slabs_free list and we are |
| 751 | * being forced to refill the cache, mark this one !pfmemalloc. |
| 752 | */ |
| 753 | n = get_node(cachep, numa_mem_id()); |
| 754 | if (!list_empty(&n->slabs_free) && force_refill) { |
| 755 | struct page *page = virt_to_head_page(objp); |
| 756 | ClearPageSlabPfmemalloc(page); |
| 757 | clear_obj_pfmemalloc(&objp); |
| 758 | recheck_pfmemalloc_active(cachep, ac); |
| 759 | return objp; |
| 760 | } |
| 761 | |
| 762 | /* No !PFMEMALLOC objects available */ |
| 763 | ac->avail++; |
| 764 | objp = NULL; |
| 765 | } |
| 766 | |
| 767 | return objp; |
| 768 | } |
| 769 | |
| 770 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
| 771 | struct array_cache *ac, gfp_t flags, bool force_refill) |
| 772 | { |
| 773 | void *objp; |
| 774 | |
| 775 | if (unlikely(sk_memalloc_socks())) |
| 776 | objp = __ac_get_obj(cachep, ac, flags, force_refill); |
| 777 | else |
| 778 | objp = ac->entry[--ac->avail]; |
| 779 | |
| 780 | return objp; |
| 781 | } |
| 782 | |
| 783 | static noinline void *__ac_put_obj(struct kmem_cache *cachep, |
| 784 | struct array_cache *ac, void *objp) |
| 785 | { |
| 786 | if (unlikely(pfmemalloc_active)) { |
| 787 | /* Some pfmemalloc slabs exist, check if this is one */ |
| 788 | struct page *page = virt_to_head_page(objp); |
| 789 | if (PageSlabPfmemalloc(page)) |
| 790 | set_obj_pfmemalloc(&objp); |
| 791 | } |
| 792 | |
| 793 | return objp; |
| 794 | } |
| 795 | |
| 796 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, |
| 797 | void *objp) |
| 798 | { |
| 799 | if (unlikely(sk_memalloc_socks())) |
| 800 | objp = __ac_put_obj(cachep, ac, objp); |
| 801 | |
| 802 | ac->entry[ac->avail++] = objp; |
| 803 | } |
| 804 | |
| 805 | /* |
| 806 | * Transfer objects in one arraycache to another. |
| 807 | * Locking must be handled by the caller. |
| 808 | * |
| 809 | * Return the number of entries transferred. |
| 810 | */ |
| 811 | static int transfer_objects(struct array_cache *to, |
| 812 | struct array_cache *from, unsigned int max) |
| 813 | { |
| 814 | /* Figure out how many entries to transfer */ |
| 815 | int nr = min3(from->avail, max, to->limit - to->avail); |
| 816 | |
| 817 | if (!nr) |
| 818 | return 0; |
| 819 | |
| 820 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, |
| 821 | sizeof(void *) *nr); |
| 822 | |
| 823 | from->avail -= nr; |
| 824 | to->avail += nr; |
| 825 | return nr; |
| 826 | } |
| 827 | |
| 828 | #ifndef CONFIG_NUMA |
| 829 | |
| 830 | #define drain_alien_cache(cachep, alien) do { } while (0) |
| 831 | #define reap_alien(cachep, n) do { } while (0) |
| 832 | |
| 833 | static inline struct alien_cache **alloc_alien_cache(int node, |
| 834 | int limit, gfp_t gfp) |
| 835 | { |
| 836 | return (struct alien_cache **)BAD_ALIEN_MAGIC; |
| 837 | } |
| 838 | |
| 839 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
| 840 | { |
| 841 | } |
| 842 | |
| 843 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
| 844 | { |
| 845 | return 0; |
| 846 | } |
| 847 | |
| 848 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, |
| 849 | gfp_t flags) |
| 850 | { |
| 851 | return NULL; |
| 852 | } |
| 853 | |
| 854 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
| 855 | gfp_t flags, int nodeid) |
| 856 | { |
| 857 | return NULL; |
| 858 | } |
| 859 | |
| 860 | static inline gfp_t gfp_exact_node(gfp_t flags) |
| 861 | { |
| 862 | return flags; |
| 863 | } |
| 864 | |
| 865 | #else /* CONFIG_NUMA */ |
| 866 | |
| 867 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
| 868 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
| 869 | |
| 870 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
| 871 | int batch, gfp_t gfp) |
| 872 | { |
| 873 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
| 874 | struct alien_cache *alc = NULL; |
| 875 | |
| 876 | alc = kmalloc_node(memsize, gfp, node); |
| 877 | init_arraycache(&alc->ac, entries, batch); |
| 878 | spin_lock_init(&alc->lock); |
| 879 | return alc; |
| 880 | } |
| 881 | |
| 882 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
| 883 | { |
| 884 | struct alien_cache **alc_ptr; |
| 885 | size_t memsize = sizeof(void *) * nr_node_ids; |
| 886 | int i; |
| 887 | |
| 888 | if (limit > 1) |
| 889 | limit = 12; |
| 890 | alc_ptr = kzalloc_node(memsize, gfp, node); |
| 891 | if (!alc_ptr) |
| 892 | return NULL; |
| 893 | |
| 894 | for_each_node(i) { |
| 895 | if (i == node || !node_online(i)) |
| 896 | continue; |
| 897 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); |
| 898 | if (!alc_ptr[i]) { |
| 899 | for (i--; i >= 0; i--) |
| 900 | kfree(alc_ptr[i]); |
| 901 | kfree(alc_ptr); |
| 902 | return NULL; |
| 903 | } |
| 904 | } |
| 905 | return alc_ptr; |
| 906 | } |
| 907 | |
| 908 | static void free_alien_cache(struct alien_cache **alc_ptr) |
| 909 | { |
| 910 | int i; |
| 911 | |
| 912 | if (!alc_ptr) |
| 913 | return; |
| 914 | for_each_node(i) |
| 915 | kfree(alc_ptr[i]); |
| 916 | kfree(alc_ptr); |
| 917 | } |
| 918 | |
| 919 | static void __drain_alien_cache(struct kmem_cache *cachep, |
| 920 | struct array_cache *ac, int node, |
| 921 | struct list_head *list) |
| 922 | { |
| 923 | struct kmem_cache_node *n = get_node(cachep, node); |
| 924 | |
| 925 | if (ac->avail) { |
| 926 | spin_lock(&n->list_lock); |
| 927 | /* |
| 928 | * Stuff objects into the remote nodes shared array first. |
| 929 | * That way we could avoid the overhead of putting the objects |
| 930 | * into the free lists and getting them back later. |
| 931 | */ |
| 932 | if (n->shared) |
| 933 | transfer_objects(n->shared, ac, ac->limit); |
| 934 | |
| 935 | free_block(cachep, ac->entry, ac->avail, node, list); |
| 936 | ac->avail = 0; |
| 937 | spin_unlock(&n->list_lock); |
| 938 | } |
| 939 | } |
| 940 | |
| 941 | /* |
| 942 | * Called from cache_reap() to regularly drain alien caches round robin. |
| 943 | */ |
| 944 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
| 945 | { |
| 946 | int node = __this_cpu_read(slab_reap_node); |
| 947 | |
| 948 | if (n->alien) { |
| 949 | struct alien_cache *alc = n->alien[node]; |
| 950 | struct array_cache *ac; |
| 951 | |
| 952 | if (alc) { |
| 953 | ac = &alc->ac; |
| 954 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
| 955 | LIST_HEAD(list); |
| 956 | |
| 957 | __drain_alien_cache(cachep, ac, node, &list); |
| 958 | spin_unlock_irq(&alc->lock); |
| 959 | slabs_destroy(cachep, &list); |
| 960 | } |
| 961 | } |
| 962 | } |
| 963 | } |
| 964 | |
| 965 | static void drain_alien_cache(struct kmem_cache *cachep, |
| 966 | struct alien_cache **alien) |
| 967 | { |
| 968 | int i = 0; |
| 969 | struct alien_cache *alc; |
| 970 | struct array_cache *ac; |
| 971 | unsigned long flags; |
| 972 | |
| 973 | for_each_online_node(i) { |
| 974 | alc = alien[i]; |
| 975 | if (alc) { |
| 976 | LIST_HEAD(list); |
| 977 | |
| 978 | ac = &alc->ac; |
| 979 | spin_lock_irqsave(&alc->lock, flags); |
| 980 | __drain_alien_cache(cachep, ac, i, &list); |
| 981 | spin_unlock_irqrestore(&alc->lock, flags); |
| 982 | slabs_destroy(cachep, &list); |
| 983 | } |
| 984 | } |
| 985 | } |
| 986 | |
| 987 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
| 988 | int node, int page_node) |
| 989 | { |
| 990 | struct kmem_cache_node *n; |
| 991 | struct alien_cache *alien = NULL; |
| 992 | struct array_cache *ac; |
| 993 | LIST_HEAD(list); |
| 994 | |
| 995 | n = get_node(cachep, node); |
| 996 | STATS_INC_NODEFREES(cachep); |
| 997 | if (n->alien && n->alien[page_node]) { |
| 998 | alien = n->alien[page_node]; |
| 999 | ac = &alien->ac; |
| 1000 | spin_lock(&alien->lock); |
| 1001 | if (unlikely(ac->avail == ac->limit)) { |
| 1002 | STATS_INC_ACOVERFLOW(cachep); |
| 1003 | __drain_alien_cache(cachep, ac, page_node, &list); |
| 1004 | } |
| 1005 | ac_put_obj(cachep, ac, objp); |
| 1006 | spin_unlock(&alien->lock); |
| 1007 | slabs_destroy(cachep, &list); |
| 1008 | } else { |
| 1009 | n = get_node(cachep, page_node); |
| 1010 | spin_lock(&n->list_lock); |
| 1011 | free_block(cachep, &objp, 1, page_node, &list); |
| 1012 | spin_unlock(&n->list_lock); |
| 1013 | slabs_destroy(cachep, &list); |
| 1014 | } |
| 1015 | return 1; |
| 1016 | } |
| 1017 | |
| 1018 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
| 1019 | { |
| 1020 | int page_node = page_to_nid(virt_to_page(objp)); |
| 1021 | int node = numa_mem_id(); |
| 1022 | /* |
| 1023 | * Make sure we are not freeing a object from another node to the array |
| 1024 | * cache on this cpu. |
| 1025 | */ |
| 1026 | if (likely(node == page_node)) |
| 1027 | return 0; |
| 1028 | |
| 1029 | return __cache_free_alien(cachep, objp, node, page_node); |
| 1030 | } |
| 1031 | |
| 1032 | /* |
| 1033 | * Construct gfp mask to allocate from a specific node but do not invoke reclaim |
| 1034 | * or warn about failures. |
| 1035 | */ |
| 1036 | static inline gfp_t gfp_exact_node(gfp_t flags) |
| 1037 | { |
| 1038 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~__GFP_WAIT; |
| 1039 | } |
| 1040 | #endif |
| 1041 | |
| 1042 | /* |
| 1043 | * Allocates and initializes node for a node on each slab cache, used for |
| 1044 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
| 1045 | * will be allocated off-node since memory is not yet online for the new node. |
| 1046 | * When hotplugging memory or a cpu, existing node are not replaced if |
| 1047 | * already in use. |
| 1048 | * |
| 1049 | * Must hold slab_mutex. |
| 1050 | */ |
| 1051 | static int init_cache_node_node(int node) |
| 1052 | { |
| 1053 | struct kmem_cache *cachep; |
| 1054 | struct kmem_cache_node *n; |
| 1055 | const size_t memsize = sizeof(struct kmem_cache_node); |
| 1056 | |
| 1057 | list_for_each_entry(cachep, &slab_caches, list) { |
| 1058 | /* |
| 1059 | * Set up the kmem_cache_node for cpu before we can |
| 1060 | * begin anything. Make sure some other cpu on this |
| 1061 | * node has not already allocated this |
| 1062 | */ |
| 1063 | n = get_node(cachep, node); |
| 1064 | if (!n) { |
| 1065 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
| 1066 | if (!n) |
| 1067 | return -ENOMEM; |
| 1068 | kmem_cache_node_init(n); |
| 1069 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
| 1070 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; |
| 1071 | |
| 1072 | /* |
| 1073 | * The kmem_cache_nodes don't come and go as CPUs |
| 1074 | * come and go. slab_mutex is sufficient |
| 1075 | * protection here. |
| 1076 | */ |
| 1077 | cachep->node[node] = n; |
| 1078 | } |
| 1079 | |
| 1080 | spin_lock_irq(&n->list_lock); |
| 1081 | n->free_limit = |
| 1082 | (1 + nr_cpus_node(node)) * |
| 1083 | cachep->batchcount + cachep->num; |
| 1084 | spin_unlock_irq(&n->list_lock); |
| 1085 | } |
| 1086 | return 0; |
| 1087 | } |
| 1088 | |
| 1089 | static inline int slabs_tofree(struct kmem_cache *cachep, |
| 1090 | struct kmem_cache_node *n) |
| 1091 | { |
| 1092 | return (n->free_objects + cachep->num - 1) / cachep->num; |
| 1093 | } |
| 1094 | |
| 1095 | static void cpuup_canceled(long cpu) |
| 1096 | { |
| 1097 | struct kmem_cache *cachep; |
| 1098 | struct kmem_cache_node *n = NULL; |
| 1099 | int node = cpu_to_mem(cpu); |
| 1100 | const struct cpumask *mask = cpumask_of_node(node); |
| 1101 | |
| 1102 | list_for_each_entry(cachep, &slab_caches, list) { |
| 1103 | struct array_cache *nc; |
| 1104 | struct array_cache *shared; |
| 1105 | struct alien_cache **alien; |
| 1106 | LIST_HEAD(list); |
| 1107 | |
| 1108 | n = get_node(cachep, node); |
| 1109 | if (!n) |
| 1110 | continue; |
| 1111 | |
| 1112 | spin_lock_irq(&n->list_lock); |
| 1113 | |
| 1114 | /* Free limit for this kmem_cache_node */ |
| 1115 | n->free_limit -= cachep->batchcount; |
| 1116 | |
| 1117 | /* cpu is dead; no one can alloc from it. */ |
| 1118 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); |
| 1119 | if (nc) { |
| 1120 | free_block(cachep, nc->entry, nc->avail, node, &list); |
| 1121 | nc->avail = 0; |
| 1122 | } |
| 1123 | |
| 1124 | if (!cpumask_empty(mask)) { |
| 1125 | spin_unlock_irq(&n->list_lock); |
| 1126 | goto free_slab; |
| 1127 | } |
| 1128 | |
| 1129 | shared = n->shared; |
| 1130 | if (shared) { |
| 1131 | free_block(cachep, shared->entry, |
| 1132 | shared->avail, node, &list); |
| 1133 | n->shared = NULL; |
| 1134 | } |
| 1135 | |
| 1136 | alien = n->alien; |
| 1137 | n->alien = NULL; |
| 1138 | |
| 1139 | spin_unlock_irq(&n->list_lock); |
| 1140 | |
| 1141 | kfree(shared); |
| 1142 | if (alien) { |
| 1143 | drain_alien_cache(cachep, alien); |
| 1144 | free_alien_cache(alien); |
| 1145 | } |
| 1146 | |
| 1147 | free_slab: |
| 1148 | slabs_destroy(cachep, &list); |
| 1149 | } |
| 1150 | /* |
| 1151 | * In the previous loop, all the objects were freed to |
| 1152 | * the respective cache's slabs, now we can go ahead and |
| 1153 | * shrink each nodelist to its limit. |
| 1154 | */ |
| 1155 | list_for_each_entry(cachep, &slab_caches, list) { |
| 1156 | n = get_node(cachep, node); |
| 1157 | if (!n) |
| 1158 | continue; |
| 1159 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
| 1160 | } |
| 1161 | } |
| 1162 | |
| 1163 | static int cpuup_prepare(long cpu) |
| 1164 | { |
| 1165 | struct kmem_cache *cachep; |
| 1166 | struct kmem_cache_node *n = NULL; |
| 1167 | int node = cpu_to_mem(cpu); |
| 1168 | int err; |
| 1169 | |
| 1170 | /* |
| 1171 | * We need to do this right in the beginning since |
| 1172 | * alloc_arraycache's are going to use this list. |
| 1173 | * kmalloc_node allows us to add the slab to the right |
| 1174 | * kmem_cache_node and not this cpu's kmem_cache_node |
| 1175 | */ |
| 1176 | err = init_cache_node_node(node); |
| 1177 | if (err < 0) |
| 1178 | goto bad; |
| 1179 | |
| 1180 | /* |
| 1181 | * Now we can go ahead with allocating the shared arrays and |
| 1182 | * array caches |
| 1183 | */ |
| 1184 | list_for_each_entry(cachep, &slab_caches, list) { |
| 1185 | struct array_cache *shared = NULL; |
| 1186 | struct alien_cache **alien = NULL; |
| 1187 | |
| 1188 | if (cachep->shared) { |
| 1189 | shared = alloc_arraycache(node, |
| 1190 | cachep->shared * cachep->batchcount, |
| 1191 | 0xbaadf00d, GFP_KERNEL); |
| 1192 | if (!shared) |
| 1193 | goto bad; |
| 1194 | } |
| 1195 | if (use_alien_caches) { |
| 1196 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
| 1197 | if (!alien) { |
| 1198 | kfree(shared); |
| 1199 | goto bad; |
| 1200 | } |
| 1201 | } |
| 1202 | n = get_node(cachep, node); |
| 1203 | BUG_ON(!n); |
| 1204 | |
| 1205 | spin_lock_irq(&n->list_lock); |
| 1206 | if (!n->shared) { |
| 1207 | /* |
| 1208 | * We are serialised from CPU_DEAD or |
| 1209 | * CPU_UP_CANCELLED by the cpucontrol lock |
| 1210 | */ |
| 1211 | n->shared = shared; |
| 1212 | shared = NULL; |
| 1213 | } |
| 1214 | #ifdef CONFIG_NUMA |
| 1215 | if (!n->alien) { |
| 1216 | n->alien = alien; |
| 1217 | alien = NULL; |
| 1218 | } |
| 1219 | #endif |
| 1220 | spin_unlock_irq(&n->list_lock); |
| 1221 | kfree(shared); |
| 1222 | free_alien_cache(alien); |
| 1223 | } |
| 1224 | |
| 1225 | return 0; |
| 1226 | bad: |
| 1227 | cpuup_canceled(cpu); |
| 1228 | return -ENOMEM; |
| 1229 | } |
| 1230 | |
| 1231 | static int cpuup_callback(struct notifier_block *nfb, |
| 1232 | unsigned long action, void *hcpu) |
| 1233 | { |
| 1234 | long cpu = (long)hcpu; |
| 1235 | int err = 0; |
| 1236 | |
| 1237 | switch (action) { |
| 1238 | case CPU_UP_PREPARE: |
| 1239 | case CPU_UP_PREPARE_FROZEN: |
| 1240 | mutex_lock(&slab_mutex); |
| 1241 | err = cpuup_prepare(cpu); |
| 1242 | mutex_unlock(&slab_mutex); |
| 1243 | break; |
| 1244 | case CPU_ONLINE: |
| 1245 | case CPU_ONLINE_FROZEN: |
| 1246 | start_cpu_timer(cpu); |
| 1247 | break; |
| 1248 | #ifdef CONFIG_HOTPLUG_CPU |
| 1249 | case CPU_DOWN_PREPARE: |
| 1250 | case CPU_DOWN_PREPARE_FROZEN: |
| 1251 | /* |
| 1252 | * Shutdown cache reaper. Note that the slab_mutex is |
| 1253 | * held so that if cache_reap() is invoked it cannot do |
| 1254 | * anything expensive but will only modify reap_work |
| 1255 | * and reschedule the timer. |
| 1256 | */ |
| 1257 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
| 1258 | /* Now the cache_reaper is guaranteed to be not running. */ |
| 1259 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
| 1260 | break; |
| 1261 | case CPU_DOWN_FAILED: |
| 1262 | case CPU_DOWN_FAILED_FROZEN: |
| 1263 | start_cpu_timer(cpu); |
| 1264 | break; |
| 1265 | case CPU_DEAD: |
| 1266 | case CPU_DEAD_FROZEN: |
| 1267 | /* |
| 1268 | * Even if all the cpus of a node are down, we don't free the |
| 1269 | * kmem_cache_node of any cache. This to avoid a race between |
| 1270 | * cpu_down, and a kmalloc allocation from another cpu for |
| 1271 | * memory from the node of the cpu going down. The node |
| 1272 | * structure is usually allocated from kmem_cache_create() and |
| 1273 | * gets destroyed at kmem_cache_destroy(). |
| 1274 | */ |
| 1275 | /* fall through */ |
| 1276 | #endif |
| 1277 | case CPU_UP_CANCELED: |
| 1278 | case CPU_UP_CANCELED_FROZEN: |
| 1279 | mutex_lock(&slab_mutex); |
| 1280 | cpuup_canceled(cpu); |
| 1281 | mutex_unlock(&slab_mutex); |
| 1282 | break; |
| 1283 | } |
| 1284 | return notifier_from_errno(err); |
| 1285 | } |
| 1286 | |
| 1287 | static struct notifier_block cpucache_notifier = { |
| 1288 | &cpuup_callback, NULL, 0 |
| 1289 | }; |
| 1290 | |
| 1291 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
| 1292 | /* |
| 1293 | * Drains freelist for a node on each slab cache, used for memory hot-remove. |
| 1294 | * Returns -EBUSY if all objects cannot be drained so that the node is not |
| 1295 | * removed. |
| 1296 | * |
| 1297 | * Must hold slab_mutex. |
| 1298 | */ |
| 1299 | static int __meminit drain_cache_node_node(int node) |
| 1300 | { |
| 1301 | struct kmem_cache *cachep; |
| 1302 | int ret = 0; |
| 1303 | |
| 1304 | list_for_each_entry(cachep, &slab_caches, list) { |
| 1305 | struct kmem_cache_node *n; |
| 1306 | |
| 1307 | n = get_node(cachep, node); |
| 1308 | if (!n) |
| 1309 | continue; |
| 1310 | |
| 1311 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
| 1312 | |
| 1313 | if (!list_empty(&n->slabs_full) || |
| 1314 | !list_empty(&n->slabs_partial)) { |
| 1315 | ret = -EBUSY; |
| 1316 | break; |
| 1317 | } |
| 1318 | } |
| 1319 | return ret; |
| 1320 | } |
| 1321 | |
| 1322 | static int __meminit slab_memory_callback(struct notifier_block *self, |
| 1323 | unsigned long action, void *arg) |
| 1324 | { |
| 1325 | struct memory_notify *mnb = arg; |
| 1326 | int ret = 0; |
| 1327 | int nid; |
| 1328 | |
| 1329 | nid = mnb->status_change_nid; |
| 1330 | if (nid < 0) |
| 1331 | goto out; |
| 1332 | |
| 1333 | switch (action) { |
| 1334 | case MEM_GOING_ONLINE: |
| 1335 | mutex_lock(&slab_mutex); |
| 1336 | ret = init_cache_node_node(nid); |
| 1337 | mutex_unlock(&slab_mutex); |
| 1338 | break; |
| 1339 | case MEM_GOING_OFFLINE: |
| 1340 | mutex_lock(&slab_mutex); |
| 1341 | ret = drain_cache_node_node(nid); |
| 1342 | mutex_unlock(&slab_mutex); |
| 1343 | break; |
| 1344 | case MEM_ONLINE: |
| 1345 | case MEM_OFFLINE: |
| 1346 | case MEM_CANCEL_ONLINE: |
| 1347 | case MEM_CANCEL_OFFLINE: |
| 1348 | break; |
| 1349 | } |
| 1350 | out: |
| 1351 | return notifier_from_errno(ret); |
| 1352 | } |
| 1353 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ |
| 1354 | |
| 1355 | /* |
| 1356 | * swap the static kmem_cache_node with kmalloced memory |
| 1357 | */ |
| 1358 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
| 1359 | int nodeid) |
| 1360 | { |
| 1361 | struct kmem_cache_node *ptr; |
| 1362 | |
| 1363 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
| 1364 | BUG_ON(!ptr); |
| 1365 | |
| 1366 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
| 1367 | /* |
| 1368 | * Do not assume that spinlocks can be initialized via memcpy: |
| 1369 | */ |
| 1370 | spin_lock_init(&ptr->list_lock); |
| 1371 | |
| 1372 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
| 1373 | cachep->node[nodeid] = ptr; |
| 1374 | } |
| 1375 | |
| 1376 | /* |
| 1377 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
| 1378 | * size of kmem_cache_node. |
| 1379 | */ |
| 1380 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
| 1381 | { |
| 1382 | int node; |
| 1383 | |
| 1384 | for_each_online_node(node) { |
| 1385 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
| 1386 | cachep->node[node]->next_reap = jiffies + |
| 1387 | REAPTIMEOUT_NODE + |
| 1388 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; |
| 1389 | } |
| 1390 | } |
| 1391 | |
| 1392 | /* |
| 1393 | * Initialisation. Called after the page allocator have been initialised and |
| 1394 | * before smp_init(). |
| 1395 | */ |
| 1396 | void __init kmem_cache_init(void) |
| 1397 | { |
| 1398 | int i; |
| 1399 | |
| 1400 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
| 1401 | sizeof(struct rcu_head)); |
| 1402 | kmem_cache = &kmem_cache_boot; |
| 1403 | |
| 1404 | if (num_possible_nodes() == 1) |
| 1405 | use_alien_caches = 0; |
| 1406 | |
| 1407 | for (i = 0; i < NUM_INIT_LISTS; i++) |
| 1408 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
| 1409 | |
| 1410 | /* |
| 1411 | * Fragmentation resistance on low memory - only use bigger |
| 1412 | * page orders on machines with more than 32MB of memory if |
| 1413 | * not overridden on the command line. |
| 1414 | */ |
| 1415 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
| 1416 | slab_max_order = SLAB_MAX_ORDER_HI; |
| 1417 | |
| 1418 | /* Bootstrap is tricky, because several objects are allocated |
| 1419 | * from caches that do not exist yet: |
| 1420 | * 1) initialize the kmem_cache cache: it contains the struct |
| 1421 | * kmem_cache structures of all caches, except kmem_cache itself: |
| 1422 | * kmem_cache is statically allocated. |
| 1423 | * Initially an __init data area is used for the head array and the |
| 1424 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
| 1425 | * array at the end of the bootstrap. |
| 1426 | * 2) Create the first kmalloc cache. |
| 1427 | * The struct kmem_cache for the new cache is allocated normally. |
| 1428 | * An __init data area is used for the head array. |
| 1429 | * 3) Create the remaining kmalloc caches, with minimally sized |
| 1430 | * head arrays. |
| 1431 | * 4) Replace the __init data head arrays for kmem_cache and the first |
| 1432 | * kmalloc cache with kmalloc allocated arrays. |
| 1433 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
| 1434 | * the other cache's with kmalloc allocated memory. |
| 1435 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. |
| 1436 | */ |
| 1437 | |
| 1438 | /* 1) create the kmem_cache */ |
| 1439 | |
| 1440 | /* |
| 1441 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
| 1442 | */ |
| 1443 | create_boot_cache(kmem_cache, "kmem_cache", |
| 1444 | offsetof(struct kmem_cache, node) + |
| 1445 | nr_node_ids * sizeof(struct kmem_cache_node *), |
| 1446 | SLAB_HWCACHE_ALIGN); |
| 1447 | list_add(&kmem_cache->list, &slab_caches); |
| 1448 | slab_state = PARTIAL; |
| 1449 | |
| 1450 | /* |
| 1451 | * Initialize the caches that provide memory for the kmem_cache_node |
| 1452 | * structures first. Without this, further allocations will bug. |
| 1453 | */ |
| 1454 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
| 1455 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
| 1456 | slab_state = PARTIAL_NODE; |
| 1457 | |
| 1458 | slab_early_init = 0; |
| 1459 | |
| 1460 | /* 5) Replace the bootstrap kmem_cache_node */ |
| 1461 | { |
| 1462 | int nid; |
| 1463 | |
| 1464 | for_each_online_node(nid) { |
| 1465 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
| 1466 | |
| 1467 | init_list(kmalloc_caches[INDEX_NODE], |
| 1468 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
| 1469 | } |
| 1470 | } |
| 1471 | |
| 1472 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
| 1473 | } |
| 1474 | |
| 1475 | void __init kmem_cache_init_late(void) |
| 1476 | { |
| 1477 | struct kmem_cache *cachep; |
| 1478 | |
| 1479 | slab_state = UP; |
| 1480 | |
| 1481 | /* 6) resize the head arrays to their final sizes */ |
| 1482 | mutex_lock(&slab_mutex); |
| 1483 | list_for_each_entry(cachep, &slab_caches, list) |
| 1484 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
| 1485 | BUG(); |
| 1486 | mutex_unlock(&slab_mutex); |
| 1487 | |
| 1488 | /* Done! */ |
| 1489 | slab_state = FULL; |
| 1490 | |
| 1491 | /* |
| 1492 | * Register a cpu startup notifier callback that initializes |
| 1493 | * cpu_cache_get for all new cpus |
| 1494 | */ |
| 1495 | register_cpu_notifier(&cpucache_notifier); |
| 1496 | |
| 1497 | #ifdef CONFIG_NUMA |
| 1498 | /* |
| 1499 | * Register a memory hotplug callback that initializes and frees |
| 1500 | * node. |
| 1501 | */ |
| 1502 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
| 1503 | #endif |
| 1504 | |
| 1505 | /* |
| 1506 | * The reap timers are started later, with a module init call: That part |
| 1507 | * of the kernel is not yet operational. |
| 1508 | */ |
| 1509 | } |
| 1510 | |
| 1511 | static int __init cpucache_init(void) |
| 1512 | { |
| 1513 | int cpu; |
| 1514 | |
| 1515 | /* |
| 1516 | * Register the timers that return unneeded pages to the page allocator |
| 1517 | */ |
| 1518 | for_each_online_cpu(cpu) |
| 1519 | start_cpu_timer(cpu); |
| 1520 | |
| 1521 | /* Done! */ |
| 1522 | slab_state = FULL; |
| 1523 | return 0; |
| 1524 | } |
| 1525 | __initcall(cpucache_init); |
| 1526 | |
| 1527 | static noinline void |
| 1528 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) |
| 1529 | { |
| 1530 | #if DEBUG |
| 1531 | struct kmem_cache_node *n; |
| 1532 | struct page *page; |
| 1533 | unsigned long flags; |
| 1534 | int node; |
| 1535 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
| 1536 | DEFAULT_RATELIMIT_BURST); |
| 1537 | |
| 1538 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) |
| 1539 | return; |
| 1540 | |
| 1541 | printk(KERN_WARNING |
| 1542 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", |
| 1543 | nodeid, gfpflags); |
| 1544 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", |
| 1545 | cachep->name, cachep->size, cachep->gfporder); |
| 1546 | |
| 1547 | for_each_kmem_cache_node(cachep, node, n) { |
| 1548 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
| 1549 | unsigned long active_slabs = 0, num_slabs = 0; |
| 1550 | |
| 1551 | spin_lock_irqsave(&n->list_lock, flags); |
| 1552 | list_for_each_entry(page, &n->slabs_full, lru) { |
| 1553 | active_objs += cachep->num; |
| 1554 | active_slabs++; |
| 1555 | } |
| 1556 | list_for_each_entry(page, &n->slabs_partial, lru) { |
| 1557 | active_objs += page->active; |
| 1558 | active_slabs++; |
| 1559 | } |
| 1560 | list_for_each_entry(page, &n->slabs_free, lru) |
| 1561 | num_slabs++; |
| 1562 | |
| 1563 | free_objects += n->free_objects; |
| 1564 | spin_unlock_irqrestore(&n->list_lock, flags); |
| 1565 | |
| 1566 | num_slabs += active_slabs; |
| 1567 | num_objs = num_slabs * cachep->num; |
| 1568 | printk(KERN_WARNING |
| 1569 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", |
| 1570 | node, active_slabs, num_slabs, active_objs, num_objs, |
| 1571 | free_objects); |
| 1572 | } |
| 1573 | #endif |
| 1574 | } |
| 1575 | |
| 1576 | /* |
| 1577 | * Interface to system's page allocator. No need to hold the |
| 1578 | * kmem_cache_node ->list_lock. |
| 1579 | * |
| 1580 | * If we requested dmaable memory, we will get it. Even if we |
| 1581 | * did not request dmaable memory, we might get it, but that |
| 1582 | * would be relatively rare and ignorable. |
| 1583 | */ |
| 1584 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
| 1585 | int nodeid) |
| 1586 | { |
| 1587 | struct page *page; |
| 1588 | int nr_pages; |
| 1589 | |
| 1590 | flags |= cachep->allocflags; |
| 1591 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
| 1592 | flags |= __GFP_RECLAIMABLE; |
| 1593 | |
| 1594 | if (memcg_charge_slab(cachep, flags, cachep->gfporder)) |
| 1595 | return NULL; |
| 1596 | |
| 1597 | page = alloc_pages_exact_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
| 1598 | if (!page) { |
| 1599 | memcg_uncharge_slab(cachep, cachep->gfporder); |
| 1600 | slab_out_of_memory(cachep, flags, nodeid); |
| 1601 | return NULL; |
| 1602 | } |
| 1603 | |
| 1604 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
| 1605 | if (unlikely(page->pfmemalloc)) |
| 1606 | pfmemalloc_active = true; |
| 1607 | |
| 1608 | nr_pages = (1 << cachep->gfporder); |
| 1609 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
| 1610 | add_zone_page_state(page_zone(page), |
| 1611 | NR_SLAB_RECLAIMABLE, nr_pages); |
| 1612 | else |
| 1613 | add_zone_page_state(page_zone(page), |
| 1614 | NR_SLAB_UNRECLAIMABLE, nr_pages); |
| 1615 | __SetPageSlab(page); |
| 1616 | if (page->pfmemalloc) |
| 1617 | SetPageSlabPfmemalloc(page); |
| 1618 | |
| 1619 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
| 1620 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); |
| 1621 | |
| 1622 | if (cachep->ctor) |
| 1623 | kmemcheck_mark_uninitialized_pages(page, nr_pages); |
| 1624 | else |
| 1625 | kmemcheck_mark_unallocated_pages(page, nr_pages); |
| 1626 | } |
| 1627 | |
| 1628 | return page; |
| 1629 | } |
| 1630 | |
| 1631 | /* |
| 1632 | * Interface to system's page release. |
| 1633 | */ |
| 1634 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
| 1635 | { |
| 1636 | const unsigned long nr_freed = (1 << cachep->gfporder); |
| 1637 | |
| 1638 | kmemcheck_free_shadow(page, cachep->gfporder); |
| 1639 | |
| 1640 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
| 1641 | sub_zone_page_state(page_zone(page), |
| 1642 | NR_SLAB_RECLAIMABLE, nr_freed); |
| 1643 | else |
| 1644 | sub_zone_page_state(page_zone(page), |
| 1645 | NR_SLAB_UNRECLAIMABLE, nr_freed); |
| 1646 | |
| 1647 | BUG_ON(!PageSlab(page)); |
| 1648 | __ClearPageSlabPfmemalloc(page); |
| 1649 | __ClearPageSlab(page); |
| 1650 | page_mapcount_reset(page); |
| 1651 | page->mapping = NULL; |
| 1652 | |
| 1653 | if (current->reclaim_state) |
| 1654 | current->reclaim_state->reclaimed_slab += nr_freed; |
| 1655 | __free_pages(page, cachep->gfporder); |
| 1656 | memcg_uncharge_slab(cachep, cachep->gfporder); |
| 1657 | } |
| 1658 | |
| 1659 | static void kmem_rcu_free(struct rcu_head *head) |
| 1660 | { |
| 1661 | struct kmem_cache *cachep; |
| 1662 | struct page *page; |
| 1663 | |
| 1664 | page = container_of(head, struct page, rcu_head); |
| 1665 | cachep = page->slab_cache; |
| 1666 | |
| 1667 | kmem_freepages(cachep, page); |
| 1668 | } |
| 1669 | |
| 1670 | #if DEBUG |
| 1671 | |
| 1672 | #ifdef CONFIG_DEBUG_PAGEALLOC |
| 1673 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
| 1674 | unsigned long caller) |
| 1675 | { |
| 1676 | int size = cachep->object_size; |
| 1677 | |
| 1678 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
| 1679 | |
| 1680 | if (size < 5 * sizeof(unsigned long)) |
| 1681 | return; |
| 1682 | |
| 1683 | *addr++ = 0x12345678; |
| 1684 | *addr++ = caller; |
| 1685 | *addr++ = smp_processor_id(); |
| 1686 | size -= 3 * sizeof(unsigned long); |
| 1687 | { |
| 1688 | unsigned long *sptr = &caller; |
| 1689 | unsigned long svalue; |
| 1690 | |
| 1691 | while (!kstack_end(sptr)) { |
| 1692 | svalue = *sptr++; |
| 1693 | if (kernel_text_address(svalue)) { |
| 1694 | *addr++ = svalue; |
| 1695 | size -= sizeof(unsigned long); |
| 1696 | if (size <= sizeof(unsigned long)) |
| 1697 | break; |
| 1698 | } |
| 1699 | } |
| 1700 | |
| 1701 | } |
| 1702 | *addr++ = 0x87654321; |
| 1703 | } |
| 1704 | #endif |
| 1705 | |
| 1706 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
| 1707 | { |
| 1708 | int size = cachep->object_size; |
| 1709 | addr = &((char *)addr)[obj_offset(cachep)]; |
| 1710 | |
| 1711 | memset(addr, val, size); |
| 1712 | *(unsigned char *)(addr + size - 1) = POISON_END; |
| 1713 | } |
| 1714 | |
| 1715 | static void dump_line(char *data, int offset, int limit) |
| 1716 | { |
| 1717 | int i; |
| 1718 | unsigned char error = 0; |
| 1719 | int bad_count = 0; |
| 1720 | |
| 1721 | printk(KERN_ERR "%03x: ", offset); |
| 1722 | for (i = 0; i < limit; i++) { |
| 1723 | if (data[offset + i] != POISON_FREE) { |
| 1724 | error = data[offset + i]; |
| 1725 | bad_count++; |
| 1726 | } |
| 1727 | } |
| 1728 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
| 1729 | &data[offset], limit, 1); |
| 1730 | |
| 1731 | if (bad_count == 1) { |
| 1732 | error ^= POISON_FREE; |
| 1733 | if (!(error & (error - 1))) { |
| 1734 | printk(KERN_ERR "Single bit error detected. Probably " |
| 1735 | "bad RAM.\n"); |
| 1736 | #ifdef CONFIG_X86 |
| 1737 | printk(KERN_ERR "Run memtest86+ or a similar memory " |
| 1738 | "test tool.\n"); |
| 1739 | #else |
| 1740 | printk(KERN_ERR "Run a memory test tool.\n"); |
| 1741 | #endif |
| 1742 | } |
| 1743 | } |
| 1744 | } |
| 1745 | #endif |
| 1746 | |
| 1747 | #if DEBUG |
| 1748 | |
| 1749 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
| 1750 | { |
| 1751 | int i, size; |
| 1752 | char *realobj; |
| 1753 | |
| 1754 | if (cachep->flags & SLAB_RED_ZONE) { |
| 1755 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
| 1756 | *dbg_redzone1(cachep, objp), |
| 1757 | *dbg_redzone2(cachep, objp)); |
| 1758 | } |
| 1759 | |
| 1760 | if (cachep->flags & SLAB_STORE_USER) { |
| 1761 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
| 1762 | *dbg_userword(cachep, objp), |
| 1763 | *dbg_userword(cachep, objp)); |
| 1764 | } |
| 1765 | realobj = (char *)objp + obj_offset(cachep); |
| 1766 | size = cachep->object_size; |
| 1767 | for (i = 0; i < size && lines; i += 16, lines--) { |
| 1768 | int limit; |
| 1769 | limit = 16; |
| 1770 | if (i + limit > size) |
| 1771 | limit = size - i; |
| 1772 | dump_line(realobj, i, limit); |
| 1773 | } |
| 1774 | } |
| 1775 | |
| 1776 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
| 1777 | { |
| 1778 | char *realobj; |
| 1779 | int size, i; |
| 1780 | int lines = 0; |
| 1781 | |
| 1782 | realobj = (char *)objp + obj_offset(cachep); |
| 1783 | size = cachep->object_size; |
| 1784 | |
| 1785 | for (i = 0; i < size; i++) { |
| 1786 | char exp = POISON_FREE; |
| 1787 | if (i == size - 1) |
| 1788 | exp = POISON_END; |
| 1789 | if (realobj[i] != exp) { |
| 1790 | int limit; |
| 1791 | /* Mismatch ! */ |
| 1792 | /* Print header */ |
| 1793 | if (lines == 0) { |
| 1794 | printk(KERN_ERR |
| 1795 | "Slab corruption (%s): %s start=%p, len=%d\n", |
| 1796 | print_tainted(), cachep->name, realobj, size); |
| 1797 | print_objinfo(cachep, objp, 0); |
| 1798 | } |
| 1799 | /* Hexdump the affected line */ |
| 1800 | i = (i / 16) * 16; |
| 1801 | limit = 16; |
| 1802 | if (i + limit > size) |
| 1803 | limit = size - i; |
| 1804 | dump_line(realobj, i, limit); |
| 1805 | i += 16; |
| 1806 | lines++; |
| 1807 | /* Limit to 5 lines */ |
| 1808 | if (lines > 5) |
| 1809 | break; |
| 1810 | } |
| 1811 | } |
| 1812 | if (lines != 0) { |
| 1813 | /* Print some data about the neighboring objects, if they |
| 1814 | * exist: |
| 1815 | */ |
| 1816 | struct page *page = virt_to_head_page(objp); |
| 1817 | unsigned int objnr; |
| 1818 | |
| 1819 | objnr = obj_to_index(cachep, page, objp); |
| 1820 | if (objnr) { |
| 1821 | objp = index_to_obj(cachep, page, objnr - 1); |
| 1822 | realobj = (char *)objp + obj_offset(cachep); |
| 1823 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
| 1824 | realobj, size); |
| 1825 | print_objinfo(cachep, objp, 2); |
| 1826 | } |
| 1827 | if (objnr + 1 < cachep->num) { |
| 1828 | objp = index_to_obj(cachep, page, objnr + 1); |
| 1829 | realobj = (char *)objp + obj_offset(cachep); |
| 1830 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
| 1831 | realobj, size); |
| 1832 | print_objinfo(cachep, objp, 2); |
| 1833 | } |
| 1834 | } |
| 1835 | } |
| 1836 | #endif |
| 1837 | |
| 1838 | #if DEBUG |
| 1839 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
| 1840 | struct page *page) |
| 1841 | { |
| 1842 | int i; |
| 1843 | for (i = 0; i < cachep->num; i++) { |
| 1844 | void *objp = index_to_obj(cachep, page, i); |
| 1845 | |
| 1846 | if (cachep->flags & SLAB_POISON) { |
| 1847 | #ifdef CONFIG_DEBUG_PAGEALLOC |
| 1848 | if (cachep->size % PAGE_SIZE == 0 && |
| 1849 | OFF_SLAB(cachep)) |
| 1850 | kernel_map_pages(virt_to_page(objp), |
| 1851 | cachep->size / PAGE_SIZE, 1); |
| 1852 | else |
| 1853 | check_poison_obj(cachep, objp); |
| 1854 | #else |
| 1855 | check_poison_obj(cachep, objp); |
| 1856 | #endif |
| 1857 | } |
| 1858 | if (cachep->flags & SLAB_RED_ZONE) { |
| 1859 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
| 1860 | slab_error(cachep, "start of a freed object " |
| 1861 | "was overwritten"); |
| 1862 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
| 1863 | slab_error(cachep, "end of a freed object " |
| 1864 | "was overwritten"); |
| 1865 | } |
| 1866 | } |
| 1867 | } |
| 1868 | #else |
| 1869 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
| 1870 | struct page *page) |
| 1871 | { |
| 1872 | } |
| 1873 | #endif |
| 1874 | |
| 1875 | /** |
| 1876 | * slab_destroy - destroy and release all objects in a slab |
| 1877 | * @cachep: cache pointer being destroyed |
| 1878 | * @page: page pointer being destroyed |
| 1879 | * |
| 1880 | * Destroy all the objs in a slab page, and release the mem back to the system. |
| 1881 | * Before calling the slab page must have been unlinked from the cache. The |
| 1882 | * kmem_cache_node ->list_lock is not held/needed. |
| 1883 | */ |
| 1884 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
| 1885 | { |
| 1886 | void *freelist; |
| 1887 | |
| 1888 | freelist = page->freelist; |
| 1889 | slab_destroy_debugcheck(cachep, page); |
| 1890 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
| 1891 | struct rcu_head *head; |
| 1892 | |
| 1893 | /* |
| 1894 | * RCU free overloads the RCU head over the LRU. |
| 1895 | * slab_page has been overloeaded over the LRU, |
| 1896 | * however it is not used from now on so that |
| 1897 | * we can use it safely. |
| 1898 | */ |
| 1899 | head = (void *)&page->rcu_head; |
| 1900 | call_rcu(head, kmem_rcu_free); |
| 1901 | |
| 1902 | } else { |
| 1903 | kmem_freepages(cachep, page); |
| 1904 | } |
| 1905 | |
| 1906 | /* |
| 1907 | * From now on, we don't use freelist |
| 1908 | * although actual page can be freed in rcu context |
| 1909 | */ |
| 1910 | if (OFF_SLAB(cachep)) |
| 1911 | kmem_cache_free(cachep->freelist_cache, freelist); |
| 1912 | } |
| 1913 | |
| 1914 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
| 1915 | { |
| 1916 | struct page *page, *n; |
| 1917 | |
| 1918 | list_for_each_entry_safe(page, n, list, lru) { |
| 1919 | list_del(&page->lru); |
| 1920 | slab_destroy(cachep, page); |
| 1921 | } |
| 1922 | } |
| 1923 | |
| 1924 | /** |
| 1925 | * calculate_slab_order - calculate size (page order) of slabs |
| 1926 | * @cachep: pointer to the cache that is being created |
| 1927 | * @size: size of objects to be created in this cache. |
| 1928 | * @align: required alignment for the objects. |
| 1929 | * @flags: slab allocation flags |
| 1930 | * |
| 1931 | * Also calculates the number of objects per slab. |
| 1932 | * |
| 1933 | * This could be made much more intelligent. For now, try to avoid using |
| 1934 | * high order pages for slabs. When the gfp() functions are more friendly |
| 1935 | * towards high-order requests, this should be changed. |
| 1936 | */ |
| 1937 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
| 1938 | size_t size, size_t align, unsigned long flags) |
| 1939 | { |
| 1940 | unsigned long offslab_limit; |
| 1941 | size_t left_over = 0; |
| 1942 | int gfporder; |
| 1943 | |
| 1944 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
| 1945 | unsigned int num; |
| 1946 | size_t remainder; |
| 1947 | |
| 1948 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
| 1949 | if (!num) |
| 1950 | continue; |
| 1951 | |
| 1952 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
| 1953 | if (num > SLAB_OBJ_MAX_NUM) |
| 1954 | break; |
| 1955 | |
| 1956 | if (flags & CFLGS_OFF_SLAB) { |
| 1957 | size_t freelist_size_per_obj = sizeof(freelist_idx_t); |
| 1958 | /* |
| 1959 | * Max number of objs-per-slab for caches which |
| 1960 | * use off-slab slabs. Needed to avoid a possible |
| 1961 | * looping condition in cache_grow(). |
| 1962 | */ |
| 1963 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
| 1964 | freelist_size_per_obj += sizeof(char); |
| 1965 | offslab_limit = size; |
| 1966 | offslab_limit /= freelist_size_per_obj; |
| 1967 | |
| 1968 | if (num > offslab_limit) |
| 1969 | break; |
| 1970 | } |
| 1971 | |
| 1972 | /* Found something acceptable - save it away */ |
| 1973 | cachep->num = num; |
| 1974 | cachep->gfporder = gfporder; |
| 1975 | left_over = remainder; |
| 1976 | |
| 1977 | /* |
| 1978 | * A VFS-reclaimable slab tends to have most allocations |
| 1979 | * as GFP_NOFS and we really don't want to have to be allocating |
| 1980 | * higher-order pages when we are unable to shrink dcache. |
| 1981 | */ |
| 1982 | if (flags & SLAB_RECLAIM_ACCOUNT) |
| 1983 | break; |
| 1984 | |
| 1985 | /* |
| 1986 | * Large number of objects is good, but very large slabs are |
| 1987 | * currently bad for the gfp()s. |
| 1988 | */ |
| 1989 | if (gfporder >= slab_max_order) |
| 1990 | break; |
| 1991 | |
| 1992 | /* |
| 1993 | * Acceptable internal fragmentation? |
| 1994 | */ |
| 1995 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
| 1996 | break; |
| 1997 | } |
| 1998 | return left_over; |
| 1999 | } |
| 2000 | |
| 2001 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
| 2002 | struct kmem_cache *cachep, int entries, int batchcount) |
| 2003 | { |
| 2004 | int cpu; |
| 2005 | size_t size; |
| 2006 | struct array_cache __percpu *cpu_cache; |
| 2007 | |
| 2008 | size = sizeof(void *) * entries + sizeof(struct array_cache); |
| 2009 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
| 2010 | |
| 2011 | if (!cpu_cache) |
| 2012 | return NULL; |
| 2013 | |
| 2014 | for_each_possible_cpu(cpu) { |
| 2015 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), |
| 2016 | entries, batchcount); |
| 2017 | } |
| 2018 | |
| 2019 | return cpu_cache; |
| 2020 | } |
| 2021 | |
| 2022 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
| 2023 | { |
| 2024 | if (slab_state >= FULL) |
| 2025 | return enable_cpucache(cachep, gfp); |
| 2026 | |
| 2027 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
| 2028 | if (!cachep->cpu_cache) |
| 2029 | return 1; |
| 2030 | |
| 2031 | if (slab_state == DOWN) { |
| 2032 | /* Creation of first cache (kmem_cache). */ |
| 2033 | set_up_node(kmem_cache, CACHE_CACHE); |
| 2034 | } else if (slab_state == PARTIAL) { |
| 2035 | /* For kmem_cache_node */ |
| 2036 | set_up_node(cachep, SIZE_NODE); |
| 2037 | } else { |
| 2038 | int node; |
| 2039 | |
| 2040 | for_each_online_node(node) { |
| 2041 | cachep->node[node] = kmalloc_node( |
| 2042 | sizeof(struct kmem_cache_node), gfp, node); |
| 2043 | BUG_ON(!cachep->node[node]); |
| 2044 | kmem_cache_node_init(cachep->node[node]); |
| 2045 | } |
| 2046 | } |
| 2047 | |
| 2048 | cachep->node[numa_mem_id()]->next_reap = |
| 2049 | jiffies + REAPTIMEOUT_NODE + |
| 2050 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; |
| 2051 | |
| 2052 | cpu_cache_get(cachep)->avail = 0; |
| 2053 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; |
| 2054 | cpu_cache_get(cachep)->batchcount = 1; |
| 2055 | cpu_cache_get(cachep)->touched = 0; |
| 2056 | cachep->batchcount = 1; |
| 2057 | cachep->limit = BOOT_CPUCACHE_ENTRIES; |
| 2058 | return 0; |
| 2059 | } |
| 2060 | |
| 2061 | unsigned long kmem_cache_flags(unsigned long object_size, |
| 2062 | unsigned long flags, const char *name, |
| 2063 | void (*ctor)(void *)) |
| 2064 | { |
| 2065 | return flags; |
| 2066 | } |
| 2067 | |
| 2068 | struct kmem_cache * |
| 2069 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
| 2070 | unsigned long flags, void (*ctor)(void *)) |
| 2071 | { |
| 2072 | struct kmem_cache *cachep; |
| 2073 | |
| 2074 | cachep = find_mergeable(size, align, flags, name, ctor); |
| 2075 | if (cachep) { |
| 2076 | cachep->refcount++; |
| 2077 | |
| 2078 | /* |
| 2079 | * Adjust the object sizes so that we clear |
| 2080 | * the complete object on kzalloc. |
| 2081 | */ |
| 2082 | cachep->object_size = max_t(int, cachep->object_size, size); |
| 2083 | } |
| 2084 | return cachep; |
| 2085 | } |
| 2086 | |
| 2087 | /** |
| 2088 | * __kmem_cache_create - Create a cache. |
| 2089 | * @cachep: cache management descriptor |
| 2090 | * @flags: SLAB flags |
| 2091 | * |
| 2092 | * Returns a ptr to the cache on success, NULL on failure. |
| 2093 | * Cannot be called within a int, but can be interrupted. |
| 2094 | * The @ctor is run when new pages are allocated by the cache. |
| 2095 | * |
| 2096 | * The flags are |
| 2097 | * |
| 2098 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) |
| 2099 | * to catch references to uninitialised memory. |
| 2100 | * |
| 2101 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check |
| 2102 | * for buffer overruns. |
| 2103 | * |
| 2104 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
| 2105 | * cacheline. This can be beneficial if you're counting cycles as closely |
| 2106 | * as davem. |
| 2107 | */ |
| 2108 | int |
| 2109 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
| 2110 | { |
| 2111 | size_t left_over, freelist_size; |
| 2112 | size_t ralign = BYTES_PER_WORD; |
| 2113 | gfp_t gfp; |
| 2114 | int err; |
| 2115 | size_t size = cachep->size; |
| 2116 | |
| 2117 | #if DEBUG |
| 2118 | #if FORCED_DEBUG |
| 2119 | /* |
| 2120 | * Enable redzoning and last user accounting, except for caches with |
| 2121 | * large objects, if the increased size would increase the object size |
| 2122 | * above the next power of two: caches with object sizes just above a |
| 2123 | * power of two have a significant amount of internal fragmentation. |
| 2124 | */ |
| 2125 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
| 2126 | 2 * sizeof(unsigned long long))) |
| 2127 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
| 2128 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
| 2129 | flags |= SLAB_POISON; |
| 2130 | #endif |
| 2131 | if (flags & SLAB_DESTROY_BY_RCU) |
| 2132 | BUG_ON(flags & SLAB_POISON); |
| 2133 | #endif |
| 2134 | |
| 2135 | /* |
| 2136 | * Check that size is in terms of words. This is needed to avoid |
| 2137 | * unaligned accesses for some archs when redzoning is used, and makes |
| 2138 | * sure any on-slab bufctl's are also correctly aligned. |
| 2139 | */ |
| 2140 | if (size & (BYTES_PER_WORD - 1)) { |
| 2141 | size += (BYTES_PER_WORD - 1); |
| 2142 | size &= ~(BYTES_PER_WORD - 1); |
| 2143 | } |
| 2144 | |
| 2145 | if (flags & SLAB_RED_ZONE) { |
| 2146 | ralign = REDZONE_ALIGN; |
| 2147 | /* If redzoning, ensure that the second redzone is suitably |
| 2148 | * aligned, by adjusting the object size accordingly. */ |
| 2149 | size += REDZONE_ALIGN - 1; |
| 2150 | size &= ~(REDZONE_ALIGN - 1); |
| 2151 | } |
| 2152 | |
| 2153 | /* 3) caller mandated alignment */ |
| 2154 | if (ralign < cachep->align) { |
| 2155 | ralign = cachep->align; |
| 2156 | } |
| 2157 | /* disable debug if necessary */ |
| 2158 | if (ralign > __alignof__(unsigned long long)) |
| 2159 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
| 2160 | /* |
| 2161 | * 4) Store it. |
| 2162 | */ |
| 2163 | cachep->align = ralign; |
| 2164 | |
| 2165 | if (slab_is_available()) |
| 2166 | gfp = GFP_KERNEL; |
| 2167 | else |
| 2168 | gfp = GFP_NOWAIT; |
| 2169 | |
| 2170 | #if DEBUG |
| 2171 | |
| 2172 | /* |
| 2173 | * Both debugging options require word-alignment which is calculated |
| 2174 | * into align above. |
| 2175 | */ |
| 2176 | if (flags & SLAB_RED_ZONE) { |
| 2177 | /* add space for red zone words */ |
| 2178 | cachep->obj_offset += sizeof(unsigned long long); |
| 2179 | size += 2 * sizeof(unsigned long long); |
| 2180 | } |
| 2181 | if (flags & SLAB_STORE_USER) { |
| 2182 | /* user store requires one word storage behind the end of |
| 2183 | * the real object. But if the second red zone needs to be |
| 2184 | * aligned to 64 bits, we must allow that much space. |
| 2185 | */ |
| 2186 | if (flags & SLAB_RED_ZONE) |
| 2187 | size += REDZONE_ALIGN; |
| 2188 | else |
| 2189 | size += BYTES_PER_WORD; |
| 2190 | } |
| 2191 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) |
| 2192 | if (size >= kmalloc_size(INDEX_NODE + 1) |
| 2193 | && cachep->object_size > cache_line_size() |
| 2194 | && ALIGN(size, cachep->align) < PAGE_SIZE) { |
| 2195 | cachep->obj_offset += PAGE_SIZE - ALIGN(size, cachep->align); |
| 2196 | size = PAGE_SIZE; |
| 2197 | } |
| 2198 | #endif |
| 2199 | #endif |
| 2200 | |
| 2201 | /* |
| 2202 | * Determine if the slab management is 'on' or 'off' slab. |
| 2203 | * (bootstrapping cannot cope with offslab caches so don't do |
| 2204 | * it too early on. Always use on-slab management when |
| 2205 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) |
| 2206 | */ |
| 2207 | if ((size >= (PAGE_SIZE >> 5)) && !slab_early_init && |
| 2208 | !(flags & SLAB_NOLEAKTRACE)) |
| 2209 | /* |
| 2210 | * Size is large, assume best to place the slab management obj |
| 2211 | * off-slab (should allow better packing of objs). |
| 2212 | */ |
| 2213 | flags |= CFLGS_OFF_SLAB; |
| 2214 | |
| 2215 | size = ALIGN(size, cachep->align); |
| 2216 | /* |
| 2217 | * We should restrict the number of objects in a slab to implement |
| 2218 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. |
| 2219 | */ |
| 2220 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) |
| 2221 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); |
| 2222 | |
| 2223 | left_over = calculate_slab_order(cachep, size, cachep->align, flags); |
| 2224 | |
| 2225 | if (!cachep->num) |
| 2226 | return -E2BIG; |
| 2227 | |
| 2228 | freelist_size = calculate_freelist_size(cachep->num, cachep->align); |
| 2229 | |
| 2230 | /* |
| 2231 | * If the slab has been placed off-slab, and we have enough space then |
| 2232 | * move it on-slab. This is at the expense of any extra colouring. |
| 2233 | */ |
| 2234 | if (flags & CFLGS_OFF_SLAB && left_over >= freelist_size) { |
| 2235 | flags &= ~CFLGS_OFF_SLAB; |
| 2236 | left_over -= freelist_size; |
| 2237 | } |
| 2238 | |
| 2239 | if (flags & CFLGS_OFF_SLAB) { |
| 2240 | /* really off slab. No need for manual alignment */ |
| 2241 | freelist_size = calculate_freelist_size(cachep->num, 0); |
| 2242 | |
| 2243 | #ifdef CONFIG_PAGE_POISONING |
| 2244 | /* If we're going to use the generic kernel_map_pages() |
| 2245 | * poisoning, then it's going to smash the contents of |
| 2246 | * the redzone and userword anyhow, so switch them off. |
| 2247 | */ |
| 2248 | if (size % PAGE_SIZE == 0 && flags & SLAB_POISON) |
| 2249 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
| 2250 | #endif |
| 2251 | } |
| 2252 | |
| 2253 | cachep->colour_off = cache_line_size(); |
| 2254 | /* Offset must be a multiple of the alignment. */ |
| 2255 | if (cachep->colour_off < cachep->align) |
| 2256 | cachep->colour_off = cachep->align; |
| 2257 | cachep->colour = left_over / cachep->colour_off; |
| 2258 | cachep->freelist_size = freelist_size; |
| 2259 | cachep->flags = flags; |
| 2260 | cachep->allocflags = __GFP_COMP; |
| 2261 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
| 2262 | cachep->allocflags |= GFP_DMA; |
| 2263 | cachep->size = size; |
| 2264 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
| 2265 | |
| 2266 | if (flags & CFLGS_OFF_SLAB) { |
| 2267 | cachep->freelist_cache = kmalloc_slab(freelist_size, 0u); |
| 2268 | /* |
| 2269 | * This is a possibility for one of the kmalloc_{dma,}_caches. |
| 2270 | * But since we go off slab only for object size greater than |
| 2271 | * PAGE_SIZE/8, and kmalloc_{dma,}_caches get created |
| 2272 | * in ascending order,this should not happen at all. |
| 2273 | * But leave a BUG_ON for some lucky dude. |
| 2274 | */ |
| 2275 | BUG_ON(ZERO_OR_NULL_PTR(cachep->freelist_cache)); |
| 2276 | } |
| 2277 | |
| 2278 | err = setup_cpu_cache(cachep, gfp); |
| 2279 | if (err) { |
| 2280 | __kmem_cache_shutdown(cachep); |
| 2281 | return err; |
| 2282 | } |
| 2283 | |
| 2284 | return 0; |
| 2285 | } |
| 2286 | |
| 2287 | #if DEBUG |
| 2288 | static void check_irq_off(void) |
| 2289 | { |
| 2290 | BUG_ON(!irqs_disabled()); |
| 2291 | } |
| 2292 | |
| 2293 | static void check_irq_on(void) |
| 2294 | { |
| 2295 | BUG_ON(irqs_disabled()); |
| 2296 | } |
| 2297 | |
| 2298 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
| 2299 | { |
| 2300 | #ifdef CONFIG_SMP |
| 2301 | check_irq_off(); |
| 2302 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
| 2303 | #endif |
| 2304 | } |
| 2305 | |
| 2306 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
| 2307 | { |
| 2308 | #ifdef CONFIG_SMP |
| 2309 | check_irq_off(); |
| 2310 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
| 2311 | #endif |
| 2312 | } |
| 2313 | |
| 2314 | #else |
| 2315 | #define check_irq_off() do { } while(0) |
| 2316 | #define check_irq_on() do { } while(0) |
| 2317 | #define check_spinlock_acquired(x) do { } while(0) |
| 2318 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
| 2319 | #endif |
| 2320 | |
| 2321 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
| 2322 | struct array_cache *ac, |
| 2323 | int force, int node); |
| 2324 | |
| 2325 | static void do_drain(void *arg) |
| 2326 | { |
| 2327 | struct kmem_cache *cachep = arg; |
| 2328 | struct array_cache *ac; |
| 2329 | int node = numa_mem_id(); |
| 2330 | struct kmem_cache_node *n; |
| 2331 | LIST_HEAD(list); |
| 2332 | |
| 2333 | check_irq_off(); |
| 2334 | ac = cpu_cache_get(cachep); |
| 2335 | n = get_node(cachep, node); |
| 2336 | spin_lock(&n->list_lock); |
| 2337 | free_block(cachep, ac->entry, ac->avail, node, &list); |
| 2338 | spin_unlock(&n->list_lock); |
| 2339 | slabs_destroy(cachep, &list); |
| 2340 | ac->avail = 0; |
| 2341 | } |
| 2342 | |
| 2343 | static void drain_cpu_caches(struct kmem_cache *cachep) |
| 2344 | { |
| 2345 | struct kmem_cache_node *n; |
| 2346 | int node; |
| 2347 | |
| 2348 | on_each_cpu(do_drain, cachep, 1); |
| 2349 | check_irq_on(); |
| 2350 | for_each_kmem_cache_node(cachep, node, n) |
| 2351 | if (n->alien) |
| 2352 | drain_alien_cache(cachep, n->alien); |
| 2353 | |
| 2354 | for_each_kmem_cache_node(cachep, node, n) |
| 2355 | drain_array(cachep, n, n->shared, 1, node); |
| 2356 | } |
| 2357 | |
| 2358 | /* |
| 2359 | * Remove slabs from the list of free slabs. |
| 2360 | * Specify the number of slabs to drain in tofree. |
| 2361 | * |
| 2362 | * Returns the actual number of slabs released. |
| 2363 | */ |
| 2364 | static int drain_freelist(struct kmem_cache *cache, |
| 2365 | struct kmem_cache_node *n, int tofree) |
| 2366 | { |
| 2367 | struct list_head *p; |
| 2368 | int nr_freed; |
| 2369 | struct page *page; |
| 2370 | |
| 2371 | nr_freed = 0; |
| 2372 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
| 2373 | |
| 2374 | spin_lock_irq(&n->list_lock); |
| 2375 | p = n->slabs_free.prev; |
| 2376 | if (p == &n->slabs_free) { |
| 2377 | spin_unlock_irq(&n->list_lock); |
| 2378 | goto out; |
| 2379 | } |
| 2380 | |
| 2381 | page = list_entry(p, struct page, lru); |
| 2382 | #if DEBUG |
| 2383 | BUG_ON(page->active); |
| 2384 | #endif |
| 2385 | list_del(&page->lru); |
| 2386 | /* |
| 2387 | * Safe to drop the lock. The slab is no longer linked |
| 2388 | * to the cache. |
| 2389 | */ |
| 2390 | n->free_objects -= cache->num; |
| 2391 | spin_unlock_irq(&n->list_lock); |
| 2392 | slab_destroy(cache, page); |
| 2393 | nr_freed++; |
| 2394 | } |
| 2395 | out: |
| 2396 | return nr_freed; |
| 2397 | } |
| 2398 | |
| 2399 | int __kmem_cache_shrink(struct kmem_cache *cachep, bool deactivate) |
| 2400 | { |
| 2401 | int ret = 0; |
| 2402 | int node; |
| 2403 | struct kmem_cache_node *n; |
| 2404 | |
| 2405 | drain_cpu_caches(cachep); |
| 2406 | |
| 2407 | check_irq_on(); |
| 2408 | for_each_kmem_cache_node(cachep, node, n) { |
| 2409 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
| 2410 | |
| 2411 | ret += !list_empty(&n->slabs_full) || |
| 2412 | !list_empty(&n->slabs_partial); |
| 2413 | } |
| 2414 | return (ret ? 1 : 0); |
| 2415 | } |
| 2416 | |
| 2417 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
| 2418 | { |
| 2419 | int i; |
| 2420 | struct kmem_cache_node *n; |
| 2421 | int rc = __kmem_cache_shrink(cachep, false); |
| 2422 | |
| 2423 | if (rc) |
| 2424 | return rc; |
| 2425 | |
| 2426 | free_percpu(cachep->cpu_cache); |
| 2427 | |
| 2428 | /* NUMA: free the node structures */ |
| 2429 | for_each_kmem_cache_node(cachep, i, n) { |
| 2430 | kfree(n->shared); |
| 2431 | free_alien_cache(n->alien); |
| 2432 | kfree(n); |
| 2433 | cachep->node[i] = NULL; |
| 2434 | } |
| 2435 | return 0; |
| 2436 | } |
| 2437 | |
| 2438 | /* |
| 2439 | * Get the memory for a slab management obj. |
| 2440 | * |
| 2441 | * For a slab cache when the slab descriptor is off-slab, the |
| 2442 | * slab descriptor can't come from the same cache which is being created, |
| 2443 | * Because if it is the case, that means we defer the creation of |
| 2444 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. |
| 2445 | * And we eventually call down to __kmem_cache_create(), which |
| 2446 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. |
| 2447 | * This is a "chicken-and-egg" problem. |
| 2448 | * |
| 2449 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, |
| 2450 | * which are all initialized during kmem_cache_init(). |
| 2451 | */ |
| 2452 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
| 2453 | struct page *page, int colour_off, |
| 2454 | gfp_t local_flags, int nodeid) |
| 2455 | { |
| 2456 | void *freelist; |
| 2457 | void *addr = page_address(page); |
| 2458 | |
| 2459 | if (OFF_SLAB(cachep)) { |
| 2460 | /* Slab management obj is off-slab. */ |
| 2461 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
| 2462 | local_flags, nodeid); |
| 2463 | if (!freelist) |
| 2464 | return NULL; |
| 2465 | } else { |
| 2466 | freelist = addr + colour_off; |
| 2467 | colour_off += cachep->freelist_size; |
| 2468 | } |
| 2469 | page->active = 0; |
| 2470 | page->s_mem = addr + colour_off; |
| 2471 | return freelist; |
| 2472 | } |
| 2473 | |
| 2474 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
| 2475 | { |
| 2476 | return ((freelist_idx_t *)page->freelist)[idx]; |
| 2477 | } |
| 2478 | |
| 2479 | static inline void set_free_obj(struct page *page, |
| 2480 | unsigned int idx, freelist_idx_t val) |
| 2481 | { |
| 2482 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
| 2483 | } |
| 2484 | |
| 2485 | static void cache_init_objs(struct kmem_cache *cachep, |
| 2486 | struct page *page) |
| 2487 | { |
| 2488 | int i; |
| 2489 | |
| 2490 | for (i = 0; i < cachep->num; i++) { |
| 2491 | void *objp = index_to_obj(cachep, page, i); |
| 2492 | #if DEBUG |
| 2493 | /* need to poison the objs? */ |
| 2494 | if (cachep->flags & SLAB_POISON) |
| 2495 | poison_obj(cachep, objp, POISON_FREE); |
| 2496 | if (cachep->flags & SLAB_STORE_USER) |
| 2497 | *dbg_userword(cachep, objp) = NULL; |
| 2498 | |
| 2499 | if (cachep->flags & SLAB_RED_ZONE) { |
| 2500 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
| 2501 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; |
| 2502 | } |
| 2503 | /* |
| 2504 | * Constructors are not allowed to allocate memory from the same |
| 2505 | * cache which they are a constructor for. Otherwise, deadlock. |
| 2506 | * They must also be threaded. |
| 2507 | */ |
| 2508 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) |
| 2509 | cachep->ctor(objp + obj_offset(cachep)); |
| 2510 | |
| 2511 | if (cachep->flags & SLAB_RED_ZONE) { |
| 2512 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
| 2513 | slab_error(cachep, "constructor overwrote the" |
| 2514 | " end of an object"); |
| 2515 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
| 2516 | slab_error(cachep, "constructor overwrote the" |
| 2517 | " start of an object"); |
| 2518 | } |
| 2519 | if ((cachep->size % PAGE_SIZE) == 0 && |
| 2520 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) |
| 2521 | kernel_map_pages(virt_to_page(objp), |
| 2522 | cachep->size / PAGE_SIZE, 0); |
| 2523 | #else |
| 2524 | if (cachep->ctor) |
| 2525 | cachep->ctor(objp); |
| 2526 | #endif |
| 2527 | set_obj_status(page, i, OBJECT_FREE); |
| 2528 | set_free_obj(page, i, i); |
| 2529 | } |
| 2530 | } |
| 2531 | |
| 2532 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
| 2533 | { |
| 2534 | if (CONFIG_ZONE_DMA_FLAG) { |
| 2535 | if (flags & GFP_DMA) |
| 2536 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
| 2537 | else |
| 2538 | BUG_ON(cachep->allocflags & GFP_DMA); |
| 2539 | } |
| 2540 | } |
| 2541 | |
| 2542 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page, |
| 2543 | int nodeid) |
| 2544 | { |
| 2545 | void *objp; |
| 2546 | |
| 2547 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
| 2548 | page->active++; |
| 2549 | #if DEBUG |
| 2550 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
| 2551 | #endif |
| 2552 | |
| 2553 | return objp; |
| 2554 | } |
| 2555 | |
| 2556 | static void slab_put_obj(struct kmem_cache *cachep, struct page *page, |
| 2557 | void *objp, int nodeid) |
| 2558 | { |
| 2559 | unsigned int objnr = obj_to_index(cachep, page, objp); |
| 2560 | #if DEBUG |
| 2561 | unsigned int i; |
| 2562 | |
| 2563 | /* Verify that the slab belongs to the intended node */ |
| 2564 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
| 2565 | |
| 2566 | /* Verify double free bug */ |
| 2567 | for (i = page->active; i < cachep->num; i++) { |
| 2568 | if (get_free_obj(page, i) == objnr) { |
| 2569 | printk(KERN_ERR "slab: double free detected in cache " |
| 2570 | "'%s', objp %p\n", cachep->name, objp); |
| 2571 | BUG(); |
| 2572 | } |
| 2573 | } |
| 2574 | #endif |
| 2575 | page->active--; |
| 2576 | set_free_obj(page, page->active, objnr); |
| 2577 | } |
| 2578 | |
| 2579 | /* |
| 2580 | * Map pages beginning at addr to the given cache and slab. This is required |
| 2581 | * for the slab allocator to be able to lookup the cache and slab of a |
| 2582 | * virtual address for kfree, ksize, and slab debugging. |
| 2583 | */ |
| 2584 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
| 2585 | void *freelist) |
| 2586 | { |
| 2587 | page->slab_cache = cache; |
| 2588 | page->freelist = freelist; |
| 2589 | } |
| 2590 | |
| 2591 | /* |
| 2592 | * Grow (by 1) the number of slabs within a cache. This is called by |
| 2593 | * kmem_cache_alloc() when there are no active objs left in a cache. |
| 2594 | */ |
| 2595 | static int cache_grow(struct kmem_cache *cachep, |
| 2596 | gfp_t flags, int nodeid, struct page *page) |
| 2597 | { |
| 2598 | void *freelist; |
| 2599 | size_t offset; |
| 2600 | gfp_t local_flags; |
| 2601 | struct kmem_cache_node *n; |
| 2602 | |
| 2603 | /* |
| 2604 | * Be lazy and only check for valid flags here, keeping it out of the |
| 2605 | * critical path in kmem_cache_alloc(). |
| 2606 | */ |
| 2607 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
| 2608 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); |
| 2609 | BUG(); |
| 2610 | } |
| 2611 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
| 2612 | |
| 2613 | /* Take the node list lock to change the colour_next on this node */ |
| 2614 | check_irq_off(); |
| 2615 | n = get_node(cachep, nodeid); |
| 2616 | spin_lock(&n->list_lock); |
| 2617 | |
| 2618 | /* Get colour for the slab, and cal the next value. */ |
| 2619 | offset = n->colour_next; |
| 2620 | n->colour_next++; |
| 2621 | if (n->colour_next >= cachep->colour) |
| 2622 | n->colour_next = 0; |
| 2623 | spin_unlock(&n->list_lock); |
| 2624 | |
| 2625 | offset *= cachep->colour_off; |
| 2626 | |
| 2627 | if (local_flags & __GFP_WAIT) |
| 2628 | local_irq_enable(); |
| 2629 | |
| 2630 | /* |
| 2631 | * The test for missing atomic flag is performed here, rather than |
| 2632 | * the more obvious place, simply to reduce the critical path length |
| 2633 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they |
| 2634 | * will eventually be caught here (where it matters). |
| 2635 | */ |
| 2636 | kmem_flagcheck(cachep, flags); |
| 2637 | |
| 2638 | /* |
| 2639 | * Get mem for the objs. Attempt to allocate a physical page from |
| 2640 | * 'nodeid'. |
| 2641 | */ |
| 2642 | if (!page) |
| 2643 | page = kmem_getpages(cachep, local_flags, nodeid); |
| 2644 | if (!page) |
| 2645 | goto failed; |
| 2646 | |
| 2647 | /* Get slab management. */ |
| 2648 | freelist = alloc_slabmgmt(cachep, page, offset, |
| 2649 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
| 2650 | if (!freelist) |
| 2651 | goto opps1; |
| 2652 | |
| 2653 | slab_map_pages(cachep, page, freelist); |
| 2654 | |
| 2655 | cache_init_objs(cachep, page); |
| 2656 | |
| 2657 | if (local_flags & __GFP_WAIT) |
| 2658 | local_irq_disable(); |
| 2659 | check_irq_off(); |
| 2660 | spin_lock(&n->list_lock); |
| 2661 | |
| 2662 | /* Make slab active. */ |
| 2663 | list_add_tail(&page->lru, &(n->slabs_free)); |
| 2664 | STATS_INC_GROWN(cachep); |
| 2665 | n->free_objects += cachep->num; |
| 2666 | spin_unlock(&n->list_lock); |
| 2667 | return 1; |
| 2668 | opps1: |
| 2669 | kmem_freepages(cachep, page); |
| 2670 | failed: |
| 2671 | if (local_flags & __GFP_WAIT) |
| 2672 | local_irq_disable(); |
| 2673 | return 0; |
| 2674 | } |
| 2675 | |
| 2676 | #if DEBUG |
| 2677 | |
| 2678 | /* |
| 2679 | * Perform extra freeing checks: |
| 2680 | * - detect bad pointers. |
| 2681 | * - POISON/RED_ZONE checking |
| 2682 | */ |
| 2683 | static void kfree_debugcheck(const void *objp) |
| 2684 | { |
| 2685 | if (!virt_addr_valid(objp)) { |
| 2686 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", |
| 2687 | (unsigned long)objp); |
| 2688 | BUG(); |
| 2689 | } |
| 2690 | } |
| 2691 | |
| 2692 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
| 2693 | { |
| 2694 | unsigned long long redzone1, redzone2; |
| 2695 | |
| 2696 | redzone1 = *dbg_redzone1(cache, obj); |
| 2697 | redzone2 = *dbg_redzone2(cache, obj); |
| 2698 | |
| 2699 | /* |
| 2700 | * Redzone is ok. |
| 2701 | */ |
| 2702 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) |
| 2703 | return; |
| 2704 | |
| 2705 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) |
| 2706 | slab_error(cache, "double free detected"); |
| 2707 | else |
| 2708 | slab_error(cache, "memory outside object was overwritten"); |
| 2709 | |
| 2710 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
| 2711 | obj, redzone1, redzone2); |
| 2712 | } |
| 2713 | |
| 2714 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
| 2715 | unsigned long caller) |
| 2716 | { |
| 2717 | unsigned int objnr; |
| 2718 | struct page *page; |
| 2719 | |
| 2720 | BUG_ON(virt_to_cache(objp) != cachep); |
| 2721 | |
| 2722 | objp -= obj_offset(cachep); |
| 2723 | kfree_debugcheck(objp); |
| 2724 | page = virt_to_head_page(objp); |
| 2725 | |
| 2726 | if (cachep->flags & SLAB_RED_ZONE) { |
| 2727 | verify_redzone_free(cachep, objp); |
| 2728 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
| 2729 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; |
| 2730 | } |
| 2731 | if (cachep->flags & SLAB_STORE_USER) |
| 2732 | *dbg_userword(cachep, objp) = (void *)caller; |
| 2733 | |
| 2734 | objnr = obj_to_index(cachep, page, objp); |
| 2735 | |
| 2736 | BUG_ON(objnr >= cachep->num); |
| 2737 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
| 2738 | |
| 2739 | set_obj_status(page, objnr, OBJECT_FREE); |
| 2740 | if (cachep->flags & SLAB_POISON) { |
| 2741 | #ifdef CONFIG_DEBUG_PAGEALLOC |
| 2742 | if ((cachep->size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
| 2743 | store_stackinfo(cachep, objp, caller); |
| 2744 | kernel_map_pages(virt_to_page(objp), |
| 2745 | cachep->size / PAGE_SIZE, 0); |
| 2746 | } else { |
| 2747 | poison_obj(cachep, objp, POISON_FREE); |
| 2748 | } |
| 2749 | #else |
| 2750 | poison_obj(cachep, objp, POISON_FREE); |
| 2751 | #endif |
| 2752 | } |
| 2753 | return objp; |
| 2754 | } |
| 2755 | |
| 2756 | #else |
| 2757 | #define kfree_debugcheck(x) do { } while(0) |
| 2758 | #define cache_free_debugcheck(x,objp,z) (objp) |
| 2759 | #endif |
| 2760 | |
| 2761 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
| 2762 | bool force_refill) |
| 2763 | { |
| 2764 | int batchcount; |
| 2765 | struct kmem_cache_node *n; |
| 2766 | struct array_cache *ac; |
| 2767 | int node; |
| 2768 | |
| 2769 | check_irq_off(); |
| 2770 | node = numa_mem_id(); |
| 2771 | if (unlikely(force_refill)) |
| 2772 | goto force_grow; |
| 2773 | retry: |
| 2774 | ac = cpu_cache_get(cachep); |
| 2775 | batchcount = ac->batchcount; |
| 2776 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { |
| 2777 | /* |
| 2778 | * If there was little recent activity on this cache, then |
| 2779 | * perform only a partial refill. Otherwise we could generate |
| 2780 | * refill bouncing. |
| 2781 | */ |
| 2782 | batchcount = BATCHREFILL_LIMIT; |
| 2783 | } |
| 2784 | n = get_node(cachep, node); |
| 2785 | |
| 2786 | BUG_ON(ac->avail > 0 || !n); |
| 2787 | spin_lock(&n->list_lock); |
| 2788 | |
| 2789 | /* See if we can refill from the shared array */ |
| 2790 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
| 2791 | n->shared->touched = 1; |
| 2792 | goto alloc_done; |
| 2793 | } |
| 2794 | |
| 2795 | while (batchcount > 0) { |
| 2796 | struct list_head *entry; |
| 2797 | struct page *page; |
| 2798 | /* Get slab alloc is to come from. */ |
| 2799 | entry = n->slabs_partial.next; |
| 2800 | if (entry == &n->slabs_partial) { |
| 2801 | n->free_touched = 1; |
| 2802 | entry = n->slabs_free.next; |
| 2803 | if (entry == &n->slabs_free) |
| 2804 | goto must_grow; |
| 2805 | } |
| 2806 | |
| 2807 | page = list_entry(entry, struct page, lru); |
| 2808 | check_spinlock_acquired(cachep); |
| 2809 | |
| 2810 | /* |
| 2811 | * The slab was either on partial or free list so |
| 2812 | * there must be at least one object available for |
| 2813 | * allocation. |
| 2814 | */ |
| 2815 | BUG_ON(page->active >= cachep->num); |
| 2816 | |
| 2817 | while (page->active < cachep->num && batchcount--) { |
| 2818 | STATS_INC_ALLOCED(cachep); |
| 2819 | STATS_INC_ACTIVE(cachep); |
| 2820 | STATS_SET_HIGH(cachep); |
| 2821 | |
| 2822 | ac_put_obj(cachep, ac, slab_get_obj(cachep, page, |
| 2823 | node)); |
| 2824 | } |
| 2825 | |
| 2826 | /* move slabp to correct slabp list: */ |
| 2827 | list_del(&page->lru); |
| 2828 | if (page->active == cachep->num) |
| 2829 | list_add(&page->lru, &n->slabs_full); |
| 2830 | else |
| 2831 | list_add(&page->lru, &n->slabs_partial); |
| 2832 | } |
| 2833 | |
| 2834 | must_grow: |
| 2835 | n->free_objects -= ac->avail; |
| 2836 | alloc_done: |
| 2837 | spin_unlock(&n->list_lock); |
| 2838 | |
| 2839 | if (unlikely(!ac->avail)) { |
| 2840 | int x; |
| 2841 | force_grow: |
| 2842 | x = cache_grow(cachep, gfp_exact_node(flags), node, NULL); |
| 2843 | |
| 2844 | /* cache_grow can reenable interrupts, then ac could change. */ |
| 2845 | ac = cpu_cache_get(cachep); |
| 2846 | node = numa_mem_id(); |
| 2847 | |
| 2848 | /* no objects in sight? abort */ |
| 2849 | if (!x && (ac->avail == 0 || force_refill)) |
| 2850 | return NULL; |
| 2851 | |
| 2852 | if (!ac->avail) /* objects refilled by interrupt? */ |
| 2853 | goto retry; |
| 2854 | } |
| 2855 | ac->touched = 1; |
| 2856 | |
| 2857 | return ac_get_obj(cachep, ac, flags, force_refill); |
| 2858 | } |
| 2859 | |
| 2860 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
| 2861 | gfp_t flags) |
| 2862 | { |
| 2863 | might_sleep_if(flags & __GFP_WAIT); |
| 2864 | #if DEBUG |
| 2865 | kmem_flagcheck(cachep, flags); |
| 2866 | #endif |
| 2867 | } |
| 2868 | |
| 2869 | #if DEBUG |
| 2870 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
| 2871 | gfp_t flags, void *objp, unsigned long caller) |
| 2872 | { |
| 2873 | struct page *page; |
| 2874 | |
| 2875 | if (!objp) |
| 2876 | return objp; |
| 2877 | if (cachep->flags & SLAB_POISON) { |
| 2878 | #ifdef CONFIG_DEBUG_PAGEALLOC |
| 2879 | if ((cachep->size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
| 2880 | kernel_map_pages(virt_to_page(objp), |
| 2881 | cachep->size / PAGE_SIZE, 1); |
| 2882 | else |
| 2883 | check_poison_obj(cachep, objp); |
| 2884 | #else |
| 2885 | check_poison_obj(cachep, objp); |
| 2886 | #endif |
| 2887 | poison_obj(cachep, objp, POISON_INUSE); |
| 2888 | } |
| 2889 | if (cachep->flags & SLAB_STORE_USER) |
| 2890 | *dbg_userword(cachep, objp) = (void *)caller; |
| 2891 | |
| 2892 | if (cachep->flags & SLAB_RED_ZONE) { |
| 2893 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
| 2894 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { |
| 2895 | slab_error(cachep, "double free, or memory outside" |
| 2896 | " object was overwritten"); |
| 2897 | printk(KERN_ERR |
| 2898 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
| 2899 | objp, *dbg_redzone1(cachep, objp), |
| 2900 | *dbg_redzone2(cachep, objp)); |
| 2901 | } |
| 2902 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; |
| 2903 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; |
| 2904 | } |
| 2905 | |
| 2906 | page = virt_to_head_page(objp); |
| 2907 | set_obj_status(page, obj_to_index(cachep, page, objp), OBJECT_ACTIVE); |
| 2908 | objp += obj_offset(cachep); |
| 2909 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
| 2910 | cachep->ctor(objp); |
| 2911 | if (ARCH_SLAB_MINALIGN && |
| 2912 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { |
| 2913 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
| 2914 | objp, (int)ARCH_SLAB_MINALIGN); |
| 2915 | } |
| 2916 | return objp; |
| 2917 | } |
| 2918 | #else |
| 2919 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) |
| 2920 | #endif |
| 2921 | |
| 2922 | static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) |
| 2923 | { |
| 2924 | if (unlikely(cachep == kmem_cache)) |
| 2925 | return false; |
| 2926 | |
| 2927 | return should_failslab(cachep->object_size, flags, cachep->flags); |
| 2928 | } |
| 2929 | |
| 2930 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
| 2931 | { |
| 2932 | void *objp; |
| 2933 | struct array_cache *ac; |
| 2934 | bool force_refill = false; |
| 2935 | |
| 2936 | check_irq_off(); |
| 2937 | |
| 2938 | ac = cpu_cache_get(cachep); |
| 2939 | if (likely(ac->avail)) { |
| 2940 | ac->touched = 1; |
| 2941 | objp = ac_get_obj(cachep, ac, flags, false); |
| 2942 | |
| 2943 | /* |
| 2944 | * Allow for the possibility all avail objects are not allowed |
| 2945 | * by the current flags |
| 2946 | */ |
| 2947 | if (objp) { |
| 2948 | STATS_INC_ALLOCHIT(cachep); |
| 2949 | goto out; |
| 2950 | } |
| 2951 | force_refill = true; |
| 2952 | } |
| 2953 | |
| 2954 | STATS_INC_ALLOCMISS(cachep); |
| 2955 | objp = cache_alloc_refill(cachep, flags, force_refill); |
| 2956 | /* |
| 2957 | * the 'ac' may be updated by cache_alloc_refill(), |
| 2958 | * and kmemleak_erase() requires its correct value. |
| 2959 | */ |
| 2960 | ac = cpu_cache_get(cachep); |
| 2961 | |
| 2962 | out: |
| 2963 | /* |
| 2964 | * To avoid a false negative, if an object that is in one of the |
| 2965 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't |
| 2966 | * treat the array pointers as a reference to the object. |
| 2967 | */ |
| 2968 | if (objp) |
| 2969 | kmemleak_erase(&ac->entry[ac->avail]); |
| 2970 | return objp; |
| 2971 | } |
| 2972 | |
| 2973 | #ifdef CONFIG_NUMA |
| 2974 | /* |
| 2975 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
| 2976 | * |
| 2977 | * If we are in_interrupt, then process context, including cpusets and |
| 2978 | * mempolicy, may not apply and should not be used for allocation policy. |
| 2979 | */ |
| 2980 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) |
| 2981 | { |
| 2982 | int nid_alloc, nid_here; |
| 2983 | |
| 2984 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
| 2985 | return NULL; |
| 2986 | nid_alloc = nid_here = numa_mem_id(); |
| 2987 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
| 2988 | nid_alloc = cpuset_slab_spread_node(); |
| 2989 | else if (current->mempolicy) |
| 2990 | nid_alloc = mempolicy_slab_node(); |
| 2991 | if (nid_alloc != nid_here) |
| 2992 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
| 2993 | return NULL; |
| 2994 | } |
| 2995 | |
| 2996 | /* |
| 2997 | * Fallback function if there was no memory available and no objects on a |
| 2998 | * certain node and fall back is permitted. First we scan all the |
| 2999 | * available node for available objects. If that fails then we |
| 3000 | * perform an allocation without specifying a node. This allows the page |
| 3001 | * allocator to do its reclaim / fallback magic. We then insert the |
| 3002 | * slab into the proper nodelist and then allocate from it. |
| 3003 | */ |
| 3004 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
| 3005 | { |
| 3006 | struct zonelist *zonelist; |
| 3007 | gfp_t local_flags; |
| 3008 | struct zoneref *z; |
| 3009 | struct zone *zone; |
| 3010 | enum zone_type high_zoneidx = gfp_zone(flags); |
| 3011 | void *obj = NULL; |
| 3012 | int nid; |
| 3013 | unsigned int cpuset_mems_cookie; |
| 3014 | |
| 3015 | if (flags & __GFP_THISNODE) |
| 3016 | return NULL; |
| 3017 | |
| 3018 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
| 3019 | |
| 3020 | retry_cpuset: |
| 3021 | cpuset_mems_cookie = read_mems_allowed_begin(); |
| 3022 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
| 3023 | |
| 3024 | retry: |
| 3025 | /* |
| 3026 | * Look through allowed nodes for objects available |
| 3027 | * from existing per node queues. |
| 3028 | */ |
| 3029 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
| 3030 | nid = zone_to_nid(zone); |
| 3031 | |
| 3032 | if (cpuset_zone_allowed(zone, flags) && |
| 3033 | get_node(cache, nid) && |
| 3034 | get_node(cache, nid)->free_objects) { |
| 3035 | obj = ____cache_alloc_node(cache, |
| 3036 | gfp_exact_node(flags), nid); |
| 3037 | if (obj) |
| 3038 | break; |
| 3039 | } |
| 3040 | } |
| 3041 | |
| 3042 | if (!obj) { |
| 3043 | /* |
| 3044 | * This allocation will be performed within the constraints |
| 3045 | * of the current cpuset / memory policy requirements. |
| 3046 | * We may trigger various forms of reclaim on the allowed |
| 3047 | * set and go into memory reserves if necessary. |
| 3048 | */ |
| 3049 | struct page *page; |
| 3050 | |
| 3051 | if (local_flags & __GFP_WAIT) |
| 3052 | local_irq_enable(); |
| 3053 | kmem_flagcheck(cache, flags); |
| 3054 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
| 3055 | if (local_flags & __GFP_WAIT) |
| 3056 | local_irq_disable(); |
| 3057 | if (page) { |
| 3058 | /* |
| 3059 | * Insert into the appropriate per node queues |
| 3060 | */ |
| 3061 | nid = page_to_nid(page); |
| 3062 | if (cache_grow(cache, flags, nid, page)) { |
| 3063 | obj = ____cache_alloc_node(cache, |
| 3064 | gfp_exact_node(flags), nid); |
| 3065 | if (!obj) |
| 3066 | /* |
| 3067 | * Another processor may allocate the |
| 3068 | * objects in the slab since we are |
| 3069 | * not holding any locks. |
| 3070 | */ |
| 3071 | goto retry; |
| 3072 | } else { |
| 3073 | /* cache_grow already freed obj */ |
| 3074 | obj = NULL; |
| 3075 | } |
| 3076 | } |
| 3077 | } |
| 3078 | |
| 3079 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
| 3080 | goto retry_cpuset; |
| 3081 | return obj; |
| 3082 | } |
| 3083 | |
| 3084 | /* |
| 3085 | * A interface to enable slab creation on nodeid |
| 3086 | */ |
| 3087 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
| 3088 | int nodeid) |
| 3089 | { |
| 3090 | struct list_head *entry; |
| 3091 | struct page *page; |
| 3092 | struct kmem_cache_node *n; |
| 3093 | void *obj; |
| 3094 | int x; |
| 3095 | |
| 3096 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
| 3097 | n = get_node(cachep, nodeid); |
| 3098 | BUG_ON(!n); |
| 3099 | |
| 3100 | retry: |
| 3101 | check_irq_off(); |
| 3102 | spin_lock(&n->list_lock); |
| 3103 | entry = n->slabs_partial.next; |
| 3104 | if (entry == &n->slabs_partial) { |
| 3105 | n->free_touched = 1; |
| 3106 | entry = n->slabs_free.next; |
| 3107 | if (entry == &n->slabs_free) |
| 3108 | goto must_grow; |
| 3109 | } |
| 3110 | |
| 3111 | page = list_entry(entry, struct page, lru); |
| 3112 | check_spinlock_acquired_node(cachep, nodeid); |
| 3113 | |
| 3114 | STATS_INC_NODEALLOCS(cachep); |
| 3115 | STATS_INC_ACTIVE(cachep); |
| 3116 | STATS_SET_HIGH(cachep); |
| 3117 | |
| 3118 | BUG_ON(page->active == cachep->num); |
| 3119 | |
| 3120 | obj = slab_get_obj(cachep, page, nodeid); |
| 3121 | n->free_objects--; |
| 3122 | /* move slabp to correct slabp list: */ |
| 3123 | list_del(&page->lru); |
| 3124 | |
| 3125 | if (page->active == cachep->num) |
| 3126 | list_add(&page->lru, &n->slabs_full); |
| 3127 | else |
| 3128 | list_add(&page->lru, &n->slabs_partial); |
| 3129 | |
| 3130 | spin_unlock(&n->list_lock); |
| 3131 | goto done; |
| 3132 | |
| 3133 | must_grow: |
| 3134 | spin_unlock(&n->list_lock); |
| 3135 | x = cache_grow(cachep, gfp_exact_node(flags), nodeid, NULL); |
| 3136 | if (x) |
| 3137 | goto retry; |
| 3138 | |
| 3139 | return fallback_alloc(cachep, flags); |
| 3140 | |
| 3141 | done: |
| 3142 | return obj; |
| 3143 | } |
| 3144 | |
| 3145 | static __always_inline void * |
| 3146 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
| 3147 | unsigned long caller) |
| 3148 | { |
| 3149 | unsigned long save_flags; |
| 3150 | void *ptr; |
| 3151 | int slab_node = numa_mem_id(); |
| 3152 | |
| 3153 | flags &= gfp_allowed_mask; |
| 3154 | |
| 3155 | lockdep_trace_alloc(flags); |
| 3156 | |
| 3157 | if (slab_should_failslab(cachep, flags)) |
| 3158 | return NULL; |
| 3159 | |
| 3160 | cachep = memcg_kmem_get_cache(cachep, flags); |
| 3161 | |
| 3162 | cache_alloc_debugcheck_before(cachep, flags); |
| 3163 | local_irq_save(save_flags); |
| 3164 | |
| 3165 | if (nodeid == NUMA_NO_NODE) |
| 3166 | nodeid = slab_node; |
| 3167 | |
| 3168 | if (unlikely(!get_node(cachep, nodeid))) { |
| 3169 | /* Node not bootstrapped yet */ |
| 3170 | ptr = fallback_alloc(cachep, flags); |
| 3171 | goto out; |
| 3172 | } |
| 3173 | |
| 3174 | if (nodeid == slab_node) { |
| 3175 | /* |
| 3176 | * Use the locally cached objects if possible. |
| 3177 | * However ____cache_alloc does not allow fallback |
| 3178 | * to other nodes. It may fail while we still have |
| 3179 | * objects on other nodes available. |
| 3180 | */ |
| 3181 | ptr = ____cache_alloc(cachep, flags); |
| 3182 | if (ptr) |
| 3183 | goto out; |
| 3184 | } |
| 3185 | /* ___cache_alloc_node can fall back to other nodes */ |
| 3186 | ptr = ____cache_alloc_node(cachep, flags, nodeid); |
| 3187 | out: |
| 3188 | local_irq_restore(save_flags); |
| 3189 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); |
| 3190 | kmemleak_alloc_recursive(ptr, cachep->object_size, 1, cachep->flags, |
| 3191 | flags); |
| 3192 | |
| 3193 | if (likely(ptr)) { |
| 3194 | kmemcheck_slab_alloc(cachep, flags, ptr, cachep->object_size); |
| 3195 | if (unlikely(flags & __GFP_ZERO)) |
| 3196 | memset(ptr, 0, cachep->object_size); |
| 3197 | } |
| 3198 | |
| 3199 | memcg_kmem_put_cache(cachep); |
| 3200 | return ptr; |
| 3201 | } |
| 3202 | |
| 3203 | static __always_inline void * |
| 3204 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) |
| 3205 | { |
| 3206 | void *objp; |
| 3207 | |
| 3208 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
| 3209 | objp = alternate_node_alloc(cache, flags); |
| 3210 | if (objp) |
| 3211 | goto out; |
| 3212 | } |
| 3213 | objp = ____cache_alloc(cache, flags); |
| 3214 | |
| 3215 | /* |
| 3216 | * We may just have run out of memory on the local node. |
| 3217 | * ____cache_alloc_node() knows how to locate memory on other nodes |
| 3218 | */ |
| 3219 | if (!objp) |
| 3220 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); |
| 3221 | |
| 3222 | out: |
| 3223 | return objp; |
| 3224 | } |
| 3225 | #else |
| 3226 | |
| 3227 | static __always_inline void * |
| 3228 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
| 3229 | { |
| 3230 | return ____cache_alloc(cachep, flags); |
| 3231 | } |
| 3232 | |
| 3233 | #endif /* CONFIG_NUMA */ |
| 3234 | |
| 3235 | static __always_inline void * |
| 3236 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
| 3237 | { |
| 3238 | unsigned long save_flags; |
| 3239 | void *objp; |
| 3240 | |
| 3241 | flags &= gfp_allowed_mask; |
| 3242 | |
| 3243 | lockdep_trace_alloc(flags); |
| 3244 | |
| 3245 | if (slab_should_failslab(cachep, flags)) |
| 3246 | return NULL; |
| 3247 | |
| 3248 | cachep = memcg_kmem_get_cache(cachep, flags); |
| 3249 | |
| 3250 | cache_alloc_debugcheck_before(cachep, flags); |
| 3251 | local_irq_save(save_flags); |
| 3252 | objp = __do_cache_alloc(cachep, flags); |
| 3253 | local_irq_restore(save_flags); |
| 3254 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); |
| 3255 | kmemleak_alloc_recursive(objp, cachep->object_size, 1, cachep->flags, |
| 3256 | flags); |
| 3257 | prefetchw(objp); |
| 3258 | |
| 3259 | if (likely(objp)) { |
| 3260 | kmemcheck_slab_alloc(cachep, flags, objp, cachep->object_size); |
| 3261 | if (unlikely(flags & __GFP_ZERO)) |
| 3262 | memset(objp, 0, cachep->object_size); |
| 3263 | } |
| 3264 | |
| 3265 | memcg_kmem_put_cache(cachep); |
| 3266 | return objp; |
| 3267 | } |
| 3268 | |
| 3269 | /* |
| 3270 | * Caller needs to acquire correct kmem_cache_node's list_lock |
| 3271 | * @list: List of detached free slabs should be freed by caller |
| 3272 | */ |
| 3273 | static void free_block(struct kmem_cache *cachep, void **objpp, |
| 3274 | int nr_objects, int node, struct list_head *list) |
| 3275 | { |
| 3276 | int i; |
| 3277 | struct kmem_cache_node *n = get_node(cachep, node); |
| 3278 | |
| 3279 | for (i = 0; i < nr_objects; i++) { |
| 3280 | void *objp; |
| 3281 | struct page *page; |
| 3282 | |
| 3283 | clear_obj_pfmemalloc(&objpp[i]); |
| 3284 | objp = objpp[i]; |
| 3285 | |
| 3286 | page = virt_to_head_page(objp); |
| 3287 | list_del(&page->lru); |
| 3288 | check_spinlock_acquired_node(cachep, node); |
| 3289 | slab_put_obj(cachep, page, objp, node); |
| 3290 | STATS_DEC_ACTIVE(cachep); |
| 3291 | n->free_objects++; |
| 3292 | |
| 3293 | /* fixup slab chains */ |
| 3294 | if (page->active == 0) { |
| 3295 | if (n->free_objects > n->free_limit) { |
| 3296 | n->free_objects -= cachep->num; |
| 3297 | list_add_tail(&page->lru, list); |
| 3298 | } else { |
| 3299 | list_add(&page->lru, &n->slabs_free); |
| 3300 | } |
| 3301 | } else { |
| 3302 | /* Unconditionally move a slab to the end of the |
| 3303 | * partial list on free - maximum time for the |
| 3304 | * other objects to be freed, too. |
| 3305 | */ |
| 3306 | list_add_tail(&page->lru, &n->slabs_partial); |
| 3307 | } |
| 3308 | } |
| 3309 | } |
| 3310 | |
| 3311 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
| 3312 | { |
| 3313 | int batchcount; |
| 3314 | struct kmem_cache_node *n; |
| 3315 | int node = numa_mem_id(); |
| 3316 | LIST_HEAD(list); |
| 3317 | |
| 3318 | batchcount = ac->batchcount; |
| 3319 | #if DEBUG |
| 3320 | BUG_ON(!batchcount || batchcount > ac->avail); |
| 3321 | #endif |
| 3322 | check_irq_off(); |
| 3323 | n = get_node(cachep, node); |
| 3324 | spin_lock(&n->list_lock); |
| 3325 | if (n->shared) { |
| 3326 | struct array_cache *shared_array = n->shared; |
| 3327 | int max = shared_array->limit - shared_array->avail; |
| 3328 | if (max) { |
| 3329 | if (batchcount > max) |
| 3330 | batchcount = max; |
| 3331 | memcpy(&(shared_array->entry[shared_array->avail]), |
| 3332 | ac->entry, sizeof(void *) * batchcount); |
| 3333 | shared_array->avail += batchcount; |
| 3334 | goto free_done; |
| 3335 | } |
| 3336 | } |
| 3337 | |
| 3338 | free_block(cachep, ac->entry, batchcount, node, &list); |
| 3339 | free_done: |
| 3340 | #if STATS |
| 3341 | { |
| 3342 | int i = 0; |
| 3343 | struct list_head *p; |
| 3344 | |
| 3345 | p = n->slabs_free.next; |
| 3346 | while (p != &(n->slabs_free)) { |
| 3347 | struct page *page; |
| 3348 | |
| 3349 | page = list_entry(p, struct page, lru); |
| 3350 | BUG_ON(page->active); |
| 3351 | |
| 3352 | i++; |
| 3353 | p = p->next; |
| 3354 | } |
| 3355 | STATS_SET_FREEABLE(cachep, i); |
| 3356 | } |
| 3357 | #endif |
| 3358 | spin_unlock(&n->list_lock); |
| 3359 | slabs_destroy(cachep, &list); |
| 3360 | ac->avail -= batchcount; |
| 3361 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
| 3362 | } |
| 3363 | |
| 3364 | /* |
| 3365 | * Release an obj back to its cache. If the obj has a constructed state, it must |
| 3366 | * be in this state _before_ it is released. Called with disabled ints. |
| 3367 | */ |
| 3368 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
| 3369 | unsigned long caller) |
| 3370 | { |
| 3371 | struct array_cache *ac = cpu_cache_get(cachep); |
| 3372 | |
| 3373 | check_irq_off(); |
| 3374 | kmemleak_free_recursive(objp, cachep->flags); |
| 3375 | objp = cache_free_debugcheck(cachep, objp, caller); |
| 3376 | |
| 3377 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
| 3378 | |
| 3379 | /* |
| 3380 | * Skip calling cache_free_alien() when the platform is not numa. |
| 3381 | * This will avoid cache misses that happen while accessing slabp (which |
| 3382 | * is per page memory reference) to get nodeid. Instead use a global |
| 3383 | * variable to skip the call, which is mostly likely to be present in |
| 3384 | * the cache. |
| 3385 | */ |
| 3386 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
| 3387 | return; |
| 3388 | |
| 3389 | if (ac->avail < ac->limit) { |
| 3390 | STATS_INC_FREEHIT(cachep); |
| 3391 | } else { |
| 3392 | STATS_INC_FREEMISS(cachep); |
| 3393 | cache_flusharray(cachep, ac); |
| 3394 | } |
| 3395 | |
| 3396 | ac_put_obj(cachep, ac, objp); |
| 3397 | } |
| 3398 | |
| 3399 | /** |
| 3400 | * kmem_cache_alloc - Allocate an object |
| 3401 | * @cachep: The cache to allocate from. |
| 3402 | * @flags: See kmalloc(). |
| 3403 | * |
| 3404 | * Allocate an object from this cache. The flags are only relevant |
| 3405 | * if the cache has no available objects. |
| 3406 | */ |
| 3407 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
| 3408 | { |
| 3409 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
| 3410 | |
| 3411 | trace_kmem_cache_alloc(_RET_IP_, ret, |
| 3412 | cachep->object_size, cachep->size, flags); |
| 3413 | |
| 3414 | return ret; |
| 3415 | } |
| 3416 | EXPORT_SYMBOL(kmem_cache_alloc); |
| 3417 | |
| 3418 | #ifdef CONFIG_TRACING |
| 3419 | void * |
| 3420 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
| 3421 | { |
| 3422 | void *ret; |
| 3423 | |
| 3424 | ret = slab_alloc(cachep, flags, _RET_IP_); |
| 3425 | |
| 3426 | trace_kmalloc(_RET_IP_, ret, |
| 3427 | size, cachep->size, flags); |
| 3428 | return ret; |
| 3429 | } |
| 3430 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
| 3431 | #endif |
| 3432 | |
| 3433 | #ifdef CONFIG_NUMA |
| 3434 | /** |
| 3435 | * kmem_cache_alloc_node - Allocate an object on the specified node |
| 3436 | * @cachep: The cache to allocate from. |
| 3437 | * @flags: See kmalloc(). |
| 3438 | * @nodeid: node number of the target node. |
| 3439 | * |
| 3440 | * Identical to kmem_cache_alloc but it will allocate memory on the given |
| 3441 | * node, which can improve the performance for cpu bound structures. |
| 3442 | * |
| 3443 | * Fallback to other node is possible if __GFP_THISNODE is not set. |
| 3444 | */ |
| 3445 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
| 3446 | { |
| 3447 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
| 3448 | |
| 3449 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
| 3450 | cachep->object_size, cachep->size, |
| 3451 | flags, nodeid); |
| 3452 | |
| 3453 | return ret; |
| 3454 | } |
| 3455 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
| 3456 | |
| 3457 | #ifdef CONFIG_TRACING |
| 3458 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
| 3459 | gfp_t flags, |
| 3460 | int nodeid, |
| 3461 | size_t size) |
| 3462 | { |
| 3463 | void *ret; |
| 3464 | |
| 3465 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
| 3466 | |
| 3467 | trace_kmalloc_node(_RET_IP_, ret, |
| 3468 | size, cachep->size, |
| 3469 | flags, nodeid); |
| 3470 | return ret; |
| 3471 | } |
| 3472 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
| 3473 | #endif |
| 3474 | |
| 3475 | static __always_inline void * |
| 3476 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
| 3477 | { |
| 3478 | struct kmem_cache *cachep; |
| 3479 | |
| 3480 | cachep = kmalloc_slab(size, flags); |
| 3481 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
| 3482 | return cachep; |
| 3483 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
| 3484 | } |
| 3485 | |
| 3486 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
| 3487 | { |
| 3488 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
| 3489 | } |
| 3490 | EXPORT_SYMBOL(__kmalloc_node); |
| 3491 | |
| 3492 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, |
| 3493 | int node, unsigned long caller) |
| 3494 | { |
| 3495 | return __do_kmalloc_node(size, flags, node, caller); |
| 3496 | } |
| 3497 | EXPORT_SYMBOL(__kmalloc_node_track_caller); |
| 3498 | #endif /* CONFIG_NUMA */ |
| 3499 | |
| 3500 | /** |
| 3501 | * __do_kmalloc - allocate memory |
| 3502 | * @size: how many bytes of memory are required. |
| 3503 | * @flags: the type of memory to allocate (see kmalloc). |
| 3504 | * @caller: function caller for debug tracking of the caller |
| 3505 | */ |
| 3506 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
| 3507 | unsigned long caller) |
| 3508 | { |
| 3509 | struct kmem_cache *cachep; |
| 3510 | void *ret; |
| 3511 | |
| 3512 | cachep = kmalloc_slab(size, flags); |
| 3513 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
| 3514 | return cachep; |
| 3515 | ret = slab_alloc(cachep, flags, caller); |
| 3516 | |
| 3517 | trace_kmalloc(caller, ret, |
| 3518 | size, cachep->size, flags); |
| 3519 | |
| 3520 | return ret; |
| 3521 | } |
| 3522 | |
| 3523 | void *__kmalloc(size_t size, gfp_t flags) |
| 3524 | { |
| 3525 | return __do_kmalloc(size, flags, _RET_IP_); |
| 3526 | } |
| 3527 | EXPORT_SYMBOL(__kmalloc); |
| 3528 | |
| 3529 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
| 3530 | { |
| 3531 | return __do_kmalloc(size, flags, caller); |
| 3532 | } |
| 3533 | EXPORT_SYMBOL(__kmalloc_track_caller); |
| 3534 | |
| 3535 | /** |
| 3536 | * kmem_cache_free - Deallocate an object |
| 3537 | * @cachep: The cache the allocation was from. |
| 3538 | * @objp: The previously allocated object. |
| 3539 | * |
| 3540 | * Free an object which was previously allocated from this |
| 3541 | * cache. |
| 3542 | */ |
| 3543 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
| 3544 | { |
| 3545 | unsigned long flags; |
| 3546 | cachep = cache_from_obj(cachep, objp); |
| 3547 | if (!cachep) |
| 3548 | return; |
| 3549 | |
| 3550 | local_irq_save(flags); |
| 3551 | debug_check_no_locks_freed(objp, cachep->object_size); |
| 3552 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
| 3553 | debug_check_no_obj_freed(objp, cachep->object_size); |
| 3554 | __cache_free(cachep, objp, _RET_IP_); |
| 3555 | local_irq_restore(flags); |
| 3556 | |
| 3557 | trace_kmem_cache_free(_RET_IP_, objp); |
| 3558 | } |
| 3559 | EXPORT_SYMBOL(kmem_cache_free); |
| 3560 | |
| 3561 | /** |
| 3562 | * kfree - free previously allocated memory |
| 3563 | * @objp: pointer returned by kmalloc. |
| 3564 | * |
| 3565 | * If @objp is NULL, no operation is performed. |
| 3566 | * |
| 3567 | * Don't free memory not originally allocated by kmalloc() |
| 3568 | * or you will run into trouble. |
| 3569 | */ |
| 3570 | void kfree(const void *objp) |
| 3571 | { |
| 3572 | struct kmem_cache *c; |
| 3573 | unsigned long flags; |
| 3574 | |
| 3575 | trace_kfree(_RET_IP_, objp); |
| 3576 | |
| 3577 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
| 3578 | return; |
| 3579 | local_irq_save(flags); |
| 3580 | kfree_debugcheck(objp); |
| 3581 | c = virt_to_cache(objp); |
| 3582 | debug_check_no_locks_freed(objp, c->object_size); |
| 3583 | |
| 3584 | debug_check_no_obj_freed(objp, c->object_size); |
| 3585 | __cache_free(c, (void *)objp, _RET_IP_); |
| 3586 | local_irq_restore(flags); |
| 3587 | } |
| 3588 | EXPORT_SYMBOL(kfree); |
| 3589 | |
| 3590 | /* |
| 3591 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
| 3592 | */ |
| 3593 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
| 3594 | { |
| 3595 | int node; |
| 3596 | struct kmem_cache_node *n; |
| 3597 | struct array_cache *new_shared; |
| 3598 | struct alien_cache **new_alien = NULL; |
| 3599 | |
| 3600 | for_each_online_node(node) { |
| 3601 | |
| 3602 | if (use_alien_caches) { |
| 3603 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); |
| 3604 | if (!new_alien) |
| 3605 | goto fail; |
| 3606 | } |
| 3607 | |
| 3608 | new_shared = NULL; |
| 3609 | if (cachep->shared) { |
| 3610 | new_shared = alloc_arraycache(node, |
| 3611 | cachep->shared*cachep->batchcount, |
| 3612 | 0xbaadf00d, gfp); |
| 3613 | if (!new_shared) { |
| 3614 | free_alien_cache(new_alien); |
| 3615 | goto fail; |
| 3616 | } |
| 3617 | } |
| 3618 | |
| 3619 | n = get_node(cachep, node); |
| 3620 | if (n) { |
| 3621 | struct array_cache *shared = n->shared; |
| 3622 | LIST_HEAD(list); |
| 3623 | |
| 3624 | spin_lock_irq(&n->list_lock); |
| 3625 | |
| 3626 | if (shared) |
| 3627 | free_block(cachep, shared->entry, |
| 3628 | shared->avail, node, &list); |
| 3629 | |
| 3630 | n->shared = new_shared; |
| 3631 | if (!n->alien) { |
| 3632 | n->alien = new_alien; |
| 3633 | new_alien = NULL; |
| 3634 | } |
| 3635 | n->free_limit = (1 + nr_cpus_node(node)) * |
| 3636 | cachep->batchcount + cachep->num; |
| 3637 | spin_unlock_irq(&n->list_lock); |
| 3638 | slabs_destroy(cachep, &list); |
| 3639 | kfree(shared); |
| 3640 | free_alien_cache(new_alien); |
| 3641 | continue; |
| 3642 | } |
| 3643 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
| 3644 | if (!n) { |
| 3645 | free_alien_cache(new_alien); |
| 3646 | kfree(new_shared); |
| 3647 | goto fail; |
| 3648 | } |
| 3649 | |
| 3650 | kmem_cache_node_init(n); |
| 3651 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
| 3652 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; |
| 3653 | n->shared = new_shared; |
| 3654 | n->alien = new_alien; |
| 3655 | n->free_limit = (1 + nr_cpus_node(node)) * |
| 3656 | cachep->batchcount + cachep->num; |
| 3657 | cachep->node[node] = n; |
| 3658 | } |
| 3659 | return 0; |
| 3660 | |
| 3661 | fail: |
| 3662 | if (!cachep->list.next) { |
| 3663 | /* Cache is not active yet. Roll back what we did */ |
| 3664 | node--; |
| 3665 | while (node >= 0) { |
| 3666 | n = get_node(cachep, node); |
| 3667 | if (n) { |
| 3668 | kfree(n->shared); |
| 3669 | free_alien_cache(n->alien); |
| 3670 | kfree(n); |
| 3671 | cachep->node[node] = NULL; |
| 3672 | } |
| 3673 | node--; |
| 3674 | } |
| 3675 | } |
| 3676 | return -ENOMEM; |
| 3677 | } |
| 3678 | |
| 3679 | /* Always called with the slab_mutex held */ |
| 3680 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
| 3681 | int batchcount, int shared, gfp_t gfp) |
| 3682 | { |
| 3683 | struct array_cache __percpu *cpu_cache, *prev; |
| 3684 | int cpu; |
| 3685 | |
| 3686 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
| 3687 | if (!cpu_cache) |
| 3688 | return -ENOMEM; |
| 3689 | |
| 3690 | prev = cachep->cpu_cache; |
| 3691 | cachep->cpu_cache = cpu_cache; |
| 3692 | kick_all_cpus_sync(); |
| 3693 | |
| 3694 | check_irq_on(); |
| 3695 | cachep->batchcount = batchcount; |
| 3696 | cachep->limit = limit; |
| 3697 | cachep->shared = shared; |
| 3698 | |
| 3699 | if (!prev) |
| 3700 | goto alloc_node; |
| 3701 | |
| 3702 | for_each_online_cpu(cpu) { |
| 3703 | LIST_HEAD(list); |
| 3704 | int node; |
| 3705 | struct kmem_cache_node *n; |
| 3706 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
| 3707 | |
| 3708 | node = cpu_to_mem(cpu); |
| 3709 | n = get_node(cachep, node); |
| 3710 | spin_lock_irq(&n->list_lock); |
| 3711 | free_block(cachep, ac->entry, ac->avail, node, &list); |
| 3712 | spin_unlock_irq(&n->list_lock); |
| 3713 | slabs_destroy(cachep, &list); |
| 3714 | } |
| 3715 | free_percpu(prev); |
| 3716 | |
| 3717 | alloc_node: |
| 3718 | return alloc_kmem_cache_node(cachep, gfp); |
| 3719 | } |
| 3720 | |
| 3721 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
| 3722 | int batchcount, int shared, gfp_t gfp) |
| 3723 | { |
| 3724 | int ret; |
| 3725 | struct kmem_cache *c; |
| 3726 | |
| 3727 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); |
| 3728 | |
| 3729 | if (slab_state < FULL) |
| 3730 | return ret; |
| 3731 | |
| 3732 | if ((ret < 0) || !is_root_cache(cachep)) |
| 3733 | return ret; |
| 3734 | |
| 3735 | lockdep_assert_held(&slab_mutex); |
| 3736 | for_each_memcg_cache(c, cachep) { |
| 3737 | /* return value determined by the root cache only */ |
| 3738 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); |
| 3739 | } |
| 3740 | |
| 3741 | return ret; |
| 3742 | } |
| 3743 | |
| 3744 | /* Called with slab_mutex held always */ |
| 3745 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
| 3746 | { |
| 3747 | int err; |
| 3748 | int limit = 0; |
| 3749 | int shared = 0; |
| 3750 | int batchcount = 0; |
| 3751 | |
| 3752 | if (!is_root_cache(cachep)) { |
| 3753 | struct kmem_cache *root = memcg_root_cache(cachep); |
| 3754 | limit = root->limit; |
| 3755 | shared = root->shared; |
| 3756 | batchcount = root->batchcount; |
| 3757 | } |
| 3758 | |
| 3759 | if (limit && shared && batchcount) |
| 3760 | goto skip_setup; |
| 3761 | /* |
| 3762 | * The head array serves three purposes: |
| 3763 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
| 3764 | * - reduce the number of spinlock operations. |
| 3765 | * - reduce the number of linked list operations on the slab and |
| 3766 | * bufctl chains: array operations are cheaper. |
| 3767 | * The numbers are guessed, we should auto-tune as described by |
| 3768 | * Bonwick. |
| 3769 | */ |
| 3770 | if (cachep->size > 131072) |
| 3771 | limit = 1; |
| 3772 | else if (cachep->size > PAGE_SIZE) |
| 3773 | limit = 8; |
| 3774 | else if (cachep->size > 1024) |
| 3775 | limit = 24; |
| 3776 | else if (cachep->size > 256) |
| 3777 | limit = 54; |
| 3778 | else |
| 3779 | limit = 120; |
| 3780 | |
| 3781 | /* |
| 3782 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound |
| 3783 | * allocation behaviour: Most allocs on one cpu, most free operations |
| 3784 | * on another cpu. For these cases, an efficient object passing between |
| 3785 | * cpus is necessary. This is provided by a shared array. The array |
| 3786 | * replaces Bonwick's magazine layer. |
| 3787 | * On uniprocessor, it's functionally equivalent (but less efficient) |
| 3788 | * to a larger limit. Thus disabled by default. |
| 3789 | */ |
| 3790 | shared = 0; |
| 3791 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
| 3792 | shared = 8; |
| 3793 | |
| 3794 | #if DEBUG |
| 3795 | /* |
| 3796 | * With debugging enabled, large batchcount lead to excessively long |
| 3797 | * periods with disabled local interrupts. Limit the batchcount |
| 3798 | */ |
| 3799 | if (limit > 32) |
| 3800 | limit = 32; |
| 3801 | #endif |
| 3802 | batchcount = (limit + 1) / 2; |
| 3803 | skip_setup: |
| 3804 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); |
| 3805 | if (err) |
| 3806 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", |
| 3807 | cachep->name, -err); |
| 3808 | return err; |
| 3809 | } |
| 3810 | |
| 3811 | /* |
| 3812 | * Drain an array if it contains any elements taking the node lock only if |
| 3813 | * necessary. Note that the node listlock also protects the array_cache |
| 3814 | * if drain_array() is used on the shared array. |
| 3815 | */ |
| 3816 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
| 3817 | struct array_cache *ac, int force, int node) |
| 3818 | { |
| 3819 | LIST_HEAD(list); |
| 3820 | int tofree; |
| 3821 | |
| 3822 | if (!ac || !ac->avail) |
| 3823 | return; |
| 3824 | if (ac->touched && !force) { |
| 3825 | ac->touched = 0; |
| 3826 | } else { |
| 3827 | spin_lock_irq(&n->list_lock); |
| 3828 | if (ac->avail) { |
| 3829 | tofree = force ? ac->avail : (ac->limit + 4) / 5; |
| 3830 | if (tofree > ac->avail) |
| 3831 | tofree = (ac->avail + 1) / 2; |
| 3832 | free_block(cachep, ac->entry, tofree, node, &list); |
| 3833 | ac->avail -= tofree; |
| 3834 | memmove(ac->entry, &(ac->entry[tofree]), |
| 3835 | sizeof(void *) * ac->avail); |
| 3836 | } |
| 3837 | spin_unlock_irq(&n->list_lock); |
| 3838 | slabs_destroy(cachep, &list); |
| 3839 | } |
| 3840 | } |
| 3841 | |
| 3842 | /** |
| 3843 | * cache_reap - Reclaim memory from caches. |
| 3844 | * @w: work descriptor |
| 3845 | * |
| 3846 | * Called from workqueue/eventd every few seconds. |
| 3847 | * Purpose: |
| 3848 | * - clear the per-cpu caches for this CPU. |
| 3849 | * - return freeable pages to the main free memory pool. |
| 3850 | * |
| 3851 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
| 3852 | * again on the next iteration. |
| 3853 | */ |
| 3854 | static void cache_reap(struct work_struct *w) |
| 3855 | { |
| 3856 | struct kmem_cache *searchp; |
| 3857 | struct kmem_cache_node *n; |
| 3858 | int node = numa_mem_id(); |
| 3859 | struct delayed_work *work = to_delayed_work(w); |
| 3860 | |
| 3861 | if (!mutex_trylock(&slab_mutex)) |
| 3862 | /* Give up. Setup the next iteration. */ |
| 3863 | goto out; |
| 3864 | |
| 3865 | list_for_each_entry(searchp, &slab_caches, list) { |
| 3866 | check_irq_on(); |
| 3867 | |
| 3868 | /* |
| 3869 | * We only take the node lock if absolutely necessary and we |
| 3870 | * have established with reasonable certainty that |
| 3871 | * we can do some work if the lock was obtained. |
| 3872 | */ |
| 3873 | n = get_node(searchp, node); |
| 3874 | |
| 3875 | reap_alien(searchp, n); |
| 3876 | |
| 3877 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
| 3878 | |
| 3879 | /* |
| 3880 | * These are racy checks but it does not matter |
| 3881 | * if we skip one check or scan twice. |
| 3882 | */ |
| 3883 | if (time_after(n->next_reap, jiffies)) |
| 3884 | goto next; |
| 3885 | |
| 3886 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
| 3887 | |
| 3888 | drain_array(searchp, n, n->shared, 0, node); |
| 3889 | |
| 3890 | if (n->free_touched) |
| 3891 | n->free_touched = 0; |
| 3892 | else { |
| 3893 | int freed; |
| 3894 | |
| 3895 | freed = drain_freelist(searchp, n, (n->free_limit + |
| 3896 | 5 * searchp->num - 1) / (5 * searchp->num)); |
| 3897 | STATS_ADD_REAPED(searchp, freed); |
| 3898 | } |
| 3899 | next: |
| 3900 | cond_resched(); |
| 3901 | } |
| 3902 | check_irq_on(); |
| 3903 | mutex_unlock(&slab_mutex); |
| 3904 | next_reap_node(); |
| 3905 | out: |
| 3906 | /* Set up the next iteration */ |
| 3907 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
| 3908 | } |
| 3909 | |
| 3910 | #ifdef CONFIG_SLABINFO |
| 3911 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
| 3912 | { |
| 3913 | struct page *page; |
| 3914 | unsigned long active_objs; |
| 3915 | unsigned long num_objs; |
| 3916 | unsigned long active_slabs = 0; |
| 3917 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; |
| 3918 | const char *name; |
| 3919 | char *error = NULL; |
| 3920 | int node; |
| 3921 | struct kmem_cache_node *n; |
| 3922 | |
| 3923 | active_objs = 0; |
| 3924 | num_slabs = 0; |
| 3925 | for_each_kmem_cache_node(cachep, node, n) { |
| 3926 | |
| 3927 | check_irq_on(); |
| 3928 | spin_lock_irq(&n->list_lock); |
| 3929 | |
| 3930 | list_for_each_entry(page, &n->slabs_full, lru) { |
| 3931 | if (page->active != cachep->num && !error) |
| 3932 | error = "slabs_full accounting error"; |
| 3933 | active_objs += cachep->num; |
| 3934 | active_slabs++; |
| 3935 | } |
| 3936 | list_for_each_entry(page, &n->slabs_partial, lru) { |
| 3937 | if (page->active == cachep->num && !error) |
| 3938 | error = "slabs_partial accounting error"; |
| 3939 | if (!page->active && !error) |
| 3940 | error = "slabs_partial accounting error"; |
| 3941 | active_objs += page->active; |
| 3942 | active_slabs++; |
| 3943 | } |
| 3944 | list_for_each_entry(page, &n->slabs_free, lru) { |
| 3945 | if (page->active && !error) |
| 3946 | error = "slabs_free accounting error"; |
| 3947 | num_slabs++; |
| 3948 | } |
| 3949 | free_objects += n->free_objects; |
| 3950 | if (n->shared) |
| 3951 | shared_avail += n->shared->avail; |
| 3952 | |
| 3953 | spin_unlock_irq(&n->list_lock); |
| 3954 | } |
| 3955 | num_slabs += active_slabs; |
| 3956 | num_objs = num_slabs * cachep->num; |
| 3957 | if (num_objs - active_objs != free_objects && !error) |
| 3958 | error = "free_objects accounting error"; |
| 3959 | |
| 3960 | name = cachep->name; |
| 3961 | if (error) |
| 3962 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); |
| 3963 | |
| 3964 | sinfo->active_objs = active_objs; |
| 3965 | sinfo->num_objs = num_objs; |
| 3966 | sinfo->active_slabs = active_slabs; |
| 3967 | sinfo->num_slabs = num_slabs; |
| 3968 | sinfo->shared_avail = shared_avail; |
| 3969 | sinfo->limit = cachep->limit; |
| 3970 | sinfo->batchcount = cachep->batchcount; |
| 3971 | sinfo->shared = cachep->shared; |
| 3972 | sinfo->objects_per_slab = cachep->num; |
| 3973 | sinfo->cache_order = cachep->gfporder; |
| 3974 | } |
| 3975 | |
| 3976 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) |
| 3977 | { |
| 3978 | #if STATS |
| 3979 | { /* node stats */ |
| 3980 | unsigned long high = cachep->high_mark; |
| 3981 | unsigned long allocs = cachep->num_allocations; |
| 3982 | unsigned long grown = cachep->grown; |
| 3983 | unsigned long reaped = cachep->reaped; |
| 3984 | unsigned long errors = cachep->errors; |
| 3985 | unsigned long max_freeable = cachep->max_freeable; |
| 3986 | unsigned long node_allocs = cachep->node_allocs; |
| 3987 | unsigned long node_frees = cachep->node_frees; |
| 3988 | unsigned long overflows = cachep->node_overflow; |
| 3989 | |
| 3990 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
| 3991 | "%4lu %4lu %4lu %4lu %4lu", |
| 3992 | allocs, high, grown, |
| 3993 | reaped, errors, max_freeable, node_allocs, |
| 3994 | node_frees, overflows); |
| 3995 | } |
| 3996 | /* cpu stats */ |
| 3997 | { |
| 3998 | unsigned long allochit = atomic_read(&cachep->allochit); |
| 3999 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); |
| 4000 | unsigned long freehit = atomic_read(&cachep->freehit); |
| 4001 | unsigned long freemiss = atomic_read(&cachep->freemiss); |
| 4002 | |
| 4003 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", |
| 4004 | allochit, allocmiss, freehit, freemiss); |
| 4005 | } |
| 4006 | #endif |
| 4007 | } |
| 4008 | |
| 4009 | #define MAX_SLABINFO_WRITE 128 |
| 4010 | /** |
| 4011 | * slabinfo_write - Tuning for the slab allocator |
| 4012 | * @file: unused |
| 4013 | * @buffer: user buffer |
| 4014 | * @count: data length |
| 4015 | * @ppos: unused |
| 4016 | */ |
| 4017 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
| 4018 | size_t count, loff_t *ppos) |
| 4019 | { |
| 4020 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
| 4021 | int limit, batchcount, shared, res; |
| 4022 | struct kmem_cache *cachep; |
| 4023 | |
| 4024 | if (count > MAX_SLABINFO_WRITE) |
| 4025 | return -EINVAL; |
| 4026 | if (copy_from_user(&kbuf, buffer, count)) |
| 4027 | return -EFAULT; |
| 4028 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
| 4029 | |
| 4030 | tmp = strchr(kbuf, ' '); |
| 4031 | if (!tmp) |
| 4032 | return -EINVAL; |
| 4033 | *tmp = '\0'; |
| 4034 | tmp++; |
| 4035 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) |
| 4036 | return -EINVAL; |
| 4037 | |
| 4038 | /* Find the cache in the chain of caches. */ |
| 4039 | mutex_lock(&slab_mutex); |
| 4040 | res = -EINVAL; |
| 4041 | list_for_each_entry(cachep, &slab_caches, list) { |
| 4042 | if (!strcmp(cachep->name, kbuf)) { |
| 4043 | if (limit < 1 || batchcount < 1 || |
| 4044 | batchcount > limit || shared < 0) { |
| 4045 | res = 0; |
| 4046 | } else { |
| 4047 | res = do_tune_cpucache(cachep, limit, |
| 4048 | batchcount, shared, |
| 4049 | GFP_KERNEL); |
| 4050 | } |
| 4051 | break; |
| 4052 | } |
| 4053 | } |
| 4054 | mutex_unlock(&slab_mutex); |
| 4055 | if (res >= 0) |
| 4056 | res = count; |
| 4057 | return res; |
| 4058 | } |
| 4059 | |
| 4060 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
| 4061 | |
| 4062 | static inline int add_caller(unsigned long *n, unsigned long v) |
| 4063 | { |
| 4064 | unsigned long *p; |
| 4065 | int l; |
| 4066 | if (!v) |
| 4067 | return 1; |
| 4068 | l = n[1]; |
| 4069 | p = n + 2; |
| 4070 | while (l) { |
| 4071 | int i = l/2; |
| 4072 | unsigned long *q = p + 2 * i; |
| 4073 | if (*q == v) { |
| 4074 | q[1]++; |
| 4075 | return 1; |
| 4076 | } |
| 4077 | if (*q > v) { |
| 4078 | l = i; |
| 4079 | } else { |
| 4080 | p = q + 2; |
| 4081 | l -= i + 1; |
| 4082 | } |
| 4083 | } |
| 4084 | if (++n[1] == n[0]) |
| 4085 | return 0; |
| 4086 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); |
| 4087 | p[0] = v; |
| 4088 | p[1] = 1; |
| 4089 | return 1; |
| 4090 | } |
| 4091 | |
| 4092 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
| 4093 | struct page *page) |
| 4094 | { |
| 4095 | void *p; |
| 4096 | int i; |
| 4097 | |
| 4098 | if (n[0] == n[1]) |
| 4099 | return; |
| 4100 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
| 4101 | if (get_obj_status(page, i) != OBJECT_ACTIVE) |
| 4102 | continue; |
| 4103 | |
| 4104 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) |
| 4105 | return; |
| 4106 | } |
| 4107 | } |
| 4108 | |
| 4109 | static void show_symbol(struct seq_file *m, unsigned long address) |
| 4110 | { |
| 4111 | #ifdef CONFIG_KALLSYMS |
| 4112 | unsigned long offset, size; |
| 4113 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
| 4114 | |
| 4115 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
| 4116 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
| 4117 | if (modname[0]) |
| 4118 | seq_printf(m, " [%s]", modname); |
| 4119 | return; |
| 4120 | } |
| 4121 | #endif |
| 4122 | seq_printf(m, "%p", (void *)address); |
| 4123 | } |
| 4124 | |
| 4125 | static int leaks_show(struct seq_file *m, void *p) |
| 4126 | { |
| 4127 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
| 4128 | struct page *page; |
| 4129 | struct kmem_cache_node *n; |
| 4130 | const char *name; |
| 4131 | unsigned long *x = m->private; |
| 4132 | int node; |
| 4133 | int i; |
| 4134 | |
| 4135 | if (!(cachep->flags & SLAB_STORE_USER)) |
| 4136 | return 0; |
| 4137 | if (!(cachep->flags & SLAB_RED_ZONE)) |
| 4138 | return 0; |
| 4139 | |
| 4140 | /* OK, we can do it */ |
| 4141 | |
| 4142 | x[1] = 0; |
| 4143 | |
| 4144 | for_each_kmem_cache_node(cachep, node, n) { |
| 4145 | |
| 4146 | check_irq_on(); |
| 4147 | spin_lock_irq(&n->list_lock); |
| 4148 | |
| 4149 | list_for_each_entry(page, &n->slabs_full, lru) |
| 4150 | handle_slab(x, cachep, page); |
| 4151 | list_for_each_entry(page, &n->slabs_partial, lru) |
| 4152 | handle_slab(x, cachep, page); |
| 4153 | spin_unlock_irq(&n->list_lock); |
| 4154 | } |
| 4155 | name = cachep->name; |
| 4156 | if (x[0] == x[1]) { |
| 4157 | /* Increase the buffer size */ |
| 4158 | mutex_unlock(&slab_mutex); |
| 4159 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
| 4160 | if (!m->private) { |
| 4161 | /* Too bad, we are really out */ |
| 4162 | m->private = x; |
| 4163 | mutex_lock(&slab_mutex); |
| 4164 | return -ENOMEM; |
| 4165 | } |
| 4166 | *(unsigned long *)m->private = x[0] * 2; |
| 4167 | kfree(x); |
| 4168 | mutex_lock(&slab_mutex); |
| 4169 | /* Now make sure this entry will be retried */ |
| 4170 | m->count = m->size; |
| 4171 | return 0; |
| 4172 | } |
| 4173 | for (i = 0; i < x[1]; i++) { |
| 4174 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); |
| 4175 | show_symbol(m, x[2*i+2]); |
| 4176 | seq_putc(m, '\n'); |
| 4177 | } |
| 4178 | |
| 4179 | return 0; |
| 4180 | } |
| 4181 | |
| 4182 | static const struct seq_operations slabstats_op = { |
| 4183 | .start = slab_start, |
| 4184 | .next = slab_next, |
| 4185 | .stop = slab_stop, |
| 4186 | .show = leaks_show, |
| 4187 | }; |
| 4188 | |
| 4189 | static int slabstats_open(struct inode *inode, struct file *file) |
| 4190 | { |
| 4191 | unsigned long *n; |
| 4192 | |
| 4193 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); |
| 4194 | if (!n) |
| 4195 | return -ENOMEM; |
| 4196 | |
| 4197 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); |
| 4198 | |
| 4199 | return 0; |
| 4200 | } |
| 4201 | |
| 4202 | static const struct file_operations proc_slabstats_operations = { |
| 4203 | .open = slabstats_open, |
| 4204 | .read = seq_read, |
| 4205 | .llseek = seq_lseek, |
| 4206 | .release = seq_release_private, |
| 4207 | }; |
| 4208 | #endif |
| 4209 | |
| 4210 | static int __init slab_proc_init(void) |
| 4211 | { |
| 4212 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
| 4213 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); |
| 4214 | #endif |
| 4215 | return 0; |
| 4216 | } |
| 4217 | module_init(slab_proc_init); |
| 4218 | #endif |
| 4219 | |
| 4220 | /** |
| 4221 | * ksize - get the actual amount of memory allocated for a given object |
| 4222 | * @objp: Pointer to the object |
| 4223 | * |
| 4224 | * kmalloc may internally round up allocations and return more memory |
| 4225 | * than requested. ksize() can be used to determine the actual amount of |
| 4226 | * memory allocated. The caller may use this additional memory, even though |
| 4227 | * a smaller amount of memory was initially specified with the kmalloc call. |
| 4228 | * The caller must guarantee that objp points to a valid object previously |
| 4229 | * allocated with either kmalloc() or kmem_cache_alloc(). The object |
| 4230 | * must not be freed during the duration of the call. |
| 4231 | */ |
| 4232 | size_t ksize(const void *objp) |
| 4233 | { |
| 4234 | BUG_ON(!objp); |
| 4235 | if (unlikely(objp == ZERO_SIZE_PTR)) |
| 4236 | return 0; |
| 4237 | |
| 4238 | return virt_to_cache(objp)->object_size; |
| 4239 | } |
| 4240 | EXPORT_SYMBOL(ksize); |