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> |
5a896d9e | 20 | #include <linux/kmemcheck.h> |
81819f0f CL |
21 | #include <linux/cpu.h> |
22 | #include <linux/cpuset.h> | |
23 | #include <linux/mempolicy.h> | |
24 | #include <linux/ctype.h> | |
3ac7fe5a | 25 | #include <linux/debugobjects.h> |
81819f0f | 26 | #include <linux/kallsyms.h> |
b9049e23 | 27 | #include <linux/memory.h> |
f8bd2258 | 28 | #include <linux/math64.h> |
773ff60e | 29 | #include <linux/fault-inject.h> |
81819f0f CL |
30 | |
31 | /* | |
32 | * Lock order: | |
33 | * 1. slab_lock(page) | |
34 | * 2. slab->list_lock | |
35 | * | |
36 | * The slab_lock protects operations on the object of a particular | |
37 | * slab and its metadata in the page struct. If the slab lock | |
38 | * has been taken then no allocations nor frees can be performed | |
39 | * on the objects in the slab nor can the slab be added or removed | |
40 | * from the partial or full lists since this would mean modifying | |
41 | * the page_struct of the slab. | |
42 | * | |
43 | * The list_lock protects the partial and full list on each node and | |
44 | * the partial slab counter. If taken then no new slabs may be added or | |
45 | * removed from the lists nor make the number of partial slabs be modified. | |
46 | * (Note that the total number of slabs is an atomic value that may be | |
47 | * modified without taking the list lock). | |
48 | * | |
49 | * The list_lock is a centralized lock and thus we avoid taking it as | |
50 | * much as possible. As long as SLUB does not have to handle partial | |
51 | * slabs, operations can continue without any centralized lock. F.e. | |
52 | * allocating a long series of objects that fill up slabs does not require | |
53 | * the list lock. | |
54 | * | |
55 | * The lock order is sometimes inverted when we are trying to get a slab | |
56 | * off a list. We take the list_lock and then look for a page on the list | |
57 | * to use. While we do that objects in the slabs may be freed. We can | |
58 | * only operate on the slab if we have also taken the slab_lock. So we use | |
59 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
60 | * can occur anymore and we can use the slab for allocations etc. If the | |
61 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
62 | * we must stay away from it for a while since we may cause a bouncing | |
63 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
64 | * If all pages are busy then we may allocate a new slab instead of reusing | |
65 | * a partial slab. A new slab has noone operating on it and thus there is | |
66 | * no danger of cacheline contention. | |
67 | * | |
68 | * Interrupts are disabled during allocation and deallocation in order to | |
69 | * make the slab allocator safe to use in the context of an irq. In addition | |
70 | * interrupts are disabled to ensure that the processor does not change | |
71 | * while handling per_cpu slabs, due to kernel preemption. | |
72 | * | |
73 | * SLUB assigns one slab for allocation to each processor. | |
74 | * Allocations only occur from these slabs called cpu slabs. | |
75 | * | |
672bba3a CL |
76 | * Slabs with free elements are kept on a partial list and during regular |
77 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 78 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
79 | * We track full slabs for debugging purposes though because otherwise we |
80 | * cannot scan all objects. | |
81819f0f CL |
81 | * |
82 | * Slabs are freed when they become empty. Teardown and setup is | |
83 | * minimal so we rely on the page allocators per cpu caches for | |
84 | * fast frees and allocs. | |
85 | * | |
86 | * Overloading of page flags that are otherwise used for LRU management. | |
87 | * | |
4b6f0750 CL |
88 | * PageActive The slab is frozen and exempt from list processing. |
89 | * This means that the slab is dedicated to a purpose | |
90 | * such as satisfying allocations for a specific | |
91 | * processor. Objects may be freed in the slab while | |
92 | * it is frozen but slab_free will then skip the usual | |
93 | * list operations. It is up to the processor holding | |
94 | * the slab to integrate the slab into the slab lists | |
95 | * when the slab is no longer needed. | |
96 | * | |
97 | * One use of this flag is to mark slabs that are | |
98 | * used for allocations. Then such a slab becomes a cpu | |
99 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 100 | * freelist that allows lockless access to |
894b8788 CL |
101 | * free objects in addition to the regular freelist |
102 | * that requires the slab lock. | |
81819f0f CL |
103 | * |
104 | * PageError Slab requires special handling due to debug | |
105 | * options set. This moves slab handling out of | |
894b8788 | 106 | * the fast path and disables lockless freelists. |
81819f0f CL |
107 | */ |
108 | ||
af537b0a CL |
109 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
110 | SLAB_TRACE | SLAB_DEBUG_FREE) | |
111 | ||
112 | static inline int kmem_cache_debug(struct kmem_cache *s) | |
113 | { | |
5577bd8a | 114 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 115 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 116 | #else |
af537b0a | 117 | return 0; |
5577bd8a | 118 | #endif |
af537b0a | 119 | } |
5577bd8a | 120 | |
81819f0f CL |
121 | /* |
122 | * Issues still to be resolved: | |
123 | * | |
81819f0f CL |
124 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
125 | * | |
81819f0f CL |
126 | * - Variable sizing of the per node arrays |
127 | */ | |
128 | ||
129 | /* Enable to test recovery from slab corruption on boot */ | |
130 | #undef SLUB_RESILIENCY_TEST | |
131 | ||
2086d26a CL |
132 | /* |
133 | * Mininum number of partial slabs. These will be left on the partial | |
134 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
135 | */ | |
76be8950 | 136 | #define MIN_PARTIAL 5 |
e95eed57 | 137 | |
2086d26a CL |
138 | /* |
139 | * Maximum number of desirable partial slabs. | |
140 | * The existence of more partial slabs makes kmem_cache_shrink | |
141 | * sort the partial list by the number of objects in the. | |
142 | */ | |
143 | #define MAX_PARTIAL 10 | |
144 | ||
81819f0f CL |
145 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
146 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 147 | |
fa5ec8a1 | 148 | /* |
3de47213 DR |
149 | * Debugging flags that require metadata to be stored in the slab. These get |
150 | * disabled when slub_debug=O is used and a cache's min order increases with | |
151 | * metadata. | |
fa5ec8a1 | 152 | */ |
3de47213 | 153 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 154 | |
81819f0f CL |
155 | /* |
156 | * Set of flags that will prevent slab merging | |
157 | */ | |
158 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
159 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
160 | SLAB_FAILSLAB) | |
81819f0f CL |
161 | |
162 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 163 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 164 | |
210b5c06 CG |
165 | #define OO_SHIFT 16 |
166 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
167 | #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */ | |
168 | ||
81819f0f | 169 | /* Internal SLUB flags */ |
f90ec390 | 170 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
81819f0f CL |
171 | |
172 | static int kmem_size = sizeof(struct kmem_cache); | |
173 | ||
174 | #ifdef CONFIG_SMP | |
175 | static struct notifier_block slab_notifier; | |
176 | #endif | |
177 | ||
178 | static enum { | |
179 | DOWN, /* No slab functionality available */ | |
51df1142 | 180 | PARTIAL, /* Kmem_cache_node works */ |
672bba3a | 181 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
182 | SYSFS /* Sysfs up */ |
183 | } slab_state = DOWN; | |
184 | ||
185 | /* A list of all slab caches on the system */ | |
186 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 187 | static LIST_HEAD(slab_caches); |
81819f0f | 188 | |
02cbc874 CL |
189 | /* |
190 | * Tracking user of a slab. | |
191 | */ | |
192 | struct track { | |
ce71e27c | 193 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
194 | int cpu; /* Was running on cpu */ |
195 | int pid; /* Pid context */ | |
196 | unsigned long when; /* When did the operation occur */ | |
197 | }; | |
198 | ||
199 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
200 | ||
ab4d5ed5 | 201 | #ifdef CONFIG_SYSFS |
81819f0f CL |
202 | static int sysfs_slab_add(struct kmem_cache *); |
203 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
204 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 205 | |
81819f0f | 206 | #else |
0c710013 CL |
207 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
208 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
209 | { return 0; } | |
151c602f CL |
210 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
211 | { | |
84c1cf62 | 212 | kfree(s->name); |
151c602f CL |
213 | kfree(s); |
214 | } | |
8ff12cfc | 215 | |
81819f0f CL |
216 | #endif |
217 | ||
84e554e6 | 218 | static inline void stat(struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
219 | { |
220 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 221 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
222 | #endif |
223 | } | |
224 | ||
81819f0f CL |
225 | /******************************************************************** |
226 | * Core slab cache functions | |
227 | *******************************************************************/ | |
228 | ||
229 | int slab_is_available(void) | |
230 | { | |
231 | return slab_state >= UP; | |
232 | } | |
233 | ||
234 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
235 | { | |
81819f0f | 236 | return s->node[node]; |
81819f0f CL |
237 | } |
238 | ||
6446faa2 | 239 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
240 | static inline int check_valid_pointer(struct kmem_cache *s, |
241 | struct page *page, const void *object) | |
242 | { | |
243 | void *base; | |
244 | ||
a973e9dd | 245 | if (!object) |
02cbc874 CL |
246 | return 1; |
247 | ||
a973e9dd | 248 | base = page_address(page); |
39b26464 | 249 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
250 | (object - base) % s->size) { |
251 | return 0; | |
252 | } | |
253 | ||
254 | return 1; | |
255 | } | |
256 | ||
7656c72b CL |
257 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
258 | { | |
259 | return *(void **)(object + s->offset); | |
260 | } | |
261 | ||
262 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
263 | { | |
264 | *(void **)(object + s->offset) = fp; | |
265 | } | |
266 | ||
267 | /* Loop over all objects in a slab */ | |
224a88be CL |
268 | #define for_each_object(__p, __s, __addr, __objects) \ |
269 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
270 | __p += (__s)->size) |
271 | ||
272 | /* Scan freelist */ | |
273 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 274 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
275 | |
276 | /* Determine object index from a given position */ | |
277 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
278 | { | |
279 | return (p - addr) / s->size; | |
280 | } | |
281 | ||
834f3d11 CL |
282 | static inline struct kmem_cache_order_objects oo_make(int order, |
283 | unsigned long size) | |
284 | { | |
285 | struct kmem_cache_order_objects x = { | |
210b5c06 | 286 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size |
834f3d11 CL |
287 | }; |
288 | ||
289 | return x; | |
290 | } | |
291 | ||
292 | static inline int oo_order(struct kmem_cache_order_objects x) | |
293 | { | |
210b5c06 | 294 | return x.x >> OO_SHIFT; |
834f3d11 CL |
295 | } |
296 | ||
297 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
298 | { | |
210b5c06 | 299 | return x.x & OO_MASK; |
834f3d11 CL |
300 | } |
301 | ||
41ecc55b CL |
302 | #ifdef CONFIG_SLUB_DEBUG |
303 | /* | |
304 | * Debug settings: | |
305 | */ | |
f0630fff CL |
306 | #ifdef CONFIG_SLUB_DEBUG_ON |
307 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
308 | #else | |
41ecc55b | 309 | static int slub_debug; |
f0630fff | 310 | #endif |
41ecc55b CL |
311 | |
312 | static char *slub_debug_slabs; | |
fa5ec8a1 | 313 | static int disable_higher_order_debug; |
41ecc55b | 314 | |
81819f0f CL |
315 | /* |
316 | * Object debugging | |
317 | */ | |
318 | static void print_section(char *text, u8 *addr, unsigned int length) | |
319 | { | |
320 | int i, offset; | |
321 | int newline = 1; | |
322 | char ascii[17]; | |
323 | ||
324 | ascii[16] = 0; | |
325 | ||
326 | for (i = 0; i < length; i++) { | |
327 | if (newline) { | |
24922684 | 328 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
329 | newline = 0; |
330 | } | |
06428780 | 331 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
332 | offset = i % 16; |
333 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
334 | if (offset == 15) { | |
06428780 | 335 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
336 | newline = 1; |
337 | } | |
338 | } | |
339 | if (!newline) { | |
340 | i %= 16; | |
341 | while (i < 16) { | |
06428780 | 342 | printk(KERN_CONT " "); |
81819f0f CL |
343 | ascii[i] = ' '; |
344 | i++; | |
345 | } | |
06428780 | 346 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
347 | } |
348 | } | |
349 | ||
81819f0f CL |
350 | static struct track *get_track(struct kmem_cache *s, void *object, |
351 | enum track_item alloc) | |
352 | { | |
353 | struct track *p; | |
354 | ||
355 | if (s->offset) | |
356 | p = object + s->offset + sizeof(void *); | |
357 | else | |
358 | p = object + s->inuse; | |
359 | ||
360 | return p + alloc; | |
361 | } | |
362 | ||
363 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 364 | enum track_item alloc, unsigned long addr) |
81819f0f | 365 | { |
1a00df4a | 366 | struct track *p = get_track(s, object, alloc); |
81819f0f | 367 | |
81819f0f CL |
368 | if (addr) { |
369 | p->addr = addr; | |
370 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 371 | p->pid = current->pid; |
81819f0f CL |
372 | p->when = jiffies; |
373 | } else | |
374 | memset(p, 0, sizeof(struct track)); | |
375 | } | |
376 | ||
81819f0f CL |
377 | static void init_tracking(struct kmem_cache *s, void *object) |
378 | { | |
24922684 CL |
379 | if (!(s->flags & SLAB_STORE_USER)) |
380 | return; | |
381 | ||
ce71e27c EGM |
382 | set_track(s, object, TRACK_FREE, 0UL); |
383 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
384 | } |
385 | ||
386 | static void print_track(const char *s, struct track *t) | |
387 | { | |
388 | if (!t->addr) | |
389 | return; | |
390 | ||
7daf705f | 391 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 392 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
393 | } |
394 | ||
395 | static void print_tracking(struct kmem_cache *s, void *object) | |
396 | { | |
397 | if (!(s->flags & SLAB_STORE_USER)) | |
398 | return; | |
399 | ||
400 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
401 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
402 | } | |
403 | ||
404 | static void print_page_info(struct page *page) | |
405 | { | |
39b26464 CL |
406 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
407 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
408 | |
409 | } | |
410 | ||
411 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
412 | { | |
413 | va_list args; | |
414 | char buf[100]; | |
415 | ||
416 | va_start(args, fmt); | |
417 | vsnprintf(buf, sizeof(buf), fmt, args); | |
418 | va_end(args); | |
419 | printk(KERN_ERR "========================================" | |
420 | "=====================================\n"); | |
421 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
422 | printk(KERN_ERR "----------------------------------------" | |
423 | "-------------------------------------\n\n"); | |
81819f0f CL |
424 | } |
425 | ||
24922684 CL |
426 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
427 | { | |
428 | va_list args; | |
429 | char buf[100]; | |
430 | ||
431 | va_start(args, fmt); | |
432 | vsnprintf(buf, sizeof(buf), fmt, args); | |
433 | va_end(args); | |
434 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
435 | } | |
436 | ||
437 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
438 | { |
439 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 440 | u8 *addr = page_address(page); |
24922684 CL |
441 | |
442 | print_tracking(s, p); | |
443 | ||
444 | print_page_info(page); | |
445 | ||
446 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
447 | p, p - addr, get_freepointer(s, p)); | |
448 | ||
449 | if (p > addr + 16) | |
450 | print_section("Bytes b4", p - 16, 16); | |
451 | ||
0ebd652b | 452 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
453 | |
454 | if (s->flags & SLAB_RED_ZONE) | |
455 | print_section("Redzone", p + s->objsize, | |
456 | s->inuse - s->objsize); | |
457 | ||
81819f0f CL |
458 | if (s->offset) |
459 | off = s->offset + sizeof(void *); | |
460 | else | |
461 | off = s->inuse; | |
462 | ||
24922684 | 463 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 464 | off += 2 * sizeof(struct track); |
81819f0f CL |
465 | |
466 | if (off != s->size) | |
467 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
468 | print_section("Padding", p + off, s->size - off); |
469 | ||
470 | dump_stack(); | |
81819f0f CL |
471 | } |
472 | ||
473 | static void object_err(struct kmem_cache *s, struct page *page, | |
474 | u8 *object, char *reason) | |
475 | { | |
3dc50637 | 476 | slab_bug(s, "%s", reason); |
24922684 | 477 | print_trailer(s, page, object); |
81819f0f CL |
478 | } |
479 | ||
24922684 | 480 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
481 | { |
482 | va_list args; | |
483 | char buf[100]; | |
484 | ||
24922684 CL |
485 | va_start(args, fmt); |
486 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 487 | va_end(args); |
3dc50637 | 488 | slab_bug(s, "%s", buf); |
24922684 | 489 | print_page_info(page); |
81819f0f CL |
490 | dump_stack(); |
491 | } | |
492 | ||
f7cb1933 | 493 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
494 | { |
495 | u8 *p = object; | |
496 | ||
497 | if (s->flags & __OBJECT_POISON) { | |
498 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 499 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
500 | } |
501 | ||
502 | if (s->flags & SLAB_RED_ZONE) | |
f7cb1933 | 503 | memset(p + s->objsize, val, s->inuse - s->objsize); |
81819f0f CL |
504 | } |
505 | ||
24922684 | 506 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
507 | { |
508 | while (bytes) { | |
509 | if (*start != (u8)value) | |
24922684 | 510 | return start; |
81819f0f CL |
511 | start++; |
512 | bytes--; | |
513 | } | |
24922684 CL |
514 | return NULL; |
515 | } | |
516 | ||
517 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
518 | void *from, void *to) | |
519 | { | |
520 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
521 | memset(from, data, to - from); | |
522 | } | |
523 | ||
524 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
525 | u8 *object, char *what, | |
06428780 | 526 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
527 | { |
528 | u8 *fault; | |
529 | u8 *end; | |
530 | ||
531 | fault = check_bytes(start, value, bytes); | |
532 | if (!fault) | |
533 | return 1; | |
534 | ||
535 | end = start + bytes; | |
536 | while (end > fault && end[-1] == value) | |
537 | end--; | |
538 | ||
539 | slab_bug(s, "%s overwritten", what); | |
540 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
541 | fault, end - 1, fault[0], value); | |
542 | print_trailer(s, page, object); | |
543 | ||
544 | restore_bytes(s, what, value, fault, end); | |
545 | return 0; | |
81819f0f CL |
546 | } |
547 | ||
81819f0f CL |
548 | /* |
549 | * Object layout: | |
550 | * | |
551 | * object address | |
552 | * Bytes of the object to be managed. | |
553 | * If the freepointer may overlay the object then the free | |
554 | * pointer is the first word of the object. | |
672bba3a | 555 | * |
81819f0f CL |
556 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
557 | * 0xa5 (POISON_END) | |
558 | * | |
559 | * object + s->objsize | |
560 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
561 | * Padding is extended by another word if Redzoning is enabled and |
562 | * objsize == inuse. | |
563 | * | |
81819f0f CL |
564 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
565 | * 0xcc (RED_ACTIVE) for objects in use. | |
566 | * | |
567 | * object + s->inuse | |
672bba3a CL |
568 | * Meta data starts here. |
569 | * | |
81819f0f CL |
570 | * A. Free pointer (if we cannot overwrite object on free) |
571 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 572 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 573 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
574 | * before the word boundary. |
575 | * | |
576 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
577 | * |
578 | * object + s->size | |
672bba3a | 579 | * Nothing is used beyond s->size. |
81819f0f | 580 | * |
672bba3a CL |
581 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
582 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
583 | * may be used with merged slabcaches. |
584 | */ | |
585 | ||
81819f0f CL |
586 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
587 | { | |
588 | unsigned long off = s->inuse; /* The end of info */ | |
589 | ||
590 | if (s->offset) | |
591 | /* Freepointer is placed after the object. */ | |
592 | off += sizeof(void *); | |
593 | ||
594 | if (s->flags & SLAB_STORE_USER) | |
595 | /* We also have user information there */ | |
596 | off += 2 * sizeof(struct track); | |
597 | ||
598 | if (s->size == off) | |
599 | return 1; | |
600 | ||
24922684 CL |
601 | return check_bytes_and_report(s, page, p, "Object padding", |
602 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
603 | } |
604 | ||
39b26464 | 605 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
606 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
607 | { | |
24922684 CL |
608 | u8 *start; |
609 | u8 *fault; | |
610 | u8 *end; | |
611 | int length; | |
612 | int remainder; | |
81819f0f CL |
613 | |
614 | if (!(s->flags & SLAB_POISON)) | |
615 | return 1; | |
616 | ||
a973e9dd | 617 | start = page_address(page); |
834f3d11 | 618 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
619 | end = start + length; |
620 | remainder = length % s->size; | |
81819f0f CL |
621 | if (!remainder) |
622 | return 1; | |
623 | ||
39b26464 | 624 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
625 | if (!fault) |
626 | return 1; | |
627 | while (end > fault && end[-1] == POISON_INUSE) | |
628 | end--; | |
629 | ||
630 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 631 | print_section("Padding", end - remainder, remainder); |
24922684 | 632 | |
8a3d271d | 633 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 634 | return 0; |
81819f0f CL |
635 | } |
636 | ||
637 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 638 | void *object, u8 val) |
81819f0f CL |
639 | { |
640 | u8 *p = object; | |
641 | u8 *endobject = object + s->objsize; | |
642 | ||
643 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 644 | if (!check_bytes_and_report(s, page, object, "Redzone", |
f7cb1933 | 645 | endobject, val, s->inuse - s->objsize)) |
81819f0f | 646 | return 0; |
81819f0f | 647 | } else { |
3adbefee IM |
648 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
649 | check_bytes_and_report(s, page, p, "Alignment padding", | |
650 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
651 | } | |
81819f0f CL |
652 | } |
653 | ||
654 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 655 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 CL |
656 | (!check_bytes_and_report(s, page, p, "Poison", p, |
657 | POISON_FREE, s->objsize - 1) || | |
658 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 659 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 660 | return 0; |
81819f0f CL |
661 | /* |
662 | * check_pad_bytes cleans up on its own. | |
663 | */ | |
664 | check_pad_bytes(s, page, p); | |
665 | } | |
666 | ||
f7cb1933 | 667 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
668 | /* |
669 | * Object and freepointer overlap. Cannot check | |
670 | * freepointer while object is allocated. | |
671 | */ | |
672 | return 1; | |
673 | ||
674 | /* Check free pointer validity */ | |
675 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
676 | object_err(s, page, p, "Freepointer corrupt"); | |
677 | /* | |
9f6c708e | 678 | * No choice but to zap it and thus lose the remainder |
81819f0f | 679 | * of the free objects in this slab. May cause |
672bba3a | 680 | * another error because the object count is now wrong. |
81819f0f | 681 | */ |
a973e9dd | 682 | set_freepointer(s, p, NULL); |
81819f0f CL |
683 | return 0; |
684 | } | |
685 | return 1; | |
686 | } | |
687 | ||
688 | static int check_slab(struct kmem_cache *s, struct page *page) | |
689 | { | |
39b26464 CL |
690 | int maxobj; |
691 | ||
81819f0f CL |
692 | VM_BUG_ON(!irqs_disabled()); |
693 | ||
694 | if (!PageSlab(page)) { | |
24922684 | 695 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
696 | return 0; |
697 | } | |
39b26464 CL |
698 | |
699 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
700 | if (page->objects > maxobj) { | |
701 | slab_err(s, page, "objects %u > max %u", | |
702 | s->name, page->objects, maxobj); | |
703 | return 0; | |
704 | } | |
705 | if (page->inuse > page->objects) { | |
24922684 | 706 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 707 | s->name, page->inuse, page->objects); |
81819f0f CL |
708 | return 0; |
709 | } | |
710 | /* Slab_pad_check fixes things up after itself */ | |
711 | slab_pad_check(s, page); | |
712 | return 1; | |
713 | } | |
714 | ||
715 | /* | |
672bba3a CL |
716 | * Determine if a certain object on a page is on the freelist. Must hold the |
717 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
718 | */ |
719 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
720 | { | |
721 | int nr = 0; | |
722 | void *fp = page->freelist; | |
723 | void *object = NULL; | |
224a88be | 724 | unsigned long max_objects; |
81819f0f | 725 | |
39b26464 | 726 | while (fp && nr <= page->objects) { |
81819f0f CL |
727 | if (fp == search) |
728 | return 1; | |
729 | if (!check_valid_pointer(s, page, fp)) { | |
730 | if (object) { | |
731 | object_err(s, page, object, | |
732 | "Freechain corrupt"); | |
a973e9dd | 733 | set_freepointer(s, object, NULL); |
81819f0f CL |
734 | break; |
735 | } else { | |
24922684 | 736 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 737 | page->freelist = NULL; |
39b26464 | 738 | page->inuse = page->objects; |
24922684 | 739 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
740 | return 0; |
741 | } | |
742 | break; | |
743 | } | |
744 | object = fp; | |
745 | fp = get_freepointer(s, object); | |
746 | nr++; | |
747 | } | |
748 | ||
224a88be | 749 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
210b5c06 CG |
750 | if (max_objects > MAX_OBJS_PER_PAGE) |
751 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
752 | |
753 | if (page->objects != max_objects) { | |
754 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
755 | "should be %d", page->objects, max_objects); | |
756 | page->objects = max_objects; | |
757 | slab_fix(s, "Number of objects adjusted."); | |
758 | } | |
39b26464 | 759 | if (page->inuse != page->objects - nr) { |
70d71228 | 760 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
761 | "counted were %d", page->inuse, page->objects - nr); |
762 | page->inuse = page->objects - nr; | |
24922684 | 763 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
764 | } |
765 | return search == NULL; | |
766 | } | |
767 | ||
0121c619 CL |
768 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
769 | int alloc) | |
3ec09742 CL |
770 | { |
771 | if (s->flags & SLAB_TRACE) { | |
772 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
773 | s->name, | |
774 | alloc ? "alloc" : "free", | |
775 | object, page->inuse, | |
776 | page->freelist); | |
777 | ||
778 | if (!alloc) | |
779 | print_section("Object", (void *)object, s->objsize); | |
780 | ||
781 | dump_stack(); | |
782 | } | |
783 | } | |
784 | ||
c016b0bd CL |
785 | /* |
786 | * Hooks for other subsystems that check memory allocations. In a typical | |
787 | * production configuration these hooks all should produce no code at all. | |
788 | */ | |
789 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
790 | { | |
c1d50836 | 791 | flags &= gfp_allowed_mask; |
c016b0bd CL |
792 | lockdep_trace_alloc(flags); |
793 | might_sleep_if(flags & __GFP_WAIT); | |
794 | ||
795 | return should_failslab(s->objsize, flags, s->flags); | |
796 | } | |
797 | ||
798 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
799 | { | |
c1d50836 | 800 | flags &= gfp_allowed_mask; |
c016b0bd CL |
801 | kmemcheck_slab_alloc(s, flags, object, s->objsize); |
802 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); | |
803 | } | |
804 | ||
805 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
806 | { | |
807 | kmemleak_free_recursive(x, s->flags); | |
808 | } | |
809 | ||
810 | static inline void slab_free_hook_irq(struct kmem_cache *s, void *object) | |
811 | { | |
812 | kmemcheck_slab_free(s, object, s->objsize); | |
813 | debug_check_no_locks_freed(object, s->objsize); | |
814 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
815 | debug_check_no_obj_freed(object, s->objsize); | |
816 | } | |
817 | ||
643b1138 | 818 | /* |
672bba3a | 819 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 820 | */ |
e95eed57 | 821 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 822 | { |
643b1138 CL |
823 | spin_lock(&n->list_lock); |
824 | list_add(&page->lru, &n->full); | |
825 | spin_unlock(&n->list_lock); | |
826 | } | |
827 | ||
828 | static void remove_full(struct kmem_cache *s, struct page *page) | |
829 | { | |
830 | struct kmem_cache_node *n; | |
831 | ||
832 | if (!(s->flags & SLAB_STORE_USER)) | |
833 | return; | |
834 | ||
835 | n = get_node(s, page_to_nid(page)); | |
836 | ||
837 | spin_lock(&n->list_lock); | |
838 | list_del(&page->lru); | |
839 | spin_unlock(&n->list_lock); | |
840 | } | |
841 | ||
0f389ec6 CL |
842 | /* Tracking of the number of slabs for debugging purposes */ |
843 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
844 | { | |
845 | struct kmem_cache_node *n = get_node(s, node); | |
846 | ||
847 | return atomic_long_read(&n->nr_slabs); | |
848 | } | |
849 | ||
26c02cf0 AB |
850 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
851 | { | |
852 | return atomic_long_read(&n->nr_slabs); | |
853 | } | |
854 | ||
205ab99d | 855 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
856 | { |
857 | struct kmem_cache_node *n = get_node(s, node); | |
858 | ||
859 | /* | |
860 | * May be called early in order to allocate a slab for the | |
861 | * kmem_cache_node structure. Solve the chicken-egg | |
862 | * dilemma by deferring the increment of the count during | |
863 | * bootstrap (see early_kmem_cache_node_alloc). | |
864 | */ | |
7340cc84 | 865 | if (n) { |
0f389ec6 | 866 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
867 | atomic_long_add(objects, &n->total_objects); |
868 | } | |
0f389ec6 | 869 | } |
205ab99d | 870 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
871 | { |
872 | struct kmem_cache_node *n = get_node(s, node); | |
873 | ||
874 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 875 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
876 | } |
877 | ||
878 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
879 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
880 | void *object) | |
881 | { | |
882 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
883 | return; | |
884 | ||
f7cb1933 | 885 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
886 | init_tracking(s, object); |
887 | } | |
888 | ||
1537066c | 889 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 890 | void *object, unsigned long addr) |
81819f0f CL |
891 | { |
892 | if (!check_slab(s, page)) | |
893 | goto bad; | |
894 | ||
d692ef6d | 895 | if (!on_freelist(s, page, object)) { |
24922684 | 896 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 897 | goto bad; |
81819f0f CL |
898 | } |
899 | ||
900 | if (!check_valid_pointer(s, page, object)) { | |
901 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 902 | goto bad; |
81819f0f CL |
903 | } |
904 | ||
f7cb1933 | 905 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 906 | goto bad; |
81819f0f | 907 | |
3ec09742 CL |
908 | /* Success perform special debug activities for allocs */ |
909 | if (s->flags & SLAB_STORE_USER) | |
910 | set_track(s, object, TRACK_ALLOC, addr); | |
911 | trace(s, page, object, 1); | |
f7cb1933 | 912 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 913 | return 1; |
3ec09742 | 914 | |
81819f0f CL |
915 | bad: |
916 | if (PageSlab(page)) { | |
917 | /* | |
918 | * If this is a slab page then lets do the best we can | |
919 | * to avoid issues in the future. Marking all objects | |
672bba3a | 920 | * as used avoids touching the remaining objects. |
81819f0f | 921 | */ |
24922684 | 922 | slab_fix(s, "Marking all objects used"); |
39b26464 | 923 | page->inuse = page->objects; |
a973e9dd | 924 | page->freelist = NULL; |
81819f0f CL |
925 | } |
926 | return 0; | |
927 | } | |
928 | ||
1537066c CL |
929 | static noinline int free_debug_processing(struct kmem_cache *s, |
930 | struct page *page, void *object, unsigned long addr) | |
81819f0f CL |
931 | { |
932 | if (!check_slab(s, page)) | |
933 | goto fail; | |
934 | ||
935 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 936 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
937 | goto fail; |
938 | } | |
939 | ||
940 | if (on_freelist(s, page, object)) { | |
24922684 | 941 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
942 | goto fail; |
943 | } | |
944 | ||
f7cb1933 | 945 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
81819f0f CL |
946 | return 0; |
947 | ||
948 | if (unlikely(s != page->slab)) { | |
3adbefee | 949 | if (!PageSlab(page)) { |
70d71228 CL |
950 | slab_err(s, page, "Attempt to free object(0x%p) " |
951 | "outside of slab", object); | |
3adbefee | 952 | } else if (!page->slab) { |
81819f0f | 953 | printk(KERN_ERR |
70d71228 | 954 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 955 | object); |
70d71228 | 956 | dump_stack(); |
06428780 | 957 | } else |
24922684 CL |
958 | object_err(s, page, object, |
959 | "page slab pointer corrupt."); | |
81819f0f CL |
960 | goto fail; |
961 | } | |
3ec09742 CL |
962 | |
963 | /* Special debug activities for freeing objects */ | |
8a38082d | 964 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
965 | remove_full(s, page); |
966 | if (s->flags & SLAB_STORE_USER) | |
967 | set_track(s, object, TRACK_FREE, addr); | |
968 | trace(s, page, object, 0); | |
f7cb1933 | 969 | init_object(s, object, SLUB_RED_INACTIVE); |
81819f0f | 970 | return 1; |
3ec09742 | 971 | |
81819f0f | 972 | fail: |
24922684 | 973 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
974 | return 0; |
975 | } | |
976 | ||
41ecc55b CL |
977 | static int __init setup_slub_debug(char *str) |
978 | { | |
f0630fff CL |
979 | slub_debug = DEBUG_DEFAULT_FLAGS; |
980 | if (*str++ != '=' || !*str) | |
981 | /* | |
982 | * No options specified. Switch on full debugging. | |
983 | */ | |
984 | goto out; | |
985 | ||
986 | if (*str == ',') | |
987 | /* | |
988 | * No options but restriction on slabs. This means full | |
989 | * debugging for slabs matching a pattern. | |
990 | */ | |
991 | goto check_slabs; | |
992 | ||
fa5ec8a1 DR |
993 | if (tolower(*str) == 'o') { |
994 | /* | |
995 | * Avoid enabling debugging on caches if its minimum order | |
996 | * would increase as a result. | |
997 | */ | |
998 | disable_higher_order_debug = 1; | |
999 | goto out; | |
1000 | } | |
1001 | ||
f0630fff CL |
1002 | slub_debug = 0; |
1003 | if (*str == '-') | |
1004 | /* | |
1005 | * Switch off all debugging measures. | |
1006 | */ | |
1007 | goto out; | |
1008 | ||
1009 | /* | |
1010 | * Determine which debug features should be switched on | |
1011 | */ | |
06428780 | 1012 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1013 | switch (tolower(*str)) { |
1014 | case 'f': | |
1015 | slub_debug |= SLAB_DEBUG_FREE; | |
1016 | break; | |
1017 | case 'z': | |
1018 | slub_debug |= SLAB_RED_ZONE; | |
1019 | break; | |
1020 | case 'p': | |
1021 | slub_debug |= SLAB_POISON; | |
1022 | break; | |
1023 | case 'u': | |
1024 | slub_debug |= SLAB_STORE_USER; | |
1025 | break; | |
1026 | case 't': | |
1027 | slub_debug |= SLAB_TRACE; | |
1028 | break; | |
4c13dd3b DM |
1029 | case 'a': |
1030 | slub_debug |= SLAB_FAILSLAB; | |
1031 | break; | |
f0630fff CL |
1032 | default: |
1033 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1034 | "unknown. skipped\n", *str); |
f0630fff | 1035 | } |
41ecc55b CL |
1036 | } |
1037 | ||
f0630fff | 1038 | check_slabs: |
41ecc55b CL |
1039 | if (*str == ',') |
1040 | slub_debug_slabs = str + 1; | |
f0630fff | 1041 | out: |
41ecc55b CL |
1042 | return 1; |
1043 | } | |
1044 | ||
1045 | __setup("slub_debug", setup_slub_debug); | |
1046 | ||
ba0268a8 CL |
1047 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1048 | unsigned long flags, const char *name, | |
51cc5068 | 1049 | void (*ctor)(void *)) |
41ecc55b CL |
1050 | { |
1051 | /* | |
e153362a | 1052 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1053 | */ |
e153362a | 1054 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1055 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1056 | flags |= slub_debug; | |
ba0268a8 CL |
1057 | |
1058 | return flags; | |
41ecc55b CL |
1059 | } |
1060 | #else | |
3ec09742 CL |
1061 | static inline void setup_object_debug(struct kmem_cache *s, |
1062 | struct page *page, void *object) {} | |
41ecc55b | 1063 | |
3ec09742 | 1064 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1065 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1066 | |
3ec09742 | 1067 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1068 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1069 | |
41ecc55b CL |
1070 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1071 | { return 1; } | |
1072 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1073 | void *object, u8 val) { return 1; } |
3ec09742 | 1074 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1075 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1076 | unsigned long flags, const char *name, | |
51cc5068 | 1077 | void (*ctor)(void *)) |
ba0268a8 CL |
1078 | { |
1079 | return flags; | |
1080 | } | |
41ecc55b | 1081 | #define slub_debug 0 |
0f389ec6 | 1082 | |
fdaa45e9 IM |
1083 | #define disable_higher_order_debug 0 |
1084 | ||
0f389ec6 CL |
1085 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1086 | { return 0; } | |
26c02cf0 AB |
1087 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1088 | { return 0; } | |
205ab99d CL |
1089 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1090 | int objects) {} | |
1091 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1092 | int objects) {} | |
7d550c56 CL |
1093 | |
1094 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1095 | { return 0; } | |
1096 | ||
1097 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1098 | void *object) {} | |
1099 | ||
1100 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1101 | ||
1102 | static inline void slab_free_hook_irq(struct kmem_cache *s, | |
1103 | void *object) {} | |
1104 | ||
ab4d5ed5 | 1105 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1106 | |
81819f0f CL |
1107 | /* |
1108 | * Slab allocation and freeing | |
1109 | */ | |
65c3376a CL |
1110 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1111 | struct kmem_cache_order_objects oo) | |
1112 | { | |
1113 | int order = oo_order(oo); | |
1114 | ||
b1eeab67 VN |
1115 | flags |= __GFP_NOTRACK; |
1116 | ||
2154a336 | 1117 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1118 | return alloc_pages(flags, order); |
1119 | else | |
6b65aaf3 | 1120 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1121 | } |
1122 | ||
81819f0f CL |
1123 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1124 | { | |
06428780 | 1125 | struct page *page; |
834f3d11 | 1126 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1127 | gfp_t alloc_gfp; |
81819f0f | 1128 | |
b7a49f0d | 1129 | flags |= s->allocflags; |
e12ba74d | 1130 | |
ba52270d PE |
1131 | /* |
1132 | * Let the initial higher-order allocation fail under memory pressure | |
1133 | * so we fall-back to the minimum order allocation. | |
1134 | */ | |
1135 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1136 | ||
1137 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1138 | if (unlikely(!page)) { |
1139 | oo = s->min; | |
1140 | /* | |
1141 | * Allocation may have failed due to fragmentation. | |
1142 | * Try a lower order alloc if possible | |
1143 | */ | |
1144 | page = alloc_slab_page(flags, node, oo); | |
1145 | if (!page) | |
1146 | return NULL; | |
81819f0f | 1147 | |
84e554e6 | 1148 | stat(s, ORDER_FALLBACK); |
65c3376a | 1149 | } |
5a896d9e VN |
1150 | |
1151 | if (kmemcheck_enabled | |
5086c389 | 1152 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1153 | int pages = 1 << oo_order(oo); |
1154 | ||
1155 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1156 | ||
1157 | /* | |
1158 | * Objects from caches that have a constructor don't get | |
1159 | * cleared when they're allocated, so we need to do it here. | |
1160 | */ | |
1161 | if (s->ctor) | |
1162 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1163 | else | |
1164 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1165 | } |
1166 | ||
834f3d11 | 1167 | page->objects = oo_objects(oo); |
81819f0f CL |
1168 | mod_zone_page_state(page_zone(page), |
1169 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1170 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1171 | 1 << oo_order(oo)); |
81819f0f CL |
1172 | |
1173 | return page; | |
1174 | } | |
1175 | ||
1176 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1177 | void *object) | |
1178 | { | |
3ec09742 | 1179 | setup_object_debug(s, page, object); |
4f104934 | 1180 | if (unlikely(s->ctor)) |
51cc5068 | 1181 | s->ctor(object); |
81819f0f CL |
1182 | } |
1183 | ||
1184 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1185 | { | |
1186 | struct page *page; | |
81819f0f | 1187 | void *start; |
81819f0f CL |
1188 | void *last; |
1189 | void *p; | |
1190 | ||
6cb06229 | 1191 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1192 | |
6cb06229 CL |
1193 | page = allocate_slab(s, |
1194 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1195 | if (!page) |
1196 | goto out; | |
1197 | ||
205ab99d | 1198 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1199 | page->slab = s; |
1200 | page->flags |= 1 << PG_slab; | |
81819f0f CL |
1201 | |
1202 | start = page_address(page); | |
81819f0f CL |
1203 | |
1204 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1205 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1206 | |
1207 | last = start; | |
224a88be | 1208 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1209 | setup_object(s, page, last); |
1210 | set_freepointer(s, last, p); | |
1211 | last = p; | |
1212 | } | |
1213 | setup_object(s, page, last); | |
a973e9dd | 1214 | set_freepointer(s, last, NULL); |
81819f0f CL |
1215 | |
1216 | page->freelist = start; | |
1217 | page->inuse = 0; | |
1218 | out: | |
81819f0f CL |
1219 | return page; |
1220 | } | |
1221 | ||
1222 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1223 | { | |
834f3d11 CL |
1224 | int order = compound_order(page); |
1225 | int pages = 1 << order; | |
81819f0f | 1226 | |
af537b0a | 1227 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1228 | void *p; |
1229 | ||
1230 | slab_pad_check(s, page); | |
224a88be CL |
1231 | for_each_object(p, s, page_address(page), |
1232 | page->objects) | |
f7cb1933 | 1233 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1234 | } |
1235 | ||
b1eeab67 | 1236 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1237 | |
81819f0f CL |
1238 | mod_zone_page_state(page_zone(page), |
1239 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1240 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1241 | -pages); |
81819f0f | 1242 | |
49bd5221 CL |
1243 | __ClearPageSlab(page); |
1244 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1245 | if (current->reclaim_state) |
1246 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1247 | __free_pages(page, order); |
81819f0f CL |
1248 | } |
1249 | ||
1250 | static void rcu_free_slab(struct rcu_head *h) | |
1251 | { | |
1252 | struct page *page; | |
1253 | ||
1254 | page = container_of((struct list_head *)h, struct page, lru); | |
1255 | __free_slab(page->slab, page); | |
1256 | } | |
1257 | ||
1258 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1259 | { | |
1260 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1261 | /* | |
1262 | * RCU free overloads the RCU head over the LRU | |
1263 | */ | |
1264 | struct rcu_head *head = (void *)&page->lru; | |
1265 | ||
1266 | call_rcu(head, rcu_free_slab); | |
1267 | } else | |
1268 | __free_slab(s, page); | |
1269 | } | |
1270 | ||
1271 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1272 | { | |
205ab99d | 1273 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1274 | free_slab(s, page); |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Per slab locking using the pagelock | |
1279 | */ | |
1280 | static __always_inline void slab_lock(struct page *page) | |
1281 | { | |
1282 | bit_spin_lock(PG_locked, &page->flags); | |
1283 | } | |
1284 | ||
1285 | static __always_inline void slab_unlock(struct page *page) | |
1286 | { | |
a76d3546 | 1287 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1288 | } |
1289 | ||
1290 | static __always_inline int slab_trylock(struct page *page) | |
1291 | { | |
1292 | int rc = 1; | |
1293 | ||
1294 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1295 | return rc; | |
1296 | } | |
1297 | ||
1298 | /* | |
1299 | * Management of partially allocated slabs | |
1300 | */ | |
7c2e132c CL |
1301 | static void add_partial(struct kmem_cache_node *n, |
1302 | struct page *page, int tail) | |
81819f0f | 1303 | { |
e95eed57 CL |
1304 | spin_lock(&n->list_lock); |
1305 | n->nr_partial++; | |
7c2e132c CL |
1306 | if (tail) |
1307 | list_add_tail(&page->lru, &n->partial); | |
1308 | else | |
1309 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1310 | spin_unlock(&n->list_lock); |
1311 | } | |
1312 | ||
62e346a8 CL |
1313 | static inline void __remove_partial(struct kmem_cache_node *n, |
1314 | struct page *page) | |
1315 | { | |
1316 | list_del(&page->lru); | |
1317 | n->nr_partial--; | |
1318 | } | |
1319 | ||
0121c619 | 1320 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1321 | { |
1322 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1323 | ||
1324 | spin_lock(&n->list_lock); | |
62e346a8 | 1325 | __remove_partial(n, page); |
81819f0f CL |
1326 | spin_unlock(&n->list_lock); |
1327 | } | |
1328 | ||
1329 | /* | |
672bba3a | 1330 | * Lock slab and remove from the partial list. |
81819f0f | 1331 | * |
672bba3a | 1332 | * Must hold list_lock. |
81819f0f | 1333 | */ |
0121c619 CL |
1334 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1335 | struct page *page) | |
81819f0f CL |
1336 | { |
1337 | if (slab_trylock(page)) { | |
62e346a8 | 1338 | __remove_partial(n, page); |
8a38082d | 1339 | __SetPageSlubFrozen(page); |
81819f0f CL |
1340 | return 1; |
1341 | } | |
1342 | return 0; | |
1343 | } | |
1344 | ||
1345 | /* | |
672bba3a | 1346 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1347 | */ |
1348 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1349 | { | |
1350 | struct page *page; | |
1351 | ||
1352 | /* | |
1353 | * Racy check. If we mistakenly see no partial slabs then we | |
1354 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1355 | * partial slab and there is none available then get_partials() |
1356 | * will return NULL. | |
81819f0f CL |
1357 | */ |
1358 | if (!n || !n->nr_partial) | |
1359 | return NULL; | |
1360 | ||
1361 | spin_lock(&n->list_lock); | |
1362 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1363 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1364 | goto out; |
1365 | page = NULL; | |
1366 | out: | |
1367 | spin_unlock(&n->list_lock); | |
1368 | return page; | |
1369 | } | |
1370 | ||
1371 | /* | |
672bba3a | 1372 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1373 | */ |
1374 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1375 | { | |
1376 | #ifdef CONFIG_NUMA | |
1377 | struct zonelist *zonelist; | |
dd1a239f | 1378 | struct zoneref *z; |
54a6eb5c MG |
1379 | struct zone *zone; |
1380 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1381 | struct page *page; |
1382 | ||
1383 | /* | |
672bba3a CL |
1384 | * The defrag ratio allows a configuration of the tradeoffs between |
1385 | * inter node defragmentation and node local allocations. A lower | |
1386 | * defrag_ratio increases the tendency to do local allocations | |
1387 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1388 | * |
672bba3a CL |
1389 | * If the defrag_ratio is set to 0 then kmalloc() always |
1390 | * returns node local objects. If the ratio is higher then kmalloc() | |
1391 | * may return off node objects because partial slabs are obtained | |
1392 | * from other nodes and filled up. | |
81819f0f | 1393 | * |
6446faa2 | 1394 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1395 | * defrag_ratio = 1000) then every (well almost) allocation will |
1396 | * first attempt to defrag slab caches on other nodes. This means | |
1397 | * scanning over all nodes to look for partial slabs which may be | |
1398 | * expensive if we do it every time we are trying to find a slab | |
1399 | * with available objects. | |
81819f0f | 1400 | */ |
9824601e CL |
1401 | if (!s->remote_node_defrag_ratio || |
1402 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1403 | return NULL; |
1404 | ||
c0ff7453 | 1405 | get_mems_allowed(); |
0e88460d | 1406 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1407 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1408 | struct kmem_cache_node *n; |
1409 | ||
54a6eb5c | 1410 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1411 | |
54a6eb5c | 1412 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1413 | n->nr_partial > s->min_partial) { |
81819f0f | 1414 | page = get_partial_node(n); |
c0ff7453 MX |
1415 | if (page) { |
1416 | put_mems_allowed(); | |
81819f0f | 1417 | return page; |
c0ff7453 | 1418 | } |
81819f0f CL |
1419 | } |
1420 | } | |
c0ff7453 | 1421 | put_mems_allowed(); |
81819f0f CL |
1422 | #endif |
1423 | return NULL; | |
1424 | } | |
1425 | ||
1426 | /* | |
1427 | * Get a partial page, lock it and return it. | |
1428 | */ | |
1429 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1430 | { | |
1431 | struct page *page; | |
2154a336 | 1432 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f CL |
1433 | |
1434 | page = get_partial_node(get_node(s, searchnode)); | |
bc6488e9 | 1435 | if (page || node != -1) |
81819f0f CL |
1436 | return page; |
1437 | ||
1438 | return get_any_partial(s, flags); | |
1439 | } | |
1440 | ||
1441 | /* | |
1442 | * Move a page back to the lists. | |
1443 | * | |
1444 | * Must be called with the slab lock held. | |
1445 | * | |
1446 | * On exit the slab lock will have been dropped. | |
1447 | */ | |
7c2e132c | 1448 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
3478973d | 1449 | __releases(bitlock) |
81819f0f | 1450 | { |
e95eed57 CL |
1451 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1452 | ||
8a38082d | 1453 | __ClearPageSlubFrozen(page); |
81819f0f | 1454 | if (page->inuse) { |
e95eed57 | 1455 | |
a973e9dd | 1456 | if (page->freelist) { |
7c2e132c | 1457 | add_partial(n, page, tail); |
84e554e6 | 1458 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
8ff12cfc | 1459 | } else { |
84e554e6 | 1460 | stat(s, DEACTIVATE_FULL); |
af537b0a | 1461 | if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER)) |
8ff12cfc CL |
1462 | add_full(n, page); |
1463 | } | |
81819f0f CL |
1464 | slab_unlock(page); |
1465 | } else { | |
84e554e6 | 1466 | stat(s, DEACTIVATE_EMPTY); |
3b89d7d8 | 1467 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1468 | /* |
672bba3a CL |
1469 | * Adding an empty slab to the partial slabs in order |
1470 | * to avoid page allocator overhead. This slab needs | |
1471 | * to come after the other slabs with objects in | |
6446faa2 CL |
1472 | * so that the others get filled first. That way the |
1473 | * size of the partial list stays small. | |
1474 | * | |
0121c619 CL |
1475 | * kmem_cache_shrink can reclaim any empty slabs from |
1476 | * the partial list. | |
e95eed57 | 1477 | */ |
7c2e132c | 1478 | add_partial(n, page, 1); |
e95eed57 CL |
1479 | slab_unlock(page); |
1480 | } else { | |
1481 | slab_unlock(page); | |
84e554e6 | 1482 | stat(s, FREE_SLAB); |
e95eed57 CL |
1483 | discard_slab(s, page); |
1484 | } | |
81819f0f CL |
1485 | } |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * Remove the cpu slab | |
1490 | */ | |
dfb4f096 | 1491 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
3478973d | 1492 | __releases(bitlock) |
81819f0f | 1493 | { |
dfb4f096 | 1494 | struct page *page = c->page; |
7c2e132c | 1495 | int tail = 1; |
8ff12cfc | 1496 | |
b773ad73 | 1497 | if (page->freelist) |
84e554e6 | 1498 | stat(s, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1499 | /* |
6446faa2 | 1500 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1501 | * because both freelists are empty. So this is unlikely |
1502 | * to occur. | |
1503 | */ | |
a973e9dd | 1504 | while (unlikely(c->freelist)) { |
894b8788 CL |
1505 | void **object; |
1506 | ||
7c2e132c CL |
1507 | tail = 0; /* Hot objects. Put the slab first */ |
1508 | ||
894b8788 | 1509 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1510 | object = c->freelist; |
ff12059e | 1511 | c->freelist = get_freepointer(s, c->freelist); |
894b8788 CL |
1512 | |
1513 | /* And put onto the regular freelist */ | |
ff12059e | 1514 | set_freepointer(s, object, page->freelist); |
894b8788 CL |
1515 | page->freelist = object; |
1516 | page->inuse--; | |
1517 | } | |
dfb4f096 | 1518 | c->page = NULL; |
7c2e132c | 1519 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1520 | } |
1521 | ||
dfb4f096 | 1522 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1523 | { |
84e554e6 | 1524 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 CL |
1525 | slab_lock(c->page); |
1526 | deactivate_slab(s, c); | |
81819f0f CL |
1527 | } |
1528 | ||
1529 | /* | |
1530 | * Flush cpu slab. | |
6446faa2 | 1531 | * |
81819f0f CL |
1532 | * Called from IPI handler with interrupts disabled. |
1533 | */ | |
0c710013 | 1534 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1535 | { |
9dfc6e68 | 1536 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1537 | |
dfb4f096 CL |
1538 | if (likely(c && c->page)) |
1539 | flush_slab(s, c); | |
81819f0f CL |
1540 | } |
1541 | ||
1542 | static void flush_cpu_slab(void *d) | |
1543 | { | |
1544 | struct kmem_cache *s = d; | |
81819f0f | 1545 | |
dfb4f096 | 1546 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1547 | } |
1548 | ||
1549 | static void flush_all(struct kmem_cache *s) | |
1550 | { | |
15c8b6c1 | 1551 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1552 | } |
1553 | ||
dfb4f096 CL |
1554 | /* |
1555 | * Check if the objects in a per cpu structure fit numa | |
1556 | * locality expectations. | |
1557 | */ | |
1558 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1559 | { | |
1560 | #ifdef CONFIG_NUMA | |
2154a336 | 1561 | if (node != NUMA_NO_NODE && c->node != node) |
dfb4f096 CL |
1562 | return 0; |
1563 | #endif | |
1564 | return 1; | |
1565 | } | |
1566 | ||
781b2ba6 PE |
1567 | static int count_free(struct page *page) |
1568 | { | |
1569 | return page->objects - page->inuse; | |
1570 | } | |
1571 | ||
1572 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1573 | int (*get_count)(struct page *)) | |
1574 | { | |
1575 | unsigned long flags; | |
1576 | unsigned long x = 0; | |
1577 | struct page *page; | |
1578 | ||
1579 | spin_lock_irqsave(&n->list_lock, flags); | |
1580 | list_for_each_entry(page, &n->partial, lru) | |
1581 | x += get_count(page); | |
1582 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1583 | return x; | |
1584 | } | |
1585 | ||
26c02cf0 AB |
1586 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1587 | { | |
1588 | #ifdef CONFIG_SLUB_DEBUG | |
1589 | return atomic_long_read(&n->total_objects); | |
1590 | #else | |
1591 | return 0; | |
1592 | #endif | |
1593 | } | |
1594 | ||
781b2ba6 PE |
1595 | static noinline void |
1596 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1597 | { | |
1598 | int node; | |
1599 | ||
1600 | printk(KERN_WARNING | |
1601 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1602 | nid, gfpflags); | |
1603 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1604 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1605 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1606 | ||
fa5ec8a1 DR |
1607 | if (oo_order(s->min) > get_order(s->objsize)) |
1608 | printk(KERN_WARNING " %s debugging increased min order, use " | |
1609 | "slub_debug=O to disable.\n", s->name); | |
1610 | ||
781b2ba6 PE |
1611 | for_each_online_node(node) { |
1612 | struct kmem_cache_node *n = get_node(s, node); | |
1613 | unsigned long nr_slabs; | |
1614 | unsigned long nr_objs; | |
1615 | unsigned long nr_free; | |
1616 | ||
1617 | if (!n) | |
1618 | continue; | |
1619 | ||
26c02cf0 AB |
1620 | nr_free = count_partial(n, count_free); |
1621 | nr_slabs = node_nr_slabs(n); | |
1622 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1623 | |
1624 | printk(KERN_WARNING | |
1625 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1626 | node, nr_slabs, nr_objs, nr_free); | |
1627 | } | |
1628 | } | |
1629 | ||
81819f0f | 1630 | /* |
894b8788 CL |
1631 | * Slow path. The lockless freelist is empty or we need to perform |
1632 | * debugging duties. | |
1633 | * | |
1634 | * Interrupts are disabled. | |
81819f0f | 1635 | * |
894b8788 CL |
1636 | * Processing is still very fast if new objects have been freed to the |
1637 | * regular freelist. In that case we simply take over the regular freelist | |
1638 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1639 | * |
894b8788 CL |
1640 | * If that is not working then we fall back to the partial lists. We take the |
1641 | * first element of the freelist as the object to allocate now and move the | |
1642 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1643 | * |
894b8788 | 1644 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1645 | * we need to allocate a new slab. This is the slowest path since it involves |
1646 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1647 | */ |
ce71e27c EGM |
1648 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1649 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1650 | { |
81819f0f | 1651 | void **object; |
dfb4f096 | 1652 | struct page *new; |
81819f0f | 1653 | |
e72e9c23 LT |
1654 | /* We handle __GFP_ZERO in the caller */ |
1655 | gfpflags &= ~__GFP_ZERO; | |
1656 | ||
dfb4f096 | 1657 | if (!c->page) |
81819f0f CL |
1658 | goto new_slab; |
1659 | ||
dfb4f096 CL |
1660 | slab_lock(c->page); |
1661 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1662 | goto another_slab; |
6446faa2 | 1663 | |
84e554e6 | 1664 | stat(s, ALLOC_REFILL); |
6446faa2 | 1665 | |
894b8788 | 1666 | load_freelist: |
dfb4f096 | 1667 | object = c->page->freelist; |
a973e9dd | 1668 | if (unlikely(!object)) |
81819f0f | 1669 | goto another_slab; |
af537b0a | 1670 | if (kmem_cache_debug(s)) |
81819f0f CL |
1671 | goto debug; |
1672 | ||
ff12059e | 1673 | c->freelist = get_freepointer(s, object); |
39b26464 | 1674 | c->page->inuse = c->page->objects; |
a973e9dd | 1675 | c->page->freelist = NULL; |
dfb4f096 | 1676 | c->node = page_to_nid(c->page); |
1f84260c | 1677 | unlock_out: |
dfb4f096 | 1678 | slab_unlock(c->page); |
84e554e6 | 1679 | stat(s, ALLOC_SLOWPATH); |
81819f0f CL |
1680 | return object; |
1681 | ||
1682 | another_slab: | |
dfb4f096 | 1683 | deactivate_slab(s, c); |
81819f0f CL |
1684 | |
1685 | new_slab: | |
dfb4f096 CL |
1686 | new = get_partial(s, gfpflags, node); |
1687 | if (new) { | |
1688 | c->page = new; | |
84e554e6 | 1689 | stat(s, ALLOC_FROM_PARTIAL); |
894b8788 | 1690 | goto load_freelist; |
81819f0f CL |
1691 | } |
1692 | ||
c1d50836 | 1693 | gfpflags &= gfp_allowed_mask; |
b811c202 CL |
1694 | if (gfpflags & __GFP_WAIT) |
1695 | local_irq_enable(); | |
1696 | ||
dfb4f096 | 1697 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1698 | |
1699 | if (gfpflags & __GFP_WAIT) | |
1700 | local_irq_disable(); | |
1701 | ||
dfb4f096 | 1702 | if (new) { |
9dfc6e68 | 1703 | c = __this_cpu_ptr(s->cpu_slab); |
84e554e6 | 1704 | stat(s, ALLOC_SLAB); |
05aa3450 | 1705 | if (c->page) |
dfb4f096 | 1706 | flush_slab(s, c); |
dfb4f096 | 1707 | slab_lock(new); |
8a38082d | 1708 | __SetPageSlubFrozen(new); |
dfb4f096 | 1709 | c->page = new; |
4b6f0750 | 1710 | goto load_freelist; |
81819f0f | 1711 | } |
95f85989 PE |
1712 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
1713 | slab_out_of_memory(s, gfpflags, node); | |
71c7a06f | 1714 | return NULL; |
81819f0f | 1715 | debug: |
dfb4f096 | 1716 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1717 | goto another_slab; |
894b8788 | 1718 | |
dfb4f096 | 1719 | c->page->inuse++; |
ff12059e | 1720 | c->page->freelist = get_freepointer(s, object); |
15b7c514 | 1721 | c->node = NUMA_NO_NODE; |
1f84260c | 1722 | goto unlock_out; |
894b8788 CL |
1723 | } |
1724 | ||
1725 | /* | |
1726 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1727 | * have the fastpath folded into their functions. So no function call | |
1728 | * overhead for requests that can be satisfied on the fastpath. | |
1729 | * | |
1730 | * The fastpath works by first checking if the lockless freelist can be used. | |
1731 | * If not then __slab_alloc is called for slow processing. | |
1732 | * | |
1733 | * Otherwise we can simply pick the next object from the lockless free list. | |
1734 | */ | |
06428780 | 1735 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1736 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1737 | { |
894b8788 | 1738 | void **object; |
dfb4f096 | 1739 | struct kmem_cache_cpu *c; |
1f84260c CL |
1740 | unsigned long flags; |
1741 | ||
c016b0bd | 1742 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 1743 | return NULL; |
1f84260c | 1744 | |
894b8788 | 1745 | local_irq_save(flags); |
9dfc6e68 CL |
1746 | c = __this_cpu_ptr(s->cpu_slab); |
1747 | object = c->freelist; | |
9dfc6e68 | 1748 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 1749 | |
dfb4f096 | 1750 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1751 | |
1752 | else { | |
ff12059e | 1753 | c->freelist = get_freepointer(s, object); |
84e554e6 | 1754 | stat(s, ALLOC_FASTPATH); |
894b8788 CL |
1755 | } |
1756 | local_irq_restore(flags); | |
d07dbea4 | 1757 | |
74e2134f | 1758 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 1759 | memset(object, 0, s->objsize); |
d07dbea4 | 1760 | |
c016b0bd | 1761 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 1762 | |
894b8788 | 1763 | return object; |
81819f0f CL |
1764 | } |
1765 | ||
1766 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1767 | { | |
2154a336 | 1768 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 1769 | |
ca2b84cb | 1770 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1771 | |
1772 | return ret; | |
81819f0f CL |
1773 | } |
1774 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1775 | ||
0f24f128 | 1776 | #ifdef CONFIG_TRACING |
5b882be4 EGM |
1777 | void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags) |
1778 | { | |
2154a336 | 1779 | return slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 EGM |
1780 | } |
1781 | EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |
1782 | #endif | |
1783 | ||
81819f0f CL |
1784 | #ifdef CONFIG_NUMA |
1785 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1786 | { | |
5b882be4 EGM |
1787 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1788 | ||
ca2b84cb EGM |
1789 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
1790 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
1791 | |
1792 | return ret; | |
81819f0f CL |
1793 | } |
1794 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 1795 | |
0f24f128 | 1796 | #ifdef CONFIG_TRACING |
5b882be4 EGM |
1797 | void *kmem_cache_alloc_node_notrace(struct kmem_cache *s, |
1798 | gfp_t gfpflags, | |
1799 | int node) | |
1800 | { | |
1801 | return slab_alloc(s, gfpflags, node, _RET_IP_); | |
1802 | } | |
1803 | EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); | |
1804 | #endif | |
5d1f57e4 | 1805 | #endif |
5b882be4 | 1806 | |
81819f0f | 1807 | /* |
894b8788 CL |
1808 | * Slow patch handling. This may still be called frequently since objects |
1809 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1810 | * |
894b8788 CL |
1811 | * So we still attempt to reduce cache line usage. Just take the slab |
1812 | * lock and free the item. If there is no additional partial page | |
1813 | * handling required then we can return immediately. | |
81819f0f | 1814 | */ |
894b8788 | 1815 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 1816 | void *x, unsigned long addr) |
81819f0f CL |
1817 | { |
1818 | void *prior; | |
1819 | void **object = (void *)x; | |
81819f0f | 1820 | |
84e554e6 | 1821 | stat(s, FREE_SLOWPATH); |
81819f0f CL |
1822 | slab_lock(page); |
1823 | ||
af537b0a | 1824 | if (kmem_cache_debug(s)) |
81819f0f | 1825 | goto debug; |
6446faa2 | 1826 | |
81819f0f | 1827 | checks_ok: |
ff12059e CL |
1828 | prior = page->freelist; |
1829 | set_freepointer(s, object, prior); | |
81819f0f CL |
1830 | page->freelist = object; |
1831 | page->inuse--; | |
1832 | ||
8a38082d | 1833 | if (unlikely(PageSlubFrozen(page))) { |
84e554e6 | 1834 | stat(s, FREE_FROZEN); |
81819f0f | 1835 | goto out_unlock; |
8ff12cfc | 1836 | } |
81819f0f CL |
1837 | |
1838 | if (unlikely(!page->inuse)) | |
1839 | goto slab_empty; | |
1840 | ||
1841 | /* | |
6446faa2 | 1842 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1843 | * then add it. |
1844 | */ | |
a973e9dd | 1845 | if (unlikely(!prior)) { |
7c2e132c | 1846 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
84e554e6 | 1847 | stat(s, FREE_ADD_PARTIAL); |
8ff12cfc | 1848 | } |
81819f0f CL |
1849 | |
1850 | out_unlock: | |
1851 | slab_unlock(page); | |
81819f0f CL |
1852 | return; |
1853 | ||
1854 | slab_empty: | |
a973e9dd | 1855 | if (prior) { |
81819f0f | 1856 | /* |
672bba3a | 1857 | * Slab still on the partial list. |
81819f0f CL |
1858 | */ |
1859 | remove_partial(s, page); | |
84e554e6 | 1860 | stat(s, FREE_REMOVE_PARTIAL); |
8ff12cfc | 1861 | } |
81819f0f | 1862 | slab_unlock(page); |
84e554e6 | 1863 | stat(s, FREE_SLAB); |
81819f0f | 1864 | discard_slab(s, page); |
81819f0f CL |
1865 | return; |
1866 | ||
1867 | debug: | |
3ec09742 | 1868 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1869 | goto out_unlock; |
77c5e2d0 | 1870 | goto checks_ok; |
81819f0f CL |
1871 | } |
1872 | ||
894b8788 CL |
1873 | /* |
1874 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1875 | * can perform fastpath freeing without additional function calls. | |
1876 | * | |
1877 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1878 | * of this processor. This typically the case if we have just allocated | |
1879 | * the item before. | |
1880 | * | |
1881 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1882 | * with all sorts of special processing. | |
1883 | */ | |
06428780 | 1884 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1885 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1886 | { |
1887 | void **object = (void *)x; | |
dfb4f096 | 1888 | struct kmem_cache_cpu *c; |
1f84260c CL |
1889 | unsigned long flags; |
1890 | ||
c016b0bd CL |
1891 | slab_free_hook(s, x); |
1892 | ||
894b8788 | 1893 | local_irq_save(flags); |
9dfc6e68 | 1894 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd CL |
1895 | |
1896 | slab_free_hook_irq(s, x); | |
1897 | ||
15b7c514 | 1898 | if (likely(page == c->page && c->node != NUMA_NO_NODE)) { |
ff12059e | 1899 | set_freepointer(s, object, c->freelist); |
dfb4f096 | 1900 | c->freelist = object; |
84e554e6 | 1901 | stat(s, FREE_FASTPATH); |
894b8788 | 1902 | } else |
ff12059e | 1903 | __slab_free(s, page, x, addr); |
894b8788 CL |
1904 | |
1905 | local_irq_restore(flags); | |
1906 | } | |
1907 | ||
81819f0f CL |
1908 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1909 | { | |
77c5e2d0 | 1910 | struct page *page; |
81819f0f | 1911 | |
b49af68f | 1912 | page = virt_to_head_page(x); |
81819f0f | 1913 | |
ce71e27c | 1914 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 1915 | |
ca2b84cb | 1916 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
1917 | } |
1918 | EXPORT_SYMBOL(kmem_cache_free); | |
1919 | ||
81819f0f | 1920 | /* |
672bba3a CL |
1921 | * Object placement in a slab is made very easy because we always start at |
1922 | * offset 0. If we tune the size of the object to the alignment then we can | |
1923 | * get the required alignment by putting one properly sized object after | |
1924 | * another. | |
81819f0f CL |
1925 | * |
1926 | * Notice that the allocation order determines the sizes of the per cpu | |
1927 | * caches. Each processor has always one slab available for allocations. | |
1928 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1929 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1930 | * locking overhead. |
81819f0f CL |
1931 | */ |
1932 | ||
1933 | /* | |
1934 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1935 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1936 | * and increases the number of allocations possible without having to | |
1937 | * take the list_lock. | |
1938 | */ | |
1939 | static int slub_min_order; | |
114e9e89 | 1940 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1941 | static int slub_min_objects; |
81819f0f CL |
1942 | |
1943 | /* | |
1944 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1945 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1946 | */ |
1947 | static int slub_nomerge; | |
1948 | ||
81819f0f CL |
1949 | /* |
1950 | * Calculate the order of allocation given an slab object size. | |
1951 | * | |
672bba3a CL |
1952 | * The order of allocation has significant impact on performance and other |
1953 | * system components. Generally order 0 allocations should be preferred since | |
1954 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1955 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1956 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1957 | * would be wasted. |
1958 | * | |
1959 | * In order to reach satisfactory performance we must ensure that a minimum | |
1960 | * number of objects is in one slab. Otherwise we may generate too much | |
1961 | * activity on the partial lists which requires taking the list_lock. This is | |
1962 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1963 | * |
672bba3a CL |
1964 | * slub_max_order specifies the order where we begin to stop considering the |
1965 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1966 | * we try to keep the page order as low as possible. So we accept more waste | |
1967 | * of space in favor of a small page order. | |
81819f0f | 1968 | * |
672bba3a CL |
1969 | * Higher order allocations also allow the placement of more objects in a |
1970 | * slab and thereby reduce object handling overhead. If the user has | |
1971 | * requested a higher mininum order then we start with that one instead of | |
1972 | * the smallest order which will fit the object. | |
81819f0f | 1973 | */ |
5e6d444e CL |
1974 | static inline int slab_order(int size, int min_objects, |
1975 | int max_order, int fract_leftover) | |
81819f0f CL |
1976 | { |
1977 | int order; | |
1978 | int rem; | |
6300ea75 | 1979 | int min_order = slub_min_order; |
81819f0f | 1980 | |
210b5c06 CG |
1981 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) |
1982 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
39b26464 | 1983 | |
6300ea75 | 1984 | for (order = max(min_order, |
5e6d444e CL |
1985 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1986 | order <= max_order; order++) { | |
81819f0f | 1987 | |
5e6d444e | 1988 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1989 | |
5e6d444e | 1990 | if (slab_size < min_objects * size) |
81819f0f CL |
1991 | continue; |
1992 | ||
1993 | rem = slab_size % size; | |
1994 | ||
5e6d444e | 1995 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1996 | break; |
1997 | ||
1998 | } | |
672bba3a | 1999 | |
81819f0f CL |
2000 | return order; |
2001 | } | |
2002 | ||
5e6d444e CL |
2003 | static inline int calculate_order(int size) |
2004 | { | |
2005 | int order; | |
2006 | int min_objects; | |
2007 | int fraction; | |
e8120ff1 | 2008 | int max_objects; |
5e6d444e CL |
2009 | |
2010 | /* | |
2011 | * Attempt to find best configuration for a slab. This | |
2012 | * works by first attempting to generate a layout with | |
2013 | * the best configuration and backing off gradually. | |
2014 | * | |
2015 | * First we reduce the acceptable waste in a slab. Then | |
2016 | * we reduce the minimum objects required in a slab. | |
2017 | */ | |
2018 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2019 | if (!min_objects) |
2020 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
e8120ff1 ZY |
2021 | max_objects = (PAGE_SIZE << slub_max_order)/size; |
2022 | min_objects = min(min_objects, max_objects); | |
2023 | ||
5e6d444e | 2024 | while (min_objects > 1) { |
c124f5b5 | 2025 | fraction = 16; |
5e6d444e CL |
2026 | while (fraction >= 4) { |
2027 | order = slab_order(size, min_objects, | |
2028 | slub_max_order, fraction); | |
2029 | if (order <= slub_max_order) | |
2030 | return order; | |
2031 | fraction /= 2; | |
2032 | } | |
5086c389 | 2033 | min_objects--; |
5e6d444e CL |
2034 | } |
2035 | ||
2036 | /* | |
2037 | * We were unable to place multiple objects in a slab. Now | |
2038 | * lets see if we can place a single object there. | |
2039 | */ | |
2040 | order = slab_order(size, 1, slub_max_order, 1); | |
2041 | if (order <= slub_max_order) | |
2042 | return order; | |
2043 | ||
2044 | /* | |
2045 | * Doh this slab cannot be placed using slub_max_order. | |
2046 | */ | |
2047 | order = slab_order(size, 1, MAX_ORDER, 1); | |
818cf590 | 2048 | if (order < MAX_ORDER) |
5e6d444e CL |
2049 | return order; |
2050 | return -ENOSYS; | |
2051 | } | |
2052 | ||
81819f0f | 2053 | /* |
672bba3a | 2054 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2055 | */ |
2056 | static unsigned long calculate_alignment(unsigned long flags, | |
2057 | unsigned long align, unsigned long size) | |
2058 | { | |
2059 | /* | |
6446faa2 CL |
2060 | * If the user wants hardware cache aligned objects then follow that |
2061 | * suggestion if the object is sufficiently large. | |
81819f0f | 2062 | * |
6446faa2 CL |
2063 | * The hardware cache alignment cannot override the specified |
2064 | * alignment though. If that is greater then use it. | |
81819f0f | 2065 | */ |
b6210386 NP |
2066 | if (flags & SLAB_HWCACHE_ALIGN) { |
2067 | unsigned long ralign = cache_line_size(); | |
2068 | while (size <= ralign / 2) | |
2069 | ralign /= 2; | |
2070 | align = max(align, ralign); | |
2071 | } | |
81819f0f CL |
2072 | |
2073 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2074 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2075 | |
2076 | return ALIGN(align, sizeof(void *)); | |
2077 | } | |
2078 | ||
5595cffc PE |
2079 | static void |
2080 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2081 | { |
2082 | n->nr_partial = 0; | |
81819f0f CL |
2083 | spin_lock_init(&n->list_lock); |
2084 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2085 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2086 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2087 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2088 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2089 | #endif |
81819f0f CL |
2090 | } |
2091 | ||
55136592 | 2092 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2093 | { |
6c182dc0 CL |
2094 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2095 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2096 | |
6c182dc0 | 2097 | s->cpu_slab = alloc_percpu(struct kmem_cache_cpu); |
4c93c355 | 2098 | |
6c182dc0 | 2099 | return s->cpu_slab != NULL; |
4c93c355 | 2100 | } |
4c93c355 | 2101 | |
51df1142 CL |
2102 | static struct kmem_cache *kmem_cache_node; |
2103 | ||
81819f0f CL |
2104 | /* |
2105 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2106 | * slab on the node for this slabcache. There are no concurrent accesses | |
2107 | * possible. | |
2108 | * | |
2109 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2110 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2111 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2112 | */ |
55136592 | 2113 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2114 | { |
2115 | struct page *page; | |
2116 | struct kmem_cache_node *n; | |
ba84c73c | 2117 | unsigned long flags; |
81819f0f | 2118 | |
51df1142 | 2119 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2120 | |
51df1142 | 2121 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2122 | |
2123 | BUG_ON(!page); | |
a2f92ee7 CL |
2124 | if (page_to_nid(page) != node) { |
2125 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2126 | "node %d\n", node); | |
2127 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2128 | "in order to be able to continue\n"); | |
2129 | } | |
2130 | ||
81819f0f CL |
2131 | n = page->freelist; |
2132 | BUG_ON(!n); | |
51df1142 | 2133 | page->freelist = get_freepointer(kmem_cache_node, n); |
81819f0f | 2134 | page->inuse++; |
51df1142 | 2135 | kmem_cache_node->node[node] = n; |
8ab1372f | 2136 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2137 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2138 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2139 | #endif |
51df1142 CL |
2140 | init_kmem_cache_node(n, kmem_cache_node); |
2141 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
6446faa2 | 2142 | |
ba84c73c | 2143 | /* |
2144 | * lockdep requires consistent irq usage for each lock | |
2145 | * so even though there cannot be a race this early in | |
2146 | * the boot sequence, we still disable irqs. | |
2147 | */ | |
2148 | local_irq_save(flags); | |
7c2e132c | 2149 | add_partial(n, page, 0); |
ba84c73c | 2150 | local_irq_restore(flags); |
81819f0f CL |
2151 | } |
2152 | ||
2153 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2154 | { | |
2155 | int node; | |
2156 | ||
f64dc58c | 2157 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2158 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2159 | |
73367bd8 | 2160 | if (n) |
51df1142 CL |
2161 | kmem_cache_free(kmem_cache_node, n); |
2162 | ||
81819f0f CL |
2163 | s->node[node] = NULL; |
2164 | } | |
2165 | } | |
2166 | ||
55136592 | 2167 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2168 | { |
2169 | int node; | |
81819f0f | 2170 | |
f64dc58c | 2171 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2172 | struct kmem_cache_node *n; |
2173 | ||
73367bd8 | 2174 | if (slab_state == DOWN) { |
55136592 | 2175 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2176 | continue; |
2177 | } | |
51df1142 | 2178 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2179 | GFP_KERNEL, node); |
81819f0f | 2180 | |
73367bd8 AD |
2181 | if (!n) { |
2182 | free_kmem_cache_nodes(s); | |
2183 | return 0; | |
81819f0f | 2184 | } |
73367bd8 | 2185 | |
81819f0f | 2186 | s->node[node] = n; |
5595cffc | 2187 | init_kmem_cache_node(n, s); |
81819f0f CL |
2188 | } |
2189 | return 1; | |
2190 | } | |
81819f0f | 2191 | |
c0bdb232 | 2192 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2193 | { |
2194 | if (min < MIN_PARTIAL) | |
2195 | min = MIN_PARTIAL; | |
2196 | else if (min > MAX_PARTIAL) | |
2197 | min = MAX_PARTIAL; | |
2198 | s->min_partial = min; | |
2199 | } | |
2200 | ||
81819f0f CL |
2201 | /* |
2202 | * calculate_sizes() determines the order and the distribution of data within | |
2203 | * a slab object. | |
2204 | */ | |
06b285dc | 2205 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2206 | { |
2207 | unsigned long flags = s->flags; | |
2208 | unsigned long size = s->objsize; | |
2209 | unsigned long align = s->align; | |
834f3d11 | 2210 | int order; |
81819f0f | 2211 | |
d8b42bf5 CL |
2212 | /* |
2213 | * Round up object size to the next word boundary. We can only | |
2214 | * place the free pointer at word boundaries and this determines | |
2215 | * the possible location of the free pointer. | |
2216 | */ | |
2217 | size = ALIGN(size, sizeof(void *)); | |
2218 | ||
2219 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2220 | /* |
2221 | * Determine if we can poison the object itself. If the user of | |
2222 | * the slab may touch the object after free or before allocation | |
2223 | * then we should never poison the object itself. | |
2224 | */ | |
2225 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2226 | !s->ctor) |
81819f0f CL |
2227 | s->flags |= __OBJECT_POISON; |
2228 | else | |
2229 | s->flags &= ~__OBJECT_POISON; | |
2230 | ||
81819f0f CL |
2231 | |
2232 | /* | |
672bba3a | 2233 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2234 | * end of the object and the free pointer. If not then add an |
672bba3a | 2235 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2236 | */ |
2237 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2238 | size += sizeof(void *); | |
41ecc55b | 2239 | #endif |
81819f0f CL |
2240 | |
2241 | /* | |
672bba3a CL |
2242 | * With that we have determined the number of bytes in actual use |
2243 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2244 | */ |
2245 | s->inuse = size; | |
2246 | ||
2247 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2248 | s->ctor)) { |
81819f0f CL |
2249 | /* |
2250 | * Relocate free pointer after the object if it is not | |
2251 | * permitted to overwrite the first word of the object on | |
2252 | * kmem_cache_free. | |
2253 | * | |
2254 | * This is the case if we do RCU, have a constructor or | |
2255 | * destructor or are poisoning the objects. | |
2256 | */ | |
2257 | s->offset = size; | |
2258 | size += sizeof(void *); | |
2259 | } | |
2260 | ||
c12b3c62 | 2261 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2262 | if (flags & SLAB_STORE_USER) |
2263 | /* | |
2264 | * Need to store information about allocs and frees after | |
2265 | * the object. | |
2266 | */ | |
2267 | size += 2 * sizeof(struct track); | |
2268 | ||
be7b3fbc | 2269 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2270 | /* |
2271 | * Add some empty padding so that we can catch | |
2272 | * overwrites from earlier objects rather than let | |
2273 | * tracking information or the free pointer be | |
0211a9c8 | 2274 | * corrupted if a user writes before the start |
81819f0f CL |
2275 | * of the object. |
2276 | */ | |
2277 | size += sizeof(void *); | |
41ecc55b | 2278 | #endif |
672bba3a | 2279 | |
81819f0f CL |
2280 | /* |
2281 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2282 | * user specified and the dynamic determination of cache line size |
2283 | * on bootup. | |
81819f0f CL |
2284 | */ |
2285 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2286 | s->align = align; |
81819f0f CL |
2287 | |
2288 | /* | |
2289 | * SLUB stores one object immediately after another beginning from | |
2290 | * offset 0. In order to align the objects we have to simply size | |
2291 | * each object to conform to the alignment. | |
2292 | */ | |
2293 | size = ALIGN(size, align); | |
2294 | s->size = size; | |
06b285dc CL |
2295 | if (forced_order >= 0) |
2296 | order = forced_order; | |
2297 | else | |
2298 | order = calculate_order(size); | |
81819f0f | 2299 | |
834f3d11 | 2300 | if (order < 0) |
81819f0f CL |
2301 | return 0; |
2302 | ||
b7a49f0d | 2303 | s->allocflags = 0; |
834f3d11 | 2304 | if (order) |
b7a49f0d CL |
2305 | s->allocflags |= __GFP_COMP; |
2306 | ||
2307 | if (s->flags & SLAB_CACHE_DMA) | |
2308 | s->allocflags |= SLUB_DMA; | |
2309 | ||
2310 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2311 | s->allocflags |= __GFP_RECLAIMABLE; | |
2312 | ||
81819f0f CL |
2313 | /* |
2314 | * Determine the number of objects per slab | |
2315 | */ | |
834f3d11 | 2316 | s->oo = oo_make(order, size); |
65c3376a | 2317 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2318 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2319 | s->max = s->oo; | |
81819f0f | 2320 | |
834f3d11 | 2321 | return !!oo_objects(s->oo); |
81819f0f CL |
2322 | |
2323 | } | |
2324 | ||
55136592 | 2325 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
2326 | const char *name, size_t size, |
2327 | size_t align, unsigned long flags, | |
51cc5068 | 2328 | void (*ctor)(void *)) |
81819f0f CL |
2329 | { |
2330 | memset(s, 0, kmem_size); | |
2331 | s->name = name; | |
2332 | s->ctor = ctor; | |
81819f0f | 2333 | s->objsize = size; |
81819f0f | 2334 | s->align = align; |
ba0268a8 | 2335 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2336 | |
06b285dc | 2337 | if (!calculate_sizes(s, -1)) |
81819f0f | 2338 | goto error; |
3de47213 DR |
2339 | if (disable_higher_order_debug) { |
2340 | /* | |
2341 | * Disable debugging flags that store metadata if the min slab | |
2342 | * order increased. | |
2343 | */ | |
2344 | if (get_order(s->size) > get_order(s->objsize)) { | |
2345 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2346 | s->offset = 0; | |
2347 | if (!calculate_sizes(s, -1)) | |
2348 | goto error; | |
2349 | } | |
2350 | } | |
81819f0f | 2351 | |
3b89d7d8 DR |
2352 | /* |
2353 | * The larger the object size is, the more pages we want on the partial | |
2354 | * list to avoid pounding the page allocator excessively. | |
2355 | */ | |
c0bdb232 | 2356 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2357 | s->refcount = 1; |
2358 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2359 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2360 | #endif |
55136592 | 2361 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 2362 | goto error; |
81819f0f | 2363 | |
55136592 | 2364 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 2365 | return 1; |
ff12059e | 2366 | |
4c93c355 | 2367 | free_kmem_cache_nodes(s); |
81819f0f CL |
2368 | error: |
2369 | if (flags & SLAB_PANIC) | |
2370 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2371 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2372 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2373 | s->offset, flags); |
2374 | return 0; | |
2375 | } | |
81819f0f | 2376 | |
81819f0f CL |
2377 | /* |
2378 | * Determine the size of a slab object | |
2379 | */ | |
2380 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2381 | { | |
2382 | return s->objsize; | |
2383 | } | |
2384 | EXPORT_SYMBOL(kmem_cache_size); | |
2385 | ||
2386 | const char *kmem_cache_name(struct kmem_cache *s) | |
2387 | { | |
2388 | return s->name; | |
2389 | } | |
2390 | EXPORT_SYMBOL(kmem_cache_name); | |
2391 | ||
33b12c38 CL |
2392 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2393 | const char *text) | |
2394 | { | |
2395 | #ifdef CONFIG_SLUB_DEBUG | |
2396 | void *addr = page_address(page); | |
2397 | void *p; | |
a5dd5c11 NK |
2398 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
2399 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
2400 | if (!map) |
2401 | return; | |
33b12c38 CL |
2402 | slab_err(s, page, "%s", text); |
2403 | slab_lock(page); | |
2404 | for_each_free_object(p, s, page->freelist) | |
2405 | set_bit(slab_index(p, s, addr), map); | |
2406 | ||
2407 | for_each_object(p, s, addr, page->objects) { | |
2408 | ||
2409 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2410 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2411 | p, p - addr); | |
2412 | print_tracking(s, p); | |
2413 | } | |
2414 | } | |
2415 | slab_unlock(page); | |
bbd7d57b | 2416 | kfree(map); |
33b12c38 CL |
2417 | #endif |
2418 | } | |
2419 | ||
81819f0f | 2420 | /* |
599870b1 | 2421 | * Attempt to free all partial slabs on a node. |
81819f0f | 2422 | */ |
599870b1 | 2423 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2424 | { |
81819f0f CL |
2425 | unsigned long flags; |
2426 | struct page *page, *h; | |
2427 | ||
2428 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2429 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 2430 | if (!page->inuse) { |
62e346a8 | 2431 | __remove_partial(n, page); |
81819f0f | 2432 | discard_slab(s, page); |
33b12c38 CL |
2433 | } else { |
2434 | list_slab_objects(s, page, | |
2435 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2436 | } |
33b12c38 | 2437 | } |
81819f0f | 2438 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2439 | } |
2440 | ||
2441 | /* | |
672bba3a | 2442 | * Release all resources used by a slab cache. |
81819f0f | 2443 | */ |
0c710013 | 2444 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2445 | { |
2446 | int node; | |
2447 | ||
2448 | flush_all(s); | |
9dfc6e68 | 2449 | free_percpu(s->cpu_slab); |
81819f0f | 2450 | /* Attempt to free all objects */ |
f64dc58c | 2451 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2452 | struct kmem_cache_node *n = get_node(s, node); |
2453 | ||
599870b1 CL |
2454 | free_partial(s, n); |
2455 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2456 | return 1; |
2457 | } | |
2458 | free_kmem_cache_nodes(s); | |
2459 | return 0; | |
2460 | } | |
2461 | ||
2462 | /* | |
2463 | * Close a cache and release the kmem_cache structure | |
2464 | * (must be used for caches created using kmem_cache_create) | |
2465 | */ | |
2466 | void kmem_cache_destroy(struct kmem_cache *s) | |
2467 | { | |
2468 | down_write(&slub_lock); | |
2469 | s->refcount--; | |
2470 | if (!s->refcount) { | |
2471 | list_del(&s->list); | |
d629d819 PE |
2472 | if (kmem_cache_close(s)) { |
2473 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2474 | "still has objects.\n", s->name, __func__); | |
2475 | dump_stack(); | |
2476 | } | |
d76b1590 ED |
2477 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2478 | rcu_barrier(); | |
81819f0f | 2479 | sysfs_slab_remove(s); |
2bce6485 CL |
2480 | } |
2481 | up_write(&slub_lock); | |
81819f0f CL |
2482 | } |
2483 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2484 | ||
2485 | /******************************************************************** | |
2486 | * Kmalloc subsystem | |
2487 | *******************************************************************/ | |
2488 | ||
51df1142 | 2489 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
2490 | EXPORT_SYMBOL(kmalloc_caches); |
2491 | ||
51df1142 CL |
2492 | static struct kmem_cache *kmem_cache; |
2493 | ||
55136592 | 2494 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 2495 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
2496 | #endif |
2497 | ||
81819f0f CL |
2498 | static int __init setup_slub_min_order(char *str) |
2499 | { | |
06428780 | 2500 | get_option(&str, &slub_min_order); |
81819f0f CL |
2501 | |
2502 | return 1; | |
2503 | } | |
2504 | ||
2505 | __setup("slub_min_order=", setup_slub_min_order); | |
2506 | ||
2507 | static int __init setup_slub_max_order(char *str) | |
2508 | { | |
06428780 | 2509 | get_option(&str, &slub_max_order); |
818cf590 | 2510 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2511 | |
2512 | return 1; | |
2513 | } | |
2514 | ||
2515 | __setup("slub_max_order=", setup_slub_max_order); | |
2516 | ||
2517 | static int __init setup_slub_min_objects(char *str) | |
2518 | { | |
06428780 | 2519 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2520 | |
2521 | return 1; | |
2522 | } | |
2523 | ||
2524 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2525 | ||
2526 | static int __init setup_slub_nomerge(char *str) | |
2527 | { | |
2528 | slub_nomerge = 1; | |
2529 | return 1; | |
2530 | } | |
2531 | ||
2532 | __setup("slub_nomerge", setup_slub_nomerge); | |
2533 | ||
51df1142 CL |
2534 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
2535 | int size, unsigned int flags) | |
81819f0f | 2536 | { |
51df1142 CL |
2537 | struct kmem_cache *s; |
2538 | ||
2539 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
2540 | ||
83b519e8 PE |
2541 | /* |
2542 | * This function is called with IRQs disabled during early-boot on | |
2543 | * single CPU so there's no need to take slub_lock here. | |
2544 | */ | |
55136592 | 2545 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2546 | flags, NULL)) |
81819f0f CL |
2547 | goto panic; |
2548 | ||
2549 | list_add(&s->list, &slab_caches); | |
51df1142 | 2550 | return s; |
81819f0f CL |
2551 | |
2552 | panic: | |
2553 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 2554 | return NULL; |
81819f0f CL |
2555 | } |
2556 | ||
f1b26339 CL |
2557 | /* |
2558 | * Conversion table for small slabs sizes / 8 to the index in the | |
2559 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2560 | * of two cache sizes there. The size of larger slabs can be determined using | |
2561 | * fls. | |
2562 | */ | |
2563 | static s8 size_index[24] = { | |
2564 | 3, /* 8 */ | |
2565 | 4, /* 16 */ | |
2566 | 5, /* 24 */ | |
2567 | 5, /* 32 */ | |
2568 | 6, /* 40 */ | |
2569 | 6, /* 48 */ | |
2570 | 6, /* 56 */ | |
2571 | 6, /* 64 */ | |
2572 | 1, /* 72 */ | |
2573 | 1, /* 80 */ | |
2574 | 1, /* 88 */ | |
2575 | 1, /* 96 */ | |
2576 | 7, /* 104 */ | |
2577 | 7, /* 112 */ | |
2578 | 7, /* 120 */ | |
2579 | 7, /* 128 */ | |
2580 | 2, /* 136 */ | |
2581 | 2, /* 144 */ | |
2582 | 2, /* 152 */ | |
2583 | 2, /* 160 */ | |
2584 | 2, /* 168 */ | |
2585 | 2, /* 176 */ | |
2586 | 2, /* 184 */ | |
2587 | 2 /* 192 */ | |
2588 | }; | |
2589 | ||
acdfcd04 AK |
2590 | static inline int size_index_elem(size_t bytes) |
2591 | { | |
2592 | return (bytes - 1) / 8; | |
2593 | } | |
2594 | ||
81819f0f CL |
2595 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2596 | { | |
f1b26339 | 2597 | int index; |
81819f0f | 2598 | |
f1b26339 CL |
2599 | if (size <= 192) { |
2600 | if (!size) | |
2601 | return ZERO_SIZE_PTR; | |
81819f0f | 2602 | |
acdfcd04 | 2603 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 2604 | } else |
f1b26339 | 2605 | index = fls(size - 1); |
81819f0f CL |
2606 | |
2607 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2608 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 2609 | return kmalloc_dma_caches[index]; |
f1b26339 | 2610 | |
81819f0f | 2611 | #endif |
51df1142 | 2612 | return kmalloc_caches[index]; |
81819f0f CL |
2613 | } |
2614 | ||
2615 | void *__kmalloc(size_t size, gfp_t flags) | |
2616 | { | |
aadb4bc4 | 2617 | struct kmem_cache *s; |
5b882be4 | 2618 | void *ret; |
81819f0f | 2619 | |
ffadd4d0 | 2620 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2621 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2622 | |
2623 | s = get_slab(size, flags); | |
2624 | ||
2625 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2626 | return s; |
2627 | ||
2154a336 | 2628 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2629 | |
ca2b84cb | 2630 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2631 | |
2632 | return ret; | |
81819f0f CL |
2633 | } |
2634 | EXPORT_SYMBOL(__kmalloc); | |
2635 | ||
5d1f57e4 | 2636 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
2637 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2638 | { | |
b1eeab67 | 2639 | struct page *page; |
e4f7c0b4 | 2640 | void *ptr = NULL; |
f619cfe1 | 2641 | |
b1eeab67 VN |
2642 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2643 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 2644 | if (page) |
e4f7c0b4 CM |
2645 | ptr = page_address(page); |
2646 | ||
2647 | kmemleak_alloc(ptr, size, 1, flags); | |
2648 | return ptr; | |
f619cfe1 CL |
2649 | } |
2650 | ||
81819f0f CL |
2651 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
2652 | { | |
aadb4bc4 | 2653 | struct kmem_cache *s; |
5b882be4 | 2654 | void *ret; |
81819f0f | 2655 | |
057685cf | 2656 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2657 | ret = kmalloc_large_node(size, flags, node); |
2658 | ||
ca2b84cb EGM |
2659 | trace_kmalloc_node(_RET_IP_, ret, |
2660 | size, PAGE_SIZE << get_order(size), | |
2661 | flags, node); | |
5b882be4 EGM |
2662 | |
2663 | return ret; | |
2664 | } | |
aadb4bc4 CL |
2665 | |
2666 | s = get_slab(size, flags); | |
2667 | ||
2668 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2669 | return s; |
2670 | ||
5b882be4 EGM |
2671 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2672 | ||
ca2b84cb | 2673 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2674 | |
2675 | return ret; | |
81819f0f CL |
2676 | } |
2677 | EXPORT_SYMBOL(__kmalloc_node); | |
2678 | #endif | |
2679 | ||
2680 | size_t ksize(const void *object) | |
2681 | { | |
272c1d21 | 2682 | struct page *page; |
81819f0f CL |
2683 | struct kmem_cache *s; |
2684 | ||
ef8b4520 | 2685 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2686 | return 0; |
2687 | ||
294a80a8 | 2688 | page = virt_to_head_page(object); |
294a80a8 | 2689 | |
76994412 PE |
2690 | if (unlikely(!PageSlab(page))) { |
2691 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2692 | return PAGE_SIZE << compound_order(page); |
76994412 | 2693 | } |
81819f0f | 2694 | s = page->slab; |
81819f0f | 2695 | |
ae20bfda | 2696 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2697 | /* |
2698 | * Debugging requires use of the padding between object | |
2699 | * and whatever may come after it. | |
2700 | */ | |
2701 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2702 | return s->objsize; | |
2703 | ||
ae20bfda | 2704 | #endif |
81819f0f CL |
2705 | /* |
2706 | * If we have the need to store the freelist pointer | |
2707 | * back there or track user information then we can | |
2708 | * only use the space before that information. | |
2709 | */ | |
2710 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2711 | return s->inuse; | |
81819f0f CL |
2712 | /* |
2713 | * Else we can use all the padding etc for the allocation | |
2714 | */ | |
2715 | return s->size; | |
2716 | } | |
b1aabecd | 2717 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2718 | |
2719 | void kfree(const void *x) | |
2720 | { | |
81819f0f | 2721 | struct page *page; |
5bb983b0 | 2722 | void *object = (void *)x; |
81819f0f | 2723 | |
2121db74 PE |
2724 | trace_kfree(_RET_IP_, x); |
2725 | ||
2408c550 | 2726 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2727 | return; |
2728 | ||
b49af68f | 2729 | page = virt_to_head_page(x); |
aadb4bc4 | 2730 | if (unlikely(!PageSlab(page))) { |
0937502a | 2731 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 2732 | kmemleak_free(x); |
aadb4bc4 CL |
2733 | put_page(page); |
2734 | return; | |
2735 | } | |
ce71e27c | 2736 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2737 | } |
2738 | EXPORT_SYMBOL(kfree); | |
2739 | ||
2086d26a | 2740 | /* |
672bba3a CL |
2741 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2742 | * the remaining slabs by the number of items in use. The slabs with the | |
2743 | * most items in use come first. New allocations will then fill those up | |
2744 | * and thus they can be removed from the partial lists. | |
2745 | * | |
2746 | * The slabs with the least items are placed last. This results in them | |
2747 | * being allocated from last increasing the chance that the last objects | |
2748 | * are freed in them. | |
2086d26a CL |
2749 | */ |
2750 | int kmem_cache_shrink(struct kmem_cache *s) | |
2751 | { | |
2752 | int node; | |
2753 | int i; | |
2754 | struct kmem_cache_node *n; | |
2755 | struct page *page; | |
2756 | struct page *t; | |
205ab99d | 2757 | int objects = oo_objects(s->max); |
2086d26a | 2758 | struct list_head *slabs_by_inuse = |
834f3d11 | 2759 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2760 | unsigned long flags; |
2761 | ||
2762 | if (!slabs_by_inuse) | |
2763 | return -ENOMEM; | |
2764 | ||
2765 | flush_all(s); | |
f64dc58c | 2766 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2767 | n = get_node(s, node); |
2768 | ||
2769 | if (!n->nr_partial) | |
2770 | continue; | |
2771 | ||
834f3d11 | 2772 | for (i = 0; i < objects; i++) |
2086d26a CL |
2773 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2774 | ||
2775 | spin_lock_irqsave(&n->list_lock, flags); | |
2776 | ||
2777 | /* | |
672bba3a | 2778 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2779 | * |
672bba3a CL |
2780 | * Note that concurrent frees may occur while we hold the |
2781 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2782 | */ |
2783 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2784 | if (!page->inuse && slab_trylock(page)) { | |
2785 | /* | |
2786 | * Must hold slab lock here because slab_free | |
2787 | * may have freed the last object and be | |
2788 | * waiting to release the slab. | |
2789 | */ | |
62e346a8 | 2790 | __remove_partial(n, page); |
2086d26a CL |
2791 | slab_unlock(page); |
2792 | discard_slab(s, page); | |
2793 | } else { | |
fcda3d89 CL |
2794 | list_move(&page->lru, |
2795 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2796 | } |
2797 | } | |
2798 | ||
2086d26a | 2799 | /* |
672bba3a CL |
2800 | * Rebuild the partial list with the slabs filled up most |
2801 | * first and the least used slabs at the end. | |
2086d26a | 2802 | */ |
834f3d11 | 2803 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2804 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2805 | ||
2086d26a CL |
2806 | spin_unlock_irqrestore(&n->list_lock, flags); |
2807 | } | |
2808 | ||
2809 | kfree(slabs_by_inuse); | |
2810 | return 0; | |
2811 | } | |
2812 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2813 | ||
92a5bbc1 | 2814 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
2815 | static int slab_mem_going_offline_callback(void *arg) |
2816 | { | |
2817 | struct kmem_cache *s; | |
2818 | ||
2819 | down_read(&slub_lock); | |
2820 | list_for_each_entry(s, &slab_caches, list) | |
2821 | kmem_cache_shrink(s); | |
2822 | up_read(&slub_lock); | |
2823 | ||
2824 | return 0; | |
2825 | } | |
2826 | ||
2827 | static void slab_mem_offline_callback(void *arg) | |
2828 | { | |
2829 | struct kmem_cache_node *n; | |
2830 | struct kmem_cache *s; | |
2831 | struct memory_notify *marg = arg; | |
2832 | int offline_node; | |
2833 | ||
2834 | offline_node = marg->status_change_nid; | |
2835 | ||
2836 | /* | |
2837 | * If the node still has available memory. we need kmem_cache_node | |
2838 | * for it yet. | |
2839 | */ | |
2840 | if (offline_node < 0) | |
2841 | return; | |
2842 | ||
2843 | down_read(&slub_lock); | |
2844 | list_for_each_entry(s, &slab_caches, list) { | |
2845 | n = get_node(s, offline_node); | |
2846 | if (n) { | |
2847 | /* | |
2848 | * if n->nr_slabs > 0, slabs still exist on the node | |
2849 | * that is going down. We were unable to free them, | |
c9404c9c | 2850 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
2851 | * callback. So, we must fail. |
2852 | */ | |
0f389ec6 | 2853 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2854 | |
2855 | s->node[offline_node] = NULL; | |
8de66a0c | 2856 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
2857 | } |
2858 | } | |
2859 | up_read(&slub_lock); | |
2860 | } | |
2861 | ||
2862 | static int slab_mem_going_online_callback(void *arg) | |
2863 | { | |
2864 | struct kmem_cache_node *n; | |
2865 | struct kmem_cache *s; | |
2866 | struct memory_notify *marg = arg; | |
2867 | int nid = marg->status_change_nid; | |
2868 | int ret = 0; | |
2869 | ||
2870 | /* | |
2871 | * If the node's memory is already available, then kmem_cache_node is | |
2872 | * already created. Nothing to do. | |
2873 | */ | |
2874 | if (nid < 0) | |
2875 | return 0; | |
2876 | ||
2877 | /* | |
0121c619 | 2878 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
2879 | * allocate a kmem_cache_node structure in order to bring the node |
2880 | * online. | |
2881 | */ | |
2882 | down_read(&slub_lock); | |
2883 | list_for_each_entry(s, &slab_caches, list) { | |
2884 | /* | |
2885 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2886 | * since memory is not yet available from the node that | |
2887 | * is brought up. | |
2888 | */ | |
8de66a0c | 2889 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
2890 | if (!n) { |
2891 | ret = -ENOMEM; | |
2892 | goto out; | |
2893 | } | |
5595cffc | 2894 | init_kmem_cache_node(n, s); |
b9049e23 YG |
2895 | s->node[nid] = n; |
2896 | } | |
2897 | out: | |
2898 | up_read(&slub_lock); | |
2899 | return ret; | |
2900 | } | |
2901 | ||
2902 | static int slab_memory_callback(struct notifier_block *self, | |
2903 | unsigned long action, void *arg) | |
2904 | { | |
2905 | int ret = 0; | |
2906 | ||
2907 | switch (action) { | |
2908 | case MEM_GOING_ONLINE: | |
2909 | ret = slab_mem_going_online_callback(arg); | |
2910 | break; | |
2911 | case MEM_GOING_OFFLINE: | |
2912 | ret = slab_mem_going_offline_callback(arg); | |
2913 | break; | |
2914 | case MEM_OFFLINE: | |
2915 | case MEM_CANCEL_ONLINE: | |
2916 | slab_mem_offline_callback(arg); | |
2917 | break; | |
2918 | case MEM_ONLINE: | |
2919 | case MEM_CANCEL_OFFLINE: | |
2920 | break; | |
2921 | } | |
dc19f9db KH |
2922 | if (ret) |
2923 | ret = notifier_from_errno(ret); | |
2924 | else | |
2925 | ret = NOTIFY_OK; | |
b9049e23 YG |
2926 | return ret; |
2927 | } | |
2928 | ||
2929 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2930 | ||
81819f0f CL |
2931 | /******************************************************************** |
2932 | * Basic setup of slabs | |
2933 | *******************************************************************/ | |
2934 | ||
51df1142 CL |
2935 | /* |
2936 | * Used for early kmem_cache structures that were allocated using | |
2937 | * the page allocator | |
2938 | */ | |
2939 | ||
2940 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
2941 | { | |
2942 | int node; | |
2943 | ||
2944 | list_add(&s->list, &slab_caches); | |
2945 | s->refcount = -1; | |
2946 | ||
2947 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
2948 | struct kmem_cache_node *n = get_node(s, node); | |
2949 | struct page *p; | |
2950 | ||
2951 | if (n) { | |
2952 | list_for_each_entry(p, &n->partial, lru) | |
2953 | p->slab = s; | |
2954 | ||
2955 | #ifdef CONFIG_SLAB_DEBUG | |
2956 | list_for_each_entry(p, &n->full, lru) | |
2957 | p->slab = s; | |
2958 | #endif | |
2959 | } | |
2960 | } | |
2961 | } | |
2962 | ||
81819f0f CL |
2963 | void __init kmem_cache_init(void) |
2964 | { | |
2965 | int i; | |
4b356be0 | 2966 | int caches = 0; |
51df1142 CL |
2967 | struct kmem_cache *temp_kmem_cache; |
2968 | int order; | |
51df1142 CL |
2969 | struct kmem_cache *temp_kmem_cache_node; |
2970 | unsigned long kmalloc_size; | |
2971 | ||
2972 | kmem_size = offsetof(struct kmem_cache, node) + | |
2973 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
2974 | ||
2975 | /* Allocate two kmem_caches from the page allocator */ | |
2976 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
2977 | order = get_order(2 * kmalloc_size); | |
2978 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
2979 | ||
81819f0f CL |
2980 | /* |
2981 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 2982 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
2983 | * kmem_cache_open for slab_state == DOWN. |
2984 | */ | |
51df1142 CL |
2985 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
2986 | ||
2987 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
2988 | sizeof(struct kmem_cache_node), | |
2989 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 2990 | |
0c40ba4f | 2991 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
2992 | |
2993 | /* Able to allocate the per node structures */ | |
2994 | slab_state = PARTIAL; | |
2995 | ||
51df1142 CL |
2996 | temp_kmem_cache = kmem_cache; |
2997 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
2998 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
2999 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3000 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3001 | |
51df1142 CL |
3002 | /* |
3003 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3004 | * kmem_cache_node is separately allocated so no need to | |
3005 | * update any list pointers. | |
3006 | */ | |
3007 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3008 | |
51df1142 CL |
3009 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3010 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3011 | ||
3012 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3013 | ||
3014 | caches++; | |
51df1142 CL |
3015 | kmem_cache_bootstrap_fixup(kmem_cache); |
3016 | caches++; | |
3017 | /* Free temporary boot structure */ | |
3018 | free_pages((unsigned long)temp_kmem_cache, order); | |
3019 | ||
3020 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3021 | |
3022 | /* | |
3023 | * Patch up the size_index table if we have strange large alignment | |
3024 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3025 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3026 | * |
3027 | * Largest permitted alignment is 256 bytes due to the way we | |
3028 | * handle the index determination for the smaller caches. | |
3029 | * | |
3030 | * Make sure that nothing crazy happens if someone starts tinkering | |
3031 | * around with ARCH_KMALLOC_MINALIGN | |
3032 | */ | |
3033 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3034 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3035 | ||
acdfcd04 AK |
3036 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3037 | int elem = size_index_elem(i); | |
3038 | if (elem >= ARRAY_SIZE(size_index)) | |
3039 | break; | |
3040 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3041 | } | |
f1b26339 | 3042 | |
acdfcd04 AK |
3043 | if (KMALLOC_MIN_SIZE == 64) { |
3044 | /* | |
3045 | * The 96 byte size cache is not used if the alignment | |
3046 | * is 64 byte. | |
3047 | */ | |
3048 | for (i = 64 + 8; i <= 96; i += 8) | |
3049 | size_index[size_index_elem(i)] = 7; | |
3050 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3051 | /* |
3052 | * The 192 byte sized cache is not used if the alignment | |
3053 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3054 | * instead. | |
3055 | */ | |
3056 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3057 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3058 | } |
3059 | ||
51df1142 CL |
3060 | /* Caches that are not of the two-to-the-power-of size */ |
3061 | if (KMALLOC_MIN_SIZE <= 32) { | |
3062 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3063 | caches++; | |
3064 | } | |
3065 | ||
3066 | if (KMALLOC_MIN_SIZE <= 64) { | |
3067 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3068 | caches++; | |
3069 | } | |
3070 | ||
3071 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3072 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3073 | caches++; | |
3074 | } | |
3075 | ||
81819f0f CL |
3076 | slab_state = UP; |
3077 | ||
3078 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3079 | if (KMALLOC_MIN_SIZE <= 32) { |
3080 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3081 | BUG_ON(!kmalloc_caches[1]->name); | |
3082 | } | |
3083 | ||
3084 | if (KMALLOC_MIN_SIZE <= 64) { | |
3085 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3086 | BUG_ON(!kmalloc_caches[2]->name); | |
3087 | } | |
3088 | ||
d7278bd7 CL |
3089 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3090 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3091 | ||
3092 | BUG_ON(!s); | |
51df1142 | 3093 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3094 | } |
81819f0f CL |
3095 | |
3096 | #ifdef CONFIG_SMP | |
3097 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3098 | #endif |
81819f0f | 3099 | |
55136592 | 3100 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3101 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3102 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3103 | |
51df1142 | 3104 | if (s && s->size) { |
55136592 CL |
3105 | char *name = kasprintf(GFP_NOWAIT, |
3106 | "dma-kmalloc-%d", s->objsize); | |
3107 | ||
3108 | BUG_ON(!name); | |
51df1142 CL |
3109 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3110 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3111 | } |
3112 | } | |
3113 | #endif | |
3adbefee IM |
3114 | printk(KERN_INFO |
3115 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3116 | " CPUs=%d, Nodes=%d\n", |
3117 | caches, cache_line_size(), | |
81819f0f CL |
3118 | slub_min_order, slub_max_order, slub_min_objects, |
3119 | nr_cpu_ids, nr_node_ids); | |
3120 | } | |
3121 | ||
7e85ee0c PE |
3122 | void __init kmem_cache_init_late(void) |
3123 | { | |
7e85ee0c PE |
3124 | } |
3125 | ||
81819f0f CL |
3126 | /* |
3127 | * Find a mergeable slab cache | |
3128 | */ | |
3129 | static int slab_unmergeable(struct kmem_cache *s) | |
3130 | { | |
3131 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3132 | return 1; | |
3133 | ||
c59def9f | 3134 | if (s->ctor) |
81819f0f CL |
3135 | return 1; |
3136 | ||
8ffa6875 CL |
3137 | /* |
3138 | * We may have set a slab to be unmergeable during bootstrap. | |
3139 | */ | |
3140 | if (s->refcount < 0) | |
3141 | return 1; | |
3142 | ||
81819f0f CL |
3143 | return 0; |
3144 | } | |
3145 | ||
3146 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3147 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3148 | void (*ctor)(void *)) |
81819f0f | 3149 | { |
5b95a4ac | 3150 | struct kmem_cache *s; |
81819f0f CL |
3151 | |
3152 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3153 | return NULL; | |
3154 | ||
c59def9f | 3155 | if (ctor) |
81819f0f CL |
3156 | return NULL; |
3157 | ||
3158 | size = ALIGN(size, sizeof(void *)); | |
3159 | align = calculate_alignment(flags, align, size); | |
3160 | size = ALIGN(size, align); | |
ba0268a8 | 3161 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3162 | |
5b95a4ac | 3163 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3164 | if (slab_unmergeable(s)) |
3165 | continue; | |
3166 | ||
3167 | if (size > s->size) | |
3168 | continue; | |
3169 | ||
ba0268a8 | 3170 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3171 | continue; |
3172 | /* | |
3173 | * Check if alignment is compatible. | |
3174 | * Courtesy of Adrian Drzewiecki | |
3175 | */ | |
06428780 | 3176 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3177 | continue; |
3178 | ||
3179 | if (s->size - size >= sizeof(void *)) | |
3180 | continue; | |
3181 | ||
3182 | return s; | |
3183 | } | |
3184 | return NULL; | |
3185 | } | |
3186 | ||
3187 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3188 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3189 | { |
3190 | struct kmem_cache *s; | |
84c1cf62 | 3191 | char *n; |
81819f0f | 3192 | |
fe1ff49d BH |
3193 | if (WARN_ON(!name)) |
3194 | return NULL; | |
3195 | ||
81819f0f | 3196 | down_write(&slub_lock); |
ba0268a8 | 3197 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3198 | if (s) { |
3199 | s->refcount++; | |
3200 | /* | |
3201 | * Adjust the object sizes so that we clear | |
3202 | * the complete object on kzalloc. | |
3203 | */ | |
3204 | s->objsize = max(s->objsize, (int)size); | |
3205 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3206 | |
7b8f3b66 | 3207 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3208 | s->refcount--; |
81819f0f | 3209 | goto err; |
7b8f3b66 | 3210 | } |
2bce6485 | 3211 | up_write(&slub_lock); |
a0e1d1be CL |
3212 | return s; |
3213 | } | |
6446faa2 | 3214 | |
84c1cf62 PE |
3215 | n = kstrdup(name, GFP_KERNEL); |
3216 | if (!n) | |
3217 | goto err; | |
3218 | ||
a0e1d1be CL |
3219 | s = kmalloc(kmem_size, GFP_KERNEL); |
3220 | if (s) { | |
84c1cf62 | 3221 | if (kmem_cache_open(s, n, |
c59def9f | 3222 | size, align, flags, ctor)) { |
81819f0f | 3223 | list_add(&s->list, &slab_caches); |
7b8f3b66 | 3224 | if (sysfs_slab_add(s)) { |
7b8f3b66 | 3225 | list_del(&s->list); |
84c1cf62 | 3226 | kfree(n); |
7b8f3b66 | 3227 | kfree(s); |
a0e1d1be | 3228 | goto err; |
7b8f3b66 | 3229 | } |
2bce6485 | 3230 | up_write(&slub_lock); |
a0e1d1be CL |
3231 | return s; |
3232 | } | |
84c1cf62 | 3233 | kfree(n); |
a0e1d1be | 3234 | kfree(s); |
81819f0f | 3235 | } |
68cee4f1 | 3236 | err: |
81819f0f | 3237 | up_write(&slub_lock); |
81819f0f | 3238 | |
81819f0f CL |
3239 | if (flags & SLAB_PANIC) |
3240 | panic("Cannot create slabcache %s\n", name); | |
3241 | else | |
3242 | s = NULL; | |
3243 | return s; | |
3244 | } | |
3245 | EXPORT_SYMBOL(kmem_cache_create); | |
3246 | ||
81819f0f | 3247 | #ifdef CONFIG_SMP |
81819f0f | 3248 | /* |
672bba3a CL |
3249 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3250 | * necessary. | |
81819f0f CL |
3251 | */ |
3252 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3253 | unsigned long action, void *hcpu) | |
3254 | { | |
3255 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3256 | struct kmem_cache *s; |
3257 | unsigned long flags; | |
81819f0f CL |
3258 | |
3259 | switch (action) { | |
3260 | case CPU_UP_CANCELED: | |
8bb78442 | 3261 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3262 | case CPU_DEAD: |
8bb78442 | 3263 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3264 | down_read(&slub_lock); |
3265 | list_for_each_entry(s, &slab_caches, list) { | |
3266 | local_irq_save(flags); | |
3267 | __flush_cpu_slab(s, cpu); | |
3268 | local_irq_restore(flags); | |
3269 | } | |
3270 | up_read(&slub_lock); | |
81819f0f CL |
3271 | break; |
3272 | default: | |
3273 | break; | |
3274 | } | |
3275 | return NOTIFY_OK; | |
3276 | } | |
3277 | ||
06428780 | 3278 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3279 | .notifier_call = slab_cpuup_callback |
06428780 | 3280 | }; |
81819f0f CL |
3281 | |
3282 | #endif | |
3283 | ||
ce71e27c | 3284 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3285 | { |
aadb4bc4 | 3286 | struct kmem_cache *s; |
94b528d0 | 3287 | void *ret; |
aadb4bc4 | 3288 | |
ffadd4d0 | 3289 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3290 | return kmalloc_large(size, gfpflags); |
3291 | ||
aadb4bc4 | 3292 | s = get_slab(size, gfpflags); |
81819f0f | 3293 | |
2408c550 | 3294 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3295 | return s; |
81819f0f | 3296 | |
2154a336 | 3297 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 EGM |
3298 | |
3299 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3300 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3301 | |
3302 | return ret; | |
81819f0f CL |
3303 | } |
3304 | ||
5d1f57e4 | 3305 | #ifdef CONFIG_NUMA |
81819f0f | 3306 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3307 | int node, unsigned long caller) |
81819f0f | 3308 | { |
aadb4bc4 | 3309 | struct kmem_cache *s; |
94b528d0 | 3310 | void *ret; |
aadb4bc4 | 3311 | |
d3e14aa3 XF |
3312 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3313 | ret = kmalloc_large_node(size, gfpflags, node); | |
3314 | ||
3315 | trace_kmalloc_node(caller, ret, | |
3316 | size, PAGE_SIZE << get_order(size), | |
3317 | gfpflags, node); | |
3318 | ||
3319 | return ret; | |
3320 | } | |
eada35ef | 3321 | |
aadb4bc4 | 3322 | s = get_slab(size, gfpflags); |
81819f0f | 3323 | |
2408c550 | 3324 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3325 | return s; |
81819f0f | 3326 | |
94b528d0 EGM |
3327 | ret = slab_alloc(s, gfpflags, node, caller); |
3328 | ||
3329 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3330 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3331 | |
3332 | return ret; | |
81819f0f | 3333 | } |
5d1f57e4 | 3334 | #endif |
81819f0f | 3335 | |
ab4d5ed5 | 3336 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3337 | static int count_inuse(struct page *page) |
3338 | { | |
3339 | return page->inuse; | |
3340 | } | |
3341 | ||
3342 | static int count_total(struct page *page) | |
3343 | { | |
3344 | return page->objects; | |
3345 | } | |
ab4d5ed5 | 3346 | #endif |
205ab99d | 3347 | |
ab4d5ed5 | 3348 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3349 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3350 | unsigned long *map) | |
53e15af0 CL |
3351 | { |
3352 | void *p; | |
a973e9dd | 3353 | void *addr = page_address(page); |
53e15af0 CL |
3354 | |
3355 | if (!check_slab(s, page) || | |
3356 | !on_freelist(s, page, NULL)) | |
3357 | return 0; | |
3358 | ||
3359 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3360 | bitmap_zero(map, page->objects); |
53e15af0 | 3361 | |
7656c72b CL |
3362 | for_each_free_object(p, s, page->freelist) { |
3363 | set_bit(slab_index(p, s, addr), map); | |
37d57443 | 3364 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) |
53e15af0 CL |
3365 | return 0; |
3366 | } | |
3367 | ||
224a88be | 3368 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3369 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3370 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3371 | return 0; |
3372 | return 1; | |
3373 | } | |
3374 | ||
434e245d CL |
3375 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3376 | unsigned long *map) | |
53e15af0 CL |
3377 | { |
3378 | if (slab_trylock(page)) { | |
434e245d | 3379 | validate_slab(s, page, map); |
53e15af0 CL |
3380 | slab_unlock(page); |
3381 | } else | |
3382 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3383 | s->name, page); | |
53e15af0 CL |
3384 | } |
3385 | ||
434e245d CL |
3386 | static int validate_slab_node(struct kmem_cache *s, |
3387 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3388 | { |
3389 | unsigned long count = 0; | |
3390 | struct page *page; | |
3391 | unsigned long flags; | |
3392 | ||
3393 | spin_lock_irqsave(&n->list_lock, flags); | |
3394 | ||
3395 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3396 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3397 | count++; |
3398 | } | |
3399 | if (count != n->nr_partial) | |
3400 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3401 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3402 | ||
3403 | if (!(s->flags & SLAB_STORE_USER)) | |
3404 | goto out; | |
3405 | ||
3406 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3407 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3408 | count++; |
3409 | } | |
3410 | if (count != atomic_long_read(&n->nr_slabs)) | |
3411 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3412 | "counter=%ld\n", s->name, count, | |
3413 | atomic_long_read(&n->nr_slabs)); | |
3414 | ||
3415 | out: | |
3416 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3417 | return count; | |
3418 | } | |
3419 | ||
434e245d | 3420 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3421 | { |
3422 | int node; | |
3423 | unsigned long count = 0; | |
205ab99d | 3424 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3425 | sizeof(unsigned long), GFP_KERNEL); |
3426 | ||
3427 | if (!map) | |
3428 | return -ENOMEM; | |
53e15af0 CL |
3429 | |
3430 | flush_all(s); | |
f64dc58c | 3431 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3432 | struct kmem_cache_node *n = get_node(s, node); |
3433 | ||
434e245d | 3434 | count += validate_slab_node(s, n, map); |
53e15af0 | 3435 | } |
434e245d | 3436 | kfree(map); |
53e15af0 CL |
3437 | return count; |
3438 | } | |
88a420e4 | 3439 | /* |
672bba3a | 3440 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3441 | * and freed. |
3442 | */ | |
3443 | ||
3444 | struct location { | |
3445 | unsigned long count; | |
ce71e27c | 3446 | unsigned long addr; |
45edfa58 CL |
3447 | long long sum_time; |
3448 | long min_time; | |
3449 | long max_time; | |
3450 | long min_pid; | |
3451 | long max_pid; | |
174596a0 | 3452 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3453 | nodemask_t nodes; |
88a420e4 CL |
3454 | }; |
3455 | ||
3456 | struct loc_track { | |
3457 | unsigned long max; | |
3458 | unsigned long count; | |
3459 | struct location *loc; | |
3460 | }; | |
3461 | ||
3462 | static void free_loc_track(struct loc_track *t) | |
3463 | { | |
3464 | if (t->max) | |
3465 | free_pages((unsigned long)t->loc, | |
3466 | get_order(sizeof(struct location) * t->max)); | |
3467 | } | |
3468 | ||
68dff6a9 | 3469 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3470 | { |
3471 | struct location *l; | |
3472 | int order; | |
3473 | ||
88a420e4 CL |
3474 | order = get_order(sizeof(struct location) * max); |
3475 | ||
68dff6a9 | 3476 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3477 | if (!l) |
3478 | return 0; | |
3479 | ||
3480 | if (t->count) { | |
3481 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3482 | free_loc_track(t); | |
3483 | } | |
3484 | t->max = max; | |
3485 | t->loc = l; | |
3486 | return 1; | |
3487 | } | |
3488 | ||
3489 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3490 | const struct track *track) |
88a420e4 CL |
3491 | { |
3492 | long start, end, pos; | |
3493 | struct location *l; | |
ce71e27c | 3494 | unsigned long caddr; |
45edfa58 | 3495 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3496 | |
3497 | start = -1; | |
3498 | end = t->count; | |
3499 | ||
3500 | for ( ; ; ) { | |
3501 | pos = start + (end - start + 1) / 2; | |
3502 | ||
3503 | /* | |
3504 | * There is nothing at "end". If we end up there | |
3505 | * we need to add something to before end. | |
3506 | */ | |
3507 | if (pos == end) | |
3508 | break; | |
3509 | ||
3510 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3511 | if (track->addr == caddr) { |
3512 | ||
3513 | l = &t->loc[pos]; | |
3514 | l->count++; | |
3515 | if (track->when) { | |
3516 | l->sum_time += age; | |
3517 | if (age < l->min_time) | |
3518 | l->min_time = age; | |
3519 | if (age > l->max_time) | |
3520 | l->max_time = age; | |
3521 | ||
3522 | if (track->pid < l->min_pid) | |
3523 | l->min_pid = track->pid; | |
3524 | if (track->pid > l->max_pid) | |
3525 | l->max_pid = track->pid; | |
3526 | ||
174596a0 RR |
3527 | cpumask_set_cpu(track->cpu, |
3528 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3529 | } |
3530 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3531 | return 1; |
3532 | } | |
3533 | ||
45edfa58 | 3534 | if (track->addr < caddr) |
88a420e4 CL |
3535 | end = pos; |
3536 | else | |
3537 | start = pos; | |
3538 | } | |
3539 | ||
3540 | /* | |
672bba3a | 3541 | * Not found. Insert new tracking element. |
88a420e4 | 3542 | */ |
68dff6a9 | 3543 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3544 | return 0; |
3545 | ||
3546 | l = t->loc + pos; | |
3547 | if (pos < t->count) | |
3548 | memmove(l + 1, l, | |
3549 | (t->count - pos) * sizeof(struct location)); | |
3550 | t->count++; | |
3551 | l->count = 1; | |
45edfa58 CL |
3552 | l->addr = track->addr; |
3553 | l->sum_time = age; | |
3554 | l->min_time = age; | |
3555 | l->max_time = age; | |
3556 | l->min_pid = track->pid; | |
3557 | l->max_pid = track->pid; | |
174596a0 RR |
3558 | cpumask_clear(to_cpumask(l->cpus)); |
3559 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3560 | nodes_clear(l->nodes); |
3561 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3562 | return 1; |
3563 | } | |
3564 | ||
3565 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 3566 | struct page *page, enum track_item alloc, |
a5dd5c11 | 3567 | unsigned long *map) |
88a420e4 | 3568 | { |
a973e9dd | 3569 | void *addr = page_address(page); |
88a420e4 CL |
3570 | void *p; |
3571 | ||
39b26464 | 3572 | bitmap_zero(map, page->objects); |
7656c72b CL |
3573 | for_each_free_object(p, s, page->freelist) |
3574 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3575 | |
224a88be | 3576 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3577 | if (!test_bit(slab_index(p, s, addr), map)) |
3578 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3579 | } |
3580 | ||
3581 | static int list_locations(struct kmem_cache *s, char *buf, | |
3582 | enum track_item alloc) | |
3583 | { | |
e374d483 | 3584 | int len = 0; |
88a420e4 | 3585 | unsigned long i; |
68dff6a9 | 3586 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 3587 | int node; |
bbd7d57b ED |
3588 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
3589 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 3590 | |
bbd7d57b ED |
3591 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
3592 | GFP_TEMPORARY)) { | |
3593 | kfree(map); | |
68dff6a9 | 3594 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 3595 | } |
88a420e4 CL |
3596 | /* Push back cpu slabs */ |
3597 | flush_all(s); | |
3598 | ||
f64dc58c | 3599 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3600 | struct kmem_cache_node *n = get_node(s, node); |
3601 | unsigned long flags; | |
3602 | struct page *page; | |
3603 | ||
9e86943b | 3604 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3605 | continue; |
3606 | ||
3607 | spin_lock_irqsave(&n->list_lock, flags); | |
3608 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 3609 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 3610 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 3611 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
3612 | spin_unlock_irqrestore(&n->list_lock, flags); |
3613 | } | |
3614 | ||
3615 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3616 | struct location *l = &t.loc[i]; |
88a420e4 | 3617 | |
9c246247 | 3618 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3619 | break; |
e374d483 | 3620 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3621 | |
3622 | if (l->addr) | |
e374d483 | 3623 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3624 | else |
e374d483 | 3625 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3626 | |
3627 | if (l->sum_time != l->min_time) { | |
e374d483 | 3628 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3629 | l->min_time, |
3630 | (long)div_u64(l->sum_time, l->count), | |
3631 | l->max_time); | |
45edfa58 | 3632 | } else |
e374d483 | 3633 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3634 | l->min_time); |
3635 | ||
3636 | if (l->min_pid != l->max_pid) | |
e374d483 | 3637 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3638 | l->min_pid, l->max_pid); |
3639 | else | |
e374d483 | 3640 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3641 | l->min_pid); |
3642 | ||
174596a0 RR |
3643 | if (num_online_cpus() > 1 && |
3644 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3645 | len < PAGE_SIZE - 60) { |
3646 | len += sprintf(buf + len, " cpus="); | |
3647 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3648 | to_cpumask(l->cpus)); |
45edfa58 CL |
3649 | } |
3650 | ||
62bc62a8 | 3651 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3652 | len < PAGE_SIZE - 60) { |
3653 | len += sprintf(buf + len, " nodes="); | |
3654 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3655 | l->nodes); |
3656 | } | |
3657 | ||
e374d483 | 3658 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3659 | } |
3660 | ||
3661 | free_loc_track(&t); | |
bbd7d57b | 3662 | kfree(map); |
88a420e4 | 3663 | if (!t.count) |
e374d483 HH |
3664 | len += sprintf(buf, "No data\n"); |
3665 | return len; | |
88a420e4 | 3666 | } |
ab4d5ed5 | 3667 | #endif |
88a420e4 | 3668 | |
a5a84755 CL |
3669 | #ifdef SLUB_RESILIENCY_TEST |
3670 | static void resiliency_test(void) | |
3671 | { | |
3672 | u8 *p; | |
3673 | ||
3674 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
3675 | ||
3676 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3677 | printk(KERN_ERR "-----------------------\n"); | |
3678 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3679 | ||
3680 | p = kzalloc(16, GFP_KERNEL); | |
3681 | p[16] = 0x12; | |
3682 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3683 | " 0x12->0x%p\n\n", p + 16); | |
3684 | ||
3685 | validate_slab_cache(kmalloc_caches[4]); | |
3686 | ||
3687 | /* Hmmm... The next two are dangerous */ | |
3688 | p = kzalloc(32, GFP_KERNEL); | |
3689 | p[32 + sizeof(void *)] = 0x34; | |
3690 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3691 | " 0x34 -> -0x%p\n", p); | |
3692 | printk(KERN_ERR | |
3693 | "If allocated object is overwritten then not detectable\n\n"); | |
3694 | ||
3695 | validate_slab_cache(kmalloc_caches[5]); | |
3696 | p = kzalloc(64, GFP_KERNEL); | |
3697 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3698 | *p = 0x56; | |
3699 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3700 | p); | |
3701 | printk(KERN_ERR | |
3702 | "If allocated object is overwritten then not detectable\n\n"); | |
3703 | validate_slab_cache(kmalloc_caches[6]); | |
3704 | ||
3705 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3706 | p = kzalloc(128, GFP_KERNEL); | |
3707 | kfree(p); | |
3708 | *p = 0x78; | |
3709 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3710 | validate_slab_cache(kmalloc_caches[7]); | |
3711 | ||
3712 | p = kzalloc(256, GFP_KERNEL); | |
3713 | kfree(p); | |
3714 | p[50] = 0x9a; | |
3715 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
3716 | p); | |
3717 | validate_slab_cache(kmalloc_caches[8]); | |
3718 | ||
3719 | p = kzalloc(512, GFP_KERNEL); | |
3720 | kfree(p); | |
3721 | p[512] = 0xab; | |
3722 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3723 | validate_slab_cache(kmalloc_caches[9]); | |
3724 | } | |
3725 | #else | |
3726 | #ifdef CONFIG_SYSFS | |
3727 | static void resiliency_test(void) {}; | |
3728 | #endif | |
3729 | #endif | |
3730 | ||
ab4d5ed5 | 3731 | #ifdef CONFIG_SYSFS |
81819f0f | 3732 | enum slab_stat_type { |
205ab99d CL |
3733 | SL_ALL, /* All slabs */ |
3734 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3735 | SL_CPU, /* Only slabs used for cpu caches */ | |
3736 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3737 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3738 | }; |
3739 | ||
205ab99d | 3740 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3741 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3742 | #define SO_CPU (1 << SL_CPU) | |
3743 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3744 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3745 | |
62e5c4b4 CG |
3746 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3747 | char *buf, unsigned long flags) | |
81819f0f CL |
3748 | { |
3749 | unsigned long total = 0; | |
81819f0f CL |
3750 | int node; |
3751 | int x; | |
3752 | unsigned long *nodes; | |
3753 | unsigned long *per_cpu; | |
3754 | ||
3755 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3756 | if (!nodes) |
3757 | return -ENOMEM; | |
81819f0f CL |
3758 | per_cpu = nodes + nr_node_ids; |
3759 | ||
205ab99d CL |
3760 | if (flags & SO_CPU) { |
3761 | int cpu; | |
81819f0f | 3762 | |
205ab99d | 3763 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 3764 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
dfb4f096 | 3765 | |
205ab99d CL |
3766 | if (!c || c->node < 0) |
3767 | continue; | |
3768 | ||
3769 | if (c->page) { | |
3770 | if (flags & SO_TOTAL) | |
3771 | x = c->page->objects; | |
3772 | else if (flags & SO_OBJECTS) | |
3773 | x = c->page->inuse; | |
81819f0f CL |
3774 | else |
3775 | x = 1; | |
205ab99d | 3776 | |
81819f0f | 3777 | total += x; |
205ab99d | 3778 | nodes[c->node] += x; |
81819f0f | 3779 | } |
205ab99d | 3780 | per_cpu[c->node]++; |
81819f0f CL |
3781 | } |
3782 | } | |
3783 | ||
ab4d5ed5 CL |
3784 | down_read(&slub_lock); |
3785 | #ifdef CONFIG_SLUB_DEBUG | |
205ab99d CL |
3786 | if (flags & SO_ALL) { |
3787 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3788 | struct kmem_cache_node *n = get_node(s, node); | |
3789 | ||
3790 | if (flags & SO_TOTAL) | |
3791 | x = atomic_long_read(&n->total_objects); | |
3792 | else if (flags & SO_OBJECTS) | |
3793 | x = atomic_long_read(&n->total_objects) - | |
3794 | count_partial(n, count_free); | |
81819f0f | 3795 | |
81819f0f | 3796 | else |
205ab99d | 3797 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3798 | total += x; |
3799 | nodes[node] += x; | |
3800 | } | |
3801 | ||
ab4d5ed5 CL |
3802 | } else |
3803 | #endif | |
3804 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
3805 | for_each_node_state(node, N_NORMAL_MEMORY) { |
3806 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3807 | |
205ab99d CL |
3808 | if (flags & SO_TOTAL) |
3809 | x = count_partial(n, count_total); | |
3810 | else if (flags & SO_OBJECTS) | |
3811 | x = count_partial(n, count_inuse); | |
81819f0f | 3812 | else |
205ab99d | 3813 | x = n->nr_partial; |
81819f0f CL |
3814 | total += x; |
3815 | nodes[node] += x; | |
3816 | } | |
3817 | } | |
81819f0f CL |
3818 | x = sprintf(buf, "%lu", total); |
3819 | #ifdef CONFIG_NUMA | |
f64dc58c | 3820 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3821 | if (nodes[node]) |
3822 | x += sprintf(buf + x, " N%d=%lu", | |
3823 | node, nodes[node]); | |
3824 | #endif | |
68cee4f1 | 3825 | up_read(&slub_lock); |
81819f0f CL |
3826 | kfree(nodes); |
3827 | return x + sprintf(buf + x, "\n"); | |
3828 | } | |
3829 | ||
ab4d5ed5 | 3830 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3831 | static int any_slab_objects(struct kmem_cache *s) |
3832 | { | |
3833 | int node; | |
81819f0f | 3834 | |
dfb4f096 | 3835 | for_each_online_node(node) { |
81819f0f CL |
3836 | struct kmem_cache_node *n = get_node(s, node); |
3837 | ||
dfb4f096 CL |
3838 | if (!n) |
3839 | continue; | |
3840 | ||
4ea33e2d | 3841 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3842 | return 1; |
3843 | } | |
3844 | return 0; | |
3845 | } | |
ab4d5ed5 | 3846 | #endif |
81819f0f CL |
3847 | |
3848 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3849 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3850 | ||
3851 | struct slab_attribute { | |
3852 | struct attribute attr; | |
3853 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3854 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3855 | }; | |
3856 | ||
3857 | #define SLAB_ATTR_RO(_name) \ | |
3858 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3859 | ||
3860 | #define SLAB_ATTR(_name) \ | |
3861 | static struct slab_attribute _name##_attr = \ | |
3862 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3863 | ||
81819f0f CL |
3864 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3865 | { | |
3866 | return sprintf(buf, "%d\n", s->size); | |
3867 | } | |
3868 | SLAB_ATTR_RO(slab_size); | |
3869 | ||
3870 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3871 | { | |
3872 | return sprintf(buf, "%d\n", s->align); | |
3873 | } | |
3874 | SLAB_ATTR_RO(align); | |
3875 | ||
3876 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3877 | { | |
3878 | return sprintf(buf, "%d\n", s->objsize); | |
3879 | } | |
3880 | SLAB_ATTR_RO(object_size); | |
3881 | ||
3882 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3883 | { | |
834f3d11 | 3884 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3885 | } |
3886 | SLAB_ATTR_RO(objs_per_slab); | |
3887 | ||
06b285dc CL |
3888 | static ssize_t order_store(struct kmem_cache *s, |
3889 | const char *buf, size_t length) | |
3890 | { | |
0121c619 CL |
3891 | unsigned long order; |
3892 | int err; | |
3893 | ||
3894 | err = strict_strtoul(buf, 10, &order); | |
3895 | if (err) | |
3896 | return err; | |
06b285dc CL |
3897 | |
3898 | if (order > slub_max_order || order < slub_min_order) | |
3899 | return -EINVAL; | |
3900 | ||
3901 | calculate_sizes(s, order); | |
3902 | return length; | |
3903 | } | |
3904 | ||
81819f0f CL |
3905 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
3906 | { | |
834f3d11 | 3907 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 3908 | } |
06b285dc | 3909 | SLAB_ATTR(order); |
81819f0f | 3910 | |
73d342b1 DR |
3911 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
3912 | { | |
3913 | return sprintf(buf, "%lu\n", s->min_partial); | |
3914 | } | |
3915 | ||
3916 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
3917 | size_t length) | |
3918 | { | |
3919 | unsigned long min; | |
3920 | int err; | |
3921 | ||
3922 | err = strict_strtoul(buf, 10, &min); | |
3923 | if (err) | |
3924 | return err; | |
3925 | ||
c0bdb232 | 3926 | set_min_partial(s, min); |
73d342b1 DR |
3927 | return length; |
3928 | } | |
3929 | SLAB_ATTR(min_partial); | |
3930 | ||
81819f0f CL |
3931 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
3932 | { | |
3933 | if (s->ctor) { | |
3934 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3935 | ||
3936 | return n + sprintf(buf + n, "\n"); | |
3937 | } | |
3938 | return 0; | |
3939 | } | |
3940 | SLAB_ATTR_RO(ctor); | |
3941 | ||
81819f0f CL |
3942 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3943 | { | |
3944 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3945 | } | |
3946 | SLAB_ATTR_RO(aliases); | |
3947 | ||
81819f0f CL |
3948 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
3949 | { | |
d9acf4b7 | 3950 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
3951 | } |
3952 | SLAB_ATTR_RO(partial); | |
3953 | ||
3954 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3955 | { | |
d9acf4b7 | 3956 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
3957 | } |
3958 | SLAB_ATTR_RO(cpu_slabs); | |
3959 | ||
3960 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3961 | { | |
205ab99d | 3962 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
3963 | } |
3964 | SLAB_ATTR_RO(objects); | |
3965 | ||
205ab99d CL |
3966 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
3967 | { | |
3968 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
3969 | } | |
3970 | SLAB_ATTR_RO(objects_partial); | |
3971 | ||
a5a84755 CL |
3972 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
3973 | { | |
3974 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3975 | } | |
3976 | ||
3977 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3978 | const char *buf, size_t length) | |
3979 | { | |
3980 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3981 | if (buf[0] == '1') | |
3982 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
3983 | return length; | |
3984 | } | |
3985 | SLAB_ATTR(reclaim_account); | |
3986 | ||
3987 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
3988 | { | |
3989 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
3990 | } | |
3991 | SLAB_ATTR_RO(hwcache_align); | |
3992 | ||
3993 | #ifdef CONFIG_ZONE_DMA | |
3994 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
3995 | { | |
3996 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
3997 | } | |
3998 | SLAB_ATTR_RO(cache_dma); | |
3999 | #endif | |
4000 | ||
4001 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4002 | { | |
4003 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4004 | } | |
4005 | SLAB_ATTR_RO(destroy_by_rcu); | |
4006 | ||
ab4d5ed5 | 4007 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4008 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4009 | { | |
4010 | return show_slab_objects(s, buf, SO_ALL); | |
4011 | } | |
4012 | SLAB_ATTR_RO(slabs); | |
4013 | ||
205ab99d CL |
4014 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4015 | { | |
4016 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4017 | } | |
4018 | SLAB_ATTR_RO(total_objects); | |
4019 | ||
81819f0f CL |
4020 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4021 | { | |
4022 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4023 | } | |
4024 | ||
4025 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4026 | const char *buf, size_t length) | |
4027 | { | |
4028 | s->flags &= ~SLAB_DEBUG_FREE; | |
4029 | if (buf[0] == '1') | |
4030 | s->flags |= SLAB_DEBUG_FREE; | |
4031 | return length; | |
4032 | } | |
4033 | SLAB_ATTR(sanity_checks); | |
4034 | ||
4035 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4036 | { | |
4037 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4038 | } | |
4039 | ||
4040 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4041 | size_t length) | |
4042 | { | |
4043 | s->flags &= ~SLAB_TRACE; | |
4044 | if (buf[0] == '1') | |
4045 | s->flags |= SLAB_TRACE; | |
4046 | return length; | |
4047 | } | |
4048 | SLAB_ATTR(trace); | |
4049 | ||
81819f0f CL |
4050 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4051 | { | |
4052 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4053 | } | |
4054 | ||
4055 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4056 | const char *buf, size_t length) | |
4057 | { | |
4058 | if (any_slab_objects(s)) | |
4059 | return -EBUSY; | |
4060 | ||
4061 | s->flags &= ~SLAB_RED_ZONE; | |
4062 | if (buf[0] == '1') | |
4063 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4064 | calculate_sizes(s, -1); |
81819f0f CL |
4065 | return length; |
4066 | } | |
4067 | SLAB_ATTR(red_zone); | |
4068 | ||
4069 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4070 | { | |
4071 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4072 | } | |
4073 | ||
4074 | static ssize_t poison_store(struct kmem_cache *s, | |
4075 | const char *buf, size_t length) | |
4076 | { | |
4077 | if (any_slab_objects(s)) | |
4078 | return -EBUSY; | |
4079 | ||
4080 | s->flags &= ~SLAB_POISON; | |
4081 | if (buf[0] == '1') | |
4082 | s->flags |= SLAB_POISON; | |
06b285dc | 4083 | calculate_sizes(s, -1); |
81819f0f CL |
4084 | return length; |
4085 | } | |
4086 | SLAB_ATTR(poison); | |
4087 | ||
4088 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4089 | { | |
4090 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4091 | } | |
4092 | ||
4093 | static ssize_t store_user_store(struct kmem_cache *s, | |
4094 | const char *buf, size_t length) | |
4095 | { | |
4096 | if (any_slab_objects(s)) | |
4097 | return -EBUSY; | |
4098 | ||
4099 | s->flags &= ~SLAB_STORE_USER; | |
4100 | if (buf[0] == '1') | |
4101 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4102 | calculate_sizes(s, -1); |
81819f0f CL |
4103 | return length; |
4104 | } | |
4105 | SLAB_ATTR(store_user); | |
4106 | ||
53e15af0 CL |
4107 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4108 | { | |
4109 | return 0; | |
4110 | } | |
4111 | ||
4112 | static ssize_t validate_store(struct kmem_cache *s, | |
4113 | const char *buf, size_t length) | |
4114 | { | |
434e245d CL |
4115 | int ret = -EINVAL; |
4116 | ||
4117 | if (buf[0] == '1') { | |
4118 | ret = validate_slab_cache(s); | |
4119 | if (ret >= 0) | |
4120 | ret = length; | |
4121 | } | |
4122 | return ret; | |
53e15af0 CL |
4123 | } |
4124 | SLAB_ATTR(validate); | |
a5a84755 CL |
4125 | |
4126 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4127 | { | |
4128 | if (!(s->flags & SLAB_STORE_USER)) | |
4129 | return -ENOSYS; | |
4130 | return list_locations(s, buf, TRACK_ALLOC); | |
4131 | } | |
4132 | SLAB_ATTR_RO(alloc_calls); | |
4133 | ||
4134 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4135 | { | |
4136 | if (!(s->flags & SLAB_STORE_USER)) | |
4137 | return -ENOSYS; | |
4138 | return list_locations(s, buf, TRACK_FREE); | |
4139 | } | |
4140 | SLAB_ATTR_RO(free_calls); | |
4141 | #endif /* CONFIG_SLUB_DEBUG */ | |
4142 | ||
4143 | #ifdef CONFIG_FAILSLAB | |
4144 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4145 | { | |
4146 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4147 | } | |
4148 | ||
4149 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4150 | size_t length) | |
4151 | { | |
4152 | s->flags &= ~SLAB_FAILSLAB; | |
4153 | if (buf[0] == '1') | |
4154 | s->flags |= SLAB_FAILSLAB; | |
4155 | return length; | |
4156 | } | |
4157 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4158 | #endif |
53e15af0 | 4159 | |
2086d26a CL |
4160 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4161 | { | |
4162 | return 0; | |
4163 | } | |
4164 | ||
4165 | static ssize_t shrink_store(struct kmem_cache *s, | |
4166 | const char *buf, size_t length) | |
4167 | { | |
4168 | if (buf[0] == '1') { | |
4169 | int rc = kmem_cache_shrink(s); | |
4170 | ||
4171 | if (rc) | |
4172 | return rc; | |
4173 | } else | |
4174 | return -EINVAL; | |
4175 | return length; | |
4176 | } | |
4177 | SLAB_ATTR(shrink); | |
4178 | ||
81819f0f | 4179 | #ifdef CONFIG_NUMA |
9824601e | 4180 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4181 | { |
9824601e | 4182 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4183 | } |
4184 | ||
9824601e | 4185 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4186 | const char *buf, size_t length) |
4187 | { | |
0121c619 CL |
4188 | unsigned long ratio; |
4189 | int err; | |
4190 | ||
4191 | err = strict_strtoul(buf, 10, &ratio); | |
4192 | if (err) | |
4193 | return err; | |
4194 | ||
e2cb96b7 | 4195 | if (ratio <= 100) |
0121c619 | 4196 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4197 | |
81819f0f CL |
4198 | return length; |
4199 | } | |
9824601e | 4200 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4201 | #endif |
4202 | ||
8ff12cfc | 4203 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4204 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4205 | { | |
4206 | unsigned long sum = 0; | |
4207 | int cpu; | |
4208 | int len; | |
4209 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4210 | ||
4211 | if (!data) | |
4212 | return -ENOMEM; | |
4213 | ||
4214 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4215 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4216 | |
4217 | data[cpu] = x; | |
4218 | sum += x; | |
4219 | } | |
4220 | ||
4221 | len = sprintf(buf, "%lu", sum); | |
4222 | ||
50ef37b9 | 4223 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4224 | for_each_online_cpu(cpu) { |
4225 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4226 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4227 | } |
50ef37b9 | 4228 | #endif |
8ff12cfc CL |
4229 | kfree(data); |
4230 | return len + sprintf(buf + len, "\n"); | |
4231 | } | |
4232 | ||
78eb00cc DR |
4233 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4234 | { | |
4235 | int cpu; | |
4236 | ||
4237 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4238 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4239 | } |
4240 | ||
8ff12cfc CL |
4241 | #define STAT_ATTR(si, text) \ |
4242 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4243 | { \ | |
4244 | return show_stat(s, buf, si); \ | |
4245 | } \ | |
78eb00cc DR |
4246 | static ssize_t text##_store(struct kmem_cache *s, \ |
4247 | const char *buf, size_t length) \ | |
4248 | { \ | |
4249 | if (buf[0] != '0') \ | |
4250 | return -EINVAL; \ | |
4251 | clear_stat(s, si); \ | |
4252 | return length; \ | |
4253 | } \ | |
4254 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4255 | |
4256 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4257 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4258 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4259 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4260 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4261 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4262 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4263 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4264 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4265 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4266 | STAT_ATTR(FREE_SLAB, free_slab); | |
4267 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4268 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4269 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4270 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4271 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4272 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4273 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4274 | #endif |
4275 | ||
06428780 | 4276 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4277 | &slab_size_attr.attr, |
4278 | &object_size_attr.attr, | |
4279 | &objs_per_slab_attr.attr, | |
4280 | &order_attr.attr, | |
73d342b1 | 4281 | &min_partial_attr.attr, |
81819f0f | 4282 | &objects_attr.attr, |
205ab99d | 4283 | &objects_partial_attr.attr, |
81819f0f CL |
4284 | &partial_attr.attr, |
4285 | &cpu_slabs_attr.attr, | |
4286 | &ctor_attr.attr, | |
81819f0f CL |
4287 | &aliases_attr.attr, |
4288 | &align_attr.attr, | |
81819f0f CL |
4289 | &hwcache_align_attr.attr, |
4290 | &reclaim_account_attr.attr, | |
4291 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4292 | &shrink_attr.attr, |
ab4d5ed5 | 4293 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4294 | &total_objects_attr.attr, |
4295 | &slabs_attr.attr, | |
4296 | &sanity_checks_attr.attr, | |
4297 | &trace_attr.attr, | |
81819f0f CL |
4298 | &red_zone_attr.attr, |
4299 | &poison_attr.attr, | |
4300 | &store_user_attr.attr, | |
53e15af0 | 4301 | &validate_attr.attr, |
88a420e4 CL |
4302 | &alloc_calls_attr.attr, |
4303 | &free_calls_attr.attr, | |
ab4d5ed5 | 4304 | #endif |
81819f0f CL |
4305 | #ifdef CONFIG_ZONE_DMA |
4306 | &cache_dma_attr.attr, | |
4307 | #endif | |
4308 | #ifdef CONFIG_NUMA | |
9824601e | 4309 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4310 | #endif |
4311 | #ifdef CONFIG_SLUB_STATS | |
4312 | &alloc_fastpath_attr.attr, | |
4313 | &alloc_slowpath_attr.attr, | |
4314 | &free_fastpath_attr.attr, | |
4315 | &free_slowpath_attr.attr, | |
4316 | &free_frozen_attr.attr, | |
4317 | &free_add_partial_attr.attr, | |
4318 | &free_remove_partial_attr.attr, | |
4319 | &alloc_from_partial_attr.attr, | |
4320 | &alloc_slab_attr.attr, | |
4321 | &alloc_refill_attr.attr, | |
4322 | &free_slab_attr.attr, | |
4323 | &cpuslab_flush_attr.attr, | |
4324 | &deactivate_full_attr.attr, | |
4325 | &deactivate_empty_attr.attr, | |
4326 | &deactivate_to_head_attr.attr, | |
4327 | &deactivate_to_tail_attr.attr, | |
4328 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4329 | &order_fallback_attr.attr, |
81819f0f | 4330 | #endif |
4c13dd3b DM |
4331 | #ifdef CONFIG_FAILSLAB |
4332 | &failslab_attr.attr, | |
4333 | #endif | |
4334 | ||
81819f0f CL |
4335 | NULL |
4336 | }; | |
4337 | ||
4338 | static struct attribute_group slab_attr_group = { | |
4339 | .attrs = slab_attrs, | |
4340 | }; | |
4341 | ||
4342 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4343 | struct attribute *attr, | |
4344 | char *buf) | |
4345 | { | |
4346 | struct slab_attribute *attribute; | |
4347 | struct kmem_cache *s; | |
4348 | int err; | |
4349 | ||
4350 | attribute = to_slab_attr(attr); | |
4351 | s = to_slab(kobj); | |
4352 | ||
4353 | if (!attribute->show) | |
4354 | return -EIO; | |
4355 | ||
4356 | err = attribute->show(s, buf); | |
4357 | ||
4358 | return err; | |
4359 | } | |
4360 | ||
4361 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4362 | struct attribute *attr, | |
4363 | const char *buf, size_t len) | |
4364 | { | |
4365 | struct slab_attribute *attribute; | |
4366 | struct kmem_cache *s; | |
4367 | int err; | |
4368 | ||
4369 | attribute = to_slab_attr(attr); | |
4370 | s = to_slab(kobj); | |
4371 | ||
4372 | if (!attribute->store) | |
4373 | return -EIO; | |
4374 | ||
4375 | err = attribute->store(s, buf, len); | |
4376 | ||
4377 | return err; | |
4378 | } | |
4379 | ||
151c602f CL |
4380 | static void kmem_cache_release(struct kobject *kobj) |
4381 | { | |
4382 | struct kmem_cache *s = to_slab(kobj); | |
4383 | ||
84c1cf62 | 4384 | kfree(s->name); |
151c602f CL |
4385 | kfree(s); |
4386 | } | |
4387 | ||
52cf25d0 | 4388 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
4389 | .show = slab_attr_show, |
4390 | .store = slab_attr_store, | |
4391 | }; | |
4392 | ||
4393 | static struct kobj_type slab_ktype = { | |
4394 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4395 | .release = kmem_cache_release |
81819f0f CL |
4396 | }; |
4397 | ||
4398 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4399 | { | |
4400 | struct kobj_type *ktype = get_ktype(kobj); | |
4401 | ||
4402 | if (ktype == &slab_ktype) | |
4403 | return 1; | |
4404 | return 0; | |
4405 | } | |
4406 | ||
9cd43611 | 4407 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
4408 | .filter = uevent_filter, |
4409 | }; | |
4410 | ||
27c3a314 | 4411 | static struct kset *slab_kset; |
81819f0f CL |
4412 | |
4413 | #define ID_STR_LENGTH 64 | |
4414 | ||
4415 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4416 | * |
4417 | * Format :[flags-]size | |
81819f0f CL |
4418 | */ |
4419 | static char *create_unique_id(struct kmem_cache *s) | |
4420 | { | |
4421 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4422 | char *p = name; | |
4423 | ||
4424 | BUG_ON(!name); | |
4425 | ||
4426 | *p++ = ':'; | |
4427 | /* | |
4428 | * First flags affecting slabcache operations. We will only | |
4429 | * get here for aliasable slabs so we do not need to support | |
4430 | * too many flags. The flags here must cover all flags that | |
4431 | * are matched during merging to guarantee that the id is | |
4432 | * unique. | |
4433 | */ | |
4434 | if (s->flags & SLAB_CACHE_DMA) | |
4435 | *p++ = 'd'; | |
4436 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4437 | *p++ = 'a'; | |
4438 | if (s->flags & SLAB_DEBUG_FREE) | |
4439 | *p++ = 'F'; | |
5a896d9e VN |
4440 | if (!(s->flags & SLAB_NOTRACK)) |
4441 | *p++ = 't'; | |
81819f0f CL |
4442 | if (p != name + 1) |
4443 | *p++ = '-'; | |
4444 | p += sprintf(p, "%07d", s->size); | |
4445 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4446 | return name; | |
4447 | } | |
4448 | ||
4449 | static int sysfs_slab_add(struct kmem_cache *s) | |
4450 | { | |
4451 | int err; | |
4452 | const char *name; | |
4453 | int unmergeable; | |
4454 | ||
4455 | if (slab_state < SYSFS) | |
4456 | /* Defer until later */ | |
4457 | return 0; | |
4458 | ||
4459 | unmergeable = slab_unmergeable(s); | |
4460 | if (unmergeable) { | |
4461 | /* | |
4462 | * Slabcache can never be merged so we can use the name proper. | |
4463 | * This is typically the case for debug situations. In that | |
4464 | * case we can catch duplicate names easily. | |
4465 | */ | |
27c3a314 | 4466 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4467 | name = s->name; |
4468 | } else { | |
4469 | /* | |
4470 | * Create a unique name for the slab as a target | |
4471 | * for the symlinks. | |
4472 | */ | |
4473 | name = create_unique_id(s); | |
4474 | } | |
4475 | ||
27c3a314 | 4476 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4477 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4478 | if (err) { | |
4479 | kobject_put(&s->kobj); | |
81819f0f | 4480 | return err; |
1eada11c | 4481 | } |
81819f0f CL |
4482 | |
4483 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
4484 | if (err) { |
4485 | kobject_del(&s->kobj); | |
4486 | kobject_put(&s->kobj); | |
81819f0f | 4487 | return err; |
5788d8ad | 4488 | } |
81819f0f CL |
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 | { | |
2bce6485 CL |
4500 | if (slab_state < SYSFS) |
4501 | /* | |
4502 | * Sysfs has not been setup yet so no need to remove the | |
4503 | * cache from sysfs. | |
4504 | */ | |
4505 | return; | |
4506 | ||
81819f0f CL |
4507 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
4508 | kobject_del(&s->kobj); | |
151c602f | 4509 | kobject_put(&s->kobj); |
81819f0f CL |
4510 | } |
4511 | ||
4512 | /* | |
4513 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4514 | * available lest we lose that information. |
81819f0f CL |
4515 | */ |
4516 | struct saved_alias { | |
4517 | struct kmem_cache *s; | |
4518 | const char *name; | |
4519 | struct saved_alias *next; | |
4520 | }; | |
4521 | ||
5af328a5 | 4522 | static struct saved_alias *alias_list; |
81819f0f CL |
4523 | |
4524 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4525 | { | |
4526 | struct saved_alias *al; | |
4527 | ||
4528 | if (slab_state == SYSFS) { | |
4529 | /* | |
4530 | * If we have a leftover link then remove it. | |
4531 | */ | |
27c3a314 GKH |
4532 | sysfs_remove_link(&slab_kset->kobj, name); |
4533 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4534 | } |
4535 | ||
4536 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4537 | if (!al) | |
4538 | return -ENOMEM; | |
4539 | ||
4540 | al->s = s; | |
4541 | al->name = name; | |
4542 | al->next = alias_list; | |
4543 | alias_list = al; | |
4544 | return 0; | |
4545 | } | |
4546 | ||
4547 | static int __init slab_sysfs_init(void) | |
4548 | { | |
5b95a4ac | 4549 | struct kmem_cache *s; |
81819f0f CL |
4550 | int err; |
4551 | ||
2bce6485 CL |
4552 | down_write(&slub_lock); |
4553 | ||
0ff21e46 | 4554 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4555 | if (!slab_kset) { |
2bce6485 | 4556 | up_write(&slub_lock); |
81819f0f CL |
4557 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4558 | return -ENOSYS; | |
4559 | } | |
4560 | ||
26a7bd03 CL |
4561 | slab_state = SYSFS; |
4562 | ||
5b95a4ac | 4563 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4564 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4565 | if (err) |
4566 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4567 | " to sysfs\n", s->name); | |
26a7bd03 | 4568 | } |
81819f0f CL |
4569 | |
4570 | while (alias_list) { | |
4571 | struct saved_alias *al = alias_list; | |
4572 | ||
4573 | alias_list = alias_list->next; | |
4574 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4575 | if (err) |
4576 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4577 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4578 | kfree(al); |
4579 | } | |
4580 | ||
2bce6485 | 4581 | up_write(&slub_lock); |
81819f0f CL |
4582 | resiliency_test(); |
4583 | return 0; | |
4584 | } | |
4585 | ||
4586 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 4587 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
4588 | |
4589 | /* | |
4590 | * The /proc/slabinfo ABI | |
4591 | */ | |
158a9624 | 4592 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4593 | static void print_slabinfo_header(struct seq_file *m) |
4594 | { | |
4595 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4596 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4597 | "<objperslab> <pagesperslab>"); | |
4598 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4599 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4600 | seq_putc(m, '\n'); | |
4601 | } | |
4602 | ||
4603 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4604 | { | |
4605 | loff_t n = *pos; | |
4606 | ||
4607 | down_read(&slub_lock); | |
4608 | if (!n) | |
4609 | print_slabinfo_header(m); | |
4610 | ||
4611 | return seq_list_start(&slab_caches, *pos); | |
4612 | } | |
4613 | ||
4614 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4615 | { | |
4616 | return seq_list_next(p, &slab_caches, pos); | |
4617 | } | |
4618 | ||
4619 | static void s_stop(struct seq_file *m, void *p) | |
4620 | { | |
4621 | up_read(&slub_lock); | |
4622 | } | |
4623 | ||
4624 | static int s_show(struct seq_file *m, void *p) | |
4625 | { | |
4626 | unsigned long nr_partials = 0; | |
4627 | unsigned long nr_slabs = 0; | |
4628 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4629 | unsigned long nr_objs = 0; |
4630 | unsigned long nr_free = 0; | |
57ed3eda PE |
4631 | struct kmem_cache *s; |
4632 | int node; | |
4633 | ||
4634 | s = list_entry(p, struct kmem_cache, list); | |
4635 | ||
4636 | for_each_online_node(node) { | |
4637 | struct kmem_cache_node *n = get_node(s, node); | |
4638 | ||
4639 | if (!n) | |
4640 | continue; | |
4641 | ||
4642 | nr_partials += n->nr_partial; | |
4643 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4644 | nr_objs += atomic_long_read(&n->total_objects); |
4645 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4646 | } |
4647 | ||
205ab99d | 4648 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4649 | |
4650 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4651 | nr_objs, s->size, oo_objects(s->oo), |
4652 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4653 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4654 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4655 | 0UL); | |
4656 | seq_putc(m, '\n'); | |
4657 | return 0; | |
4658 | } | |
4659 | ||
7b3c3a50 | 4660 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4661 | .start = s_start, |
4662 | .next = s_next, | |
4663 | .stop = s_stop, | |
4664 | .show = s_show, | |
4665 | }; | |
4666 | ||
7b3c3a50 AD |
4667 | static int slabinfo_open(struct inode *inode, struct file *file) |
4668 | { | |
4669 | return seq_open(file, &slabinfo_op); | |
4670 | } | |
4671 | ||
4672 | static const struct file_operations proc_slabinfo_operations = { | |
4673 | .open = slabinfo_open, | |
4674 | .read = seq_read, | |
4675 | .llseek = seq_lseek, | |
4676 | .release = seq_release, | |
4677 | }; | |
4678 | ||
4679 | static int __init slab_proc_init(void) | |
4680 | { | |
cf5d1131 | 4681 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
4682 | return 0; |
4683 | } | |
4684 | module_init(slab_proc_init); | |
158a9624 | 4685 | #endif /* CONFIG_SLABINFO */ |