Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
5 | * The allocator synchronizes using per slab locks and only | |
6 | * uses a centralized lock to manage a pool of partial slabs. | |
7 | * | |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
81819f0f CL |
9 | */ |
10 | ||
11 | #include <linux/mm.h> | |
1eb5ac64 | 12 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
13 | #include <linux/module.h> |
14 | #include <linux/bit_spinlock.h> | |
15 | #include <linux/interrupt.h> | |
16 | #include <linux/bitops.h> | |
17 | #include <linux/slab.h> | |
7b3c3a50 | 18 | #include <linux/proc_fs.h> |
81819f0f | 19 | #include <linux/seq_file.h> |
02af61bb | 20 | #include <linux/kmemtrace.h> |
5a896d9e | 21 | #include <linux/kmemcheck.h> |
81819f0f CL |
22 | #include <linux/cpu.h> |
23 | #include <linux/cpuset.h> | |
06f22f13 | 24 | #include <linux/kmemleak.h> |
81819f0f CL |
25 | #include <linux/mempolicy.h> |
26 | #include <linux/ctype.h> | |
3ac7fe5a | 27 | #include <linux/debugobjects.h> |
81819f0f | 28 | #include <linux/kallsyms.h> |
b9049e23 | 29 | #include <linux/memory.h> |
f8bd2258 | 30 | #include <linux/math64.h> |
773ff60e | 31 | #include <linux/fault-inject.h> |
81819f0f CL |
32 | |
33 | /* | |
34 | * Lock order: | |
35 | * 1. slab_lock(page) | |
36 | * 2. slab->list_lock | |
37 | * | |
38 | * The slab_lock protects operations on the object of a particular | |
39 | * slab and its metadata in the page struct. If the slab lock | |
40 | * has been taken then no allocations nor frees can be performed | |
41 | * on the objects in the slab nor can the slab be added or removed | |
42 | * from the partial or full lists since this would mean modifying | |
43 | * the page_struct of the slab. | |
44 | * | |
45 | * The list_lock protects the partial and full list on each node and | |
46 | * the partial slab counter. If taken then no new slabs may be added or | |
47 | * removed from the lists nor make the number of partial slabs be modified. | |
48 | * (Note that the total number of slabs is an atomic value that may be | |
49 | * modified without taking the list lock). | |
50 | * | |
51 | * The list_lock is a centralized lock and thus we avoid taking it as | |
52 | * much as possible. As long as SLUB does not have to handle partial | |
53 | * slabs, operations can continue without any centralized lock. F.e. | |
54 | * allocating a long series of objects that fill up slabs does not require | |
55 | * the list lock. | |
56 | * | |
57 | * The lock order is sometimes inverted when we are trying to get a slab | |
58 | * off a list. We take the list_lock and then look for a page on the list | |
59 | * to use. While we do that objects in the slabs may be freed. We can | |
60 | * only operate on the slab if we have also taken the slab_lock. So we use | |
61 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
62 | * can occur anymore and we can use the slab for allocations etc. If the | |
63 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
64 | * we must stay away from it for a while since we may cause a bouncing | |
65 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
66 | * If all pages are busy then we may allocate a new slab instead of reusing | |
67 | * a partial slab. A new slab has noone operating on it and thus there is | |
68 | * no danger of cacheline contention. | |
69 | * | |
70 | * Interrupts are disabled during allocation and deallocation in order to | |
71 | * make the slab allocator safe to use in the context of an irq. In addition | |
72 | * interrupts are disabled to ensure that the processor does not change | |
73 | * while handling per_cpu slabs, due to kernel preemption. | |
74 | * | |
75 | * SLUB assigns one slab for allocation to each processor. | |
76 | * Allocations only occur from these slabs called cpu slabs. | |
77 | * | |
672bba3a CL |
78 | * Slabs with free elements are kept on a partial list and during regular |
79 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 80 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
81 | * We track full slabs for debugging purposes though because otherwise we |
82 | * cannot scan all objects. | |
81819f0f CL |
83 | * |
84 | * Slabs are freed when they become empty. Teardown and setup is | |
85 | * minimal so we rely on the page allocators per cpu caches for | |
86 | * fast frees and allocs. | |
87 | * | |
88 | * Overloading of page flags that are otherwise used for LRU management. | |
89 | * | |
4b6f0750 CL |
90 | * PageActive The slab is frozen and exempt from list processing. |
91 | * This means that the slab is dedicated to a purpose | |
92 | * such as satisfying allocations for a specific | |
93 | * processor. Objects may be freed in the slab while | |
94 | * it is frozen but slab_free will then skip the usual | |
95 | * list operations. It is up to the processor holding | |
96 | * the slab to integrate the slab into the slab lists | |
97 | * when the slab is no longer needed. | |
98 | * | |
99 | * One use of this flag is to mark slabs that are | |
100 | * used for allocations. Then such a slab becomes a cpu | |
101 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 102 | * freelist that allows lockless access to |
894b8788 CL |
103 | * free objects in addition to the regular freelist |
104 | * that requires the slab lock. | |
81819f0f CL |
105 | * |
106 | * PageError Slab requires special handling due to debug | |
107 | * options set. This moves slab handling out of | |
894b8788 | 108 | * the fast path and disables lockless freelists. |
81819f0f CL |
109 | */ |
110 | ||
5577bd8a | 111 | #ifdef CONFIG_SLUB_DEBUG |
8a38082d | 112 | #define SLABDEBUG 1 |
5577bd8a CL |
113 | #else |
114 | #define SLABDEBUG 0 | |
115 | #endif | |
116 | ||
81819f0f CL |
117 | /* |
118 | * Issues still to be resolved: | |
119 | * | |
81819f0f CL |
120 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
121 | * | |
81819f0f CL |
122 | * - Variable sizing of the per node arrays |
123 | */ | |
124 | ||
125 | /* Enable to test recovery from slab corruption on boot */ | |
126 | #undef SLUB_RESILIENCY_TEST | |
127 | ||
2086d26a CL |
128 | /* |
129 | * Mininum number of partial slabs. These will be left on the partial | |
130 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
131 | */ | |
76be8950 | 132 | #define MIN_PARTIAL 5 |
e95eed57 | 133 | |
2086d26a CL |
134 | /* |
135 | * Maximum number of desirable partial slabs. | |
136 | * The existence of more partial slabs makes kmem_cache_shrink | |
137 | * sort the partial list by the number of objects in the. | |
138 | */ | |
139 | #define MAX_PARTIAL 10 | |
140 | ||
81819f0f CL |
141 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
142 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 143 | |
81819f0f CL |
144 | /* |
145 | * Set of flags that will prevent slab merging | |
146 | */ | |
147 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
06f22f13 | 148 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE) |
81819f0f CL |
149 | |
150 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 151 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f CL |
152 | |
153 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 154 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
155 | #endif |
156 | ||
157 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 158 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
159 | #endif |
160 | ||
210b5c06 CG |
161 | #define OO_SHIFT 16 |
162 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
163 | #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */ | |
164 | ||
81819f0f | 165 | /* Internal SLUB flags */ |
1ceef402 CL |
166 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
167 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
81819f0f CL |
168 | |
169 | static int kmem_size = sizeof(struct kmem_cache); | |
170 | ||
171 | #ifdef CONFIG_SMP | |
172 | static struct notifier_block slab_notifier; | |
173 | #endif | |
174 | ||
175 | static enum { | |
176 | DOWN, /* No slab functionality available */ | |
177 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 178 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
179 | SYSFS /* Sysfs up */ |
180 | } slab_state = DOWN; | |
181 | ||
7e85ee0c PE |
182 | /* |
183 | * The slab allocator is initialized with interrupts disabled. Therefore, make | |
184 | * sure early boot allocations don't accidentally enable interrupts. | |
185 | */ | |
186 | static gfp_t slab_gfp_mask __read_mostly = SLAB_GFP_BOOT_MASK; | |
187 | ||
81819f0f CL |
188 | /* A list of all slab caches on the system */ |
189 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 190 | static LIST_HEAD(slab_caches); |
81819f0f | 191 | |
02cbc874 CL |
192 | /* |
193 | * Tracking user of a slab. | |
194 | */ | |
195 | struct track { | |
ce71e27c | 196 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
197 | int cpu; /* Was running on cpu */ |
198 | int pid; /* Pid context */ | |
199 | unsigned long when; /* When did the operation occur */ | |
200 | }; | |
201 | ||
202 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
203 | ||
f6acb635 | 204 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
205 | static int sysfs_slab_add(struct kmem_cache *); |
206 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
207 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 208 | |
81819f0f | 209 | #else |
0c710013 CL |
210 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
211 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
212 | { return 0; } | |
151c602f CL |
213 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
214 | { | |
215 | kfree(s); | |
216 | } | |
8ff12cfc | 217 | |
81819f0f CL |
218 | #endif |
219 | ||
8ff12cfc CL |
220 | static inline void stat(struct kmem_cache_cpu *c, enum stat_item si) |
221 | { | |
222 | #ifdef CONFIG_SLUB_STATS | |
223 | c->stat[si]++; | |
224 | #endif | |
225 | } | |
226 | ||
81819f0f CL |
227 | /******************************************************************** |
228 | * Core slab cache functions | |
229 | *******************************************************************/ | |
230 | ||
231 | int slab_is_available(void) | |
232 | { | |
233 | return slab_state >= UP; | |
234 | } | |
235 | ||
236 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
237 | { | |
238 | #ifdef CONFIG_NUMA | |
239 | return s->node[node]; | |
240 | #else | |
241 | return &s->local_node; | |
242 | #endif | |
243 | } | |
244 | ||
dfb4f096 CL |
245 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
246 | { | |
4c93c355 CL |
247 | #ifdef CONFIG_SMP |
248 | return s->cpu_slab[cpu]; | |
249 | #else | |
250 | return &s->cpu_slab; | |
251 | #endif | |
dfb4f096 CL |
252 | } |
253 | ||
6446faa2 | 254 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
255 | static inline int check_valid_pointer(struct kmem_cache *s, |
256 | struct page *page, const void *object) | |
257 | { | |
258 | void *base; | |
259 | ||
a973e9dd | 260 | if (!object) |
02cbc874 CL |
261 | return 1; |
262 | ||
a973e9dd | 263 | base = page_address(page); |
39b26464 | 264 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
265 | (object - base) % s->size) { |
266 | return 0; | |
267 | } | |
268 | ||
269 | return 1; | |
270 | } | |
271 | ||
7656c72b CL |
272 | /* |
273 | * Slow version of get and set free pointer. | |
274 | * | |
275 | * This version requires touching the cache lines of kmem_cache which | |
276 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
277 | * from the page struct. | |
278 | */ | |
279 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
280 | { | |
281 | return *(void **)(object + s->offset); | |
282 | } | |
283 | ||
284 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
285 | { | |
286 | *(void **)(object + s->offset) = fp; | |
287 | } | |
288 | ||
289 | /* Loop over all objects in a slab */ | |
224a88be CL |
290 | #define for_each_object(__p, __s, __addr, __objects) \ |
291 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
292 | __p += (__s)->size) |
293 | ||
294 | /* Scan freelist */ | |
295 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 296 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
297 | |
298 | /* Determine object index from a given position */ | |
299 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
300 | { | |
301 | return (p - addr) / s->size; | |
302 | } | |
303 | ||
834f3d11 CL |
304 | static inline struct kmem_cache_order_objects oo_make(int order, |
305 | unsigned long size) | |
306 | { | |
307 | struct kmem_cache_order_objects x = { | |
210b5c06 | 308 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size |
834f3d11 CL |
309 | }; |
310 | ||
311 | return x; | |
312 | } | |
313 | ||
314 | static inline int oo_order(struct kmem_cache_order_objects x) | |
315 | { | |
210b5c06 | 316 | return x.x >> OO_SHIFT; |
834f3d11 CL |
317 | } |
318 | ||
319 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
320 | { | |
210b5c06 | 321 | return x.x & OO_MASK; |
834f3d11 CL |
322 | } |
323 | ||
41ecc55b CL |
324 | #ifdef CONFIG_SLUB_DEBUG |
325 | /* | |
326 | * Debug settings: | |
327 | */ | |
f0630fff CL |
328 | #ifdef CONFIG_SLUB_DEBUG_ON |
329 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
330 | #else | |
41ecc55b | 331 | static int slub_debug; |
f0630fff | 332 | #endif |
41ecc55b CL |
333 | |
334 | static char *slub_debug_slabs; | |
335 | ||
81819f0f CL |
336 | /* |
337 | * Object debugging | |
338 | */ | |
339 | static void print_section(char *text, u8 *addr, unsigned int length) | |
340 | { | |
341 | int i, offset; | |
342 | int newline = 1; | |
343 | char ascii[17]; | |
344 | ||
345 | ascii[16] = 0; | |
346 | ||
347 | for (i = 0; i < length; i++) { | |
348 | if (newline) { | |
24922684 | 349 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
350 | newline = 0; |
351 | } | |
06428780 | 352 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
353 | offset = i % 16; |
354 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
355 | if (offset == 15) { | |
06428780 | 356 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
357 | newline = 1; |
358 | } | |
359 | } | |
360 | if (!newline) { | |
361 | i %= 16; | |
362 | while (i < 16) { | |
06428780 | 363 | printk(KERN_CONT " "); |
81819f0f CL |
364 | ascii[i] = ' '; |
365 | i++; | |
366 | } | |
06428780 | 367 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
368 | } |
369 | } | |
370 | ||
81819f0f CL |
371 | static struct track *get_track(struct kmem_cache *s, void *object, |
372 | enum track_item alloc) | |
373 | { | |
374 | struct track *p; | |
375 | ||
376 | if (s->offset) | |
377 | p = object + s->offset + sizeof(void *); | |
378 | else | |
379 | p = object + s->inuse; | |
380 | ||
381 | return p + alloc; | |
382 | } | |
383 | ||
384 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 385 | enum track_item alloc, unsigned long addr) |
81819f0f | 386 | { |
1a00df4a | 387 | struct track *p = get_track(s, object, alloc); |
81819f0f | 388 | |
81819f0f CL |
389 | if (addr) { |
390 | p->addr = addr; | |
391 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 392 | p->pid = current->pid; |
81819f0f CL |
393 | p->when = jiffies; |
394 | } else | |
395 | memset(p, 0, sizeof(struct track)); | |
396 | } | |
397 | ||
81819f0f CL |
398 | static void init_tracking(struct kmem_cache *s, void *object) |
399 | { | |
24922684 CL |
400 | if (!(s->flags & SLAB_STORE_USER)) |
401 | return; | |
402 | ||
ce71e27c EGM |
403 | set_track(s, object, TRACK_FREE, 0UL); |
404 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
405 | } |
406 | ||
407 | static void print_track(const char *s, struct track *t) | |
408 | { | |
409 | if (!t->addr) | |
410 | return; | |
411 | ||
7daf705f | 412 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 413 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
414 | } |
415 | ||
416 | static void print_tracking(struct kmem_cache *s, void *object) | |
417 | { | |
418 | if (!(s->flags & SLAB_STORE_USER)) | |
419 | return; | |
420 | ||
421 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
422 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
423 | } | |
424 | ||
425 | static void print_page_info(struct page *page) | |
426 | { | |
39b26464 CL |
427 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
428 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
429 | |
430 | } | |
431 | ||
432 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
433 | { | |
434 | va_list args; | |
435 | char buf[100]; | |
436 | ||
437 | va_start(args, fmt); | |
438 | vsnprintf(buf, sizeof(buf), fmt, args); | |
439 | va_end(args); | |
440 | printk(KERN_ERR "========================================" | |
441 | "=====================================\n"); | |
442 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
443 | printk(KERN_ERR "----------------------------------------" | |
444 | "-------------------------------------\n\n"); | |
81819f0f CL |
445 | } |
446 | ||
24922684 CL |
447 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
448 | { | |
449 | va_list args; | |
450 | char buf[100]; | |
451 | ||
452 | va_start(args, fmt); | |
453 | vsnprintf(buf, sizeof(buf), fmt, args); | |
454 | va_end(args); | |
455 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
456 | } | |
457 | ||
458 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
459 | { |
460 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 461 | u8 *addr = page_address(page); |
24922684 CL |
462 | |
463 | print_tracking(s, p); | |
464 | ||
465 | print_page_info(page); | |
466 | ||
467 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
468 | p, p - addr, get_freepointer(s, p)); | |
469 | ||
470 | if (p > addr + 16) | |
471 | print_section("Bytes b4", p - 16, 16); | |
472 | ||
0ebd652b | 473 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
474 | |
475 | if (s->flags & SLAB_RED_ZONE) | |
476 | print_section("Redzone", p + s->objsize, | |
477 | s->inuse - s->objsize); | |
478 | ||
81819f0f CL |
479 | if (s->offset) |
480 | off = s->offset + sizeof(void *); | |
481 | else | |
482 | off = s->inuse; | |
483 | ||
24922684 | 484 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 485 | off += 2 * sizeof(struct track); |
81819f0f CL |
486 | |
487 | if (off != s->size) | |
488 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
489 | print_section("Padding", p + off, s->size - off); |
490 | ||
491 | dump_stack(); | |
81819f0f CL |
492 | } |
493 | ||
494 | static void object_err(struct kmem_cache *s, struct page *page, | |
495 | u8 *object, char *reason) | |
496 | { | |
3dc50637 | 497 | slab_bug(s, "%s", reason); |
24922684 | 498 | print_trailer(s, page, object); |
81819f0f CL |
499 | } |
500 | ||
24922684 | 501 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
502 | { |
503 | va_list args; | |
504 | char buf[100]; | |
505 | ||
24922684 CL |
506 | va_start(args, fmt); |
507 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 508 | va_end(args); |
3dc50637 | 509 | slab_bug(s, "%s", buf); |
24922684 | 510 | print_page_info(page); |
81819f0f CL |
511 | dump_stack(); |
512 | } | |
513 | ||
514 | static void init_object(struct kmem_cache *s, void *object, int active) | |
515 | { | |
516 | u8 *p = object; | |
517 | ||
518 | if (s->flags & __OBJECT_POISON) { | |
519 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 520 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
521 | } |
522 | ||
523 | if (s->flags & SLAB_RED_ZONE) | |
524 | memset(p + s->objsize, | |
525 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
526 | s->inuse - s->objsize); | |
527 | } | |
528 | ||
24922684 | 529 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
530 | { |
531 | while (bytes) { | |
532 | if (*start != (u8)value) | |
24922684 | 533 | return start; |
81819f0f CL |
534 | start++; |
535 | bytes--; | |
536 | } | |
24922684 CL |
537 | return NULL; |
538 | } | |
539 | ||
540 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
541 | void *from, void *to) | |
542 | { | |
543 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
544 | memset(from, data, to - from); | |
545 | } | |
546 | ||
547 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
548 | u8 *object, char *what, | |
06428780 | 549 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
550 | { |
551 | u8 *fault; | |
552 | u8 *end; | |
553 | ||
554 | fault = check_bytes(start, value, bytes); | |
555 | if (!fault) | |
556 | return 1; | |
557 | ||
558 | end = start + bytes; | |
559 | while (end > fault && end[-1] == value) | |
560 | end--; | |
561 | ||
562 | slab_bug(s, "%s overwritten", what); | |
563 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
564 | fault, end - 1, fault[0], value); | |
565 | print_trailer(s, page, object); | |
566 | ||
567 | restore_bytes(s, what, value, fault, end); | |
568 | return 0; | |
81819f0f CL |
569 | } |
570 | ||
81819f0f CL |
571 | /* |
572 | * Object layout: | |
573 | * | |
574 | * object address | |
575 | * Bytes of the object to be managed. | |
576 | * If the freepointer may overlay the object then the free | |
577 | * pointer is the first word of the object. | |
672bba3a | 578 | * |
81819f0f CL |
579 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
580 | * 0xa5 (POISON_END) | |
581 | * | |
582 | * object + s->objsize | |
583 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
584 | * Padding is extended by another word if Redzoning is enabled and |
585 | * objsize == inuse. | |
586 | * | |
81819f0f CL |
587 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
588 | * 0xcc (RED_ACTIVE) for objects in use. | |
589 | * | |
590 | * object + s->inuse | |
672bba3a CL |
591 | * Meta data starts here. |
592 | * | |
81819f0f CL |
593 | * A. Free pointer (if we cannot overwrite object on free) |
594 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 595 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 596 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
597 | * before the word boundary. |
598 | * | |
599 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
600 | * |
601 | * object + s->size | |
672bba3a | 602 | * Nothing is used beyond s->size. |
81819f0f | 603 | * |
672bba3a CL |
604 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
605 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
606 | * may be used with merged slabcaches. |
607 | */ | |
608 | ||
81819f0f CL |
609 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
610 | { | |
611 | unsigned long off = s->inuse; /* The end of info */ | |
612 | ||
613 | if (s->offset) | |
614 | /* Freepointer is placed after the object. */ | |
615 | off += sizeof(void *); | |
616 | ||
617 | if (s->flags & SLAB_STORE_USER) | |
618 | /* We also have user information there */ | |
619 | off += 2 * sizeof(struct track); | |
620 | ||
621 | if (s->size == off) | |
622 | return 1; | |
623 | ||
24922684 CL |
624 | return check_bytes_and_report(s, page, p, "Object padding", |
625 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
626 | } |
627 | ||
39b26464 | 628 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
629 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
630 | { | |
24922684 CL |
631 | u8 *start; |
632 | u8 *fault; | |
633 | u8 *end; | |
634 | int length; | |
635 | int remainder; | |
81819f0f CL |
636 | |
637 | if (!(s->flags & SLAB_POISON)) | |
638 | return 1; | |
639 | ||
a973e9dd | 640 | start = page_address(page); |
834f3d11 | 641 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
642 | end = start + length; |
643 | remainder = length % s->size; | |
81819f0f CL |
644 | if (!remainder) |
645 | return 1; | |
646 | ||
39b26464 | 647 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
648 | if (!fault) |
649 | return 1; | |
650 | while (end > fault && end[-1] == POISON_INUSE) | |
651 | end--; | |
652 | ||
653 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 654 | print_section("Padding", end - remainder, remainder); |
24922684 CL |
655 | |
656 | restore_bytes(s, "slab padding", POISON_INUSE, start, end); | |
657 | return 0; | |
81819f0f CL |
658 | } |
659 | ||
660 | static int check_object(struct kmem_cache *s, struct page *page, | |
661 | void *object, int active) | |
662 | { | |
663 | u8 *p = object; | |
664 | u8 *endobject = object + s->objsize; | |
665 | ||
666 | if (s->flags & SLAB_RED_ZONE) { | |
667 | unsigned int red = | |
668 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
669 | ||
24922684 CL |
670 | if (!check_bytes_and_report(s, page, object, "Redzone", |
671 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 672 | return 0; |
81819f0f | 673 | } else { |
3adbefee IM |
674 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
675 | check_bytes_and_report(s, page, p, "Alignment padding", | |
676 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
677 | } | |
81819f0f CL |
678 | } |
679 | ||
680 | if (s->flags & SLAB_POISON) { | |
681 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
682 | (!check_bytes_and_report(s, page, p, "Poison", p, |
683 | POISON_FREE, s->objsize - 1) || | |
684 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 685 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 686 | return 0; |
81819f0f CL |
687 | /* |
688 | * check_pad_bytes cleans up on its own. | |
689 | */ | |
690 | check_pad_bytes(s, page, p); | |
691 | } | |
692 | ||
693 | if (!s->offset && active) | |
694 | /* | |
695 | * Object and freepointer overlap. Cannot check | |
696 | * freepointer while object is allocated. | |
697 | */ | |
698 | return 1; | |
699 | ||
700 | /* Check free pointer validity */ | |
701 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
702 | object_err(s, page, p, "Freepointer corrupt"); | |
703 | /* | |
9f6c708e | 704 | * No choice but to zap it and thus lose the remainder |
81819f0f | 705 | * of the free objects in this slab. May cause |
672bba3a | 706 | * another error because the object count is now wrong. |
81819f0f | 707 | */ |
a973e9dd | 708 | set_freepointer(s, p, NULL); |
81819f0f CL |
709 | return 0; |
710 | } | |
711 | return 1; | |
712 | } | |
713 | ||
714 | static int check_slab(struct kmem_cache *s, struct page *page) | |
715 | { | |
39b26464 CL |
716 | int maxobj; |
717 | ||
81819f0f CL |
718 | VM_BUG_ON(!irqs_disabled()); |
719 | ||
720 | if (!PageSlab(page)) { | |
24922684 | 721 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
722 | return 0; |
723 | } | |
39b26464 CL |
724 | |
725 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
726 | if (page->objects > maxobj) { | |
727 | slab_err(s, page, "objects %u > max %u", | |
728 | s->name, page->objects, maxobj); | |
729 | return 0; | |
730 | } | |
731 | if (page->inuse > page->objects) { | |
24922684 | 732 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 733 | s->name, page->inuse, page->objects); |
81819f0f CL |
734 | return 0; |
735 | } | |
736 | /* Slab_pad_check fixes things up after itself */ | |
737 | slab_pad_check(s, page); | |
738 | return 1; | |
739 | } | |
740 | ||
741 | /* | |
672bba3a CL |
742 | * Determine if a certain object on a page is on the freelist. Must hold the |
743 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
744 | */ |
745 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
746 | { | |
747 | int nr = 0; | |
748 | void *fp = page->freelist; | |
749 | void *object = NULL; | |
224a88be | 750 | unsigned long max_objects; |
81819f0f | 751 | |
39b26464 | 752 | while (fp && nr <= page->objects) { |
81819f0f CL |
753 | if (fp == search) |
754 | return 1; | |
755 | if (!check_valid_pointer(s, page, fp)) { | |
756 | if (object) { | |
757 | object_err(s, page, object, | |
758 | "Freechain corrupt"); | |
a973e9dd | 759 | set_freepointer(s, object, NULL); |
81819f0f CL |
760 | break; |
761 | } else { | |
24922684 | 762 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 763 | page->freelist = NULL; |
39b26464 | 764 | page->inuse = page->objects; |
24922684 | 765 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
766 | return 0; |
767 | } | |
768 | break; | |
769 | } | |
770 | object = fp; | |
771 | fp = get_freepointer(s, object); | |
772 | nr++; | |
773 | } | |
774 | ||
224a88be | 775 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
210b5c06 CG |
776 | if (max_objects > MAX_OBJS_PER_PAGE) |
777 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
778 | |
779 | if (page->objects != max_objects) { | |
780 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
781 | "should be %d", page->objects, max_objects); | |
782 | page->objects = max_objects; | |
783 | slab_fix(s, "Number of objects adjusted."); | |
784 | } | |
39b26464 | 785 | if (page->inuse != page->objects - nr) { |
70d71228 | 786 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
787 | "counted were %d", page->inuse, page->objects - nr); |
788 | page->inuse = page->objects - nr; | |
24922684 | 789 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
790 | } |
791 | return search == NULL; | |
792 | } | |
793 | ||
0121c619 CL |
794 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
795 | int alloc) | |
3ec09742 CL |
796 | { |
797 | if (s->flags & SLAB_TRACE) { | |
798 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
799 | s->name, | |
800 | alloc ? "alloc" : "free", | |
801 | object, page->inuse, | |
802 | page->freelist); | |
803 | ||
804 | if (!alloc) | |
805 | print_section("Object", (void *)object, s->objsize); | |
806 | ||
807 | dump_stack(); | |
808 | } | |
809 | } | |
810 | ||
643b1138 | 811 | /* |
672bba3a | 812 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 813 | */ |
e95eed57 | 814 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 815 | { |
643b1138 CL |
816 | spin_lock(&n->list_lock); |
817 | list_add(&page->lru, &n->full); | |
818 | spin_unlock(&n->list_lock); | |
819 | } | |
820 | ||
821 | static void remove_full(struct kmem_cache *s, struct page *page) | |
822 | { | |
823 | struct kmem_cache_node *n; | |
824 | ||
825 | if (!(s->flags & SLAB_STORE_USER)) | |
826 | return; | |
827 | ||
828 | n = get_node(s, page_to_nid(page)); | |
829 | ||
830 | spin_lock(&n->list_lock); | |
831 | list_del(&page->lru); | |
832 | spin_unlock(&n->list_lock); | |
833 | } | |
834 | ||
0f389ec6 CL |
835 | /* Tracking of the number of slabs for debugging purposes */ |
836 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
837 | { | |
838 | struct kmem_cache_node *n = get_node(s, node); | |
839 | ||
840 | return atomic_long_read(&n->nr_slabs); | |
841 | } | |
842 | ||
205ab99d | 843 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
844 | { |
845 | struct kmem_cache_node *n = get_node(s, node); | |
846 | ||
847 | /* | |
848 | * May be called early in order to allocate a slab for the | |
849 | * kmem_cache_node structure. Solve the chicken-egg | |
850 | * dilemma by deferring the increment of the count during | |
851 | * bootstrap (see early_kmem_cache_node_alloc). | |
852 | */ | |
205ab99d | 853 | if (!NUMA_BUILD || n) { |
0f389ec6 | 854 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
855 | atomic_long_add(objects, &n->total_objects); |
856 | } | |
0f389ec6 | 857 | } |
205ab99d | 858 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
859 | { |
860 | struct kmem_cache_node *n = get_node(s, node); | |
861 | ||
862 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 863 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
864 | } |
865 | ||
866 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
867 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
868 | void *object) | |
869 | { | |
870 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
871 | return; | |
872 | ||
873 | init_object(s, object, 0); | |
874 | init_tracking(s, object); | |
875 | } | |
876 | ||
877 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
ce71e27c | 878 | void *object, unsigned long addr) |
81819f0f CL |
879 | { |
880 | if (!check_slab(s, page)) | |
881 | goto bad; | |
882 | ||
d692ef6d | 883 | if (!on_freelist(s, page, object)) { |
24922684 | 884 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 885 | goto bad; |
81819f0f CL |
886 | } |
887 | ||
888 | if (!check_valid_pointer(s, page, object)) { | |
889 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 890 | goto bad; |
81819f0f CL |
891 | } |
892 | ||
d692ef6d | 893 | if (!check_object(s, page, object, 0)) |
81819f0f | 894 | goto bad; |
81819f0f | 895 | |
3ec09742 CL |
896 | /* Success perform special debug activities for allocs */ |
897 | if (s->flags & SLAB_STORE_USER) | |
898 | set_track(s, object, TRACK_ALLOC, addr); | |
899 | trace(s, page, object, 1); | |
900 | init_object(s, object, 1); | |
81819f0f | 901 | return 1; |
3ec09742 | 902 | |
81819f0f CL |
903 | bad: |
904 | if (PageSlab(page)) { | |
905 | /* | |
906 | * If this is a slab page then lets do the best we can | |
907 | * to avoid issues in the future. Marking all objects | |
672bba3a | 908 | * as used avoids touching the remaining objects. |
81819f0f | 909 | */ |
24922684 | 910 | slab_fix(s, "Marking all objects used"); |
39b26464 | 911 | page->inuse = page->objects; |
a973e9dd | 912 | page->freelist = NULL; |
81819f0f CL |
913 | } |
914 | return 0; | |
915 | } | |
916 | ||
3ec09742 | 917 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 918 | void *object, unsigned long addr) |
81819f0f CL |
919 | { |
920 | if (!check_slab(s, page)) | |
921 | goto fail; | |
922 | ||
923 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 924 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
925 | goto fail; |
926 | } | |
927 | ||
928 | if (on_freelist(s, page, object)) { | |
24922684 | 929 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
930 | goto fail; |
931 | } | |
932 | ||
933 | if (!check_object(s, page, object, 1)) | |
934 | return 0; | |
935 | ||
936 | if (unlikely(s != page->slab)) { | |
3adbefee | 937 | if (!PageSlab(page)) { |
70d71228 CL |
938 | slab_err(s, page, "Attempt to free object(0x%p) " |
939 | "outside of slab", object); | |
3adbefee | 940 | } else if (!page->slab) { |
81819f0f | 941 | printk(KERN_ERR |
70d71228 | 942 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 943 | object); |
70d71228 | 944 | dump_stack(); |
06428780 | 945 | } else |
24922684 CL |
946 | object_err(s, page, object, |
947 | "page slab pointer corrupt."); | |
81819f0f CL |
948 | goto fail; |
949 | } | |
3ec09742 CL |
950 | |
951 | /* Special debug activities for freeing objects */ | |
8a38082d | 952 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
953 | remove_full(s, page); |
954 | if (s->flags & SLAB_STORE_USER) | |
955 | set_track(s, object, TRACK_FREE, addr); | |
956 | trace(s, page, object, 0); | |
957 | init_object(s, object, 0); | |
81819f0f | 958 | return 1; |
3ec09742 | 959 | |
81819f0f | 960 | fail: |
24922684 | 961 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
962 | return 0; |
963 | } | |
964 | ||
41ecc55b CL |
965 | static int __init setup_slub_debug(char *str) |
966 | { | |
f0630fff CL |
967 | slub_debug = DEBUG_DEFAULT_FLAGS; |
968 | if (*str++ != '=' || !*str) | |
969 | /* | |
970 | * No options specified. Switch on full debugging. | |
971 | */ | |
972 | goto out; | |
973 | ||
974 | if (*str == ',') | |
975 | /* | |
976 | * No options but restriction on slabs. This means full | |
977 | * debugging for slabs matching a pattern. | |
978 | */ | |
979 | goto check_slabs; | |
980 | ||
981 | slub_debug = 0; | |
982 | if (*str == '-') | |
983 | /* | |
984 | * Switch off all debugging measures. | |
985 | */ | |
986 | goto out; | |
987 | ||
988 | /* | |
989 | * Determine which debug features should be switched on | |
990 | */ | |
06428780 | 991 | for (; *str && *str != ','; str++) { |
f0630fff CL |
992 | switch (tolower(*str)) { |
993 | case 'f': | |
994 | slub_debug |= SLAB_DEBUG_FREE; | |
995 | break; | |
996 | case 'z': | |
997 | slub_debug |= SLAB_RED_ZONE; | |
998 | break; | |
999 | case 'p': | |
1000 | slub_debug |= SLAB_POISON; | |
1001 | break; | |
1002 | case 'u': | |
1003 | slub_debug |= SLAB_STORE_USER; | |
1004 | break; | |
1005 | case 't': | |
1006 | slub_debug |= SLAB_TRACE; | |
1007 | break; | |
1008 | default: | |
1009 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1010 | "unknown. skipped\n", *str); |
f0630fff | 1011 | } |
41ecc55b CL |
1012 | } |
1013 | ||
f0630fff | 1014 | check_slabs: |
41ecc55b CL |
1015 | if (*str == ',') |
1016 | slub_debug_slabs = str + 1; | |
f0630fff | 1017 | out: |
41ecc55b CL |
1018 | return 1; |
1019 | } | |
1020 | ||
1021 | __setup("slub_debug", setup_slub_debug); | |
1022 | ||
ba0268a8 CL |
1023 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1024 | unsigned long flags, const char *name, | |
51cc5068 | 1025 | void (*ctor)(void *)) |
41ecc55b CL |
1026 | { |
1027 | /* | |
e153362a | 1028 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1029 | */ |
e153362a CL |
1030 | if (slub_debug && (!slub_debug_slabs || |
1031 | strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)) == 0)) | |
1032 | flags |= slub_debug; | |
ba0268a8 CL |
1033 | |
1034 | return flags; | |
41ecc55b CL |
1035 | } |
1036 | #else | |
3ec09742 CL |
1037 | static inline void setup_object_debug(struct kmem_cache *s, |
1038 | struct page *page, void *object) {} | |
41ecc55b | 1039 | |
3ec09742 | 1040 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1041 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1042 | |
3ec09742 | 1043 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1044 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1045 | |
41ecc55b CL |
1046 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1047 | { return 1; } | |
1048 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1049 | void *object, int active) { return 1; } | |
3ec09742 | 1050 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1051 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1052 | unsigned long flags, const char *name, | |
51cc5068 | 1053 | void (*ctor)(void *)) |
ba0268a8 CL |
1054 | { |
1055 | return flags; | |
1056 | } | |
41ecc55b | 1057 | #define slub_debug 0 |
0f389ec6 CL |
1058 | |
1059 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1060 | { return 0; } | |
205ab99d CL |
1061 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1062 | int objects) {} | |
1063 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1064 | int objects) {} | |
41ecc55b | 1065 | #endif |
205ab99d | 1066 | |
81819f0f CL |
1067 | /* |
1068 | * Slab allocation and freeing | |
1069 | */ | |
65c3376a CL |
1070 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1071 | struct kmem_cache_order_objects oo) | |
1072 | { | |
1073 | int order = oo_order(oo); | |
1074 | ||
b1eeab67 VN |
1075 | flags |= __GFP_NOTRACK; |
1076 | ||
65c3376a CL |
1077 | if (node == -1) |
1078 | return alloc_pages(flags, order); | |
1079 | else | |
1080 | return alloc_pages_node(node, flags, order); | |
1081 | } | |
1082 | ||
81819f0f CL |
1083 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1084 | { | |
06428780 | 1085 | struct page *page; |
834f3d11 | 1086 | struct kmem_cache_order_objects oo = s->oo; |
81819f0f | 1087 | |
b7a49f0d | 1088 | flags |= s->allocflags; |
e12ba74d | 1089 | |
65c3376a CL |
1090 | page = alloc_slab_page(flags | __GFP_NOWARN | __GFP_NORETRY, node, |
1091 | oo); | |
1092 | if (unlikely(!page)) { | |
1093 | oo = s->min; | |
1094 | /* | |
1095 | * Allocation may have failed due to fragmentation. | |
1096 | * Try a lower order alloc if possible | |
1097 | */ | |
1098 | page = alloc_slab_page(flags, node, oo); | |
1099 | if (!page) | |
1100 | return NULL; | |
81819f0f | 1101 | |
65c3376a CL |
1102 | stat(get_cpu_slab(s, raw_smp_processor_id()), ORDER_FALLBACK); |
1103 | } | |
5a896d9e VN |
1104 | |
1105 | if (kmemcheck_enabled | |
1106 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) | |
1107 | { | |
b1eeab67 VN |
1108 | int pages = 1 << oo_order(oo); |
1109 | ||
1110 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1111 | ||
1112 | /* | |
1113 | * Objects from caches that have a constructor don't get | |
1114 | * cleared when they're allocated, so we need to do it here. | |
1115 | */ | |
1116 | if (s->ctor) | |
1117 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1118 | else | |
1119 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1120 | } |
1121 | ||
834f3d11 | 1122 | page->objects = oo_objects(oo); |
81819f0f CL |
1123 | mod_zone_page_state(page_zone(page), |
1124 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1125 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1126 | 1 << oo_order(oo)); |
81819f0f CL |
1127 | |
1128 | return page; | |
1129 | } | |
1130 | ||
1131 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1132 | void *object) | |
1133 | { | |
3ec09742 | 1134 | setup_object_debug(s, page, object); |
4f104934 | 1135 | if (unlikely(s->ctor)) |
51cc5068 | 1136 | s->ctor(object); |
81819f0f CL |
1137 | } |
1138 | ||
1139 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1140 | { | |
1141 | struct page *page; | |
81819f0f | 1142 | void *start; |
81819f0f CL |
1143 | void *last; |
1144 | void *p; | |
1145 | ||
6cb06229 | 1146 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1147 | |
6cb06229 CL |
1148 | page = allocate_slab(s, |
1149 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1150 | if (!page) |
1151 | goto out; | |
1152 | ||
205ab99d | 1153 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1154 | page->slab = s; |
1155 | page->flags |= 1 << PG_slab; | |
1156 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1157 | SLAB_STORE_USER | SLAB_TRACE)) | |
8a38082d | 1158 | __SetPageSlubDebug(page); |
81819f0f CL |
1159 | |
1160 | start = page_address(page); | |
81819f0f CL |
1161 | |
1162 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1163 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1164 | |
1165 | last = start; | |
224a88be | 1166 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1167 | setup_object(s, page, last); |
1168 | set_freepointer(s, last, p); | |
1169 | last = p; | |
1170 | } | |
1171 | setup_object(s, page, last); | |
a973e9dd | 1172 | set_freepointer(s, last, NULL); |
81819f0f CL |
1173 | |
1174 | page->freelist = start; | |
1175 | page->inuse = 0; | |
1176 | out: | |
81819f0f CL |
1177 | return page; |
1178 | } | |
1179 | ||
1180 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1181 | { | |
834f3d11 CL |
1182 | int order = compound_order(page); |
1183 | int pages = 1 << order; | |
81819f0f | 1184 | |
8a38082d | 1185 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) { |
81819f0f CL |
1186 | void *p; |
1187 | ||
1188 | slab_pad_check(s, page); | |
224a88be CL |
1189 | for_each_object(p, s, page_address(page), |
1190 | page->objects) | |
81819f0f | 1191 | check_object(s, page, p, 0); |
8a38082d | 1192 | __ClearPageSlubDebug(page); |
81819f0f CL |
1193 | } |
1194 | ||
b1eeab67 | 1195 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1196 | |
81819f0f CL |
1197 | mod_zone_page_state(page_zone(page), |
1198 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1199 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1200 | -pages); |
81819f0f | 1201 | |
49bd5221 CL |
1202 | __ClearPageSlab(page); |
1203 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1204 | if (current->reclaim_state) |
1205 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1206 | __free_pages(page, order); |
81819f0f CL |
1207 | } |
1208 | ||
1209 | static void rcu_free_slab(struct rcu_head *h) | |
1210 | { | |
1211 | struct page *page; | |
1212 | ||
1213 | page = container_of((struct list_head *)h, struct page, lru); | |
1214 | __free_slab(page->slab, page); | |
1215 | } | |
1216 | ||
1217 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1218 | { | |
1219 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1220 | /* | |
1221 | * RCU free overloads the RCU head over the LRU | |
1222 | */ | |
1223 | struct rcu_head *head = (void *)&page->lru; | |
1224 | ||
1225 | call_rcu(head, rcu_free_slab); | |
1226 | } else | |
1227 | __free_slab(s, page); | |
1228 | } | |
1229 | ||
1230 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1231 | { | |
205ab99d | 1232 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1233 | free_slab(s, page); |
1234 | } | |
1235 | ||
1236 | /* | |
1237 | * Per slab locking using the pagelock | |
1238 | */ | |
1239 | static __always_inline void slab_lock(struct page *page) | |
1240 | { | |
1241 | bit_spin_lock(PG_locked, &page->flags); | |
1242 | } | |
1243 | ||
1244 | static __always_inline void slab_unlock(struct page *page) | |
1245 | { | |
a76d3546 | 1246 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1247 | } |
1248 | ||
1249 | static __always_inline int slab_trylock(struct page *page) | |
1250 | { | |
1251 | int rc = 1; | |
1252 | ||
1253 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1254 | return rc; | |
1255 | } | |
1256 | ||
1257 | /* | |
1258 | * Management of partially allocated slabs | |
1259 | */ | |
7c2e132c CL |
1260 | static void add_partial(struct kmem_cache_node *n, |
1261 | struct page *page, int tail) | |
81819f0f | 1262 | { |
e95eed57 CL |
1263 | spin_lock(&n->list_lock); |
1264 | n->nr_partial++; | |
7c2e132c CL |
1265 | if (tail) |
1266 | list_add_tail(&page->lru, &n->partial); | |
1267 | else | |
1268 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1269 | spin_unlock(&n->list_lock); |
1270 | } | |
1271 | ||
0121c619 | 1272 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1273 | { |
1274 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1275 | ||
1276 | spin_lock(&n->list_lock); | |
1277 | list_del(&page->lru); | |
1278 | n->nr_partial--; | |
1279 | spin_unlock(&n->list_lock); | |
1280 | } | |
1281 | ||
1282 | /* | |
672bba3a | 1283 | * Lock slab and remove from the partial list. |
81819f0f | 1284 | * |
672bba3a | 1285 | * Must hold list_lock. |
81819f0f | 1286 | */ |
0121c619 CL |
1287 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1288 | struct page *page) | |
81819f0f CL |
1289 | { |
1290 | if (slab_trylock(page)) { | |
1291 | list_del(&page->lru); | |
1292 | n->nr_partial--; | |
8a38082d | 1293 | __SetPageSlubFrozen(page); |
81819f0f CL |
1294 | return 1; |
1295 | } | |
1296 | return 0; | |
1297 | } | |
1298 | ||
1299 | /* | |
672bba3a | 1300 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1301 | */ |
1302 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1303 | { | |
1304 | struct page *page; | |
1305 | ||
1306 | /* | |
1307 | * Racy check. If we mistakenly see no partial slabs then we | |
1308 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1309 | * partial slab and there is none available then get_partials() |
1310 | * will return NULL. | |
81819f0f CL |
1311 | */ |
1312 | if (!n || !n->nr_partial) | |
1313 | return NULL; | |
1314 | ||
1315 | spin_lock(&n->list_lock); | |
1316 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1317 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1318 | goto out; |
1319 | page = NULL; | |
1320 | out: | |
1321 | spin_unlock(&n->list_lock); | |
1322 | return page; | |
1323 | } | |
1324 | ||
1325 | /* | |
672bba3a | 1326 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1327 | */ |
1328 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1329 | { | |
1330 | #ifdef CONFIG_NUMA | |
1331 | struct zonelist *zonelist; | |
dd1a239f | 1332 | struct zoneref *z; |
54a6eb5c MG |
1333 | struct zone *zone; |
1334 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1335 | struct page *page; |
1336 | ||
1337 | /* | |
672bba3a CL |
1338 | * The defrag ratio allows a configuration of the tradeoffs between |
1339 | * inter node defragmentation and node local allocations. A lower | |
1340 | * defrag_ratio increases the tendency to do local allocations | |
1341 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1342 | * |
672bba3a CL |
1343 | * If the defrag_ratio is set to 0 then kmalloc() always |
1344 | * returns node local objects. If the ratio is higher then kmalloc() | |
1345 | * may return off node objects because partial slabs are obtained | |
1346 | * from other nodes and filled up. | |
81819f0f | 1347 | * |
6446faa2 | 1348 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1349 | * defrag_ratio = 1000) then every (well almost) allocation will |
1350 | * first attempt to defrag slab caches on other nodes. This means | |
1351 | * scanning over all nodes to look for partial slabs which may be | |
1352 | * expensive if we do it every time we are trying to find a slab | |
1353 | * with available objects. | |
81819f0f | 1354 | */ |
9824601e CL |
1355 | if (!s->remote_node_defrag_ratio || |
1356 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1357 | return NULL; |
1358 | ||
0e88460d | 1359 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1360 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1361 | struct kmem_cache_node *n; |
1362 | ||
54a6eb5c | 1363 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1364 | |
54a6eb5c | 1365 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1366 | n->nr_partial > s->min_partial) { |
81819f0f CL |
1367 | page = get_partial_node(n); |
1368 | if (page) | |
1369 | return page; | |
1370 | } | |
1371 | } | |
1372 | #endif | |
1373 | return NULL; | |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * Get a partial page, lock it and return it. | |
1378 | */ | |
1379 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1380 | { | |
1381 | struct page *page; | |
1382 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1383 | ||
1384 | page = get_partial_node(get_node(s, searchnode)); | |
1385 | if (page || (flags & __GFP_THISNODE)) | |
1386 | return page; | |
1387 | ||
1388 | return get_any_partial(s, flags); | |
1389 | } | |
1390 | ||
1391 | /* | |
1392 | * Move a page back to the lists. | |
1393 | * | |
1394 | * Must be called with the slab lock held. | |
1395 | * | |
1396 | * On exit the slab lock will have been dropped. | |
1397 | */ | |
7c2e132c | 1398 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1399 | { |
e95eed57 | 1400 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
8ff12cfc | 1401 | struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id()); |
e95eed57 | 1402 | |
8a38082d | 1403 | __ClearPageSlubFrozen(page); |
81819f0f | 1404 | if (page->inuse) { |
e95eed57 | 1405 | |
a973e9dd | 1406 | if (page->freelist) { |
7c2e132c | 1407 | add_partial(n, page, tail); |
8ff12cfc CL |
1408 | stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
1409 | } else { | |
1410 | stat(c, DEACTIVATE_FULL); | |
8a38082d AW |
1411 | if (SLABDEBUG && PageSlubDebug(page) && |
1412 | (s->flags & SLAB_STORE_USER)) | |
8ff12cfc CL |
1413 | add_full(n, page); |
1414 | } | |
81819f0f CL |
1415 | slab_unlock(page); |
1416 | } else { | |
8ff12cfc | 1417 | stat(c, DEACTIVATE_EMPTY); |
3b89d7d8 | 1418 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1419 | /* |
672bba3a CL |
1420 | * Adding an empty slab to the partial slabs in order |
1421 | * to avoid page allocator overhead. This slab needs | |
1422 | * to come after the other slabs with objects in | |
6446faa2 CL |
1423 | * so that the others get filled first. That way the |
1424 | * size of the partial list stays small. | |
1425 | * | |
0121c619 CL |
1426 | * kmem_cache_shrink can reclaim any empty slabs from |
1427 | * the partial list. | |
e95eed57 | 1428 | */ |
7c2e132c | 1429 | add_partial(n, page, 1); |
e95eed57 CL |
1430 | slab_unlock(page); |
1431 | } else { | |
1432 | slab_unlock(page); | |
8ff12cfc | 1433 | stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB); |
e95eed57 CL |
1434 | discard_slab(s, page); |
1435 | } | |
81819f0f CL |
1436 | } |
1437 | } | |
1438 | ||
1439 | /* | |
1440 | * Remove the cpu slab | |
1441 | */ | |
dfb4f096 | 1442 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1443 | { |
dfb4f096 | 1444 | struct page *page = c->page; |
7c2e132c | 1445 | int tail = 1; |
8ff12cfc | 1446 | |
b773ad73 | 1447 | if (page->freelist) |
8ff12cfc | 1448 | stat(c, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1449 | /* |
6446faa2 | 1450 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1451 | * because both freelists are empty. So this is unlikely |
1452 | * to occur. | |
1453 | */ | |
a973e9dd | 1454 | while (unlikely(c->freelist)) { |
894b8788 CL |
1455 | void **object; |
1456 | ||
7c2e132c CL |
1457 | tail = 0; /* Hot objects. Put the slab first */ |
1458 | ||
894b8788 | 1459 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1460 | object = c->freelist; |
b3fba8da | 1461 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1462 | |
1463 | /* And put onto the regular freelist */ | |
b3fba8da | 1464 | object[c->offset] = page->freelist; |
894b8788 CL |
1465 | page->freelist = object; |
1466 | page->inuse--; | |
1467 | } | |
dfb4f096 | 1468 | c->page = NULL; |
7c2e132c | 1469 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1470 | } |
1471 | ||
dfb4f096 | 1472 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1473 | { |
8ff12cfc | 1474 | stat(c, CPUSLAB_FLUSH); |
dfb4f096 CL |
1475 | slab_lock(c->page); |
1476 | deactivate_slab(s, c); | |
81819f0f CL |
1477 | } |
1478 | ||
1479 | /* | |
1480 | * Flush cpu slab. | |
6446faa2 | 1481 | * |
81819f0f CL |
1482 | * Called from IPI handler with interrupts disabled. |
1483 | */ | |
0c710013 | 1484 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1485 | { |
dfb4f096 | 1486 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1487 | |
dfb4f096 CL |
1488 | if (likely(c && c->page)) |
1489 | flush_slab(s, c); | |
81819f0f CL |
1490 | } |
1491 | ||
1492 | static void flush_cpu_slab(void *d) | |
1493 | { | |
1494 | struct kmem_cache *s = d; | |
81819f0f | 1495 | |
dfb4f096 | 1496 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1497 | } |
1498 | ||
1499 | static void flush_all(struct kmem_cache *s) | |
1500 | { | |
15c8b6c1 | 1501 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1502 | } |
1503 | ||
dfb4f096 CL |
1504 | /* |
1505 | * Check if the objects in a per cpu structure fit numa | |
1506 | * locality expectations. | |
1507 | */ | |
1508 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1509 | { | |
1510 | #ifdef CONFIG_NUMA | |
1511 | if (node != -1 && c->node != node) | |
1512 | return 0; | |
1513 | #endif | |
1514 | return 1; | |
1515 | } | |
1516 | ||
81819f0f | 1517 | /* |
894b8788 CL |
1518 | * Slow path. The lockless freelist is empty or we need to perform |
1519 | * debugging duties. | |
1520 | * | |
1521 | * Interrupts are disabled. | |
81819f0f | 1522 | * |
894b8788 CL |
1523 | * Processing is still very fast if new objects have been freed to the |
1524 | * regular freelist. In that case we simply take over the regular freelist | |
1525 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1526 | * |
894b8788 CL |
1527 | * If that is not working then we fall back to the partial lists. We take the |
1528 | * first element of the freelist as the object to allocate now and move the | |
1529 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1530 | * |
894b8788 | 1531 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1532 | * we need to allocate a new slab. This is the slowest path since it involves |
1533 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1534 | */ |
ce71e27c EGM |
1535 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1536 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1537 | { |
81819f0f | 1538 | void **object; |
dfb4f096 | 1539 | struct page *new; |
81819f0f | 1540 | |
e72e9c23 LT |
1541 | /* We handle __GFP_ZERO in the caller */ |
1542 | gfpflags &= ~__GFP_ZERO; | |
1543 | ||
dfb4f096 | 1544 | if (!c->page) |
81819f0f CL |
1545 | goto new_slab; |
1546 | ||
dfb4f096 CL |
1547 | slab_lock(c->page); |
1548 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1549 | goto another_slab; |
6446faa2 | 1550 | |
8ff12cfc | 1551 | stat(c, ALLOC_REFILL); |
6446faa2 | 1552 | |
894b8788 | 1553 | load_freelist: |
dfb4f096 | 1554 | object = c->page->freelist; |
a973e9dd | 1555 | if (unlikely(!object)) |
81819f0f | 1556 | goto another_slab; |
8a38082d | 1557 | if (unlikely(SLABDEBUG && PageSlubDebug(c->page))) |
81819f0f CL |
1558 | goto debug; |
1559 | ||
b3fba8da | 1560 | c->freelist = object[c->offset]; |
39b26464 | 1561 | c->page->inuse = c->page->objects; |
a973e9dd | 1562 | c->page->freelist = NULL; |
dfb4f096 | 1563 | c->node = page_to_nid(c->page); |
1f84260c | 1564 | unlock_out: |
dfb4f096 | 1565 | slab_unlock(c->page); |
8ff12cfc | 1566 | stat(c, ALLOC_SLOWPATH); |
81819f0f CL |
1567 | return object; |
1568 | ||
1569 | another_slab: | |
dfb4f096 | 1570 | deactivate_slab(s, c); |
81819f0f CL |
1571 | |
1572 | new_slab: | |
dfb4f096 CL |
1573 | new = get_partial(s, gfpflags, node); |
1574 | if (new) { | |
1575 | c->page = new; | |
8ff12cfc | 1576 | stat(c, ALLOC_FROM_PARTIAL); |
894b8788 | 1577 | goto load_freelist; |
81819f0f CL |
1578 | } |
1579 | ||
b811c202 CL |
1580 | if (gfpflags & __GFP_WAIT) |
1581 | local_irq_enable(); | |
1582 | ||
dfb4f096 | 1583 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1584 | |
1585 | if (gfpflags & __GFP_WAIT) | |
1586 | local_irq_disable(); | |
1587 | ||
dfb4f096 CL |
1588 | if (new) { |
1589 | c = get_cpu_slab(s, smp_processor_id()); | |
8ff12cfc | 1590 | stat(c, ALLOC_SLAB); |
05aa3450 | 1591 | if (c->page) |
dfb4f096 | 1592 | flush_slab(s, c); |
dfb4f096 | 1593 | slab_lock(new); |
8a38082d | 1594 | __SetPageSlubFrozen(new); |
dfb4f096 | 1595 | c->page = new; |
4b6f0750 | 1596 | goto load_freelist; |
81819f0f | 1597 | } |
71c7a06f | 1598 | return NULL; |
81819f0f | 1599 | debug: |
dfb4f096 | 1600 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1601 | goto another_slab; |
894b8788 | 1602 | |
dfb4f096 | 1603 | c->page->inuse++; |
b3fba8da | 1604 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1605 | c->node = -1; |
1f84260c | 1606 | goto unlock_out; |
894b8788 CL |
1607 | } |
1608 | ||
1609 | /* | |
1610 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1611 | * have the fastpath folded into their functions. So no function call | |
1612 | * overhead for requests that can be satisfied on the fastpath. | |
1613 | * | |
1614 | * The fastpath works by first checking if the lockless freelist can be used. | |
1615 | * If not then __slab_alloc is called for slow processing. | |
1616 | * | |
1617 | * Otherwise we can simply pick the next object from the lockless free list. | |
1618 | */ | |
06428780 | 1619 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1620 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1621 | { |
894b8788 | 1622 | void **object; |
dfb4f096 | 1623 | struct kmem_cache_cpu *c; |
1f84260c | 1624 | unsigned long flags; |
bdb21928 | 1625 | unsigned int objsize; |
1f84260c | 1626 | |
7e85ee0c PE |
1627 | gfpflags &= slab_gfp_mask; |
1628 | ||
cf40bd16 | 1629 | lockdep_trace_alloc(gfpflags); |
89124d70 | 1630 | might_sleep_if(gfpflags & __GFP_WAIT); |
3c506efd | 1631 | |
773ff60e AM |
1632 | if (should_failslab(s->objsize, gfpflags)) |
1633 | return NULL; | |
1f84260c | 1634 | |
894b8788 | 1635 | local_irq_save(flags); |
dfb4f096 | 1636 | c = get_cpu_slab(s, smp_processor_id()); |
bdb21928 | 1637 | objsize = c->objsize; |
a973e9dd | 1638 | if (unlikely(!c->freelist || !node_match(c, node))) |
894b8788 | 1639 | |
dfb4f096 | 1640 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1641 | |
1642 | else { | |
dfb4f096 | 1643 | object = c->freelist; |
b3fba8da | 1644 | c->freelist = object[c->offset]; |
8ff12cfc | 1645 | stat(c, ALLOC_FASTPATH); |
894b8788 CL |
1646 | } |
1647 | local_irq_restore(flags); | |
d07dbea4 CL |
1648 | |
1649 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
bdb21928 | 1650 | memset(object, 0, objsize); |
d07dbea4 | 1651 | |
5a896d9e | 1652 | kmemcheck_slab_alloc(s, gfpflags, object, c->objsize); |
06f22f13 | 1653 | kmemleak_alloc_recursive(object, objsize, 1, s->flags, gfpflags); |
5a896d9e | 1654 | |
894b8788 | 1655 | return object; |
81819f0f CL |
1656 | } |
1657 | ||
1658 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1659 | { | |
5b882be4 EGM |
1660 | void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_); |
1661 | ||
ca2b84cb | 1662 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1663 | |
1664 | return ret; | |
81819f0f CL |
1665 | } |
1666 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1667 | ||
5b882be4 EGM |
1668 | #ifdef CONFIG_KMEMTRACE |
1669 | void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags) | |
1670 | { | |
1671 | return slab_alloc(s, gfpflags, -1, _RET_IP_); | |
1672 | } | |
1673 | EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |
1674 | #endif | |
1675 | ||
81819f0f CL |
1676 | #ifdef CONFIG_NUMA |
1677 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1678 | { | |
5b882be4 EGM |
1679 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1680 | ||
ca2b84cb EGM |
1681 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
1682 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
1683 | |
1684 | return ret; | |
81819f0f CL |
1685 | } |
1686 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1687 | #endif | |
1688 | ||
5b882be4 EGM |
1689 | #ifdef CONFIG_KMEMTRACE |
1690 | void *kmem_cache_alloc_node_notrace(struct kmem_cache *s, | |
1691 | gfp_t gfpflags, | |
1692 | int node) | |
1693 | { | |
1694 | return slab_alloc(s, gfpflags, node, _RET_IP_); | |
1695 | } | |
1696 | EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); | |
1697 | #endif | |
1698 | ||
81819f0f | 1699 | /* |
894b8788 CL |
1700 | * Slow patch handling. This may still be called frequently since objects |
1701 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1702 | * |
894b8788 CL |
1703 | * So we still attempt to reduce cache line usage. Just take the slab |
1704 | * lock and free the item. If there is no additional partial page | |
1705 | * handling required then we can return immediately. | |
81819f0f | 1706 | */ |
894b8788 | 1707 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ce71e27c | 1708 | void *x, unsigned long addr, unsigned int offset) |
81819f0f CL |
1709 | { |
1710 | void *prior; | |
1711 | void **object = (void *)x; | |
8ff12cfc | 1712 | struct kmem_cache_cpu *c; |
81819f0f | 1713 | |
8ff12cfc CL |
1714 | c = get_cpu_slab(s, raw_smp_processor_id()); |
1715 | stat(c, FREE_SLOWPATH); | |
81819f0f CL |
1716 | slab_lock(page); |
1717 | ||
8a38082d | 1718 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) |
81819f0f | 1719 | goto debug; |
6446faa2 | 1720 | |
81819f0f | 1721 | checks_ok: |
b3fba8da | 1722 | prior = object[offset] = page->freelist; |
81819f0f CL |
1723 | page->freelist = object; |
1724 | page->inuse--; | |
1725 | ||
8a38082d | 1726 | if (unlikely(PageSlubFrozen(page))) { |
8ff12cfc | 1727 | stat(c, FREE_FROZEN); |
81819f0f | 1728 | goto out_unlock; |
8ff12cfc | 1729 | } |
81819f0f CL |
1730 | |
1731 | if (unlikely(!page->inuse)) | |
1732 | goto slab_empty; | |
1733 | ||
1734 | /* | |
6446faa2 | 1735 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1736 | * then add it. |
1737 | */ | |
a973e9dd | 1738 | if (unlikely(!prior)) { |
7c2e132c | 1739 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
8ff12cfc CL |
1740 | stat(c, FREE_ADD_PARTIAL); |
1741 | } | |
81819f0f CL |
1742 | |
1743 | out_unlock: | |
1744 | slab_unlock(page); | |
81819f0f CL |
1745 | return; |
1746 | ||
1747 | slab_empty: | |
a973e9dd | 1748 | if (prior) { |
81819f0f | 1749 | /* |
672bba3a | 1750 | * Slab still on the partial list. |
81819f0f CL |
1751 | */ |
1752 | remove_partial(s, page); | |
8ff12cfc CL |
1753 | stat(c, FREE_REMOVE_PARTIAL); |
1754 | } | |
81819f0f | 1755 | slab_unlock(page); |
8ff12cfc | 1756 | stat(c, FREE_SLAB); |
81819f0f | 1757 | discard_slab(s, page); |
81819f0f CL |
1758 | return; |
1759 | ||
1760 | debug: | |
3ec09742 | 1761 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1762 | goto out_unlock; |
77c5e2d0 | 1763 | goto checks_ok; |
81819f0f CL |
1764 | } |
1765 | ||
894b8788 CL |
1766 | /* |
1767 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1768 | * can perform fastpath freeing without additional function calls. | |
1769 | * | |
1770 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1771 | * of this processor. This typically the case if we have just allocated | |
1772 | * the item before. | |
1773 | * | |
1774 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1775 | * with all sorts of special processing. | |
1776 | */ | |
06428780 | 1777 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1778 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1779 | { |
1780 | void **object = (void *)x; | |
dfb4f096 | 1781 | struct kmem_cache_cpu *c; |
1f84260c CL |
1782 | unsigned long flags; |
1783 | ||
06f22f13 | 1784 | kmemleak_free_recursive(x, s->flags); |
894b8788 | 1785 | local_irq_save(flags); |
dfb4f096 | 1786 | c = get_cpu_slab(s, smp_processor_id()); |
5a896d9e | 1787 | kmemcheck_slab_free(s, object, c->objsize); |
27d9e4e9 | 1788 | debug_check_no_locks_freed(object, c->objsize); |
3ac7fe5a | 1789 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
6047a007 | 1790 | debug_check_no_obj_freed(object, c->objsize); |
ee3c72a1 | 1791 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1792 | object[c->offset] = c->freelist; |
dfb4f096 | 1793 | c->freelist = object; |
8ff12cfc | 1794 | stat(c, FREE_FASTPATH); |
894b8788 | 1795 | } else |
b3fba8da | 1796 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1797 | |
1798 | local_irq_restore(flags); | |
1799 | } | |
1800 | ||
81819f0f CL |
1801 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1802 | { | |
77c5e2d0 | 1803 | struct page *page; |
81819f0f | 1804 | |
b49af68f | 1805 | page = virt_to_head_page(x); |
81819f0f | 1806 | |
ce71e27c | 1807 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 1808 | |
ca2b84cb | 1809 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
1810 | } |
1811 | EXPORT_SYMBOL(kmem_cache_free); | |
1812 | ||
e9beef18 | 1813 | /* Figure out on which slab page the object resides */ |
81819f0f CL |
1814 | static struct page *get_object_page(const void *x) |
1815 | { | |
b49af68f | 1816 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1817 | |
1818 | if (!PageSlab(page)) | |
1819 | return NULL; | |
1820 | ||
1821 | return page; | |
1822 | } | |
1823 | ||
1824 | /* | |
672bba3a CL |
1825 | * Object placement in a slab is made very easy because we always start at |
1826 | * offset 0. If we tune the size of the object to the alignment then we can | |
1827 | * get the required alignment by putting one properly sized object after | |
1828 | * another. | |
81819f0f CL |
1829 | * |
1830 | * Notice that the allocation order determines the sizes of the per cpu | |
1831 | * caches. Each processor has always one slab available for allocations. | |
1832 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1833 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1834 | * locking overhead. |
81819f0f CL |
1835 | */ |
1836 | ||
1837 | /* | |
1838 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1839 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1840 | * and increases the number of allocations possible without having to | |
1841 | * take the list_lock. | |
1842 | */ | |
1843 | static int slub_min_order; | |
114e9e89 | 1844 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1845 | static int slub_min_objects; |
81819f0f CL |
1846 | |
1847 | /* | |
1848 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1849 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1850 | */ |
1851 | static int slub_nomerge; | |
1852 | ||
81819f0f CL |
1853 | /* |
1854 | * Calculate the order of allocation given an slab object size. | |
1855 | * | |
672bba3a CL |
1856 | * The order of allocation has significant impact on performance and other |
1857 | * system components. Generally order 0 allocations should be preferred since | |
1858 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1859 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1860 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1861 | * would be wasted. |
1862 | * | |
1863 | * In order to reach satisfactory performance we must ensure that a minimum | |
1864 | * number of objects is in one slab. Otherwise we may generate too much | |
1865 | * activity on the partial lists which requires taking the list_lock. This is | |
1866 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1867 | * |
672bba3a CL |
1868 | * slub_max_order specifies the order where we begin to stop considering the |
1869 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1870 | * we try to keep the page order as low as possible. So we accept more waste | |
1871 | * of space in favor of a small page order. | |
81819f0f | 1872 | * |
672bba3a CL |
1873 | * Higher order allocations also allow the placement of more objects in a |
1874 | * slab and thereby reduce object handling overhead. If the user has | |
1875 | * requested a higher mininum order then we start with that one instead of | |
1876 | * the smallest order which will fit the object. | |
81819f0f | 1877 | */ |
5e6d444e CL |
1878 | static inline int slab_order(int size, int min_objects, |
1879 | int max_order, int fract_leftover) | |
81819f0f CL |
1880 | { |
1881 | int order; | |
1882 | int rem; | |
6300ea75 | 1883 | int min_order = slub_min_order; |
81819f0f | 1884 | |
210b5c06 CG |
1885 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) |
1886 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
39b26464 | 1887 | |
6300ea75 | 1888 | for (order = max(min_order, |
5e6d444e CL |
1889 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1890 | order <= max_order; order++) { | |
81819f0f | 1891 | |
5e6d444e | 1892 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1893 | |
5e6d444e | 1894 | if (slab_size < min_objects * size) |
81819f0f CL |
1895 | continue; |
1896 | ||
1897 | rem = slab_size % size; | |
1898 | ||
5e6d444e | 1899 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1900 | break; |
1901 | ||
1902 | } | |
672bba3a | 1903 | |
81819f0f CL |
1904 | return order; |
1905 | } | |
1906 | ||
5e6d444e CL |
1907 | static inline int calculate_order(int size) |
1908 | { | |
1909 | int order; | |
1910 | int min_objects; | |
1911 | int fraction; | |
e8120ff1 | 1912 | int max_objects; |
5e6d444e CL |
1913 | |
1914 | /* | |
1915 | * Attempt to find best configuration for a slab. This | |
1916 | * works by first attempting to generate a layout with | |
1917 | * the best configuration and backing off gradually. | |
1918 | * | |
1919 | * First we reduce the acceptable waste in a slab. Then | |
1920 | * we reduce the minimum objects required in a slab. | |
1921 | */ | |
1922 | min_objects = slub_min_objects; | |
9b2cd506 CL |
1923 | if (!min_objects) |
1924 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
e8120ff1 ZY |
1925 | max_objects = (PAGE_SIZE << slub_max_order)/size; |
1926 | min_objects = min(min_objects, max_objects); | |
1927 | ||
5e6d444e | 1928 | while (min_objects > 1) { |
c124f5b5 | 1929 | fraction = 16; |
5e6d444e CL |
1930 | while (fraction >= 4) { |
1931 | order = slab_order(size, min_objects, | |
1932 | slub_max_order, fraction); | |
1933 | if (order <= slub_max_order) | |
1934 | return order; | |
1935 | fraction /= 2; | |
1936 | } | |
e8120ff1 | 1937 | min_objects --; |
5e6d444e CL |
1938 | } |
1939 | ||
1940 | /* | |
1941 | * We were unable to place multiple objects in a slab. Now | |
1942 | * lets see if we can place a single object there. | |
1943 | */ | |
1944 | order = slab_order(size, 1, slub_max_order, 1); | |
1945 | if (order <= slub_max_order) | |
1946 | return order; | |
1947 | ||
1948 | /* | |
1949 | * Doh this slab cannot be placed using slub_max_order. | |
1950 | */ | |
1951 | order = slab_order(size, 1, MAX_ORDER, 1); | |
818cf590 | 1952 | if (order < MAX_ORDER) |
5e6d444e CL |
1953 | return order; |
1954 | return -ENOSYS; | |
1955 | } | |
1956 | ||
81819f0f | 1957 | /* |
672bba3a | 1958 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
1959 | */ |
1960 | static unsigned long calculate_alignment(unsigned long flags, | |
1961 | unsigned long align, unsigned long size) | |
1962 | { | |
1963 | /* | |
6446faa2 CL |
1964 | * If the user wants hardware cache aligned objects then follow that |
1965 | * suggestion if the object is sufficiently large. | |
81819f0f | 1966 | * |
6446faa2 CL |
1967 | * The hardware cache alignment cannot override the specified |
1968 | * alignment though. If that is greater then use it. | |
81819f0f | 1969 | */ |
b6210386 NP |
1970 | if (flags & SLAB_HWCACHE_ALIGN) { |
1971 | unsigned long ralign = cache_line_size(); | |
1972 | while (size <= ralign / 2) | |
1973 | ralign /= 2; | |
1974 | align = max(align, ralign); | |
1975 | } | |
81819f0f CL |
1976 | |
1977 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 1978 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
1979 | |
1980 | return ALIGN(align, sizeof(void *)); | |
1981 | } | |
1982 | ||
dfb4f096 CL |
1983 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
1984 | struct kmem_cache_cpu *c) | |
1985 | { | |
1986 | c->page = NULL; | |
a973e9dd | 1987 | c->freelist = NULL; |
dfb4f096 | 1988 | c->node = 0; |
42a9fdbb CL |
1989 | c->offset = s->offset / sizeof(void *); |
1990 | c->objsize = s->objsize; | |
62f75532 PE |
1991 | #ifdef CONFIG_SLUB_STATS |
1992 | memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned)); | |
1993 | #endif | |
dfb4f096 CL |
1994 | } |
1995 | ||
5595cffc PE |
1996 | static void |
1997 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
1998 | { |
1999 | n->nr_partial = 0; | |
81819f0f CL |
2000 | spin_lock_init(&n->list_lock); |
2001 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2002 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2003 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2004 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2005 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2006 | #endif |
81819f0f CL |
2007 | } |
2008 | ||
4c93c355 CL |
2009 | #ifdef CONFIG_SMP |
2010 | /* | |
2011 | * Per cpu array for per cpu structures. | |
2012 | * | |
2013 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
2014 | * close together meaning that it becomes possible that multiple per cpu | |
2015 | * structures are contained in one cacheline. This may be particularly | |
2016 | * beneficial for the kmalloc caches. | |
2017 | * | |
2018 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
2019 | * likely able to get per cpu structures for all caches from the array defined | |
2020 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
2021 | * | |
2022 | * If the per cpu array is exhausted then fall back to kmalloc | |
2023 | * of individual cachelines. No sharing is possible then. | |
2024 | */ | |
2025 | #define NR_KMEM_CACHE_CPU 100 | |
2026 | ||
2027 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
2028 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
2029 | ||
2030 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
174596a0 | 2031 | static DECLARE_BITMAP(kmem_cach_cpu_free_init_once, CONFIG_NR_CPUS); |
4c93c355 CL |
2032 | |
2033 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
2034 | int cpu, gfp_t flags) | |
2035 | { | |
2036 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
2037 | ||
2038 | if (c) | |
2039 | per_cpu(kmem_cache_cpu_free, cpu) = | |
2040 | (void *)c->freelist; | |
2041 | else { | |
2042 | /* Table overflow: So allocate ourselves */ | |
2043 | c = kmalloc_node( | |
2044 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
2045 | flags, cpu_to_node(cpu)); | |
2046 | if (!c) | |
2047 | return NULL; | |
2048 | } | |
2049 | ||
2050 | init_kmem_cache_cpu(s, c); | |
2051 | return c; | |
2052 | } | |
2053 | ||
2054 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
2055 | { | |
2056 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
37189094 | 2057 | c >= per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { |
4c93c355 CL |
2058 | kfree(c); |
2059 | return; | |
2060 | } | |
2061 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
2062 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
2063 | } | |
2064 | ||
2065 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
2066 | { | |
2067 | int cpu; | |
2068 | ||
2069 | for_each_online_cpu(cpu) { | |
2070 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2071 | ||
2072 | if (c) { | |
2073 | s->cpu_slab[cpu] = NULL; | |
2074 | free_kmem_cache_cpu(c, cpu); | |
2075 | } | |
2076 | } | |
2077 | } | |
2078 | ||
2079 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2080 | { | |
2081 | int cpu; | |
2082 | ||
2083 | for_each_online_cpu(cpu) { | |
2084 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2085 | ||
2086 | if (c) | |
2087 | continue; | |
2088 | ||
2089 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
2090 | if (!c) { | |
2091 | free_kmem_cache_cpus(s); | |
2092 | return 0; | |
2093 | } | |
2094 | s->cpu_slab[cpu] = c; | |
2095 | } | |
2096 | return 1; | |
2097 | } | |
2098 | ||
2099 | /* | |
2100 | * Initialize the per cpu array. | |
2101 | */ | |
2102 | static void init_alloc_cpu_cpu(int cpu) | |
2103 | { | |
2104 | int i; | |
2105 | ||
174596a0 | 2106 | if (cpumask_test_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once))) |
4c93c355 CL |
2107 | return; |
2108 | ||
2109 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
2110 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
2111 | ||
174596a0 | 2112 | cpumask_set_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once)); |
4c93c355 CL |
2113 | } |
2114 | ||
2115 | static void __init init_alloc_cpu(void) | |
2116 | { | |
2117 | int cpu; | |
2118 | ||
2119 | for_each_online_cpu(cpu) | |
2120 | init_alloc_cpu_cpu(cpu); | |
2121 | } | |
2122 | ||
2123 | #else | |
2124 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
2125 | static inline void init_alloc_cpu(void) {} | |
2126 | ||
2127 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2128 | { | |
2129 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
2130 | return 1; | |
2131 | } | |
2132 | #endif | |
2133 | ||
81819f0f CL |
2134 | #ifdef CONFIG_NUMA |
2135 | /* | |
2136 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2137 | * slab on the node for this slabcache. There are no concurrent accesses | |
2138 | * possible. | |
2139 | * | |
2140 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2141 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2142 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2143 | */ |
0094de92 | 2144 | static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node) |
81819f0f CL |
2145 | { |
2146 | struct page *page; | |
2147 | struct kmem_cache_node *n; | |
ba84c73c | 2148 | unsigned long flags; |
81819f0f CL |
2149 | |
2150 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2151 | ||
a2f92ee7 | 2152 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2153 | |
2154 | BUG_ON(!page); | |
a2f92ee7 CL |
2155 | if (page_to_nid(page) != node) { |
2156 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2157 | "node %d\n", node); | |
2158 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2159 | "in order to be able to continue\n"); | |
2160 | } | |
2161 | ||
81819f0f CL |
2162 | n = page->freelist; |
2163 | BUG_ON(!n); | |
2164 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2165 | page->inuse++; | |
2166 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2167 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2168 | init_object(kmalloc_caches, n, 1); |
2169 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2170 | #endif |
5595cffc | 2171 | init_kmem_cache_node(n, kmalloc_caches); |
205ab99d | 2172 | inc_slabs_node(kmalloc_caches, node, page->objects); |
6446faa2 | 2173 | |
ba84c73c | 2174 | /* |
2175 | * lockdep requires consistent irq usage for each lock | |
2176 | * so even though there cannot be a race this early in | |
2177 | * the boot sequence, we still disable irqs. | |
2178 | */ | |
2179 | local_irq_save(flags); | |
7c2e132c | 2180 | add_partial(n, page, 0); |
ba84c73c | 2181 | local_irq_restore(flags); |
81819f0f CL |
2182 | } |
2183 | ||
2184 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2185 | { | |
2186 | int node; | |
2187 | ||
f64dc58c | 2188 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2189 | struct kmem_cache_node *n = s->node[node]; |
2190 | if (n && n != &s->local_node) | |
2191 | kmem_cache_free(kmalloc_caches, n); | |
2192 | s->node[node] = NULL; | |
2193 | } | |
2194 | } | |
2195 | ||
2196 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2197 | { | |
2198 | int node; | |
2199 | int local_node; | |
2200 | ||
2201 | if (slab_state >= UP) | |
2202 | local_node = page_to_nid(virt_to_page(s)); | |
2203 | else | |
2204 | local_node = 0; | |
2205 | ||
f64dc58c | 2206 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2207 | struct kmem_cache_node *n; |
2208 | ||
2209 | if (local_node == node) | |
2210 | n = &s->local_node; | |
2211 | else { | |
2212 | if (slab_state == DOWN) { | |
0094de92 | 2213 | early_kmem_cache_node_alloc(gfpflags, node); |
81819f0f CL |
2214 | continue; |
2215 | } | |
2216 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2217 | gfpflags, node); | |
2218 | ||
2219 | if (!n) { | |
2220 | free_kmem_cache_nodes(s); | |
2221 | return 0; | |
2222 | } | |
2223 | ||
2224 | } | |
2225 | s->node[node] = n; | |
5595cffc | 2226 | init_kmem_cache_node(n, s); |
81819f0f CL |
2227 | } |
2228 | return 1; | |
2229 | } | |
2230 | #else | |
2231 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2232 | { | |
2233 | } | |
2234 | ||
2235 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2236 | { | |
5595cffc | 2237 | init_kmem_cache_node(&s->local_node, s); |
81819f0f CL |
2238 | return 1; |
2239 | } | |
2240 | #endif | |
2241 | ||
c0bdb232 | 2242 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2243 | { |
2244 | if (min < MIN_PARTIAL) | |
2245 | min = MIN_PARTIAL; | |
2246 | else if (min > MAX_PARTIAL) | |
2247 | min = MAX_PARTIAL; | |
2248 | s->min_partial = min; | |
2249 | } | |
2250 | ||
81819f0f CL |
2251 | /* |
2252 | * calculate_sizes() determines the order and the distribution of data within | |
2253 | * a slab object. | |
2254 | */ | |
06b285dc | 2255 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2256 | { |
2257 | unsigned long flags = s->flags; | |
2258 | unsigned long size = s->objsize; | |
2259 | unsigned long align = s->align; | |
834f3d11 | 2260 | int order; |
81819f0f | 2261 | |
d8b42bf5 CL |
2262 | /* |
2263 | * Round up object size to the next word boundary. We can only | |
2264 | * place the free pointer at word boundaries and this determines | |
2265 | * the possible location of the free pointer. | |
2266 | */ | |
2267 | size = ALIGN(size, sizeof(void *)); | |
2268 | ||
2269 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2270 | /* |
2271 | * Determine if we can poison the object itself. If the user of | |
2272 | * the slab may touch the object after free or before allocation | |
2273 | * then we should never poison the object itself. | |
2274 | */ | |
2275 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2276 | !s->ctor) |
81819f0f CL |
2277 | s->flags |= __OBJECT_POISON; |
2278 | else | |
2279 | s->flags &= ~__OBJECT_POISON; | |
2280 | ||
81819f0f CL |
2281 | |
2282 | /* | |
672bba3a | 2283 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2284 | * end of the object and the free pointer. If not then add an |
672bba3a | 2285 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2286 | */ |
2287 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2288 | size += sizeof(void *); | |
41ecc55b | 2289 | #endif |
81819f0f CL |
2290 | |
2291 | /* | |
672bba3a CL |
2292 | * With that we have determined the number of bytes in actual use |
2293 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2294 | */ |
2295 | s->inuse = size; | |
2296 | ||
2297 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2298 | s->ctor)) { |
81819f0f CL |
2299 | /* |
2300 | * Relocate free pointer after the object if it is not | |
2301 | * permitted to overwrite the first word of the object on | |
2302 | * kmem_cache_free. | |
2303 | * | |
2304 | * This is the case if we do RCU, have a constructor or | |
2305 | * destructor or are poisoning the objects. | |
2306 | */ | |
2307 | s->offset = size; | |
2308 | size += sizeof(void *); | |
2309 | } | |
2310 | ||
c12b3c62 | 2311 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2312 | if (flags & SLAB_STORE_USER) |
2313 | /* | |
2314 | * Need to store information about allocs and frees after | |
2315 | * the object. | |
2316 | */ | |
2317 | size += 2 * sizeof(struct track); | |
2318 | ||
be7b3fbc | 2319 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2320 | /* |
2321 | * Add some empty padding so that we can catch | |
2322 | * overwrites from earlier objects rather than let | |
2323 | * tracking information or the free pointer be | |
0211a9c8 | 2324 | * corrupted if a user writes before the start |
81819f0f CL |
2325 | * of the object. |
2326 | */ | |
2327 | size += sizeof(void *); | |
41ecc55b | 2328 | #endif |
672bba3a | 2329 | |
81819f0f CL |
2330 | /* |
2331 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2332 | * user specified and the dynamic determination of cache line size |
2333 | * on bootup. | |
81819f0f CL |
2334 | */ |
2335 | align = calculate_alignment(flags, align, s->objsize); | |
2336 | ||
2337 | /* | |
2338 | * SLUB stores one object immediately after another beginning from | |
2339 | * offset 0. In order to align the objects we have to simply size | |
2340 | * each object to conform to the alignment. | |
2341 | */ | |
2342 | size = ALIGN(size, align); | |
2343 | s->size = size; | |
06b285dc CL |
2344 | if (forced_order >= 0) |
2345 | order = forced_order; | |
2346 | else | |
2347 | order = calculate_order(size); | |
81819f0f | 2348 | |
834f3d11 | 2349 | if (order < 0) |
81819f0f CL |
2350 | return 0; |
2351 | ||
b7a49f0d | 2352 | s->allocflags = 0; |
834f3d11 | 2353 | if (order) |
b7a49f0d CL |
2354 | s->allocflags |= __GFP_COMP; |
2355 | ||
2356 | if (s->flags & SLAB_CACHE_DMA) | |
2357 | s->allocflags |= SLUB_DMA; | |
2358 | ||
2359 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2360 | s->allocflags |= __GFP_RECLAIMABLE; | |
2361 | ||
81819f0f CL |
2362 | /* |
2363 | * Determine the number of objects per slab | |
2364 | */ | |
834f3d11 | 2365 | s->oo = oo_make(order, size); |
65c3376a | 2366 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2367 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2368 | s->max = s->oo; | |
81819f0f | 2369 | |
834f3d11 | 2370 | return !!oo_objects(s->oo); |
81819f0f CL |
2371 | |
2372 | } | |
2373 | ||
81819f0f CL |
2374 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2375 | const char *name, size_t size, | |
2376 | size_t align, unsigned long flags, | |
51cc5068 | 2377 | void (*ctor)(void *)) |
81819f0f CL |
2378 | { |
2379 | memset(s, 0, kmem_size); | |
2380 | s->name = name; | |
2381 | s->ctor = ctor; | |
81819f0f | 2382 | s->objsize = size; |
81819f0f | 2383 | s->align = align; |
ba0268a8 | 2384 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2385 | |
06b285dc | 2386 | if (!calculate_sizes(s, -1)) |
81819f0f CL |
2387 | goto error; |
2388 | ||
3b89d7d8 DR |
2389 | /* |
2390 | * The larger the object size is, the more pages we want on the partial | |
2391 | * list to avoid pounding the page allocator excessively. | |
2392 | */ | |
c0bdb232 | 2393 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2394 | s->refcount = 1; |
2395 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2396 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2397 | #endif |
dfb4f096 CL |
2398 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2399 | goto error; | |
81819f0f | 2400 | |
dfb4f096 | 2401 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2402 | return 1; |
4c93c355 | 2403 | free_kmem_cache_nodes(s); |
81819f0f CL |
2404 | error: |
2405 | if (flags & SLAB_PANIC) | |
2406 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2407 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2408 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2409 | s->offset, flags); |
2410 | return 0; | |
2411 | } | |
81819f0f CL |
2412 | |
2413 | /* | |
2414 | * Check if a given pointer is valid | |
2415 | */ | |
2416 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2417 | { | |
06428780 | 2418 | struct page *page; |
81819f0f CL |
2419 | |
2420 | page = get_object_page(object); | |
2421 | ||
2422 | if (!page || s != page->slab) | |
2423 | /* No slab or wrong slab */ | |
2424 | return 0; | |
2425 | ||
abcd08a6 | 2426 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2427 | return 0; |
2428 | ||
2429 | /* | |
2430 | * We could also check if the object is on the slabs freelist. | |
2431 | * But this would be too expensive and it seems that the main | |
6446faa2 | 2432 | * purpose of kmem_ptr_valid() is to check if the object belongs |
81819f0f CL |
2433 | * to a certain slab. |
2434 | */ | |
2435 | return 1; | |
2436 | } | |
2437 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2438 | ||
2439 | /* | |
2440 | * Determine the size of a slab object | |
2441 | */ | |
2442 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2443 | { | |
2444 | return s->objsize; | |
2445 | } | |
2446 | EXPORT_SYMBOL(kmem_cache_size); | |
2447 | ||
2448 | const char *kmem_cache_name(struct kmem_cache *s) | |
2449 | { | |
2450 | return s->name; | |
2451 | } | |
2452 | EXPORT_SYMBOL(kmem_cache_name); | |
2453 | ||
33b12c38 CL |
2454 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2455 | const char *text) | |
2456 | { | |
2457 | #ifdef CONFIG_SLUB_DEBUG | |
2458 | void *addr = page_address(page); | |
2459 | void *p; | |
2460 | DECLARE_BITMAP(map, page->objects); | |
2461 | ||
2462 | bitmap_zero(map, page->objects); | |
2463 | slab_err(s, page, "%s", text); | |
2464 | slab_lock(page); | |
2465 | for_each_free_object(p, s, page->freelist) | |
2466 | set_bit(slab_index(p, s, addr), map); | |
2467 | ||
2468 | for_each_object(p, s, addr, page->objects) { | |
2469 | ||
2470 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2471 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2472 | p, p - addr); | |
2473 | print_tracking(s, p); | |
2474 | } | |
2475 | } | |
2476 | slab_unlock(page); | |
2477 | #endif | |
2478 | } | |
2479 | ||
81819f0f | 2480 | /* |
599870b1 | 2481 | * Attempt to free all partial slabs on a node. |
81819f0f | 2482 | */ |
599870b1 | 2483 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2484 | { |
81819f0f CL |
2485 | unsigned long flags; |
2486 | struct page *page, *h; | |
2487 | ||
2488 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2489 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f CL |
2490 | if (!page->inuse) { |
2491 | list_del(&page->lru); | |
2492 | discard_slab(s, page); | |
599870b1 | 2493 | n->nr_partial--; |
33b12c38 CL |
2494 | } else { |
2495 | list_slab_objects(s, page, | |
2496 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2497 | } |
33b12c38 | 2498 | } |
81819f0f | 2499 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2500 | } |
2501 | ||
2502 | /* | |
672bba3a | 2503 | * Release all resources used by a slab cache. |
81819f0f | 2504 | */ |
0c710013 | 2505 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2506 | { |
2507 | int node; | |
2508 | ||
2509 | flush_all(s); | |
2510 | ||
2511 | /* Attempt to free all objects */ | |
4c93c355 | 2512 | free_kmem_cache_cpus(s); |
f64dc58c | 2513 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2514 | struct kmem_cache_node *n = get_node(s, node); |
2515 | ||
599870b1 CL |
2516 | free_partial(s, n); |
2517 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2518 | return 1; |
2519 | } | |
2520 | free_kmem_cache_nodes(s); | |
2521 | return 0; | |
2522 | } | |
2523 | ||
2524 | /* | |
2525 | * Close a cache and release the kmem_cache structure | |
2526 | * (must be used for caches created using kmem_cache_create) | |
2527 | */ | |
2528 | void kmem_cache_destroy(struct kmem_cache *s) | |
2529 | { | |
2530 | down_write(&slub_lock); | |
2531 | s->refcount--; | |
2532 | if (!s->refcount) { | |
2533 | list_del(&s->list); | |
a0e1d1be | 2534 | up_write(&slub_lock); |
d629d819 PE |
2535 | if (kmem_cache_close(s)) { |
2536 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2537 | "still has objects.\n", s->name, __func__); | |
2538 | dump_stack(); | |
2539 | } | |
81819f0f | 2540 | sysfs_slab_remove(s); |
a0e1d1be CL |
2541 | } else |
2542 | up_write(&slub_lock); | |
81819f0f CL |
2543 | } |
2544 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2545 | ||
2546 | /******************************************************************** | |
2547 | * Kmalloc subsystem | |
2548 | *******************************************************************/ | |
2549 | ||
ffadd4d0 | 2550 | struct kmem_cache kmalloc_caches[SLUB_PAGE_SHIFT] __cacheline_aligned; |
81819f0f CL |
2551 | EXPORT_SYMBOL(kmalloc_caches); |
2552 | ||
81819f0f CL |
2553 | static int __init setup_slub_min_order(char *str) |
2554 | { | |
06428780 | 2555 | get_option(&str, &slub_min_order); |
81819f0f CL |
2556 | |
2557 | return 1; | |
2558 | } | |
2559 | ||
2560 | __setup("slub_min_order=", setup_slub_min_order); | |
2561 | ||
2562 | static int __init setup_slub_max_order(char *str) | |
2563 | { | |
06428780 | 2564 | get_option(&str, &slub_max_order); |
818cf590 | 2565 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2566 | |
2567 | return 1; | |
2568 | } | |
2569 | ||
2570 | __setup("slub_max_order=", setup_slub_max_order); | |
2571 | ||
2572 | static int __init setup_slub_min_objects(char *str) | |
2573 | { | |
06428780 | 2574 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2575 | |
2576 | return 1; | |
2577 | } | |
2578 | ||
2579 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2580 | ||
2581 | static int __init setup_slub_nomerge(char *str) | |
2582 | { | |
2583 | slub_nomerge = 1; | |
2584 | return 1; | |
2585 | } | |
2586 | ||
2587 | __setup("slub_nomerge", setup_slub_nomerge); | |
2588 | ||
81819f0f CL |
2589 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2590 | const char *name, int size, gfp_t gfp_flags) | |
2591 | { | |
2592 | unsigned int flags = 0; | |
2593 | ||
2594 | if (gfp_flags & SLUB_DMA) | |
2595 | flags = SLAB_CACHE_DMA; | |
2596 | ||
83b519e8 PE |
2597 | /* |
2598 | * This function is called with IRQs disabled during early-boot on | |
2599 | * single CPU so there's no need to take slub_lock here. | |
2600 | */ | |
81819f0f | 2601 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2602 | flags, NULL)) |
81819f0f CL |
2603 | goto panic; |
2604 | ||
2605 | list_add(&s->list, &slab_caches); | |
83b519e8 | 2606 | |
81819f0f CL |
2607 | if (sysfs_slab_add(s)) |
2608 | goto panic; | |
2609 | return s; | |
2610 | ||
2611 | panic: | |
2612 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2613 | } | |
2614 | ||
2e443fd0 | 2615 | #ifdef CONFIG_ZONE_DMA |
ffadd4d0 | 2616 | static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT]; |
1ceef402 CL |
2617 | |
2618 | static void sysfs_add_func(struct work_struct *w) | |
2619 | { | |
2620 | struct kmem_cache *s; | |
2621 | ||
2622 | down_write(&slub_lock); | |
2623 | list_for_each_entry(s, &slab_caches, list) { | |
2624 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2625 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2626 | sysfs_slab_add(s); | |
2627 | } | |
2628 | } | |
2629 | up_write(&slub_lock); | |
2630 | } | |
2631 | ||
2632 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2633 | ||
2e443fd0 CL |
2634 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2635 | { | |
2636 | struct kmem_cache *s; | |
2e443fd0 CL |
2637 | char *text; |
2638 | size_t realsize; | |
2639 | ||
2640 | s = kmalloc_caches_dma[index]; | |
2641 | if (s) | |
2642 | return s; | |
2643 | ||
2644 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2645 | if (flags & __GFP_WAIT) |
2646 | down_write(&slub_lock); | |
2647 | else { | |
2648 | if (!down_write_trylock(&slub_lock)) | |
2649 | goto out; | |
2650 | } | |
2651 | ||
2652 | if (kmalloc_caches_dma[index]) | |
2653 | goto unlock_out; | |
2e443fd0 | 2654 | |
7b55f620 | 2655 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2656 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2657 | (unsigned int)realsize); | |
1ceef402 CL |
2658 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); |
2659 | ||
2660 | if (!s || !text || !kmem_cache_open(s, flags, text, | |
2661 | realsize, ARCH_KMALLOC_MINALIGN, | |
5a896d9e VN |
2662 | SLAB_CACHE_DMA|SLAB_NOTRACK|__SYSFS_ADD_DEFERRED, |
2663 | NULL)) { | |
1ceef402 CL |
2664 | kfree(s); |
2665 | kfree(text); | |
2666 | goto unlock_out; | |
dfce8648 | 2667 | } |
1ceef402 CL |
2668 | |
2669 | list_add(&s->list, &slab_caches); | |
2670 | kmalloc_caches_dma[index] = s; | |
2671 | ||
2672 | schedule_work(&sysfs_add_work); | |
2673 | ||
2674 | unlock_out: | |
dfce8648 | 2675 | up_write(&slub_lock); |
1ceef402 | 2676 | out: |
dfce8648 | 2677 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2678 | } |
2679 | #endif | |
2680 | ||
f1b26339 CL |
2681 | /* |
2682 | * Conversion table for small slabs sizes / 8 to the index in the | |
2683 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2684 | * of two cache sizes there. The size of larger slabs can be determined using | |
2685 | * fls. | |
2686 | */ | |
2687 | static s8 size_index[24] = { | |
2688 | 3, /* 8 */ | |
2689 | 4, /* 16 */ | |
2690 | 5, /* 24 */ | |
2691 | 5, /* 32 */ | |
2692 | 6, /* 40 */ | |
2693 | 6, /* 48 */ | |
2694 | 6, /* 56 */ | |
2695 | 6, /* 64 */ | |
2696 | 1, /* 72 */ | |
2697 | 1, /* 80 */ | |
2698 | 1, /* 88 */ | |
2699 | 1, /* 96 */ | |
2700 | 7, /* 104 */ | |
2701 | 7, /* 112 */ | |
2702 | 7, /* 120 */ | |
2703 | 7, /* 128 */ | |
2704 | 2, /* 136 */ | |
2705 | 2, /* 144 */ | |
2706 | 2, /* 152 */ | |
2707 | 2, /* 160 */ | |
2708 | 2, /* 168 */ | |
2709 | 2, /* 176 */ | |
2710 | 2, /* 184 */ | |
2711 | 2 /* 192 */ | |
2712 | }; | |
2713 | ||
81819f0f CL |
2714 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2715 | { | |
f1b26339 | 2716 | int index; |
81819f0f | 2717 | |
f1b26339 CL |
2718 | if (size <= 192) { |
2719 | if (!size) | |
2720 | return ZERO_SIZE_PTR; | |
81819f0f | 2721 | |
f1b26339 | 2722 | index = size_index[(size - 1) / 8]; |
aadb4bc4 | 2723 | } else |
f1b26339 | 2724 | index = fls(size - 1); |
81819f0f CL |
2725 | |
2726 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2727 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2728 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2729 | |
81819f0f CL |
2730 | #endif |
2731 | return &kmalloc_caches[index]; | |
2732 | } | |
2733 | ||
2734 | void *__kmalloc(size_t size, gfp_t flags) | |
2735 | { | |
aadb4bc4 | 2736 | struct kmem_cache *s; |
5b882be4 | 2737 | void *ret; |
81819f0f | 2738 | |
ffadd4d0 | 2739 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2740 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2741 | |
2742 | s = get_slab(size, flags); | |
2743 | ||
2744 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2745 | return s; |
2746 | ||
5b882be4 EGM |
2747 | ret = slab_alloc(s, flags, -1, _RET_IP_); |
2748 | ||
ca2b84cb | 2749 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2750 | |
2751 | return ret; | |
81819f0f CL |
2752 | } |
2753 | EXPORT_SYMBOL(__kmalloc); | |
2754 | ||
f619cfe1 CL |
2755 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2756 | { | |
b1eeab67 | 2757 | struct page *page; |
f619cfe1 | 2758 | |
b1eeab67 VN |
2759 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2760 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 CL |
2761 | if (page) |
2762 | return page_address(page); | |
2763 | else | |
2764 | return NULL; | |
2765 | } | |
2766 | ||
81819f0f CL |
2767 | #ifdef CONFIG_NUMA |
2768 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2769 | { | |
aadb4bc4 | 2770 | struct kmem_cache *s; |
5b882be4 | 2771 | void *ret; |
81819f0f | 2772 | |
057685cf | 2773 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2774 | ret = kmalloc_large_node(size, flags, node); |
2775 | ||
ca2b84cb EGM |
2776 | trace_kmalloc_node(_RET_IP_, ret, |
2777 | size, PAGE_SIZE << get_order(size), | |
2778 | flags, node); | |
5b882be4 EGM |
2779 | |
2780 | return ret; | |
2781 | } | |
aadb4bc4 CL |
2782 | |
2783 | s = get_slab(size, flags); | |
2784 | ||
2785 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2786 | return s; |
2787 | ||
5b882be4 EGM |
2788 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2789 | ||
ca2b84cb | 2790 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2791 | |
2792 | return ret; | |
81819f0f CL |
2793 | } |
2794 | EXPORT_SYMBOL(__kmalloc_node); | |
2795 | #endif | |
2796 | ||
2797 | size_t ksize(const void *object) | |
2798 | { | |
272c1d21 | 2799 | struct page *page; |
81819f0f CL |
2800 | struct kmem_cache *s; |
2801 | ||
ef8b4520 | 2802 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2803 | return 0; |
2804 | ||
294a80a8 | 2805 | page = virt_to_head_page(object); |
294a80a8 | 2806 | |
76994412 PE |
2807 | if (unlikely(!PageSlab(page))) { |
2808 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2809 | return PAGE_SIZE << compound_order(page); |
76994412 | 2810 | } |
81819f0f | 2811 | s = page->slab; |
81819f0f | 2812 | |
ae20bfda | 2813 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2814 | /* |
2815 | * Debugging requires use of the padding between object | |
2816 | * and whatever may come after it. | |
2817 | */ | |
2818 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2819 | return s->objsize; | |
2820 | ||
ae20bfda | 2821 | #endif |
81819f0f CL |
2822 | /* |
2823 | * If we have the need to store the freelist pointer | |
2824 | * back there or track user information then we can | |
2825 | * only use the space before that information. | |
2826 | */ | |
2827 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2828 | return s->inuse; | |
81819f0f CL |
2829 | /* |
2830 | * Else we can use all the padding etc for the allocation | |
2831 | */ | |
2832 | return s->size; | |
2833 | } | |
b1aabecd | 2834 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2835 | |
2836 | void kfree(const void *x) | |
2837 | { | |
81819f0f | 2838 | struct page *page; |
5bb983b0 | 2839 | void *object = (void *)x; |
81819f0f | 2840 | |
2121db74 PE |
2841 | trace_kfree(_RET_IP_, x); |
2842 | ||
2408c550 | 2843 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2844 | return; |
2845 | ||
b49af68f | 2846 | page = virt_to_head_page(x); |
aadb4bc4 | 2847 | if (unlikely(!PageSlab(page))) { |
0937502a | 2848 | BUG_ON(!PageCompound(page)); |
aadb4bc4 CL |
2849 | put_page(page); |
2850 | return; | |
2851 | } | |
ce71e27c | 2852 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2853 | } |
2854 | EXPORT_SYMBOL(kfree); | |
2855 | ||
2086d26a | 2856 | /* |
672bba3a CL |
2857 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2858 | * the remaining slabs by the number of items in use. The slabs with the | |
2859 | * most items in use come first. New allocations will then fill those up | |
2860 | * and thus they can be removed from the partial lists. | |
2861 | * | |
2862 | * The slabs with the least items are placed last. This results in them | |
2863 | * being allocated from last increasing the chance that the last objects | |
2864 | * are freed in them. | |
2086d26a CL |
2865 | */ |
2866 | int kmem_cache_shrink(struct kmem_cache *s) | |
2867 | { | |
2868 | int node; | |
2869 | int i; | |
2870 | struct kmem_cache_node *n; | |
2871 | struct page *page; | |
2872 | struct page *t; | |
205ab99d | 2873 | int objects = oo_objects(s->max); |
2086d26a | 2874 | struct list_head *slabs_by_inuse = |
834f3d11 | 2875 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2876 | unsigned long flags; |
2877 | ||
2878 | if (!slabs_by_inuse) | |
2879 | return -ENOMEM; | |
2880 | ||
2881 | flush_all(s); | |
f64dc58c | 2882 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2883 | n = get_node(s, node); |
2884 | ||
2885 | if (!n->nr_partial) | |
2886 | continue; | |
2887 | ||
834f3d11 | 2888 | for (i = 0; i < objects; i++) |
2086d26a CL |
2889 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2890 | ||
2891 | spin_lock_irqsave(&n->list_lock, flags); | |
2892 | ||
2893 | /* | |
672bba3a | 2894 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2895 | * |
672bba3a CL |
2896 | * Note that concurrent frees may occur while we hold the |
2897 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2898 | */ |
2899 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2900 | if (!page->inuse && slab_trylock(page)) { | |
2901 | /* | |
2902 | * Must hold slab lock here because slab_free | |
2903 | * may have freed the last object and be | |
2904 | * waiting to release the slab. | |
2905 | */ | |
2906 | list_del(&page->lru); | |
2907 | n->nr_partial--; | |
2908 | slab_unlock(page); | |
2909 | discard_slab(s, page); | |
2910 | } else { | |
fcda3d89 CL |
2911 | list_move(&page->lru, |
2912 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2913 | } |
2914 | } | |
2915 | ||
2086d26a | 2916 | /* |
672bba3a CL |
2917 | * Rebuild the partial list with the slabs filled up most |
2918 | * first and the least used slabs at the end. | |
2086d26a | 2919 | */ |
834f3d11 | 2920 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2921 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2922 | ||
2086d26a CL |
2923 | spin_unlock_irqrestore(&n->list_lock, flags); |
2924 | } | |
2925 | ||
2926 | kfree(slabs_by_inuse); | |
2927 | return 0; | |
2928 | } | |
2929 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2930 | ||
b9049e23 YG |
2931 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2932 | static int slab_mem_going_offline_callback(void *arg) | |
2933 | { | |
2934 | struct kmem_cache *s; | |
2935 | ||
2936 | down_read(&slub_lock); | |
2937 | list_for_each_entry(s, &slab_caches, list) | |
2938 | kmem_cache_shrink(s); | |
2939 | up_read(&slub_lock); | |
2940 | ||
2941 | return 0; | |
2942 | } | |
2943 | ||
2944 | static void slab_mem_offline_callback(void *arg) | |
2945 | { | |
2946 | struct kmem_cache_node *n; | |
2947 | struct kmem_cache *s; | |
2948 | struct memory_notify *marg = arg; | |
2949 | int offline_node; | |
2950 | ||
2951 | offline_node = marg->status_change_nid; | |
2952 | ||
2953 | /* | |
2954 | * If the node still has available memory. we need kmem_cache_node | |
2955 | * for it yet. | |
2956 | */ | |
2957 | if (offline_node < 0) | |
2958 | return; | |
2959 | ||
2960 | down_read(&slub_lock); | |
2961 | list_for_each_entry(s, &slab_caches, list) { | |
2962 | n = get_node(s, offline_node); | |
2963 | if (n) { | |
2964 | /* | |
2965 | * if n->nr_slabs > 0, slabs still exist on the node | |
2966 | * that is going down. We were unable to free them, | |
2967 | * and offline_pages() function shoudn't call this | |
2968 | * callback. So, we must fail. | |
2969 | */ | |
0f389ec6 | 2970 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2971 | |
2972 | s->node[offline_node] = NULL; | |
2973 | kmem_cache_free(kmalloc_caches, n); | |
2974 | } | |
2975 | } | |
2976 | up_read(&slub_lock); | |
2977 | } | |
2978 | ||
2979 | static int slab_mem_going_online_callback(void *arg) | |
2980 | { | |
2981 | struct kmem_cache_node *n; | |
2982 | struct kmem_cache *s; | |
2983 | struct memory_notify *marg = arg; | |
2984 | int nid = marg->status_change_nid; | |
2985 | int ret = 0; | |
2986 | ||
2987 | /* | |
2988 | * If the node's memory is already available, then kmem_cache_node is | |
2989 | * already created. Nothing to do. | |
2990 | */ | |
2991 | if (nid < 0) | |
2992 | return 0; | |
2993 | ||
2994 | /* | |
0121c619 | 2995 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
2996 | * allocate a kmem_cache_node structure in order to bring the node |
2997 | * online. | |
2998 | */ | |
2999 | down_read(&slub_lock); | |
3000 | list_for_each_entry(s, &slab_caches, list) { | |
3001 | /* | |
3002 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3003 | * since memory is not yet available from the node that | |
3004 | * is brought up. | |
3005 | */ | |
3006 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
3007 | if (!n) { | |
3008 | ret = -ENOMEM; | |
3009 | goto out; | |
3010 | } | |
5595cffc | 3011 | init_kmem_cache_node(n, s); |
b9049e23 YG |
3012 | s->node[nid] = n; |
3013 | } | |
3014 | out: | |
3015 | up_read(&slub_lock); | |
3016 | return ret; | |
3017 | } | |
3018 | ||
3019 | static int slab_memory_callback(struct notifier_block *self, | |
3020 | unsigned long action, void *arg) | |
3021 | { | |
3022 | int ret = 0; | |
3023 | ||
3024 | switch (action) { | |
3025 | case MEM_GOING_ONLINE: | |
3026 | ret = slab_mem_going_online_callback(arg); | |
3027 | break; | |
3028 | case MEM_GOING_OFFLINE: | |
3029 | ret = slab_mem_going_offline_callback(arg); | |
3030 | break; | |
3031 | case MEM_OFFLINE: | |
3032 | case MEM_CANCEL_ONLINE: | |
3033 | slab_mem_offline_callback(arg); | |
3034 | break; | |
3035 | case MEM_ONLINE: | |
3036 | case MEM_CANCEL_OFFLINE: | |
3037 | break; | |
3038 | } | |
dc19f9db KH |
3039 | if (ret) |
3040 | ret = notifier_from_errno(ret); | |
3041 | else | |
3042 | ret = NOTIFY_OK; | |
b9049e23 YG |
3043 | return ret; |
3044 | } | |
3045 | ||
3046 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3047 | ||
81819f0f CL |
3048 | /******************************************************************** |
3049 | * Basic setup of slabs | |
3050 | *******************************************************************/ | |
3051 | ||
3052 | void __init kmem_cache_init(void) | |
3053 | { | |
3054 | int i; | |
4b356be0 | 3055 | int caches = 0; |
81819f0f | 3056 | |
4c93c355 CL |
3057 | init_alloc_cpu(); |
3058 | ||
81819f0f CL |
3059 | #ifdef CONFIG_NUMA |
3060 | /* | |
3061 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3062 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3063 | * kmem_cache_open for slab_state == DOWN. |
3064 | */ | |
3065 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
83b519e8 | 3066 | sizeof(struct kmem_cache_node), GFP_NOWAIT); |
8ffa6875 | 3067 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 3068 | caches++; |
b9049e23 | 3069 | |
0c40ba4f | 3070 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3071 | #endif |
3072 | ||
3073 | /* Able to allocate the per node structures */ | |
3074 | slab_state = PARTIAL; | |
3075 | ||
3076 | /* Caches that are not of the two-to-the-power-of size */ | |
4b356be0 CL |
3077 | if (KMALLOC_MIN_SIZE <= 64) { |
3078 | create_kmalloc_cache(&kmalloc_caches[1], | |
83b519e8 | 3079 | "kmalloc-96", 96, GFP_NOWAIT); |
4b356be0 | 3080 | caches++; |
4b356be0 | 3081 | create_kmalloc_cache(&kmalloc_caches[2], |
83b519e8 | 3082 | "kmalloc-192", 192, GFP_NOWAIT); |
4b356be0 CL |
3083 | caches++; |
3084 | } | |
81819f0f | 3085 | |
ffadd4d0 | 3086 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
81819f0f | 3087 | create_kmalloc_cache(&kmalloc_caches[i], |
83b519e8 | 3088 | "kmalloc", 1 << i, GFP_NOWAIT); |
4b356be0 CL |
3089 | caches++; |
3090 | } | |
81819f0f | 3091 | |
f1b26339 CL |
3092 | |
3093 | /* | |
3094 | * Patch up the size_index table if we have strange large alignment | |
3095 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3096 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3097 | * |
3098 | * Largest permitted alignment is 256 bytes due to the way we | |
3099 | * handle the index determination for the smaller caches. | |
3100 | * | |
3101 | * Make sure that nothing crazy happens if someone starts tinkering | |
3102 | * around with ARCH_KMALLOC_MINALIGN | |
3103 | */ | |
3104 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3105 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3106 | ||
12ad6843 | 3107 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) |
f1b26339 CL |
3108 | size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW; |
3109 | ||
41d54d3b CL |
3110 | if (KMALLOC_MIN_SIZE == 128) { |
3111 | /* | |
3112 | * The 192 byte sized cache is not used if the alignment | |
3113 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3114 | * instead. | |
3115 | */ | |
3116 | for (i = 128 + 8; i <= 192; i += 8) | |
3117 | size_index[(i - 1) / 8] = 8; | |
3118 | } | |
3119 | ||
81819f0f CL |
3120 | slab_state = UP; |
3121 | ||
3122 | /* Provide the correct kmalloc names now that the caches are up */ | |
ffadd4d0 | 3123 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) |
81819f0f | 3124 | kmalloc_caches[i]. name = |
83b519e8 | 3125 | kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); |
81819f0f CL |
3126 | |
3127 | #ifdef CONFIG_SMP | |
3128 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
3129 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
3130 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
3131 | #else | |
3132 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
3133 | #endif |
3134 | ||
3adbefee IM |
3135 | printk(KERN_INFO |
3136 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3137 | " CPUs=%d, Nodes=%d\n", |
3138 | caches, cache_line_size(), | |
81819f0f CL |
3139 | slub_min_order, slub_max_order, slub_min_objects, |
3140 | nr_cpu_ids, nr_node_ids); | |
3141 | } | |
3142 | ||
7e85ee0c PE |
3143 | void __init kmem_cache_init_late(void) |
3144 | { | |
3145 | /* | |
3146 | * Interrupts are enabled now so all GFP allocations are safe. | |
3147 | */ | |
3148 | slab_gfp_mask = __GFP_BITS_MASK; | |
3149 | } | |
3150 | ||
81819f0f CL |
3151 | /* |
3152 | * Find a mergeable slab cache | |
3153 | */ | |
3154 | static int slab_unmergeable(struct kmem_cache *s) | |
3155 | { | |
3156 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3157 | return 1; | |
3158 | ||
c59def9f | 3159 | if (s->ctor) |
81819f0f CL |
3160 | return 1; |
3161 | ||
8ffa6875 CL |
3162 | /* |
3163 | * We may have set a slab to be unmergeable during bootstrap. | |
3164 | */ | |
3165 | if (s->refcount < 0) | |
3166 | return 1; | |
3167 | ||
81819f0f CL |
3168 | return 0; |
3169 | } | |
3170 | ||
3171 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3172 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3173 | void (*ctor)(void *)) |
81819f0f | 3174 | { |
5b95a4ac | 3175 | struct kmem_cache *s; |
81819f0f CL |
3176 | |
3177 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3178 | return NULL; | |
3179 | ||
c59def9f | 3180 | if (ctor) |
81819f0f CL |
3181 | return NULL; |
3182 | ||
3183 | size = ALIGN(size, sizeof(void *)); | |
3184 | align = calculate_alignment(flags, align, size); | |
3185 | size = ALIGN(size, align); | |
ba0268a8 | 3186 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3187 | |
5b95a4ac | 3188 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3189 | if (slab_unmergeable(s)) |
3190 | continue; | |
3191 | ||
3192 | if (size > s->size) | |
3193 | continue; | |
3194 | ||
ba0268a8 | 3195 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3196 | continue; |
3197 | /* | |
3198 | * Check if alignment is compatible. | |
3199 | * Courtesy of Adrian Drzewiecki | |
3200 | */ | |
06428780 | 3201 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3202 | continue; |
3203 | ||
3204 | if (s->size - size >= sizeof(void *)) | |
3205 | continue; | |
3206 | ||
3207 | return s; | |
3208 | } | |
3209 | return NULL; | |
3210 | } | |
3211 | ||
3212 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3213 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3214 | { |
3215 | struct kmem_cache *s; | |
3216 | ||
3217 | down_write(&slub_lock); | |
ba0268a8 | 3218 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3219 | if (s) { |
42a9fdbb CL |
3220 | int cpu; |
3221 | ||
81819f0f CL |
3222 | s->refcount++; |
3223 | /* | |
3224 | * Adjust the object sizes so that we clear | |
3225 | * the complete object on kzalloc. | |
3226 | */ | |
3227 | s->objsize = max(s->objsize, (int)size); | |
42a9fdbb CL |
3228 | |
3229 | /* | |
3230 | * And then we need to update the object size in the | |
3231 | * per cpu structures | |
3232 | */ | |
3233 | for_each_online_cpu(cpu) | |
3234 | get_cpu_slab(s, cpu)->objsize = s->objsize; | |
6446faa2 | 3235 | |
81819f0f | 3236 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 3237 | up_write(&slub_lock); |
6446faa2 | 3238 | |
7b8f3b66 DR |
3239 | if (sysfs_slab_alias(s, name)) { |
3240 | down_write(&slub_lock); | |
3241 | s->refcount--; | |
3242 | up_write(&slub_lock); | |
81819f0f | 3243 | goto err; |
7b8f3b66 | 3244 | } |
a0e1d1be CL |
3245 | return s; |
3246 | } | |
6446faa2 | 3247 | |
a0e1d1be CL |
3248 | s = kmalloc(kmem_size, GFP_KERNEL); |
3249 | if (s) { | |
3250 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3251 | size, align, flags, ctor)) { |
81819f0f | 3252 | list_add(&s->list, &slab_caches); |
a0e1d1be | 3253 | up_write(&slub_lock); |
7b8f3b66 DR |
3254 | if (sysfs_slab_add(s)) { |
3255 | down_write(&slub_lock); | |
3256 | list_del(&s->list); | |
3257 | up_write(&slub_lock); | |
3258 | kfree(s); | |
a0e1d1be | 3259 | goto err; |
7b8f3b66 | 3260 | } |
a0e1d1be CL |
3261 | return s; |
3262 | } | |
3263 | kfree(s); | |
81819f0f CL |
3264 | } |
3265 | up_write(&slub_lock); | |
81819f0f CL |
3266 | |
3267 | err: | |
81819f0f CL |
3268 | if (flags & SLAB_PANIC) |
3269 | panic("Cannot create slabcache %s\n", name); | |
3270 | else | |
3271 | s = NULL; | |
3272 | return s; | |
3273 | } | |
3274 | EXPORT_SYMBOL(kmem_cache_create); | |
3275 | ||
81819f0f | 3276 | #ifdef CONFIG_SMP |
81819f0f | 3277 | /* |
672bba3a CL |
3278 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3279 | * necessary. | |
81819f0f CL |
3280 | */ |
3281 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3282 | unsigned long action, void *hcpu) | |
3283 | { | |
3284 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3285 | struct kmem_cache *s; |
3286 | unsigned long flags; | |
81819f0f CL |
3287 | |
3288 | switch (action) { | |
4c93c355 CL |
3289 | case CPU_UP_PREPARE: |
3290 | case CPU_UP_PREPARE_FROZEN: | |
3291 | init_alloc_cpu_cpu(cpu); | |
3292 | down_read(&slub_lock); | |
3293 | list_for_each_entry(s, &slab_caches, list) | |
3294 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
3295 | GFP_KERNEL); | |
3296 | up_read(&slub_lock); | |
3297 | break; | |
3298 | ||
81819f0f | 3299 | case CPU_UP_CANCELED: |
8bb78442 | 3300 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3301 | case CPU_DEAD: |
8bb78442 | 3302 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3303 | down_read(&slub_lock); |
3304 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
3305 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
3306 | ||
5b95a4ac CL |
3307 | local_irq_save(flags); |
3308 | __flush_cpu_slab(s, cpu); | |
3309 | local_irq_restore(flags); | |
4c93c355 CL |
3310 | free_kmem_cache_cpu(c, cpu); |
3311 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
3312 | } |
3313 | up_read(&slub_lock); | |
81819f0f CL |
3314 | break; |
3315 | default: | |
3316 | break; | |
3317 | } | |
3318 | return NOTIFY_OK; | |
3319 | } | |
3320 | ||
06428780 | 3321 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3322 | .notifier_call = slab_cpuup_callback |
06428780 | 3323 | }; |
81819f0f CL |
3324 | |
3325 | #endif | |
3326 | ||
ce71e27c | 3327 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3328 | { |
aadb4bc4 | 3329 | struct kmem_cache *s; |
94b528d0 | 3330 | void *ret; |
aadb4bc4 | 3331 | |
ffadd4d0 | 3332 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3333 | return kmalloc_large(size, gfpflags); |
3334 | ||
aadb4bc4 | 3335 | s = get_slab(size, gfpflags); |
81819f0f | 3336 | |
2408c550 | 3337 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3338 | return s; |
81819f0f | 3339 | |
94b528d0 EGM |
3340 | ret = slab_alloc(s, gfpflags, -1, caller); |
3341 | ||
3342 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3343 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3344 | |
3345 | return ret; | |
81819f0f CL |
3346 | } |
3347 | ||
3348 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
ce71e27c | 3349 | int node, unsigned long caller) |
81819f0f | 3350 | { |
aadb4bc4 | 3351 | struct kmem_cache *s; |
94b528d0 | 3352 | void *ret; |
aadb4bc4 | 3353 | |
ffadd4d0 | 3354 | if (unlikely(size > SLUB_MAX_SIZE)) |
f619cfe1 | 3355 | return kmalloc_large_node(size, gfpflags, node); |
eada35ef | 3356 | |
aadb4bc4 | 3357 | s = get_slab(size, gfpflags); |
81819f0f | 3358 | |
2408c550 | 3359 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3360 | return s; |
81819f0f | 3361 | |
94b528d0 EGM |
3362 | ret = slab_alloc(s, gfpflags, node, caller); |
3363 | ||
3364 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3365 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3366 | |
3367 | return ret; | |
81819f0f CL |
3368 | } |
3369 | ||
f6acb635 | 3370 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
3371 | static unsigned long count_partial(struct kmem_cache_node *n, |
3372 | int (*get_count)(struct page *)) | |
5b06c853 CL |
3373 | { |
3374 | unsigned long flags; | |
3375 | unsigned long x = 0; | |
3376 | struct page *page; | |
3377 | ||
3378 | spin_lock_irqsave(&n->list_lock, flags); | |
3379 | list_for_each_entry(page, &n->partial, lru) | |
205ab99d | 3380 | x += get_count(page); |
5b06c853 CL |
3381 | spin_unlock_irqrestore(&n->list_lock, flags); |
3382 | return x; | |
3383 | } | |
205ab99d CL |
3384 | |
3385 | static int count_inuse(struct page *page) | |
3386 | { | |
3387 | return page->inuse; | |
3388 | } | |
3389 | ||
3390 | static int count_total(struct page *page) | |
3391 | { | |
3392 | return page->objects; | |
3393 | } | |
3394 | ||
3395 | static int count_free(struct page *page) | |
3396 | { | |
3397 | return page->objects - page->inuse; | |
3398 | } | |
5b06c853 | 3399 | |
434e245d CL |
3400 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3401 | unsigned long *map) | |
53e15af0 CL |
3402 | { |
3403 | void *p; | |
a973e9dd | 3404 | void *addr = page_address(page); |
53e15af0 CL |
3405 | |
3406 | if (!check_slab(s, page) || | |
3407 | !on_freelist(s, page, NULL)) | |
3408 | return 0; | |
3409 | ||
3410 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3411 | bitmap_zero(map, page->objects); |
53e15af0 | 3412 | |
7656c72b CL |
3413 | for_each_free_object(p, s, page->freelist) { |
3414 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3415 | if (!check_object(s, page, p, 0)) |
3416 | return 0; | |
3417 | } | |
3418 | ||
224a88be | 3419 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3420 | if (!test_bit(slab_index(p, s, addr), map)) |
53e15af0 CL |
3421 | if (!check_object(s, page, p, 1)) |
3422 | return 0; | |
3423 | return 1; | |
3424 | } | |
3425 | ||
434e245d CL |
3426 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3427 | unsigned long *map) | |
53e15af0 CL |
3428 | { |
3429 | if (slab_trylock(page)) { | |
434e245d | 3430 | validate_slab(s, page, map); |
53e15af0 CL |
3431 | slab_unlock(page); |
3432 | } else | |
3433 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3434 | s->name, page); | |
3435 | ||
3436 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
8a38082d AW |
3437 | if (!PageSlubDebug(page)) |
3438 | printk(KERN_ERR "SLUB %s: SlubDebug not set " | |
53e15af0 CL |
3439 | "on slab 0x%p\n", s->name, page); |
3440 | } else { | |
8a38082d AW |
3441 | if (PageSlubDebug(page)) |
3442 | printk(KERN_ERR "SLUB %s: SlubDebug set on " | |
53e15af0 CL |
3443 | "slab 0x%p\n", s->name, page); |
3444 | } | |
3445 | } | |
3446 | ||
434e245d CL |
3447 | static int validate_slab_node(struct kmem_cache *s, |
3448 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3449 | { |
3450 | unsigned long count = 0; | |
3451 | struct page *page; | |
3452 | unsigned long flags; | |
3453 | ||
3454 | spin_lock_irqsave(&n->list_lock, flags); | |
3455 | ||
3456 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3457 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3458 | count++; |
3459 | } | |
3460 | if (count != n->nr_partial) | |
3461 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3462 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3463 | ||
3464 | if (!(s->flags & SLAB_STORE_USER)) | |
3465 | goto out; | |
3466 | ||
3467 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3468 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3469 | count++; |
3470 | } | |
3471 | if (count != atomic_long_read(&n->nr_slabs)) | |
3472 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3473 | "counter=%ld\n", s->name, count, | |
3474 | atomic_long_read(&n->nr_slabs)); | |
3475 | ||
3476 | out: | |
3477 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3478 | return count; | |
3479 | } | |
3480 | ||
434e245d | 3481 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3482 | { |
3483 | int node; | |
3484 | unsigned long count = 0; | |
205ab99d | 3485 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3486 | sizeof(unsigned long), GFP_KERNEL); |
3487 | ||
3488 | if (!map) | |
3489 | return -ENOMEM; | |
53e15af0 CL |
3490 | |
3491 | flush_all(s); | |
f64dc58c | 3492 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3493 | struct kmem_cache_node *n = get_node(s, node); |
3494 | ||
434e245d | 3495 | count += validate_slab_node(s, n, map); |
53e15af0 | 3496 | } |
434e245d | 3497 | kfree(map); |
53e15af0 CL |
3498 | return count; |
3499 | } | |
3500 | ||
b3459709 CL |
3501 | #ifdef SLUB_RESILIENCY_TEST |
3502 | static void resiliency_test(void) | |
3503 | { | |
3504 | u8 *p; | |
3505 | ||
3506 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3507 | printk(KERN_ERR "-----------------------\n"); | |
3508 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3509 | ||
3510 | p = kzalloc(16, GFP_KERNEL); | |
3511 | p[16] = 0x12; | |
3512 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3513 | " 0x12->0x%p\n\n", p + 16); | |
3514 | ||
3515 | validate_slab_cache(kmalloc_caches + 4); | |
3516 | ||
3517 | /* Hmmm... The next two are dangerous */ | |
3518 | p = kzalloc(32, GFP_KERNEL); | |
3519 | p[32 + sizeof(void *)] = 0x34; | |
3520 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3521 | " 0x34 -> -0x%p\n", p); |
3522 | printk(KERN_ERR | |
3523 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3524 | |
3525 | validate_slab_cache(kmalloc_caches + 5); | |
3526 | p = kzalloc(64, GFP_KERNEL); | |
3527 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3528 | *p = 0x56; | |
3529 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3530 | p); | |
3adbefee IM |
3531 | printk(KERN_ERR |
3532 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3533 | validate_slab_cache(kmalloc_caches + 6); |
3534 | ||
3535 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3536 | p = kzalloc(128, GFP_KERNEL); | |
3537 | kfree(p); | |
3538 | *p = 0x78; | |
3539 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3540 | validate_slab_cache(kmalloc_caches + 7); | |
3541 | ||
3542 | p = kzalloc(256, GFP_KERNEL); | |
3543 | kfree(p); | |
3544 | p[50] = 0x9a; | |
3adbefee IM |
3545 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3546 | p); | |
b3459709 CL |
3547 | validate_slab_cache(kmalloc_caches + 8); |
3548 | ||
3549 | p = kzalloc(512, GFP_KERNEL); | |
3550 | kfree(p); | |
3551 | p[512] = 0xab; | |
3552 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3553 | validate_slab_cache(kmalloc_caches + 9); | |
3554 | } | |
3555 | #else | |
3556 | static void resiliency_test(void) {}; | |
3557 | #endif | |
3558 | ||
88a420e4 | 3559 | /* |
672bba3a | 3560 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3561 | * and freed. |
3562 | */ | |
3563 | ||
3564 | struct location { | |
3565 | unsigned long count; | |
ce71e27c | 3566 | unsigned long addr; |
45edfa58 CL |
3567 | long long sum_time; |
3568 | long min_time; | |
3569 | long max_time; | |
3570 | long min_pid; | |
3571 | long max_pid; | |
174596a0 | 3572 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3573 | nodemask_t nodes; |
88a420e4 CL |
3574 | }; |
3575 | ||
3576 | struct loc_track { | |
3577 | unsigned long max; | |
3578 | unsigned long count; | |
3579 | struct location *loc; | |
3580 | }; | |
3581 | ||
3582 | static void free_loc_track(struct loc_track *t) | |
3583 | { | |
3584 | if (t->max) | |
3585 | free_pages((unsigned long)t->loc, | |
3586 | get_order(sizeof(struct location) * t->max)); | |
3587 | } | |
3588 | ||
68dff6a9 | 3589 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3590 | { |
3591 | struct location *l; | |
3592 | int order; | |
3593 | ||
88a420e4 CL |
3594 | order = get_order(sizeof(struct location) * max); |
3595 | ||
68dff6a9 | 3596 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3597 | if (!l) |
3598 | return 0; | |
3599 | ||
3600 | if (t->count) { | |
3601 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3602 | free_loc_track(t); | |
3603 | } | |
3604 | t->max = max; | |
3605 | t->loc = l; | |
3606 | return 1; | |
3607 | } | |
3608 | ||
3609 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3610 | const struct track *track) |
88a420e4 CL |
3611 | { |
3612 | long start, end, pos; | |
3613 | struct location *l; | |
ce71e27c | 3614 | unsigned long caddr; |
45edfa58 | 3615 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3616 | |
3617 | start = -1; | |
3618 | end = t->count; | |
3619 | ||
3620 | for ( ; ; ) { | |
3621 | pos = start + (end - start + 1) / 2; | |
3622 | ||
3623 | /* | |
3624 | * There is nothing at "end". If we end up there | |
3625 | * we need to add something to before end. | |
3626 | */ | |
3627 | if (pos == end) | |
3628 | break; | |
3629 | ||
3630 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3631 | if (track->addr == caddr) { |
3632 | ||
3633 | l = &t->loc[pos]; | |
3634 | l->count++; | |
3635 | if (track->when) { | |
3636 | l->sum_time += age; | |
3637 | if (age < l->min_time) | |
3638 | l->min_time = age; | |
3639 | if (age > l->max_time) | |
3640 | l->max_time = age; | |
3641 | ||
3642 | if (track->pid < l->min_pid) | |
3643 | l->min_pid = track->pid; | |
3644 | if (track->pid > l->max_pid) | |
3645 | l->max_pid = track->pid; | |
3646 | ||
174596a0 RR |
3647 | cpumask_set_cpu(track->cpu, |
3648 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3649 | } |
3650 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3651 | return 1; |
3652 | } | |
3653 | ||
45edfa58 | 3654 | if (track->addr < caddr) |
88a420e4 CL |
3655 | end = pos; |
3656 | else | |
3657 | start = pos; | |
3658 | } | |
3659 | ||
3660 | /* | |
672bba3a | 3661 | * Not found. Insert new tracking element. |
88a420e4 | 3662 | */ |
68dff6a9 | 3663 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3664 | return 0; |
3665 | ||
3666 | l = t->loc + pos; | |
3667 | if (pos < t->count) | |
3668 | memmove(l + 1, l, | |
3669 | (t->count - pos) * sizeof(struct location)); | |
3670 | t->count++; | |
3671 | l->count = 1; | |
45edfa58 CL |
3672 | l->addr = track->addr; |
3673 | l->sum_time = age; | |
3674 | l->min_time = age; | |
3675 | l->max_time = age; | |
3676 | l->min_pid = track->pid; | |
3677 | l->max_pid = track->pid; | |
174596a0 RR |
3678 | cpumask_clear(to_cpumask(l->cpus)); |
3679 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3680 | nodes_clear(l->nodes); |
3681 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3682 | return 1; |
3683 | } | |
3684 | ||
3685 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3686 | struct page *page, enum track_item alloc) | |
3687 | { | |
a973e9dd | 3688 | void *addr = page_address(page); |
39b26464 | 3689 | DECLARE_BITMAP(map, page->objects); |
88a420e4 CL |
3690 | void *p; |
3691 | ||
39b26464 | 3692 | bitmap_zero(map, page->objects); |
7656c72b CL |
3693 | for_each_free_object(p, s, page->freelist) |
3694 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3695 | |
224a88be | 3696 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3697 | if (!test_bit(slab_index(p, s, addr), map)) |
3698 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3699 | } |
3700 | ||
3701 | static int list_locations(struct kmem_cache *s, char *buf, | |
3702 | enum track_item alloc) | |
3703 | { | |
e374d483 | 3704 | int len = 0; |
88a420e4 | 3705 | unsigned long i; |
68dff6a9 | 3706 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3707 | int node; |
3708 | ||
68dff6a9 | 3709 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
ea3061d2 | 3710 | GFP_TEMPORARY)) |
68dff6a9 | 3711 | return sprintf(buf, "Out of memory\n"); |
88a420e4 CL |
3712 | |
3713 | /* Push back cpu slabs */ | |
3714 | flush_all(s); | |
3715 | ||
f64dc58c | 3716 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3717 | struct kmem_cache_node *n = get_node(s, node); |
3718 | unsigned long flags; | |
3719 | struct page *page; | |
3720 | ||
9e86943b | 3721 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3722 | continue; |
3723 | ||
3724 | spin_lock_irqsave(&n->list_lock, flags); | |
3725 | list_for_each_entry(page, &n->partial, lru) | |
3726 | process_slab(&t, s, page, alloc); | |
3727 | list_for_each_entry(page, &n->full, lru) | |
3728 | process_slab(&t, s, page, alloc); | |
3729 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3730 | } | |
3731 | ||
3732 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3733 | struct location *l = &t.loc[i]; |
88a420e4 | 3734 | |
9c246247 | 3735 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3736 | break; |
e374d483 | 3737 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3738 | |
3739 | if (l->addr) | |
e374d483 | 3740 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3741 | else |
e374d483 | 3742 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3743 | |
3744 | if (l->sum_time != l->min_time) { | |
e374d483 | 3745 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3746 | l->min_time, |
3747 | (long)div_u64(l->sum_time, l->count), | |
3748 | l->max_time); | |
45edfa58 | 3749 | } else |
e374d483 | 3750 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3751 | l->min_time); |
3752 | ||
3753 | if (l->min_pid != l->max_pid) | |
e374d483 | 3754 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3755 | l->min_pid, l->max_pid); |
3756 | else | |
e374d483 | 3757 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3758 | l->min_pid); |
3759 | ||
174596a0 RR |
3760 | if (num_online_cpus() > 1 && |
3761 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3762 | len < PAGE_SIZE - 60) { |
3763 | len += sprintf(buf + len, " cpus="); | |
3764 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3765 | to_cpumask(l->cpus)); |
45edfa58 CL |
3766 | } |
3767 | ||
62bc62a8 | 3768 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3769 | len < PAGE_SIZE - 60) { |
3770 | len += sprintf(buf + len, " nodes="); | |
3771 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3772 | l->nodes); |
3773 | } | |
3774 | ||
e374d483 | 3775 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3776 | } |
3777 | ||
3778 | free_loc_track(&t); | |
3779 | if (!t.count) | |
e374d483 HH |
3780 | len += sprintf(buf, "No data\n"); |
3781 | return len; | |
88a420e4 CL |
3782 | } |
3783 | ||
81819f0f | 3784 | enum slab_stat_type { |
205ab99d CL |
3785 | SL_ALL, /* All slabs */ |
3786 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3787 | SL_CPU, /* Only slabs used for cpu caches */ | |
3788 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3789 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3790 | }; |
3791 | ||
205ab99d | 3792 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3793 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3794 | #define SO_CPU (1 << SL_CPU) | |
3795 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3796 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3797 | |
62e5c4b4 CG |
3798 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3799 | char *buf, unsigned long flags) | |
81819f0f CL |
3800 | { |
3801 | unsigned long total = 0; | |
81819f0f CL |
3802 | int node; |
3803 | int x; | |
3804 | unsigned long *nodes; | |
3805 | unsigned long *per_cpu; | |
3806 | ||
3807 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3808 | if (!nodes) |
3809 | return -ENOMEM; | |
81819f0f CL |
3810 | per_cpu = nodes + nr_node_ids; |
3811 | ||
205ab99d CL |
3812 | if (flags & SO_CPU) { |
3813 | int cpu; | |
81819f0f | 3814 | |
205ab99d CL |
3815 | for_each_possible_cpu(cpu) { |
3816 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
dfb4f096 | 3817 | |
205ab99d CL |
3818 | if (!c || c->node < 0) |
3819 | continue; | |
3820 | ||
3821 | if (c->page) { | |
3822 | if (flags & SO_TOTAL) | |
3823 | x = c->page->objects; | |
3824 | else if (flags & SO_OBJECTS) | |
3825 | x = c->page->inuse; | |
81819f0f CL |
3826 | else |
3827 | x = 1; | |
205ab99d | 3828 | |
81819f0f | 3829 | total += x; |
205ab99d | 3830 | nodes[c->node] += x; |
81819f0f | 3831 | } |
205ab99d | 3832 | per_cpu[c->node]++; |
81819f0f CL |
3833 | } |
3834 | } | |
3835 | ||
205ab99d CL |
3836 | if (flags & SO_ALL) { |
3837 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3838 | struct kmem_cache_node *n = get_node(s, node); | |
3839 | ||
3840 | if (flags & SO_TOTAL) | |
3841 | x = atomic_long_read(&n->total_objects); | |
3842 | else if (flags & SO_OBJECTS) | |
3843 | x = atomic_long_read(&n->total_objects) - | |
3844 | count_partial(n, count_free); | |
81819f0f | 3845 | |
81819f0f | 3846 | else |
205ab99d | 3847 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3848 | total += x; |
3849 | nodes[node] += x; | |
3850 | } | |
3851 | ||
205ab99d CL |
3852 | } else if (flags & SO_PARTIAL) { |
3853 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3854 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3855 | |
205ab99d CL |
3856 | if (flags & SO_TOTAL) |
3857 | x = count_partial(n, count_total); | |
3858 | else if (flags & SO_OBJECTS) | |
3859 | x = count_partial(n, count_inuse); | |
81819f0f | 3860 | else |
205ab99d | 3861 | x = n->nr_partial; |
81819f0f CL |
3862 | total += x; |
3863 | nodes[node] += x; | |
3864 | } | |
3865 | } | |
81819f0f CL |
3866 | x = sprintf(buf, "%lu", total); |
3867 | #ifdef CONFIG_NUMA | |
f64dc58c | 3868 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3869 | if (nodes[node]) |
3870 | x += sprintf(buf + x, " N%d=%lu", | |
3871 | node, nodes[node]); | |
3872 | #endif | |
3873 | kfree(nodes); | |
3874 | return x + sprintf(buf + x, "\n"); | |
3875 | } | |
3876 | ||
3877 | static int any_slab_objects(struct kmem_cache *s) | |
3878 | { | |
3879 | int node; | |
81819f0f | 3880 | |
dfb4f096 | 3881 | for_each_online_node(node) { |
81819f0f CL |
3882 | struct kmem_cache_node *n = get_node(s, node); |
3883 | ||
dfb4f096 CL |
3884 | if (!n) |
3885 | continue; | |
3886 | ||
4ea33e2d | 3887 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3888 | return 1; |
3889 | } | |
3890 | return 0; | |
3891 | } | |
3892 | ||
3893 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3894 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3895 | ||
3896 | struct slab_attribute { | |
3897 | struct attribute attr; | |
3898 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3899 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3900 | }; | |
3901 | ||
3902 | #define SLAB_ATTR_RO(_name) \ | |
3903 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3904 | ||
3905 | #define SLAB_ATTR(_name) \ | |
3906 | static struct slab_attribute _name##_attr = \ | |
3907 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3908 | ||
81819f0f CL |
3909 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3910 | { | |
3911 | return sprintf(buf, "%d\n", s->size); | |
3912 | } | |
3913 | SLAB_ATTR_RO(slab_size); | |
3914 | ||
3915 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3916 | { | |
3917 | return sprintf(buf, "%d\n", s->align); | |
3918 | } | |
3919 | SLAB_ATTR_RO(align); | |
3920 | ||
3921 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3922 | { | |
3923 | return sprintf(buf, "%d\n", s->objsize); | |
3924 | } | |
3925 | SLAB_ATTR_RO(object_size); | |
3926 | ||
3927 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3928 | { | |
834f3d11 | 3929 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3930 | } |
3931 | SLAB_ATTR_RO(objs_per_slab); | |
3932 | ||
06b285dc CL |
3933 | static ssize_t order_store(struct kmem_cache *s, |
3934 | const char *buf, size_t length) | |
3935 | { | |
0121c619 CL |
3936 | unsigned long order; |
3937 | int err; | |
3938 | ||
3939 | err = strict_strtoul(buf, 10, &order); | |
3940 | if (err) | |
3941 | return err; | |
06b285dc CL |
3942 | |
3943 | if (order > slub_max_order || order < slub_min_order) | |
3944 | return -EINVAL; | |
3945 | ||
3946 | calculate_sizes(s, order); | |
3947 | return length; | |
3948 | } | |
3949 | ||
81819f0f CL |
3950 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
3951 | { | |
834f3d11 | 3952 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 3953 | } |
06b285dc | 3954 | SLAB_ATTR(order); |
81819f0f | 3955 | |
73d342b1 DR |
3956 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
3957 | { | |
3958 | return sprintf(buf, "%lu\n", s->min_partial); | |
3959 | } | |
3960 | ||
3961 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
3962 | size_t length) | |
3963 | { | |
3964 | unsigned long min; | |
3965 | int err; | |
3966 | ||
3967 | err = strict_strtoul(buf, 10, &min); | |
3968 | if (err) | |
3969 | return err; | |
3970 | ||
c0bdb232 | 3971 | set_min_partial(s, min); |
73d342b1 DR |
3972 | return length; |
3973 | } | |
3974 | SLAB_ATTR(min_partial); | |
3975 | ||
81819f0f CL |
3976 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
3977 | { | |
3978 | if (s->ctor) { | |
3979 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3980 | ||
3981 | return n + sprintf(buf + n, "\n"); | |
3982 | } | |
3983 | return 0; | |
3984 | } | |
3985 | SLAB_ATTR_RO(ctor); | |
3986 | ||
81819f0f CL |
3987 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3988 | { | |
3989 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3990 | } | |
3991 | SLAB_ATTR_RO(aliases); | |
3992 | ||
3993 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3994 | { | |
205ab99d | 3995 | return show_slab_objects(s, buf, SO_ALL); |
81819f0f CL |
3996 | } |
3997 | SLAB_ATTR_RO(slabs); | |
3998 | ||
3999 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
4000 | { | |
d9acf4b7 | 4001 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4002 | } |
4003 | SLAB_ATTR_RO(partial); | |
4004 | ||
4005 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4006 | { | |
d9acf4b7 | 4007 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4008 | } |
4009 | SLAB_ATTR_RO(cpu_slabs); | |
4010 | ||
4011 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4012 | { | |
205ab99d | 4013 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4014 | } |
4015 | SLAB_ATTR_RO(objects); | |
4016 | ||
205ab99d CL |
4017 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4018 | { | |
4019 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4020 | } | |
4021 | SLAB_ATTR_RO(objects_partial); | |
4022 | ||
4023 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
4024 | { | |
4025 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4026 | } | |
4027 | SLAB_ATTR_RO(total_objects); | |
4028 | ||
81819f0f CL |
4029 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4030 | { | |
4031 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4032 | } | |
4033 | ||
4034 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4035 | const char *buf, size_t length) | |
4036 | { | |
4037 | s->flags &= ~SLAB_DEBUG_FREE; | |
4038 | if (buf[0] == '1') | |
4039 | s->flags |= SLAB_DEBUG_FREE; | |
4040 | return length; | |
4041 | } | |
4042 | SLAB_ATTR(sanity_checks); | |
4043 | ||
4044 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4045 | { | |
4046 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4047 | } | |
4048 | ||
4049 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4050 | size_t length) | |
4051 | { | |
4052 | s->flags &= ~SLAB_TRACE; | |
4053 | if (buf[0] == '1') | |
4054 | s->flags |= SLAB_TRACE; | |
4055 | return length; | |
4056 | } | |
4057 | SLAB_ATTR(trace); | |
4058 | ||
4059 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
4060 | { | |
4061 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4062 | } | |
4063 | ||
4064 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4065 | const char *buf, size_t length) | |
4066 | { | |
4067 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4068 | if (buf[0] == '1') | |
4069 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4070 | return length; | |
4071 | } | |
4072 | SLAB_ATTR(reclaim_account); | |
4073 | ||
4074 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4075 | { | |
5af60839 | 4076 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
4077 | } |
4078 | SLAB_ATTR_RO(hwcache_align); | |
4079 | ||
4080 | #ifdef CONFIG_ZONE_DMA | |
4081 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4082 | { | |
4083 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4084 | } | |
4085 | SLAB_ATTR_RO(cache_dma); | |
4086 | #endif | |
4087 | ||
4088 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4089 | { | |
4090 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4091 | } | |
4092 | SLAB_ATTR_RO(destroy_by_rcu); | |
4093 | ||
4094 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
4095 | { | |
4096 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4097 | } | |
4098 | ||
4099 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4100 | const char *buf, size_t length) | |
4101 | { | |
4102 | if (any_slab_objects(s)) | |
4103 | return -EBUSY; | |
4104 | ||
4105 | s->flags &= ~SLAB_RED_ZONE; | |
4106 | if (buf[0] == '1') | |
4107 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4108 | calculate_sizes(s, -1); |
81819f0f CL |
4109 | return length; |
4110 | } | |
4111 | SLAB_ATTR(red_zone); | |
4112 | ||
4113 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4114 | { | |
4115 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4116 | } | |
4117 | ||
4118 | static ssize_t poison_store(struct kmem_cache *s, | |
4119 | const char *buf, size_t length) | |
4120 | { | |
4121 | if (any_slab_objects(s)) | |
4122 | return -EBUSY; | |
4123 | ||
4124 | s->flags &= ~SLAB_POISON; | |
4125 | if (buf[0] == '1') | |
4126 | s->flags |= SLAB_POISON; | |
06b285dc | 4127 | calculate_sizes(s, -1); |
81819f0f CL |
4128 | return length; |
4129 | } | |
4130 | SLAB_ATTR(poison); | |
4131 | ||
4132 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4133 | { | |
4134 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4135 | } | |
4136 | ||
4137 | static ssize_t store_user_store(struct kmem_cache *s, | |
4138 | const char *buf, size_t length) | |
4139 | { | |
4140 | if (any_slab_objects(s)) | |
4141 | return -EBUSY; | |
4142 | ||
4143 | s->flags &= ~SLAB_STORE_USER; | |
4144 | if (buf[0] == '1') | |
4145 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4146 | calculate_sizes(s, -1); |
81819f0f CL |
4147 | return length; |
4148 | } | |
4149 | SLAB_ATTR(store_user); | |
4150 | ||
53e15af0 CL |
4151 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4152 | { | |
4153 | return 0; | |
4154 | } | |
4155 | ||
4156 | static ssize_t validate_store(struct kmem_cache *s, | |
4157 | const char *buf, size_t length) | |
4158 | { | |
434e245d CL |
4159 | int ret = -EINVAL; |
4160 | ||
4161 | if (buf[0] == '1') { | |
4162 | ret = validate_slab_cache(s); | |
4163 | if (ret >= 0) | |
4164 | ret = length; | |
4165 | } | |
4166 | return ret; | |
53e15af0 CL |
4167 | } |
4168 | SLAB_ATTR(validate); | |
4169 | ||
2086d26a CL |
4170 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4171 | { | |
4172 | return 0; | |
4173 | } | |
4174 | ||
4175 | static ssize_t shrink_store(struct kmem_cache *s, | |
4176 | const char *buf, size_t length) | |
4177 | { | |
4178 | if (buf[0] == '1') { | |
4179 | int rc = kmem_cache_shrink(s); | |
4180 | ||
4181 | if (rc) | |
4182 | return rc; | |
4183 | } else | |
4184 | return -EINVAL; | |
4185 | return length; | |
4186 | } | |
4187 | SLAB_ATTR(shrink); | |
4188 | ||
88a420e4 CL |
4189 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4190 | { | |
4191 | if (!(s->flags & SLAB_STORE_USER)) | |
4192 | return -ENOSYS; | |
4193 | return list_locations(s, buf, TRACK_ALLOC); | |
4194 | } | |
4195 | SLAB_ATTR_RO(alloc_calls); | |
4196 | ||
4197 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4198 | { | |
4199 | if (!(s->flags & SLAB_STORE_USER)) | |
4200 | return -ENOSYS; | |
4201 | return list_locations(s, buf, TRACK_FREE); | |
4202 | } | |
4203 | SLAB_ATTR_RO(free_calls); | |
4204 | ||
81819f0f | 4205 | #ifdef CONFIG_NUMA |
9824601e | 4206 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4207 | { |
9824601e | 4208 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4209 | } |
4210 | ||
9824601e | 4211 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4212 | const char *buf, size_t length) |
4213 | { | |
0121c619 CL |
4214 | unsigned long ratio; |
4215 | int err; | |
4216 | ||
4217 | err = strict_strtoul(buf, 10, &ratio); | |
4218 | if (err) | |
4219 | return err; | |
4220 | ||
e2cb96b7 | 4221 | if (ratio <= 100) |
0121c619 | 4222 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4223 | |
81819f0f CL |
4224 | return length; |
4225 | } | |
9824601e | 4226 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4227 | #endif |
4228 | ||
8ff12cfc | 4229 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4230 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4231 | { | |
4232 | unsigned long sum = 0; | |
4233 | int cpu; | |
4234 | int len; | |
4235 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4236 | ||
4237 | if (!data) | |
4238 | return -ENOMEM; | |
4239 | ||
4240 | for_each_online_cpu(cpu) { | |
4241 | unsigned x = get_cpu_slab(s, cpu)->stat[si]; | |
4242 | ||
4243 | data[cpu] = x; | |
4244 | sum += x; | |
4245 | } | |
4246 | ||
4247 | len = sprintf(buf, "%lu", sum); | |
4248 | ||
50ef37b9 | 4249 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4250 | for_each_online_cpu(cpu) { |
4251 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4252 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4253 | } |
50ef37b9 | 4254 | #endif |
8ff12cfc CL |
4255 | kfree(data); |
4256 | return len + sprintf(buf + len, "\n"); | |
4257 | } | |
4258 | ||
4259 | #define STAT_ATTR(si, text) \ | |
4260 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4261 | { \ | |
4262 | return show_stat(s, buf, si); \ | |
4263 | } \ | |
4264 | SLAB_ATTR_RO(text); \ | |
4265 | ||
4266 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4267 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4268 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4269 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4270 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4271 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4272 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4273 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4274 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4275 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4276 | STAT_ATTR(FREE_SLAB, free_slab); | |
4277 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4278 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4279 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4280 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4281 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4282 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4283 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4284 | #endif |
4285 | ||
06428780 | 4286 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4287 | &slab_size_attr.attr, |
4288 | &object_size_attr.attr, | |
4289 | &objs_per_slab_attr.attr, | |
4290 | &order_attr.attr, | |
73d342b1 | 4291 | &min_partial_attr.attr, |
81819f0f | 4292 | &objects_attr.attr, |
205ab99d CL |
4293 | &objects_partial_attr.attr, |
4294 | &total_objects_attr.attr, | |
81819f0f CL |
4295 | &slabs_attr.attr, |
4296 | &partial_attr.attr, | |
4297 | &cpu_slabs_attr.attr, | |
4298 | &ctor_attr.attr, | |
81819f0f CL |
4299 | &aliases_attr.attr, |
4300 | &align_attr.attr, | |
4301 | &sanity_checks_attr.attr, | |
4302 | &trace_attr.attr, | |
4303 | &hwcache_align_attr.attr, | |
4304 | &reclaim_account_attr.attr, | |
4305 | &destroy_by_rcu_attr.attr, | |
4306 | &red_zone_attr.attr, | |
4307 | &poison_attr.attr, | |
4308 | &store_user_attr.attr, | |
53e15af0 | 4309 | &validate_attr.attr, |
2086d26a | 4310 | &shrink_attr.attr, |
88a420e4 CL |
4311 | &alloc_calls_attr.attr, |
4312 | &free_calls_attr.attr, | |
81819f0f CL |
4313 | #ifdef CONFIG_ZONE_DMA |
4314 | &cache_dma_attr.attr, | |
4315 | #endif | |
4316 | #ifdef CONFIG_NUMA | |
9824601e | 4317 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4318 | #endif |
4319 | #ifdef CONFIG_SLUB_STATS | |
4320 | &alloc_fastpath_attr.attr, | |
4321 | &alloc_slowpath_attr.attr, | |
4322 | &free_fastpath_attr.attr, | |
4323 | &free_slowpath_attr.attr, | |
4324 | &free_frozen_attr.attr, | |
4325 | &free_add_partial_attr.attr, | |
4326 | &free_remove_partial_attr.attr, | |
4327 | &alloc_from_partial_attr.attr, | |
4328 | &alloc_slab_attr.attr, | |
4329 | &alloc_refill_attr.attr, | |
4330 | &free_slab_attr.attr, | |
4331 | &cpuslab_flush_attr.attr, | |
4332 | &deactivate_full_attr.attr, | |
4333 | &deactivate_empty_attr.attr, | |
4334 | &deactivate_to_head_attr.attr, | |
4335 | &deactivate_to_tail_attr.attr, | |
4336 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4337 | &order_fallback_attr.attr, |
81819f0f CL |
4338 | #endif |
4339 | NULL | |
4340 | }; | |
4341 | ||
4342 | static struct attribute_group slab_attr_group = { | |
4343 | .attrs = slab_attrs, | |
4344 | }; | |
4345 | ||
4346 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4347 | struct attribute *attr, | |
4348 | char *buf) | |
4349 | { | |
4350 | struct slab_attribute *attribute; | |
4351 | struct kmem_cache *s; | |
4352 | int err; | |
4353 | ||
4354 | attribute = to_slab_attr(attr); | |
4355 | s = to_slab(kobj); | |
4356 | ||
4357 | if (!attribute->show) | |
4358 | return -EIO; | |
4359 | ||
4360 | err = attribute->show(s, buf); | |
4361 | ||
4362 | return err; | |
4363 | } | |
4364 | ||
4365 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4366 | struct attribute *attr, | |
4367 | const char *buf, size_t len) | |
4368 | { | |
4369 | struct slab_attribute *attribute; | |
4370 | struct kmem_cache *s; | |
4371 | int err; | |
4372 | ||
4373 | attribute = to_slab_attr(attr); | |
4374 | s = to_slab(kobj); | |
4375 | ||
4376 | if (!attribute->store) | |
4377 | return -EIO; | |
4378 | ||
4379 | err = attribute->store(s, buf, len); | |
4380 | ||
4381 | return err; | |
4382 | } | |
4383 | ||
151c602f CL |
4384 | static void kmem_cache_release(struct kobject *kobj) |
4385 | { | |
4386 | struct kmem_cache *s = to_slab(kobj); | |
4387 | ||
4388 | kfree(s); | |
4389 | } | |
4390 | ||
81819f0f CL |
4391 | static struct sysfs_ops slab_sysfs_ops = { |
4392 | .show = slab_attr_show, | |
4393 | .store = slab_attr_store, | |
4394 | }; | |
4395 | ||
4396 | static struct kobj_type slab_ktype = { | |
4397 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4398 | .release = kmem_cache_release |
81819f0f CL |
4399 | }; |
4400 | ||
4401 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4402 | { | |
4403 | struct kobj_type *ktype = get_ktype(kobj); | |
4404 | ||
4405 | if (ktype == &slab_ktype) | |
4406 | return 1; | |
4407 | return 0; | |
4408 | } | |
4409 | ||
4410 | static struct kset_uevent_ops slab_uevent_ops = { | |
4411 | .filter = uevent_filter, | |
4412 | }; | |
4413 | ||
27c3a314 | 4414 | static struct kset *slab_kset; |
81819f0f CL |
4415 | |
4416 | #define ID_STR_LENGTH 64 | |
4417 | ||
4418 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4419 | * |
4420 | * Format :[flags-]size | |
81819f0f CL |
4421 | */ |
4422 | static char *create_unique_id(struct kmem_cache *s) | |
4423 | { | |
4424 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4425 | char *p = name; | |
4426 | ||
4427 | BUG_ON(!name); | |
4428 | ||
4429 | *p++ = ':'; | |
4430 | /* | |
4431 | * First flags affecting slabcache operations. We will only | |
4432 | * get here for aliasable slabs so we do not need to support | |
4433 | * too many flags. The flags here must cover all flags that | |
4434 | * are matched during merging to guarantee that the id is | |
4435 | * unique. | |
4436 | */ | |
4437 | if (s->flags & SLAB_CACHE_DMA) | |
4438 | *p++ = 'd'; | |
4439 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4440 | *p++ = 'a'; | |
4441 | if (s->flags & SLAB_DEBUG_FREE) | |
4442 | *p++ = 'F'; | |
5a896d9e VN |
4443 | if (!(s->flags & SLAB_NOTRACK)) |
4444 | *p++ = 't'; | |
81819f0f CL |
4445 | if (p != name + 1) |
4446 | *p++ = '-'; | |
4447 | p += sprintf(p, "%07d", s->size); | |
4448 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4449 | return name; | |
4450 | } | |
4451 | ||
4452 | static int sysfs_slab_add(struct kmem_cache *s) | |
4453 | { | |
4454 | int err; | |
4455 | const char *name; | |
4456 | int unmergeable; | |
4457 | ||
4458 | if (slab_state < SYSFS) | |
4459 | /* Defer until later */ | |
4460 | return 0; | |
4461 | ||
4462 | unmergeable = slab_unmergeable(s); | |
4463 | if (unmergeable) { | |
4464 | /* | |
4465 | * Slabcache can never be merged so we can use the name proper. | |
4466 | * This is typically the case for debug situations. In that | |
4467 | * case we can catch duplicate names easily. | |
4468 | */ | |
27c3a314 | 4469 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4470 | name = s->name; |
4471 | } else { | |
4472 | /* | |
4473 | * Create a unique name for the slab as a target | |
4474 | * for the symlinks. | |
4475 | */ | |
4476 | name = create_unique_id(s); | |
4477 | } | |
4478 | ||
27c3a314 | 4479 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4480 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4481 | if (err) { | |
4482 | kobject_put(&s->kobj); | |
81819f0f | 4483 | return err; |
1eada11c | 4484 | } |
81819f0f CL |
4485 | |
4486 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
4487 | if (err) | |
4488 | return err; | |
4489 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
4490 | if (!unmergeable) { | |
4491 | /* Setup first alias */ | |
4492 | sysfs_slab_alias(s, s->name); | |
4493 | kfree(name); | |
4494 | } | |
4495 | return 0; | |
4496 | } | |
4497 | ||
4498 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4499 | { | |
4500 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4501 | kobject_del(&s->kobj); | |
151c602f | 4502 | kobject_put(&s->kobj); |
81819f0f CL |
4503 | } |
4504 | ||
4505 | /* | |
4506 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4507 | * available lest we lose that information. |
81819f0f CL |
4508 | */ |
4509 | struct saved_alias { | |
4510 | struct kmem_cache *s; | |
4511 | const char *name; | |
4512 | struct saved_alias *next; | |
4513 | }; | |
4514 | ||
5af328a5 | 4515 | static struct saved_alias *alias_list; |
81819f0f CL |
4516 | |
4517 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4518 | { | |
4519 | struct saved_alias *al; | |
4520 | ||
4521 | if (slab_state == SYSFS) { | |
4522 | /* | |
4523 | * If we have a leftover link then remove it. | |
4524 | */ | |
27c3a314 GKH |
4525 | sysfs_remove_link(&slab_kset->kobj, name); |
4526 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4527 | } |
4528 | ||
4529 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4530 | if (!al) | |
4531 | return -ENOMEM; | |
4532 | ||
4533 | al->s = s; | |
4534 | al->name = name; | |
4535 | al->next = alias_list; | |
4536 | alias_list = al; | |
4537 | return 0; | |
4538 | } | |
4539 | ||
4540 | static int __init slab_sysfs_init(void) | |
4541 | { | |
5b95a4ac | 4542 | struct kmem_cache *s; |
81819f0f CL |
4543 | int err; |
4544 | ||
0ff21e46 | 4545 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4546 | if (!slab_kset) { |
81819f0f CL |
4547 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4548 | return -ENOSYS; | |
4549 | } | |
4550 | ||
26a7bd03 CL |
4551 | slab_state = SYSFS; |
4552 | ||
5b95a4ac | 4553 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4554 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4555 | if (err) |
4556 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4557 | " to sysfs\n", s->name); | |
26a7bd03 | 4558 | } |
81819f0f CL |
4559 | |
4560 | while (alias_list) { | |
4561 | struct saved_alias *al = alias_list; | |
4562 | ||
4563 | alias_list = alias_list->next; | |
4564 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4565 | if (err) |
4566 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4567 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4568 | kfree(al); |
4569 | } | |
4570 | ||
4571 | resiliency_test(); | |
4572 | return 0; | |
4573 | } | |
4574 | ||
4575 | __initcall(slab_sysfs_init); | |
81819f0f | 4576 | #endif |
57ed3eda PE |
4577 | |
4578 | /* | |
4579 | * The /proc/slabinfo ABI | |
4580 | */ | |
158a9624 | 4581 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4582 | static void print_slabinfo_header(struct seq_file *m) |
4583 | { | |
4584 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4585 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4586 | "<objperslab> <pagesperslab>"); | |
4587 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4588 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4589 | seq_putc(m, '\n'); | |
4590 | } | |
4591 | ||
4592 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4593 | { | |
4594 | loff_t n = *pos; | |
4595 | ||
4596 | down_read(&slub_lock); | |
4597 | if (!n) | |
4598 | print_slabinfo_header(m); | |
4599 | ||
4600 | return seq_list_start(&slab_caches, *pos); | |
4601 | } | |
4602 | ||
4603 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4604 | { | |
4605 | return seq_list_next(p, &slab_caches, pos); | |
4606 | } | |
4607 | ||
4608 | static void s_stop(struct seq_file *m, void *p) | |
4609 | { | |
4610 | up_read(&slub_lock); | |
4611 | } | |
4612 | ||
4613 | static int s_show(struct seq_file *m, void *p) | |
4614 | { | |
4615 | unsigned long nr_partials = 0; | |
4616 | unsigned long nr_slabs = 0; | |
4617 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4618 | unsigned long nr_objs = 0; |
4619 | unsigned long nr_free = 0; | |
57ed3eda PE |
4620 | struct kmem_cache *s; |
4621 | int node; | |
4622 | ||
4623 | s = list_entry(p, struct kmem_cache, list); | |
4624 | ||
4625 | for_each_online_node(node) { | |
4626 | struct kmem_cache_node *n = get_node(s, node); | |
4627 | ||
4628 | if (!n) | |
4629 | continue; | |
4630 | ||
4631 | nr_partials += n->nr_partial; | |
4632 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4633 | nr_objs += atomic_long_read(&n->total_objects); |
4634 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4635 | } |
4636 | ||
205ab99d | 4637 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4638 | |
4639 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4640 | nr_objs, s->size, oo_objects(s->oo), |
4641 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4642 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4643 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4644 | 0UL); | |
4645 | seq_putc(m, '\n'); | |
4646 | return 0; | |
4647 | } | |
4648 | ||
7b3c3a50 | 4649 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4650 | .start = s_start, |
4651 | .next = s_next, | |
4652 | .stop = s_stop, | |
4653 | .show = s_show, | |
4654 | }; | |
4655 | ||
7b3c3a50 AD |
4656 | static int slabinfo_open(struct inode *inode, struct file *file) |
4657 | { | |
4658 | return seq_open(file, &slabinfo_op); | |
4659 | } | |
4660 | ||
4661 | static const struct file_operations proc_slabinfo_operations = { | |
4662 | .open = slabinfo_open, | |
4663 | .read = seq_read, | |
4664 | .llseek = seq_lseek, | |
4665 | .release = seq_release, | |
4666 | }; | |
4667 | ||
4668 | static int __init slab_proc_init(void) | |
4669 | { | |
4670 | proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations); | |
4671 | return 0; | |
4672 | } | |
4673 | module_init(slab_proc_init); | |
158a9624 | 4674 | #endif /* CONFIG_SLABINFO */ |