]>
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 | 30 | |
4a92379b RK |
31 | #include <trace/events/kmem.h> |
32 | ||
81819f0f CL |
33 | /* |
34 | * Lock order: | |
35 | * 1. slab_lock(page) | |
36 | * 2. slab->list_lock | |
37 | * | |
38 | * The slab_lock protects operations on the object of a particular | |
39 | * slab and its metadata in the page struct. If the slab lock | |
40 | * has been taken then no allocations nor frees can be performed | |
41 | * on the objects in the slab nor can the slab be added or removed | |
42 | * from the partial or full lists since this would mean modifying | |
43 | * the page_struct of the slab. | |
44 | * | |
45 | * The list_lock protects the partial and full list on each node and | |
46 | * the partial slab counter. If taken then no new slabs may be added or | |
47 | * removed from the lists nor make the number of partial slabs be modified. | |
48 | * (Note that the total number of slabs is an atomic value that may be | |
49 | * modified without taking the list lock). | |
50 | * | |
51 | * The list_lock is a centralized lock and thus we avoid taking it as | |
52 | * much as possible. As long as SLUB does not have to handle partial | |
53 | * slabs, operations can continue without any centralized lock. F.e. | |
54 | * allocating a long series of objects that fill up slabs does not require | |
55 | * the list lock. | |
56 | * | |
57 | * The lock order is sometimes inverted when we are trying to get a slab | |
58 | * off a list. We take the list_lock and then look for a page on the list | |
59 | * to use. While we do that objects in the slabs may be freed. We can | |
60 | * only operate on the slab if we have also taken the slab_lock. So we use | |
61 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
62 | * can occur anymore and we can use the slab for allocations etc. If the | |
63 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
64 | * we must stay away from it for a while since we may cause a bouncing | |
65 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
66 | * If all pages are busy then we may allocate a new slab instead of reusing | |
67 | * a partial slab. A new slab has noone operating on it and thus there is | |
68 | * no danger of cacheline contention. | |
69 | * | |
70 | * Interrupts are disabled during allocation and deallocation in order to | |
71 | * make the slab allocator safe to use in the context of an irq. In addition | |
72 | * interrupts are disabled to ensure that the processor does not change | |
73 | * while handling per_cpu slabs, due to kernel preemption. | |
74 | * | |
75 | * SLUB assigns one slab for allocation to each processor. | |
76 | * Allocations only occur from these slabs called cpu slabs. | |
77 | * | |
672bba3a CL |
78 | * Slabs with free elements are kept on a partial list and during regular |
79 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 80 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
81 | * We track full slabs for debugging purposes though because otherwise we |
82 | * cannot scan all objects. | |
81819f0f CL |
83 | * |
84 | * Slabs are freed when they become empty. Teardown and setup is | |
85 | * minimal so we rely on the page allocators per cpu caches for | |
86 | * fast frees and allocs. | |
87 | * | |
88 | * Overloading of page flags that are otherwise used for LRU management. | |
89 | * | |
4b6f0750 CL |
90 | * PageActive The slab is frozen and exempt from list processing. |
91 | * This means that the slab is dedicated to a purpose | |
92 | * such as satisfying allocations for a specific | |
93 | * processor. Objects may be freed in the slab while | |
94 | * it is frozen but slab_free will then skip the usual | |
95 | * list operations. It is up to the processor holding | |
96 | * the slab to integrate the slab into the slab lists | |
97 | * when the slab is no longer needed. | |
98 | * | |
99 | * One use of this flag is to mark slabs that are | |
100 | * used for allocations. Then such a slab becomes a cpu | |
101 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 102 | * freelist that allows lockless access to |
894b8788 CL |
103 | * free objects in addition to the regular freelist |
104 | * that requires the slab lock. | |
81819f0f CL |
105 | * |
106 | * PageError Slab requires special handling due to debug | |
107 | * options set. This moves slab handling out of | |
894b8788 | 108 | * the fast path and disables lockless freelists. |
81819f0f CL |
109 | */ |
110 | ||
af537b0a CL |
111 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
112 | SLAB_TRACE | SLAB_DEBUG_FREE) | |
113 | ||
114 | static inline int kmem_cache_debug(struct kmem_cache *s) | |
115 | { | |
5577bd8a | 116 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 117 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 118 | #else |
af537b0a | 119 | return 0; |
5577bd8a | 120 | #endif |
af537b0a | 121 | } |
5577bd8a | 122 | |
81819f0f CL |
123 | /* |
124 | * Issues still to be resolved: | |
125 | * | |
81819f0f CL |
126 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
127 | * | |
81819f0f CL |
128 | * - Variable sizing of the per node arrays |
129 | */ | |
130 | ||
131 | /* Enable to test recovery from slab corruption on boot */ | |
132 | #undef SLUB_RESILIENCY_TEST | |
133 | ||
2086d26a CL |
134 | /* |
135 | * Mininum number of partial slabs. These will be left on the partial | |
136 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
137 | */ | |
76be8950 | 138 | #define MIN_PARTIAL 5 |
e95eed57 | 139 | |
2086d26a CL |
140 | /* |
141 | * Maximum number of desirable partial slabs. | |
142 | * The existence of more partial slabs makes kmem_cache_shrink | |
143 | * sort the partial list by the number of objects in the. | |
144 | */ | |
145 | #define MAX_PARTIAL 10 | |
146 | ||
81819f0f CL |
147 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
148 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 149 | |
fa5ec8a1 | 150 | /* |
3de47213 DR |
151 | * Debugging flags that require metadata to be stored in the slab. These get |
152 | * disabled when slub_debug=O is used and a cache's min order increases with | |
153 | * metadata. | |
fa5ec8a1 | 154 | */ |
3de47213 | 155 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 156 | |
81819f0f CL |
157 | /* |
158 | * Set of flags that will prevent slab merging | |
159 | */ | |
160 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
161 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
162 | SLAB_FAILSLAB) | |
81819f0f CL |
163 | |
164 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 165 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 166 | |
210b5c06 CG |
167 | #define OO_SHIFT 16 |
168 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
169 | #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */ | |
170 | ||
81819f0f | 171 | /* Internal SLUB flags */ |
f90ec390 | 172 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
81819f0f CL |
173 | |
174 | static int kmem_size = sizeof(struct kmem_cache); | |
175 | ||
176 | #ifdef CONFIG_SMP | |
177 | static struct notifier_block slab_notifier; | |
178 | #endif | |
179 | ||
180 | static enum { | |
181 | DOWN, /* No slab functionality available */ | |
51df1142 | 182 | PARTIAL, /* Kmem_cache_node works */ |
672bba3a | 183 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
184 | SYSFS /* Sysfs up */ |
185 | } slab_state = DOWN; | |
186 | ||
187 | /* A list of all slab caches on the system */ | |
188 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 189 | static LIST_HEAD(slab_caches); |
81819f0f | 190 | |
02cbc874 CL |
191 | /* |
192 | * Tracking user of a slab. | |
193 | */ | |
194 | struct track { | |
ce71e27c | 195 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
196 | int cpu; /* Was running on cpu */ |
197 | int pid; /* Pid context */ | |
198 | unsigned long when; /* When did the operation occur */ | |
199 | }; | |
200 | ||
201 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
202 | ||
ab4d5ed5 | 203 | #ifdef CONFIG_SYSFS |
81819f0f CL |
204 | static int sysfs_slab_add(struct kmem_cache *); |
205 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
206 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 207 | |
81819f0f | 208 | #else |
0c710013 CL |
209 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
210 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
211 | { return 0; } | |
151c602f CL |
212 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
213 | { | |
84c1cf62 | 214 | kfree(s->name); |
151c602f CL |
215 | kfree(s); |
216 | } | |
8ff12cfc | 217 | |
81819f0f CL |
218 | #endif |
219 | ||
84e554e6 | 220 | static inline void stat(struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
221 | { |
222 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 223 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
224 | #endif |
225 | } | |
226 | ||
81819f0f CL |
227 | /******************************************************************** |
228 | * Core slab cache functions | |
229 | *******************************************************************/ | |
230 | ||
231 | int slab_is_available(void) | |
232 | { | |
233 | return slab_state >= UP; | |
234 | } | |
235 | ||
236 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
237 | { | |
81819f0f | 238 | return s->node[node]; |
81819f0f CL |
239 | } |
240 | ||
6446faa2 | 241 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
242 | static inline int check_valid_pointer(struct kmem_cache *s, |
243 | struct page *page, const void *object) | |
244 | { | |
245 | void *base; | |
246 | ||
a973e9dd | 247 | if (!object) |
02cbc874 CL |
248 | return 1; |
249 | ||
a973e9dd | 250 | base = page_address(page); |
39b26464 | 251 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
252 | (object - base) % s->size) { |
253 | return 0; | |
254 | } | |
255 | ||
256 | return 1; | |
257 | } | |
258 | ||
7656c72b CL |
259 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
260 | { | |
261 | return *(void **)(object + s->offset); | |
262 | } | |
263 | ||
264 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
265 | { | |
266 | *(void **)(object + s->offset) = fp; | |
267 | } | |
268 | ||
269 | /* Loop over all objects in a slab */ | |
224a88be CL |
270 | #define for_each_object(__p, __s, __addr, __objects) \ |
271 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
272 | __p += (__s)->size) |
273 | ||
274 | /* Scan freelist */ | |
275 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 276 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
277 | |
278 | /* Determine object index from a given position */ | |
279 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
280 | { | |
281 | return (p - addr) / s->size; | |
282 | } | |
283 | ||
834f3d11 CL |
284 | static inline struct kmem_cache_order_objects oo_make(int order, |
285 | unsigned long size) | |
286 | { | |
287 | struct kmem_cache_order_objects x = { | |
210b5c06 | 288 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size |
834f3d11 CL |
289 | }; |
290 | ||
291 | return x; | |
292 | } | |
293 | ||
294 | static inline int oo_order(struct kmem_cache_order_objects x) | |
295 | { | |
210b5c06 | 296 | return x.x >> OO_SHIFT; |
834f3d11 CL |
297 | } |
298 | ||
299 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
300 | { | |
210b5c06 | 301 | return x.x & OO_MASK; |
834f3d11 CL |
302 | } |
303 | ||
41ecc55b CL |
304 | #ifdef CONFIG_SLUB_DEBUG |
305 | /* | |
306 | * Debug settings: | |
307 | */ | |
f0630fff CL |
308 | #ifdef CONFIG_SLUB_DEBUG_ON |
309 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
310 | #else | |
41ecc55b | 311 | static int slub_debug; |
f0630fff | 312 | #endif |
41ecc55b CL |
313 | |
314 | static char *slub_debug_slabs; | |
fa5ec8a1 | 315 | static int disable_higher_order_debug; |
41ecc55b | 316 | |
81819f0f CL |
317 | /* |
318 | * Object debugging | |
319 | */ | |
320 | static void print_section(char *text, u8 *addr, unsigned int length) | |
321 | { | |
322 | int i, offset; | |
323 | int newline = 1; | |
324 | char ascii[17]; | |
325 | ||
326 | ascii[16] = 0; | |
327 | ||
328 | for (i = 0; i < length; i++) { | |
329 | if (newline) { | |
24922684 | 330 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
331 | newline = 0; |
332 | } | |
06428780 | 333 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
334 | offset = i % 16; |
335 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
336 | if (offset == 15) { | |
06428780 | 337 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
338 | newline = 1; |
339 | } | |
340 | } | |
341 | if (!newline) { | |
342 | i %= 16; | |
343 | while (i < 16) { | |
06428780 | 344 | printk(KERN_CONT " "); |
81819f0f CL |
345 | ascii[i] = ' '; |
346 | i++; | |
347 | } | |
06428780 | 348 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
349 | } |
350 | } | |
351 | ||
81819f0f CL |
352 | static struct track *get_track(struct kmem_cache *s, void *object, |
353 | enum track_item alloc) | |
354 | { | |
355 | struct track *p; | |
356 | ||
357 | if (s->offset) | |
358 | p = object + s->offset + sizeof(void *); | |
359 | else | |
360 | p = object + s->inuse; | |
361 | ||
362 | return p + alloc; | |
363 | } | |
364 | ||
365 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 366 | enum track_item alloc, unsigned long addr) |
81819f0f | 367 | { |
1a00df4a | 368 | struct track *p = get_track(s, object, alloc); |
81819f0f | 369 | |
81819f0f CL |
370 | if (addr) { |
371 | p->addr = addr; | |
372 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 373 | p->pid = current->pid; |
81819f0f CL |
374 | p->when = jiffies; |
375 | } else | |
376 | memset(p, 0, sizeof(struct track)); | |
377 | } | |
378 | ||
81819f0f CL |
379 | static void init_tracking(struct kmem_cache *s, void *object) |
380 | { | |
24922684 CL |
381 | if (!(s->flags & SLAB_STORE_USER)) |
382 | return; | |
383 | ||
ce71e27c EGM |
384 | set_track(s, object, TRACK_FREE, 0UL); |
385 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
386 | } |
387 | ||
388 | static void print_track(const char *s, struct track *t) | |
389 | { | |
390 | if (!t->addr) | |
391 | return; | |
392 | ||
7daf705f | 393 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 394 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
395 | } |
396 | ||
397 | static void print_tracking(struct kmem_cache *s, void *object) | |
398 | { | |
399 | if (!(s->flags & SLAB_STORE_USER)) | |
400 | return; | |
401 | ||
402 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
403 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
404 | } | |
405 | ||
406 | static void print_page_info(struct page *page) | |
407 | { | |
39b26464 CL |
408 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
409 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
410 | |
411 | } | |
412 | ||
413 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
414 | { | |
415 | va_list args; | |
416 | char buf[100]; | |
417 | ||
418 | va_start(args, fmt); | |
419 | vsnprintf(buf, sizeof(buf), fmt, args); | |
420 | va_end(args); | |
421 | printk(KERN_ERR "========================================" | |
422 | "=====================================\n"); | |
423 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
424 | printk(KERN_ERR "----------------------------------------" | |
425 | "-------------------------------------\n\n"); | |
81819f0f CL |
426 | } |
427 | ||
24922684 CL |
428 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
429 | { | |
430 | va_list args; | |
431 | char buf[100]; | |
432 | ||
433 | va_start(args, fmt); | |
434 | vsnprintf(buf, sizeof(buf), fmt, args); | |
435 | va_end(args); | |
436 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
437 | } | |
438 | ||
439 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
440 | { |
441 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 442 | u8 *addr = page_address(page); |
24922684 CL |
443 | |
444 | print_tracking(s, p); | |
445 | ||
446 | print_page_info(page); | |
447 | ||
448 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
449 | p, p - addr, get_freepointer(s, p)); | |
450 | ||
451 | if (p > addr + 16) | |
452 | print_section("Bytes b4", p - 16, 16); | |
453 | ||
0ebd652b | 454 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
455 | |
456 | if (s->flags & SLAB_RED_ZONE) | |
457 | print_section("Redzone", p + s->objsize, | |
458 | s->inuse - s->objsize); | |
459 | ||
81819f0f CL |
460 | if (s->offset) |
461 | off = s->offset + sizeof(void *); | |
462 | else | |
463 | off = s->inuse; | |
464 | ||
24922684 | 465 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 466 | off += 2 * sizeof(struct track); |
81819f0f CL |
467 | |
468 | if (off != s->size) | |
469 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
470 | print_section("Padding", p + off, s->size - off); |
471 | ||
472 | dump_stack(); | |
81819f0f CL |
473 | } |
474 | ||
475 | static void object_err(struct kmem_cache *s, struct page *page, | |
476 | u8 *object, char *reason) | |
477 | { | |
3dc50637 | 478 | slab_bug(s, "%s", reason); |
24922684 | 479 | print_trailer(s, page, object); |
81819f0f CL |
480 | } |
481 | ||
24922684 | 482 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
483 | { |
484 | va_list args; | |
485 | char buf[100]; | |
486 | ||
24922684 CL |
487 | va_start(args, fmt); |
488 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 489 | va_end(args); |
3dc50637 | 490 | slab_bug(s, "%s", buf); |
24922684 | 491 | print_page_info(page); |
81819f0f CL |
492 | dump_stack(); |
493 | } | |
494 | ||
f7cb1933 | 495 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
496 | { |
497 | u8 *p = object; | |
498 | ||
499 | if (s->flags & __OBJECT_POISON) { | |
500 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 501 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
502 | } |
503 | ||
504 | if (s->flags & SLAB_RED_ZONE) | |
f7cb1933 | 505 | memset(p + s->objsize, val, s->inuse - s->objsize); |
81819f0f CL |
506 | } |
507 | ||
24922684 | 508 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
509 | { |
510 | while (bytes) { | |
511 | if (*start != (u8)value) | |
24922684 | 512 | return start; |
81819f0f CL |
513 | start++; |
514 | bytes--; | |
515 | } | |
24922684 CL |
516 | return NULL; |
517 | } | |
518 | ||
519 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
520 | void *from, void *to) | |
521 | { | |
522 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
523 | memset(from, data, to - from); | |
524 | } | |
525 | ||
526 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
527 | u8 *object, char *what, | |
06428780 | 528 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
529 | { |
530 | u8 *fault; | |
531 | u8 *end; | |
532 | ||
533 | fault = check_bytes(start, value, bytes); | |
534 | if (!fault) | |
535 | return 1; | |
536 | ||
537 | end = start + bytes; | |
538 | while (end > fault && end[-1] == value) | |
539 | end--; | |
540 | ||
541 | slab_bug(s, "%s overwritten", what); | |
542 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
543 | fault, end - 1, fault[0], value); | |
544 | print_trailer(s, page, object); | |
545 | ||
546 | restore_bytes(s, what, value, fault, end); | |
547 | return 0; | |
81819f0f CL |
548 | } |
549 | ||
81819f0f CL |
550 | /* |
551 | * Object layout: | |
552 | * | |
553 | * object address | |
554 | * Bytes of the object to be managed. | |
555 | * If the freepointer may overlay the object then the free | |
556 | * pointer is the first word of the object. | |
672bba3a | 557 | * |
81819f0f CL |
558 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
559 | * 0xa5 (POISON_END) | |
560 | * | |
561 | * object + s->objsize | |
562 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
563 | * Padding is extended by another word if Redzoning is enabled and |
564 | * objsize == inuse. | |
565 | * | |
81819f0f CL |
566 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
567 | * 0xcc (RED_ACTIVE) for objects in use. | |
568 | * | |
569 | * object + s->inuse | |
672bba3a CL |
570 | * Meta data starts here. |
571 | * | |
81819f0f CL |
572 | * A. Free pointer (if we cannot overwrite object on free) |
573 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 574 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 575 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
576 | * before the word boundary. |
577 | * | |
578 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
579 | * |
580 | * object + s->size | |
672bba3a | 581 | * Nothing is used beyond s->size. |
81819f0f | 582 | * |
672bba3a CL |
583 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
584 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
585 | * may be used with merged slabcaches. |
586 | */ | |
587 | ||
81819f0f CL |
588 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
589 | { | |
590 | unsigned long off = s->inuse; /* The end of info */ | |
591 | ||
592 | if (s->offset) | |
593 | /* Freepointer is placed after the object. */ | |
594 | off += sizeof(void *); | |
595 | ||
596 | if (s->flags & SLAB_STORE_USER) | |
597 | /* We also have user information there */ | |
598 | off += 2 * sizeof(struct track); | |
599 | ||
600 | if (s->size == off) | |
601 | return 1; | |
602 | ||
24922684 CL |
603 | return check_bytes_and_report(s, page, p, "Object padding", |
604 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
605 | } |
606 | ||
39b26464 | 607 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
608 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
609 | { | |
24922684 CL |
610 | u8 *start; |
611 | u8 *fault; | |
612 | u8 *end; | |
613 | int length; | |
614 | int remainder; | |
81819f0f CL |
615 | |
616 | if (!(s->flags & SLAB_POISON)) | |
617 | return 1; | |
618 | ||
a973e9dd | 619 | start = page_address(page); |
834f3d11 | 620 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
621 | end = start + length; |
622 | remainder = length % s->size; | |
81819f0f CL |
623 | if (!remainder) |
624 | return 1; | |
625 | ||
39b26464 | 626 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
627 | if (!fault) |
628 | return 1; | |
629 | while (end > fault && end[-1] == POISON_INUSE) | |
630 | end--; | |
631 | ||
632 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 633 | print_section("Padding", end - remainder, remainder); |
24922684 | 634 | |
8a3d271d | 635 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 636 | return 0; |
81819f0f CL |
637 | } |
638 | ||
639 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 640 | void *object, u8 val) |
81819f0f CL |
641 | { |
642 | u8 *p = object; | |
643 | u8 *endobject = object + s->objsize; | |
644 | ||
645 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 646 | if (!check_bytes_and_report(s, page, object, "Redzone", |
f7cb1933 | 647 | endobject, val, s->inuse - s->objsize)) |
81819f0f | 648 | return 0; |
81819f0f | 649 | } else { |
3adbefee IM |
650 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
651 | check_bytes_and_report(s, page, p, "Alignment padding", | |
652 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
653 | } | |
81819f0f CL |
654 | } |
655 | ||
656 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 657 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 CL |
658 | (!check_bytes_and_report(s, page, p, "Poison", p, |
659 | POISON_FREE, s->objsize - 1) || | |
660 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 661 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 662 | return 0; |
81819f0f CL |
663 | /* |
664 | * check_pad_bytes cleans up on its own. | |
665 | */ | |
666 | check_pad_bytes(s, page, p); | |
667 | } | |
668 | ||
f7cb1933 | 669 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
670 | /* |
671 | * Object and freepointer overlap. Cannot check | |
672 | * freepointer while object is allocated. | |
673 | */ | |
674 | return 1; | |
675 | ||
676 | /* Check free pointer validity */ | |
677 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
678 | object_err(s, page, p, "Freepointer corrupt"); | |
679 | /* | |
9f6c708e | 680 | * No choice but to zap it and thus lose the remainder |
81819f0f | 681 | * of the free objects in this slab. May cause |
672bba3a | 682 | * another error because the object count is now wrong. |
81819f0f | 683 | */ |
a973e9dd | 684 | set_freepointer(s, p, NULL); |
81819f0f CL |
685 | return 0; |
686 | } | |
687 | return 1; | |
688 | } | |
689 | ||
690 | static int check_slab(struct kmem_cache *s, struct page *page) | |
691 | { | |
39b26464 CL |
692 | int maxobj; |
693 | ||
81819f0f CL |
694 | VM_BUG_ON(!irqs_disabled()); |
695 | ||
696 | if (!PageSlab(page)) { | |
24922684 | 697 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
698 | return 0; |
699 | } | |
39b26464 CL |
700 | |
701 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
702 | if (page->objects > maxobj) { | |
703 | slab_err(s, page, "objects %u > max %u", | |
704 | s->name, page->objects, maxobj); | |
705 | return 0; | |
706 | } | |
707 | if (page->inuse > page->objects) { | |
24922684 | 708 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 709 | s->name, page->inuse, page->objects); |
81819f0f CL |
710 | return 0; |
711 | } | |
712 | /* Slab_pad_check fixes things up after itself */ | |
713 | slab_pad_check(s, page); | |
714 | return 1; | |
715 | } | |
716 | ||
717 | /* | |
672bba3a CL |
718 | * Determine if a certain object on a page is on the freelist. Must hold the |
719 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
720 | */ |
721 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
722 | { | |
723 | int nr = 0; | |
724 | void *fp = page->freelist; | |
725 | void *object = NULL; | |
224a88be | 726 | unsigned long max_objects; |
81819f0f | 727 | |
39b26464 | 728 | while (fp && nr <= page->objects) { |
81819f0f CL |
729 | if (fp == search) |
730 | return 1; | |
731 | if (!check_valid_pointer(s, page, fp)) { | |
732 | if (object) { | |
733 | object_err(s, page, object, | |
734 | "Freechain corrupt"); | |
a973e9dd | 735 | set_freepointer(s, object, NULL); |
81819f0f CL |
736 | break; |
737 | } else { | |
24922684 | 738 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 739 | page->freelist = NULL; |
39b26464 | 740 | page->inuse = page->objects; |
24922684 | 741 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
742 | return 0; |
743 | } | |
744 | break; | |
745 | } | |
746 | object = fp; | |
747 | fp = get_freepointer(s, object); | |
748 | nr++; | |
749 | } | |
750 | ||
224a88be | 751 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
210b5c06 CG |
752 | if (max_objects > MAX_OBJS_PER_PAGE) |
753 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
754 | |
755 | if (page->objects != max_objects) { | |
756 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
757 | "should be %d", page->objects, max_objects); | |
758 | page->objects = max_objects; | |
759 | slab_fix(s, "Number of objects adjusted."); | |
760 | } | |
39b26464 | 761 | if (page->inuse != page->objects - nr) { |
70d71228 | 762 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
763 | "counted were %d", page->inuse, page->objects - nr); |
764 | page->inuse = page->objects - nr; | |
24922684 | 765 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
766 | } |
767 | return search == NULL; | |
768 | } | |
769 | ||
0121c619 CL |
770 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
771 | int alloc) | |
3ec09742 CL |
772 | { |
773 | if (s->flags & SLAB_TRACE) { | |
774 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
775 | s->name, | |
776 | alloc ? "alloc" : "free", | |
777 | object, page->inuse, | |
778 | page->freelist); | |
779 | ||
780 | if (!alloc) | |
781 | print_section("Object", (void *)object, s->objsize); | |
782 | ||
783 | dump_stack(); | |
784 | } | |
785 | } | |
786 | ||
c016b0bd CL |
787 | /* |
788 | * Hooks for other subsystems that check memory allocations. In a typical | |
789 | * production configuration these hooks all should produce no code at all. | |
790 | */ | |
791 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
792 | { | |
c1d50836 | 793 | flags &= gfp_allowed_mask; |
c016b0bd CL |
794 | lockdep_trace_alloc(flags); |
795 | might_sleep_if(flags & __GFP_WAIT); | |
796 | ||
797 | return should_failslab(s->objsize, flags, s->flags); | |
798 | } | |
799 | ||
800 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
801 | { | |
c1d50836 | 802 | flags &= gfp_allowed_mask; |
c016b0bd CL |
803 | kmemcheck_slab_alloc(s, flags, object, s->objsize); |
804 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); | |
805 | } | |
806 | ||
807 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
808 | { | |
809 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 810 | |
d3f661d6 CL |
811 | /* |
812 | * Trouble is that we may no longer disable interupts in the fast path | |
813 | * So in order to make the debug calls that expect irqs to be | |
814 | * disabled we need to disable interrupts temporarily. | |
815 | */ | |
816 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
817 | { | |
818 | unsigned long flags; | |
819 | ||
820 | local_irq_save(flags); | |
821 | kmemcheck_slab_free(s, x, s->objsize); | |
822 | debug_check_no_locks_freed(x, s->objsize); | |
823 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
824 | debug_check_no_obj_freed(x, s->objsize); | |
825 | local_irq_restore(flags); | |
826 | } | |
827 | #endif | |
c016b0bd CL |
828 | } |
829 | ||
643b1138 | 830 | /* |
672bba3a | 831 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 832 | */ |
e95eed57 | 833 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 834 | { |
643b1138 CL |
835 | spin_lock(&n->list_lock); |
836 | list_add(&page->lru, &n->full); | |
837 | spin_unlock(&n->list_lock); | |
838 | } | |
839 | ||
840 | static void remove_full(struct kmem_cache *s, struct page *page) | |
841 | { | |
842 | struct kmem_cache_node *n; | |
843 | ||
844 | if (!(s->flags & SLAB_STORE_USER)) | |
845 | return; | |
846 | ||
847 | n = get_node(s, page_to_nid(page)); | |
848 | ||
849 | spin_lock(&n->list_lock); | |
850 | list_del(&page->lru); | |
851 | spin_unlock(&n->list_lock); | |
852 | } | |
853 | ||
0f389ec6 CL |
854 | /* Tracking of the number of slabs for debugging purposes */ |
855 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
856 | { | |
857 | struct kmem_cache_node *n = get_node(s, node); | |
858 | ||
859 | return atomic_long_read(&n->nr_slabs); | |
860 | } | |
861 | ||
26c02cf0 AB |
862 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
863 | { | |
864 | return atomic_long_read(&n->nr_slabs); | |
865 | } | |
866 | ||
205ab99d | 867 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
868 | { |
869 | struct kmem_cache_node *n = get_node(s, node); | |
870 | ||
871 | /* | |
872 | * May be called early in order to allocate a slab for the | |
873 | * kmem_cache_node structure. Solve the chicken-egg | |
874 | * dilemma by deferring the increment of the count during | |
875 | * bootstrap (see early_kmem_cache_node_alloc). | |
876 | */ | |
7340cc84 | 877 | if (n) { |
0f389ec6 | 878 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
879 | atomic_long_add(objects, &n->total_objects); |
880 | } | |
0f389ec6 | 881 | } |
205ab99d | 882 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
883 | { |
884 | struct kmem_cache_node *n = get_node(s, node); | |
885 | ||
886 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 887 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
888 | } |
889 | ||
890 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
891 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
892 | void *object) | |
893 | { | |
894 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
895 | return; | |
896 | ||
f7cb1933 | 897 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
898 | init_tracking(s, object); |
899 | } | |
900 | ||
1537066c | 901 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 902 | void *object, unsigned long addr) |
81819f0f CL |
903 | { |
904 | if (!check_slab(s, page)) | |
905 | goto bad; | |
906 | ||
d692ef6d | 907 | if (!on_freelist(s, page, object)) { |
24922684 | 908 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 909 | goto bad; |
81819f0f CL |
910 | } |
911 | ||
912 | if (!check_valid_pointer(s, page, object)) { | |
913 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 914 | goto bad; |
81819f0f CL |
915 | } |
916 | ||
f7cb1933 | 917 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 918 | goto bad; |
81819f0f | 919 | |
3ec09742 CL |
920 | /* Success perform special debug activities for allocs */ |
921 | if (s->flags & SLAB_STORE_USER) | |
922 | set_track(s, object, TRACK_ALLOC, addr); | |
923 | trace(s, page, object, 1); | |
f7cb1933 | 924 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 925 | return 1; |
3ec09742 | 926 | |
81819f0f CL |
927 | bad: |
928 | if (PageSlab(page)) { | |
929 | /* | |
930 | * If this is a slab page then lets do the best we can | |
931 | * to avoid issues in the future. Marking all objects | |
672bba3a | 932 | * as used avoids touching the remaining objects. |
81819f0f | 933 | */ |
24922684 | 934 | slab_fix(s, "Marking all objects used"); |
39b26464 | 935 | page->inuse = page->objects; |
a973e9dd | 936 | page->freelist = NULL; |
81819f0f CL |
937 | } |
938 | return 0; | |
939 | } | |
940 | ||
1537066c CL |
941 | static noinline int free_debug_processing(struct kmem_cache *s, |
942 | struct page *page, void *object, unsigned long addr) | |
81819f0f CL |
943 | { |
944 | if (!check_slab(s, page)) | |
945 | goto fail; | |
946 | ||
947 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 948 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
949 | goto fail; |
950 | } | |
951 | ||
952 | if (on_freelist(s, page, object)) { | |
24922684 | 953 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
954 | goto fail; |
955 | } | |
956 | ||
f7cb1933 | 957 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
81819f0f CL |
958 | return 0; |
959 | ||
960 | if (unlikely(s != page->slab)) { | |
3adbefee | 961 | if (!PageSlab(page)) { |
70d71228 CL |
962 | slab_err(s, page, "Attempt to free object(0x%p) " |
963 | "outside of slab", object); | |
3adbefee | 964 | } else if (!page->slab) { |
81819f0f | 965 | printk(KERN_ERR |
70d71228 | 966 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 967 | object); |
70d71228 | 968 | dump_stack(); |
06428780 | 969 | } else |
24922684 CL |
970 | object_err(s, page, object, |
971 | "page slab pointer corrupt."); | |
81819f0f CL |
972 | goto fail; |
973 | } | |
3ec09742 CL |
974 | |
975 | /* Special debug activities for freeing objects */ | |
8a38082d | 976 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
977 | remove_full(s, page); |
978 | if (s->flags & SLAB_STORE_USER) | |
979 | set_track(s, object, TRACK_FREE, addr); | |
980 | trace(s, page, object, 0); | |
f7cb1933 | 981 | init_object(s, object, SLUB_RED_INACTIVE); |
81819f0f | 982 | return 1; |
3ec09742 | 983 | |
81819f0f | 984 | fail: |
24922684 | 985 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
986 | return 0; |
987 | } | |
988 | ||
41ecc55b CL |
989 | static int __init setup_slub_debug(char *str) |
990 | { | |
f0630fff CL |
991 | slub_debug = DEBUG_DEFAULT_FLAGS; |
992 | if (*str++ != '=' || !*str) | |
993 | /* | |
994 | * No options specified. Switch on full debugging. | |
995 | */ | |
996 | goto out; | |
997 | ||
998 | if (*str == ',') | |
999 | /* | |
1000 | * No options but restriction on slabs. This means full | |
1001 | * debugging for slabs matching a pattern. | |
1002 | */ | |
1003 | goto check_slabs; | |
1004 | ||
fa5ec8a1 DR |
1005 | if (tolower(*str) == 'o') { |
1006 | /* | |
1007 | * Avoid enabling debugging on caches if its minimum order | |
1008 | * would increase as a result. | |
1009 | */ | |
1010 | disable_higher_order_debug = 1; | |
1011 | goto out; | |
1012 | } | |
1013 | ||
f0630fff CL |
1014 | slub_debug = 0; |
1015 | if (*str == '-') | |
1016 | /* | |
1017 | * Switch off all debugging measures. | |
1018 | */ | |
1019 | goto out; | |
1020 | ||
1021 | /* | |
1022 | * Determine which debug features should be switched on | |
1023 | */ | |
06428780 | 1024 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1025 | switch (tolower(*str)) { |
1026 | case 'f': | |
1027 | slub_debug |= SLAB_DEBUG_FREE; | |
1028 | break; | |
1029 | case 'z': | |
1030 | slub_debug |= SLAB_RED_ZONE; | |
1031 | break; | |
1032 | case 'p': | |
1033 | slub_debug |= SLAB_POISON; | |
1034 | break; | |
1035 | case 'u': | |
1036 | slub_debug |= SLAB_STORE_USER; | |
1037 | break; | |
1038 | case 't': | |
1039 | slub_debug |= SLAB_TRACE; | |
1040 | break; | |
4c13dd3b DM |
1041 | case 'a': |
1042 | slub_debug |= SLAB_FAILSLAB; | |
1043 | break; | |
f0630fff CL |
1044 | default: |
1045 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1046 | "unknown. skipped\n", *str); |
f0630fff | 1047 | } |
41ecc55b CL |
1048 | } |
1049 | ||
f0630fff | 1050 | check_slabs: |
41ecc55b CL |
1051 | if (*str == ',') |
1052 | slub_debug_slabs = str + 1; | |
f0630fff | 1053 | out: |
41ecc55b CL |
1054 | return 1; |
1055 | } | |
1056 | ||
1057 | __setup("slub_debug", setup_slub_debug); | |
1058 | ||
ba0268a8 CL |
1059 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1060 | unsigned long flags, const char *name, | |
51cc5068 | 1061 | void (*ctor)(void *)) |
41ecc55b CL |
1062 | { |
1063 | /* | |
e153362a | 1064 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1065 | */ |
e153362a | 1066 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1067 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1068 | flags |= slub_debug; | |
ba0268a8 CL |
1069 | |
1070 | return flags; | |
41ecc55b CL |
1071 | } |
1072 | #else | |
3ec09742 CL |
1073 | static inline void setup_object_debug(struct kmem_cache *s, |
1074 | struct page *page, void *object) {} | |
41ecc55b | 1075 | |
3ec09742 | 1076 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1077 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1078 | |
3ec09742 | 1079 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1080 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1081 | |
41ecc55b CL |
1082 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1083 | { return 1; } | |
1084 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1085 | void *object, u8 val) { return 1; } |
3ec09742 | 1086 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1087 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1088 | unsigned long flags, const char *name, | |
51cc5068 | 1089 | void (*ctor)(void *)) |
ba0268a8 CL |
1090 | { |
1091 | return flags; | |
1092 | } | |
41ecc55b | 1093 | #define slub_debug 0 |
0f389ec6 | 1094 | |
fdaa45e9 IM |
1095 | #define disable_higher_order_debug 0 |
1096 | ||
0f389ec6 CL |
1097 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1098 | { return 0; } | |
26c02cf0 AB |
1099 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1100 | { return 0; } | |
205ab99d CL |
1101 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1102 | int objects) {} | |
1103 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1104 | int objects) {} | |
7d550c56 CL |
1105 | |
1106 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1107 | { return 0; } | |
1108 | ||
1109 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1110 | void *object) {} | |
1111 | ||
1112 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1113 | ||
ab4d5ed5 | 1114 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1115 | |
81819f0f CL |
1116 | /* |
1117 | * Slab allocation and freeing | |
1118 | */ | |
65c3376a CL |
1119 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1120 | struct kmem_cache_order_objects oo) | |
1121 | { | |
1122 | int order = oo_order(oo); | |
1123 | ||
b1eeab67 VN |
1124 | flags |= __GFP_NOTRACK; |
1125 | ||
2154a336 | 1126 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1127 | return alloc_pages(flags, order); |
1128 | else | |
6b65aaf3 | 1129 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1130 | } |
1131 | ||
81819f0f CL |
1132 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1133 | { | |
06428780 | 1134 | struct page *page; |
834f3d11 | 1135 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1136 | gfp_t alloc_gfp; |
81819f0f | 1137 | |
b7a49f0d | 1138 | flags |= s->allocflags; |
e12ba74d | 1139 | |
ba52270d PE |
1140 | /* |
1141 | * Let the initial higher-order allocation fail under memory pressure | |
1142 | * so we fall-back to the minimum order allocation. | |
1143 | */ | |
1144 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1145 | ||
1146 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1147 | if (unlikely(!page)) { |
1148 | oo = s->min; | |
1149 | /* | |
1150 | * Allocation may have failed due to fragmentation. | |
1151 | * Try a lower order alloc if possible | |
1152 | */ | |
1153 | page = alloc_slab_page(flags, node, oo); | |
1154 | if (!page) | |
1155 | return NULL; | |
81819f0f | 1156 | |
84e554e6 | 1157 | stat(s, ORDER_FALLBACK); |
65c3376a | 1158 | } |
5a896d9e VN |
1159 | |
1160 | if (kmemcheck_enabled | |
5086c389 | 1161 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1162 | int pages = 1 << oo_order(oo); |
1163 | ||
1164 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1165 | ||
1166 | /* | |
1167 | * Objects from caches that have a constructor don't get | |
1168 | * cleared when they're allocated, so we need to do it here. | |
1169 | */ | |
1170 | if (s->ctor) | |
1171 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1172 | else | |
1173 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1174 | } |
1175 | ||
834f3d11 | 1176 | page->objects = oo_objects(oo); |
81819f0f CL |
1177 | mod_zone_page_state(page_zone(page), |
1178 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1179 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1180 | 1 << oo_order(oo)); |
81819f0f CL |
1181 | |
1182 | return page; | |
1183 | } | |
1184 | ||
1185 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1186 | void *object) | |
1187 | { | |
3ec09742 | 1188 | setup_object_debug(s, page, object); |
4f104934 | 1189 | if (unlikely(s->ctor)) |
51cc5068 | 1190 | s->ctor(object); |
81819f0f CL |
1191 | } |
1192 | ||
1193 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1194 | { | |
1195 | struct page *page; | |
81819f0f | 1196 | void *start; |
81819f0f CL |
1197 | void *last; |
1198 | void *p; | |
1199 | ||
6cb06229 | 1200 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1201 | |
6cb06229 CL |
1202 | page = allocate_slab(s, |
1203 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1204 | if (!page) |
1205 | goto out; | |
1206 | ||
205ab99d | 1207 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1208 | page->slab = s; |
1209 | page->flags |= 1 << PG_slab; | |
81819f0f CL |
1210 | |
1211 | start = page_address(page); | |
81819f0f CL |
1212 | |
1213 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1214 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1215 | |
1216 | last = start; | |
224a88be | 1217 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1218 | setup_object(s, page, last); |
1219 | set_freepointer(s, last, p); | |
1220 | last = p; | |
1221 | } | |
1222 | setup_object(s, page, last); | |
a973e9dd | 1223 | set_freepointer(s, last, NULL); |
81819f0f CL |
1224 | |
1225 | page->freelist = start; | |
1226 | page->inuse = 0; | |
1227 | out: | |
81819f0f CL |
1228 | return page; |
1229 | } | |
1230 | ||
1231 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1232 | { | |
834f3d11 CL |
1233 | int order = compound_order(page); |
1234 | int pages = 1 << order; | |
81819f0f | 1235 | |
af537b0a | 1236 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1237 | void *p; |
1238 | ||
1239 | slab_pad_check(s, page); | |
224a88be CL |
1240 | for_each_object(p, s, page_address(page), |
1241 | page->objects) | |
f7cb1933 | 1242 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1243 | } |
1244 | ||
b1eeab67 | 1245 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1246 | |
81819f0f CL |
1247 | mod_zone_page_state(page_zone(page), |
1248 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1249 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1250 | -pages); |
81819f0f | 1251 | |
49bd5221 CL |
1252 | __ClearPageSlab(page); |
1253 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1254 | if (current->reclaim_state) |
1255 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1256 | __free_pages(page, order); |
81819f0f CL |
1257 | } |
1258 | ||
1259 | static void rcu_free_slab(struct rcu_head *h) | |
1260 | { | |
1261 | struct page *page; | |
1262 | ||
1263 | page = container_of((struct list_head *)h, struct page, lru); | |
1264 | __free_slab(page->slab, page); | |
1265 | } | |
1266 | ||
1267 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1268 | { | |
1269 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1270 | /* | |
1271 | * RCU free overloads the RCU head over the LRU | |
1272 | */ | |
1273 | struct rcu_head *head = (void *)&page->lru; | |
1274 | ||
1275 | call_rcu(head, rcu_free_slab); | |
1276 | } else | |
1277 | __free_slab(s, page); | |
1278 | } | |
1279 | ||
1280 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1281 | { | |
205ab99d | 1282 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1283 | free_slab(s, page); |
1284 | } | |
1285 | ||
1286 | /* | |
1287 | * Per slab locking using the pagelock | |
1288 | */ | |
1289 | static __always_inline void slab_lock(struct page *page) | |
1290 | { | |
1291 | bit_spin_lock(PG_locked, &page->flags); | |
1292 | } | |
1293 | ||
1294 | static __always_inline void slab_unlock(struct page *page) | |
1295 | { | |
a76d3546 | 1296 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1297 | } |
1298 | ||
1299 | static __always_inline int slab_trylock(struct page *page) | |
1300 | { | |
1301 | int rc = 1; | |
1302 | ||
1303 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1304 | return rc; | |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * Management of partially allocated slabs | |
1309 | */ | |
7c2e132c CL |
1310 | static void add_partial(struct kmem_cache_node *n, |
1311 | struct page *page, int tail) | |
81819f0f | 1312 | { |
e95eed57 CL |
1313 | spin_lock(&n->list_lock); |
1314 | n->nr_partial++; | |
7c2e132c CL |
1315 | if (tail) |
1316 | list_add_tail(&page->lru, &n->partial); | |
1317 | else | |
1318 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1319 | spin_unlock(&n->list_lock); |
1320 | } | |
1321 | ||
62e346a8 CL |
1322 | static inline void __remove_partial(struct kmem_cache_node *n, |
1323 | struct page *page) | |
1324 | { | |
1325 | list_del(&page->lru); | |
1326 | n->nr_partial--; | |
1327 | } | |
1328 | ||
0121c619 | 1329 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1330 | { |
1331 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1332 | ||
1333 | spin_lock(&n->list_lock); | |
62e346a8 | 1334 | __remove_partial(n, page); |
81819f0f CL |
1335 | spin_unlock(&n->list_lock); |
1336 | } | |
1337 | ||
1338 | /* | |
672bba3a | 1339 | * Lock slab and remove from the partial list. |
81819f0f | 1340 | * |
672bba3a | 1341 | * Must hold list_lock. |
81819f0f | 1342 | */ |
0121c619 CL |
1343 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1344 | struct page *page) | |
81819f0f CL |
1345 | { |
1346 | if (slab_trylock(page)) { | |
62e346a8 | 1347 | __remove_partial(n, page); |
8a38082d | 1348 | __SetPageSlubFrozen(page); |
81819f0f CL |
1349 | return 1; |
1350 | } | |
1351 | return 0; | |
1352 | } | |
1353 | ||
1354 | /* | |
672bba3a | 1355 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1356 | */ |
1357 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1358 | { | |
1359 | struct page *page; | |
1360 | ||
1361 | /* | |
1362 | * Racy check. If we mistakenly see no partial slabs then we | |
1363 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1364 | * partial slab and there is none available then get_partials() |
1365 | * will return NULL. | |
81819f0f CL |
1366 | */ |
1367 | if (!n || !n->nr_partial) | |
1368 | return NULL; | |
1369 | ||
1370 | spin_lock(&n->list_lock); | |
1371 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1372 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1373 | goto out; |
1374 | page = NULL; | |
1375 | out: | |
1376 | spin_unlock(&n->list_lock); | |
1377 | return page; | |
1378 | } | |
1379 | ||
1380 | /* | |
672bba3a | 1381 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1382 | */ |
1383 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1384 | { | |
1385 | #ifdef CONFIG_NUMA | |
1386 | struct zonelist *zonelist; | |
dd1a239f | 1387 | struct zoneref *z; |
54a6eb5c MG |
1388 | struct zone *zone; |
1389 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1390 | struct page *page; |
1391 | ||
1392 | /* | |
672bba3a CL |
1393 | * The defrag ratio allows a configuration of the tradeoffs between |
1394 | * inter node defragmentation and node local allocations. A lower | |
1395 | * defrag_ratio increases the tendency to do local allocations | |
1396 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1397 | * |
672bba3a CL |
1398 | * If the defrag_ratio is set to 0 then kmalloc() always |
1399 | * returns node local objects. If the ratio is higher then kmalloc() | |
1400 | * may return off node objects because partial slabs are obtained | |
1401 | * from other nodes and filled up. | |
81819f0f | 1402 | * |
6446faa2 | 1403 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1404 | * defrag_ratio = 1000) then every (well almost) allocation will |
1405 | * first attempt to defrag slab caches on other nodes. This means | |
1406 | * scanning over all nodes to look for partial slabs which may be | |
1407 | * expensive if we do it every time we are trying to find a slab | |
1408 | * with available objects. | |
81819f0f | 1409 | */ |
9824601e CL |
1410 | if (!s->remote_node_defrag_ratio || |
1411 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1412 | return NULL; |
1413 | ||
c0ff7453 | 1414 | get_mems_allowed(); |
0e88460d | 1415 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1416 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1417 | struct kmem_cache_node *n; |
1418 | ||
54a6eb5c | 1419 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1420 | |
54a6eb5c | 1421 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1422 | n->nr_partial > s->min_partial) { |
81819f0f | 1423 | page = get_partial_node(n); |
c0ff7453 MX |
1424 | if (page) { |
1425 | put_mems_allowed(); | |
81819f0f | 1426 | return page; |
c0ff7453 | 1427 | } |
81819f0f CL |
1428 | } |
1429 | } | |
c0ff7453 | 1430 | put_mems_allowed(); |
81819f0f CL |
1431 | #endif |
1432 | return NULL; | |
1433 | } | |
1434 | ||
1435 | /* | |
1436 | * Get a partial page, lock it and return it. | |
1437 | */ | |
1438 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1439 | { | |
1440 | struct page *page; | |
2154a336 | 1441 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f CL |
1442 | |
1443 | page = get_partial_node(get_node(s, searchnode)); | |
bc6488e9 | 1444 | if (page || node != -1) |
81819f0f CL |
1445 | return page; |
1446 | ||
1447 | return get_any_partial(s, flags); | |
1448 | } | |
1449 | ||
1450 | /* | |
1451 | * Move a page back to the lists. | |
1452 | * | |
1453 | * Must be called with the slab lock held. | |
1454 | * | |
1455 | * On exit the slab lock will have been dropped. | |
1456 | */ | |
7c2e132c | 1457 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
3478973d | 1458 | __releases(bitlock) |
81819f0f | 1459 | { |
e95eed57 CL |
1460 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1461 | ||
8a38082d | 1462 | __ClearPageSlubFrozen(page); |
81819f0f | 1463 | if (page->inuse) { |
e95eed57 | 1464 | |
a973e9dd | 1465 | if (page->freelist) { |
7c2e132c | 1466 | add_partial(n, page, tail); |
84e554e6 | 1467 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
8ff12cfc | 1468 | } else { |
84e554e6 | 1469 | stat(s, DEACTIVATE_FULL); |
af537b0a | 1470 | if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER)) |
8ff12cfc CL |
1471 | add_full(n, page); |
1472 | } | |
81819f0f CL |
1473 | slab_unlock(page); |
1474 | } else { | |
84e554e6 | 1475 | stat(s, DEACTIVATE_EMPTY); |
3b89d7d8 | 1476 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1477 | /* |
672bba3a CL |
1478 | * Adding an empty slab to the partial slabs in order |
1479 | * to avoid page allocator overhead. This slab needs | |
1480 | * to come after the other slabs with objects in | |
6446faa2 CL |
1481 | * so that the others get filled first. That way the |
1482 | * size of the partial list stays small. | |
1483 | * | |
0121c619 CL |
1484 | * kmem_cache_shrink can reclaim any empty slabs from |
1485 | * the partial list. | |
e95eed57 | 1486 | */ |
7c2e132c | 1487 | add_partial(n, page, 1); |
e95eed57 CL |
1488 | slab_unlock(page); |
1489 | } else { | |
1490 | slab_unlock(page); | |
84e554e6 | 1491 | stat(s, FREE_SLAB); |
e95eed57 CL |
1492 | discard_slab(s, page); |
1493 | } | |
81819f0f CL |
1494 | } |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * Remove the cpu slab | |
1499 | */ | |
dfb4f096 | 1500 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
3478973d | 1501 | __releases(bitlock) |
81819f0f | 1502 | { |
dfb4f096 | 1503 | struct page *page = c->page; |
7c2e132c | 1504 | int tail = 1; |
8ff12cfc | 1505 | |
b773ad73 | 1506 | if (page->freelist) |
84e554e6 | 1507 | stat(s, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1508 | /* |
6446faa2 | 1509 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1510 | * because both freelists are empty. So this is unlikely |
1511 | * to occur. | |
1512 | */ | |
a973e9dd | 1513 | while (unlikely(c->freelist)) { |
894b8788 CL |
1514 | void **object; |
1515 | ||
7c2e132c CL |
1516 | tail = 0; /* Hot objects. Put the slab first */ |
1517 | ||
894b8788 | 1518 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1519 | object = c->freelist; |
ff12059e | 1520 | c->freelist = get_freepointer(s, c->freelist); |
894b8788 CL |
1521 | |
1522 | /* And put onto the regular freelist */ | |
ff12059e | 1523 | set_freepointer(s, object, page->freelist); |
894b8788 CL |
1524 | page->freelist = object; |
1525 | page->inuse--; | |
1526 | } | |
dfb4f096 | 1527 | c->page = NULL; |
7c2e132c | 1528 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1529 | } |
1530 | ||
dfb4f096 | 1531 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1532 | { |
84e554e6 | 1533 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 CL |
1534 | slab_lock(c->page); |
1535 | deactivate_slab(s, c); | |
81819f0f CL |
1536 | } |
1537 | ||
1538 | /* | |
1539 | * Flush cpu slab. | |
6446faa2 | 1540 | * |
81819f0f CL |
1541 | * Called from IPI handler with interrupts disabled. |
1542 | */ | |
0c710013 | 1543 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1544 | { |
9dfc6e68 | 1545 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1546 | |
dfb4f096 CL |
1547 | if (likely(c && c->page)) |
1548 | flush_slab(s, c); | |
81819f0f CL |
1549 | } |
1550 | ||
1551 | static void flush_cpu_slab(void *d) | |
1552 | { | |
1553 | struct kmem_cache *s = d; | |
81819f0f | 1554 | |
dfb4f096 | 1555 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1556 | } |
1557 | ||
1558 | static void flush_all(struct kmem_cache *s) | |
1559 | { | |
15c8b6c1 | 1560 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1561 | } |
1562 | ||
dfb4f096 CL |
1563 | /* |
1564 | * Check if the objects in a per cpu structure fit numa | |
1565 | * locality expectations. | |
1566 | */ | |
1567 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1568 | { | |
1569 | #ifdef CONFIG_NUMA | |
2154a336 | 1570 | if (node != NUMA_NO_NODE && c->node != node) |
dfb4f096 CL |
1571 | return 0; |
1572 | #endif | |
1573 | return 1; | |
1574 | } | |
1575 | ||
781b2ba6 PE |
1576 | static int count_free(struct page *page) |
1577 | { | |
1578 | return page->objects - page->inuse; | |
1579 | } | |
1580 | ||
1581 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1582 | int (*get_count)(struct page *)) | |
1583 | { | |
1584 | unsigned long flags; | |
1585 | unsigned long x = 0; | |
1586 | struct page *page; | |
1587 | ||
1588 | spin_lock_irqsave(&n->list_lock, flags); | |
1589 | list_for_each_entry(page, &n->partial, lru) | |
1590 | x += get_count(page); | |
1591 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1592 | return x; | |
1593 | } | |
1594 | ||
26c02cf0 AB |
1595 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1596 | { | |
1597 | #ifdef CONFIG_SLUB_DEBUG | |
1598 | return atomic_long_read(&n->total_objects); | |
1599 | #else | |
1600 | return 0; | |
1601 | #endif | |
1602 | } | |
1603 | ||
781b2ba6 PE |
1604 | static noinline void |
1605 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1606 | { | |
1607 | int node; | |
1608 | ||
1609 | printk(KERN_WARNING | |
1610 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1611 | nid, gfpflags); | |
1612 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1613 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1614 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1615 | ||
fa5ec8a1 DR |
1616 | if (oo_order(s->min) > get_order(s->objsize)) |
1617 | printk(KERN_WARNING " %s debugging increased min order, use " | |
1618 | "slub_debug=O to disable.\n", s->name); | |
1619 | ||
781b2ba6 PE |
1620 | for_each_online_node(node) { |
1621 | struct kmem_cache_node *n = get_node(s, node); | |
1622 | unsigned long nr_slabs; | |
1623 | unsigned long nr_objs; | |
1624 | unsigned long nr_free; | |
1625 | ||
1626 | if (!n) | |
1627 | continue; | |
1628 | ||
26c02cf0 AB |
1629 | nr_free = count_partial(n, count_free); |
1630 | nr_slabs = node_nr_slabs(n); | |
1631 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1632 | |
1633 | printk(KERN_WARNING | |
1634 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1635 | node, nr_slabs, nr_objs, nr_free); | |
1636 | } | |
1637 | } | |
1638 | ||
81819f0f | 1639 | /* |
894b8788 CL |
1640 | * Slow path. The lockless freelist is empty or we need to perform |
1641 | * debugging duties. | |
1642 | * | |
1643 | * Interrupts are disabled. | |
81819f0f | 1644 | * |
894b8788 CL |
1645 | * Processing is still very fast if new objects have been freed to the |
1646 | * regular freelist. In that case we simply take over the regular freelist | |
1647 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1648 | * |
894b8788 CL |
1649 | * If that is not working then we fall back to the partial lists. We take the |
1650 | * first element of the freelist as the object to allocate now and move the | |
1651 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1652 | * |
894b8788 | 1653 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1654 | * we need to allocate a new slab. This is the slowest path since it involves |
1655 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1656 | */ |
ce71e27c EGM |
1657 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1658 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1659 | { |
81819f0f | 1660 | void **object; |
dfb4f096 | 1661 | struct page *new; |
81819f0f | 1662 | |
e72e9c23 LT |
1663 | /* We handle __GFP_ZERO in the caller */ |
1664 | gfpflags &= ~__GFP_ZERO; | |
1665 | ||
dfb4f096 | 1666 | if (!c->page) |
81819f0f CL |
1667 | goto new_slab; |
1668 | ||
dfb4f096 CL |
1669 | slab_lock(c->page); |
1670 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1671 | goto another_slab; |
6446faa2 | 1672 | |
84e554e6 | 1673 | stat(s, ALLOC_REFILL); |
6446faa2 | 1674 | |
894b8788 | 1675 | load_freelist: |
dfb4f096 | 1676 | object = c->page->freelist; |
a973e9dd | 1677 | if (unlikely(!object)) |
81819f0f | 1678 | goto another_slab; |
af537b0a | 1679 | if (kmem_cache_debug(s)) |
81819f0f CL |
1680 | goto debug; |
1681 | ||
ff12059e | 1682 | c->freelist = get_freepointer(s, object); |
39b26464 | 1683 | c->page->inuse = c->page->objects; |
a973e9dd | 1684 | c->page->freelist = NULL; |
dfb4f096 | 1685 | c->node = page_to_nid(c->page); |
1f84260c | 1686 | unlock_out: |
dfb4f096 | 1687 | slab_unlock(c->page); |
84e554e6 | 1688 | stat(s, ALLOC_SLOWPATH); |
81819f0f CL |
1689 | return object; |
1690 | ||
1691 | another_slab: | |
dfb4f096 | 1692 | deactivate_slab(s, c); |
81819f0f CL |
1693 | |
1694 | new_slab: | |
dfb4f096 CL |
1695 | new = get_partial(s, gfpflags, node); |
1696 | if (new) { | |
1697 | c->page = new; | |
84e554e6 | 1698 | stat(s, ALLOC_FROM_PARTIAL); |
894b8788 | 1699 | goto load_freelist; |
81819f0f CL |
1700 | } |
1701 | ||
c1d50836 | 1702 | gfpflags &= gfp_allowed_mask; |
b811c202 CL |
1703 | if (gfpflags & __GFP_WAIT) |
1704 | local_irq_enable(); | |
1705 | ||
dfb4f096 | 1706 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1707 | |
1708 | if (gfpflags & __GFP_WAIT) | |
1709 | local_irq_disable(); | |
1710 | ||
dfb4f096 | 1711 | if (new) { |
9dfc6e68 | 1712 | c = __this_cpu_ptr(s->cpu_slab); |
84e554e6 | 1713 | stat(s, ALLOC_SLAB); |
05aa3450 | 1714 | if (c->page) |
dfb4f096 | 1715 | flush_slab(s, c); |
dfb4f096 | 1716 | slab_lock(new); |
8a38082d | 1717 | __SetPageSlubFrozen(new); |
dfb4f096 | 1718 | c->page = new; |
4b6f0750 | 1719 | goto load_freelist; |
81819f0f | 1720 | } |
95f85989 PE |
1721 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
1722 | slab_out_of_memory(s, gfpflags, node); | |
71c7a06f | 1723 | return NULL; |
81819f0f | 1724 | debug: |
dfb4f096 | 1725 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1726 | goto another_slab; |
894b8788 | 1727 | |
dfb4f096 | 1728 | c->page->inuse++; |
ff12059e | 1729 | c->page->freelist = get_freepointer(s, object); |
15b7c514 | 1730 | c->node = NUMA_NO_NODE; |
1f84260c | 1731 | goto unlock_out; |
894b8788 CL |
1732 | } |
1733 | ||
1734 | /* | |
1735 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1736 | * have the fastpath folded into their functions. So no function call | |
1737 | * overhead for requests that can be satisfied on the fastpath. | |
1738 | * | |
1739 | * The fastpath works by first checking if the lockless freelist can be used. | |
1740 | * If not then __slab_alloc is called for slow processing. | |
1741 | * | |
1742 | * Otherwise we can simply pick the next object from the lockless free list. | |
1743 | */ | |
06428780 | 1744 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1745 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1746 | { |
894b8788 | 1747 | void **object; |
dfb4f096 | 1748 | struct kmem_cache_cpu *c; |
1f84260c CL |
1749 | unsigned long flags; |
1750 | ||
c016b0bd | 1751 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 1752 | return NULL; |
1f84260c | 1753 | |
894b8788 | 1754 | local_irq_save(flags); |
9dfc6e68 CL |
1755 | c = __this_cpu_ptr(s->cpu_slab); |
1756 | object = c->freelist; | |
9dfc6e68 | 1757 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 1758 | |
dfb4f096 | 1759 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1760 | |
1761 | else { | |
ff12059e | 1762 | c->freelist = get_freepointer(s, object); |
84e554e6 | 1763 | stat(s, ALLOC_FASTPATH); |
894b8788 CL |
1764 | } |
1765 | local_irq_restore(flags); | |
d07dbea4 | 1766 | |
74e2134f | 1767 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 1768 | memset(object, 0, s->objsize); |
d07dbea4 | 1769 | |
c016b0bd | 1770 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 1771 | |
894b8788 | 1772 | return object; |
81819f0f CL |
1773 | } |
1774 | ||
1775 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1776 | { | |
2154a336 | 1777 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 1778 | |
ca2b84cb | 1779 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1780 | |
1781 | return ret; | |
81819f0f CL |
1782 | } |
1783 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1784 | ||
0f24f128 | 1785 | #ifdef CONFIG_TRACING |
4a92379b RK |
1786 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
1787 | { | |
1788 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
1789 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
1790 | return ret; | |
1791 | } | |
1792 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
1793 | ||
1794 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 1795 | { |
4a92379b RK |
1796 | void *ret = kmalloc_order(size, flags, order); |
1797 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1798 | return ret; | |
5b882be4 | 1799 | } |
4a92379b | 1800 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
1801 | #endif |
1802 | ||
81819f0f CL |
1803 | #ifdef CONFIG_NUMA |
1804 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1805 | { | |
5b882be4 EGM |
1806 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1807 | ||
ca2b84cb EGM |
1808 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
1809 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
1810 | |
1811 | return ret; | |
81819f0f CL |
1812 | } |
1813 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 1814 | |
0f24f128 | 1815 | #ifdef CONFIG_TRACING |
4a92379b | 1816 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 1817 | gfp_t gfpflags, |
4a92379b | 1818 | int node, size_t size) |
5b882be4 | 1819 | { |
4a92379b RK |
1820 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1821 | ||
1822 | trace_kmalloc_node(_RET_IP_, ret, | |
1823 | size, s->size, gfpflags, node); | |
1824 | return ret; | |
5b882be4 | 1825 | } |
4a92379b | 1826 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 1827 | #endif |
5d1f57e4 | 1828 | #endif |
5b882be4 | 1829 | |
81819f0f | 1830 | /* |
894b8788 CL |
1831 | * Slow patch handling. This may still be called frequently since objects |
1832 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1833 | * |
894b8788 CL |
1834 | * So we still attempt to reduce cache line usage. Just take the slab |
1835 | * lock and free the item. If there is no additional partial page | |
1836 | * handling required then we can return immediately. | |
81819f0f | 1837 | */ |
894b8788 | 1838 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 1839 | void *x, unsigned long addr) |
81819f0f CL |
1840 | { |
1841 | void *prior; | |
1842 | void **object = (void *)x; | |
81819f0f | 1843 | |
84e554e6 | 1844 | stat(s, FREE_SLOWPATH); |
81819f0f CL |
1845 | slab_lock(page); |
1846 | ||
af537b0a | 1847 | if (kmem_cache_debug(s)) |
81819f0f | 1848 | goto debug; |
6446faa2 | 1849 | |
81819f0f | 1850 | checks_ok: |
ff12059e CL |
1851 | prior = page->freelist; |
1852 | set_freepointer(s, object, prior); | |
81819f0f CL |
1853 | page->freelist = object; |
1854 | page->inuse--; | |
1855 | ||
8a38082d | 1856 | if (unlikely(PageSlubFrozen(page))) { |
84e554e6 | 1857 | stat(s, FREE_FROZEN); |
81819f0f | 1858 | goto out_unlock; |
8ff12cfc | 1859 | } |
81819f0f CL |
1860 | |
1861 | if (unlikely(!page->inuse)) | |
1862 | goto slab_empty; | |
1863 | ||
1864 | /* | |
6446faa2 | 1865 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1866 | * then add it. |
1867 | */ | |
a973e9dd | 1868 | if (unlikely(!prior)) { |
7c2e132c | 1869 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
84e554e6 | 1870 | stat(s, FREE_ADD_PARTIAL); |
8ff12cfc | 1871 | } |
81819f0f CL |
1872 | |
1873 | out_unlock: | |
1874 | slab_unlock(page); | |
81819f0f CL |
1875 | return; |
1876 | ||
1877 | slab_empty: | |
a973e9dd | 1878 | if (prior) { |
81819f0f | 1879 | /* |
672bba3a | 1880 | * Slab still on the partial list. |
81819f0f CL |
1881 | */ |
1882 | remove_partial(s, page); | |
84e554e6 | 1883 | stat(s, FREE_REMOVE_PARTIAL); |
8ff12cfc | 1884 | } |
81819f0f | 1885 | slab_unlock(page); |
84e554e6 | 1886 | stat(s, FREE_SLAB); |
81819f0f | 1887 | discard_slab(s, page); |
81819f0f CL |
1888 | return; |
1889 | ||
1890 | debug: | |
3ec09742 | 1891 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1892 | goto out_unlock; |
77c5e2d0 | 1893 | goto checks_ok; |
81819f0f CL |
1894 | } |
1895 | ||
894b8788 CL |
1896 | /* |
1897 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1898 | * can perform fastpath freeing without additional function calls. | |
1899 | * | |
1900 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1901 | * of this processor. This typically the case if we have just allocated | |
1902 | * the item before. | |
1903 | * | |
1904 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1905 | * with all sorts of special processing. | |
1906 | */ | |
06428780 | 1907 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1908 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1909 | { |
1910 | void **object = (void *)x; | |
dfb4f096 | 1911 | struct kmem_cache_cpu *c; |
1f84260c CL |
1912 | unsigned long flags; |
1913 | ||
c016b0bd CL |
1914 | slab_free_hook(s, x); |
1915 | ||
894b8788 | 1916 | local_irq_save(flags); |
9dfc6e68 | 1917 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 1918 | |
15b7c514 | 1919 | if (likely(page == c->page && c->node != NUMA_NO_NODE)) { |
ff12059e | 1920 | set_freepointer(s, object, c->freelist); |
dfb4f096 | 1921 | c->freelist = object; |
84e554e6 | 1922 | stat(s, FREE_FASTPATH); |
894b8788 | 1923 | } else |
ff12059e | 1924 | __slab_free(s, page, x, addr); |
894b8788 CL |
1925 | |
1926 | local_irq_restore(flags); | |
1927 | } | |
1928 | ||
81819f0f CL |
1929 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1930 | { | |
77c5e2d0 | 1931 | struct page *page; |
81819f0f | 1932 | |
b49af68f | 1933 | page = virt_to_head_page(x); |
81819f0f | 1934 | |
ce71e27c | 1935 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 1936 | |
ca2b84cb | 1937 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
1938 | } |
1939 | EXPORT_SYMBOL(kmem_cache_free); | |
1940 | ||
81819f0f | 1941 | /* |
672bba3a CL |
1942 | * Object placement in a slab is made very easy because we always start at |
1943 | * offset 0. If we tune the size of the object to the alignment then we can | |
1944 | * get the required alignment by putting one properly sized object after | |
1945 | * another. | |
81819f0f CL |
1946 | * |
1947 | * Notice that the allocation order determines the sizes of the per cpu | |
1948 | * caches. Each processor has always one slab available for allocations. | |
1949 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1950 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1951 | * locking overhead. |
81819f0f CL |
1952 | */ |
1953 | ||
1954 | /* | |
1955 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1956 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1957 | * and increases the number of allocations possible without having to | |
1958 | * take the list_lock. | |
1959 | */ | |
1960 | static int slub_min_order; | |
114e9e89 | 1961 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1962 | static int slub_min_objects; |
81819f0f CL |
1963 | |
1964 | /* | |
1965 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1966 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1967 | */ |
1968 | static int slub_nomerge; | |
1969 | ||
81819f0f CL |
1970 | /* |
1971 | * Calculate the order of allocation given an slab object size. | |
1972 | * | |
672bba3a CL |
1973 | * The order of allocation has significant impact on performance and other |
1974 | * system components. Generally order 0 allocations should be preferred since | |
1975 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1976 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1977 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1978 | * would be wasted. |
1979 | * | |
1980 | * In order to reach satisfactory performance we must ensure that a minimum | |
1981 | * number of objects is in one slab. Otherwise we may generate too much | |
1982 | * activity on the partial lists which requires taking the list_lock. This is | |
1983 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1984 | * |
672bba3a CL |
1985 | * slub_max_order specifies the order where we begin to stop considering the |
1986 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1987 | * we try to keep the page order as low as possible. So we accept more waste | |
1988 | * of space in favor of a small page order. | |
81819f0f | 1989 | * |
672bba3a CL |
1990 | * Higher order allocations also allow the placement of more objects in a |
1991 | * slab and thereby reduce object handling overhead. If the user has | |
1992 | * requested a higher mininum order then we start with that one instead of | |
1993 | * the smallest order which will fit the object. | |
81819f0f | 1994 | */ |
5e6d444e CL |
1995 | static inline int slab_order(int size, int min_objects, |
1996 | int max_order, int fract_leftover) | |
81819f0f CL |
1997 | { |
1998 | int order; | |
1999 | int rem; | |
6300ea75 | 2000 | int min_order = slub_min_order; |
81819f0f | 2001 | |
210b5c06 CG |
2002 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) |
2003 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
39b26464 | 2004 | |
6300ea75 | 2005 | for (order = max(min_order, |
5e6d444e CL |
2006 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2007 | order <= max_order; order++) { | |
81819f0f | 2008 | |
5e6d444e | 2009 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2010 | |
5e6d444e | 2011 | if (slab_size < min_objects * size) |
81819f0f CL |
2012 | continue; |
2013 | ||
2014 | rem = slab_size % size; | |
2015 | ||
5e6d444e | 2016 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2017 | break; |
2018 | ||
2019 | } | |
672bba3a | 2020 | |
81819f0f CL |
2021 | return order; |
2022 | } | |
2023 | ||
5e6d444e CL |
2024 | static inline int calculate_order(int size) |
2025 | { | |
2026 | int order; | |
2027 | int min_objects; | |
2028 | int fraction; | |
e8120ff1 | 2029 | int max_objects; |
5e6d444e CL |
2030 | |
2031 | /* | |
2032 | * Attempt to find best configuration for a slab. This | |
2033 | * works by first attempting to generate a layout with | |
2034 | * the best configuration and backing off gradually. | |
2035 | * | |
2036 | * First we reduce the acceptable waste in a slab. Then | |
2037 | * we reduce the minimum objects required in a slab. | |
2038 | */ | |
2039 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2040 | if (!min_objects) |
2041 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
e8120ff1 ZY |
2042 | max_objects = (PAGE_SIZE << slub_max_order)/size; |
2043 | min_objects = min(min_objects, max_objects); | |
2044 | ||
5e6d444e | 2045 | while (min_objects > 1) { |
c124f5b5 | 2046 | fraction = 16; |
5e6d444e CL |
2047 | while (fraction >= 4) { |
2048 | order = slab_order(size, min_objects, | |
2049 | slub_max_order, fraction); | |
2050 | if (order <= slub_max_order) | |
2051 | return order; | |
2052 | fraction /= 2; | |
2053 | } | |
5086c389 | 2054 | min_objects--; |
5e6d444e CL |
2055 | } |
2056 | ||
2057 | /* | |
2058 | * We were unable to place multiple objects in a slab. Now | |
2059 | * lets see if we can place a single object there. | |
2060 | */ | |
2061 | order = slab_order(size, 1, slub_max_order, 1); | |
2062 | if (order <= slub_max_order) | |
2063 | return order; | |
2064 | ||
2065 | /* | |
2066 | * Doh this slab cannot be placed using slub_max_order. | |
2067 | */ | |
2068 | order = slab_order(size, 1, MAX_ORDER, 1); | |
818cf590 | 2069 | if (order < MAX_ORDER) |
5e6d444e CL |
2070 | return order; |
2071 | return -ENOSYS; | |
2072 | } | |
2073 | ||
81819f0f | 2074 | /* |
672bba3a | 2075 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2076 | */ |
2077 | static unsigned long calculate_alignment(unsigned long flags, | |
2078 | unsigned long align, unsigned long size) | |
2079 | { | |
2080 | /* | |
6446faa2 CL |
2081 | * If the user wants hardware cache aligned objects then follow that |
2082 | * suggestion if the object is sufficiently large. | |
81819f0f | 2083 | * |
6446faa2 CL |
2084 | * The hardware cache alignment cannot override the specified |
2085 | * alignment though. If that is greater then use it. | |
81819f0f | 2086 | */ |
b6210386 NP |
2087 | if (flags & SLAB_HWCACHE_ALIGN) { |
2088 | unsigned long ralign = cache_line_size(); | |
2089 | while (size <= ralign / 2) | |
2090 | ralign /= 2; | |
2091 | align = max(align, ralign); | |
2092 | } | |
81819f0f CL |
2093 | |
2094 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2095 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2096 | |
2097 | return ALIGN(align, sizeof(void *)); | |
2098 | } | |
2099 | ||
5595cffc PE |
2100 | static void |
2101 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2102 | { |
2103 | n->nr_partial = 0; | |
81819f0f CL |
2104 | spin_lock_init(&n->list_lock); |
2105 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2106 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2107 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2108 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2109 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2110 | #endif |
81819f0f CL |
2111 | } |
2112 | ||
55136592 | 2113 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2114 | { |
6c182dc0 CL |
2115 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2116 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2117 | |
6c182dc0 | 2118 | s->cpu_slab = alloc_percpu(struct kmem_cache_cpu); |
4c93c355 | 2119 | |
6c182dc0 | 2120 | return s->cpu_slab != NULL; |
4c93c355 | 2121 | } |
4c93c355 | 2122 | |
51df1142 CL |
2123 | static struct kmem_cache *kmem_cache_node; |
2124 | ||
81819f0f CL |
2125 | /* |
2126 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2127 | * slab on the node for this slabcache. There are no concurrent accesses | |
2128 | * possible. | |
2129 | * | |
2130 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2131 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2132 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2133 | */ |
55136592 | 2134 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2135 | { |
2136 | struct page *page; | |
2137 | struct kmem_cache_node *n; | |
ba84c73c | 2138 | unsigned long flags; |
81819f0f | 2139 | |
51df1142 | 2140 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2141 | |
51df1142 | 2142 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2143 | |
2144 | BUG_ON(!page); | |
a2f92ee7 CL |
2145 | if (page_to_nid(page) != node) { |
2146 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2147 | "node %d\n", node); | |
2148 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2149 | "in order to be able to continue\n"); | |
2150 | } | |
2151 | ||
81819f0f CL |
2152 | n = page->freelist; |
2153 | BUG_ON(!n); | |
51df1142 | 2154 | page->freelist = get_freepointer(kmem_cache_node, n); |
81819f0f | 2155 | page->inuse++; |
51df1142 | 2156 | kmem_cache_node->node[node] = n; |
8ab1372f | 2157 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2158 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2159 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2160 | #endif |
51df1142 CL |
2161 | init_kmem_cache_node(n, kmem_cache_node); |
2162 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
6446faa2 | 2163 | |
ba84c73c | 2164 | /* |
2165 | * lockdep requires consistent irq usage for each lock | |
2166 | * so even though there cannot be a race this early in | |
2167 | * the boot sequence, we still disable irqs. | |
2168 | */ | |
2169 | local_irq_save(flags); | |
7c2e132c | 2170 | add_partial(n, page, 0); |
ba84c73c | 2171 | local_irq_restore(flags); |
81819f0f CL |
2172 | } |
2173 | ||
2174 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2175 | { | |
2176 | int node; | |
2177 | ||
f64dc58c | 2178 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2179 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2180 | |
73367bd8 | 2181 | if (n) |
51df1142 CL |
2182 | kmem_cache_free(kmem_cache_node, n); |
2183 | ||
81819f0f CL |
2184 | s->node[node] = NULL; |
2185 | } | |
2186 | } | |
2187 | ||
55136592 | 2188 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2189 | { |
2190 | int node; | |
81819f0f | 2191 | |
f64dc58c | 2192 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2193 | struct kmem_cache_node *n; |
2194 | ||
73367bd8 | 2195 | if (slab_state == DOWN) { |
55136592 | 2196 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2197 | continue; |
2198 | } | |
51df1142 | 2199 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2200 | GFP_KERNEL, node); |
81819f0f | 2201 | |
73367bd8 AD |
2202 | if (!n) { |
2203 | free_kmem_cache_nodes(s); | |
2204 | return 0; | |
81819f0f | 2205 | } |
73367bd8 | 2206 | |
81819f0f | 2207 | s->node[node] = n; |
5595cffc | 2208 | init_kmem_cache_node(n, s); |
81819f0f CL |
2209 | } |
2210 | return 1; | |
2211 | } | |
81819f0f | 2212 | |
c0bdb232 | 2213 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2214 | { |
2215 | if (min < MIN_PARTIAL) | |
2216 | min = MIN_PARTIAL; | |
2217 | else if (min > MAX_PARTIAL) | |
2218 | min = MAX_PARTIAL; | |
2219 | s->min_partial = min; | |
2220 | } | |
2221 | ||
81819f0f CL |
2222 | /* |
2223 | * calculate_sizes() determines the order and the distribution of data within | |
2224 | * a slab object. | |
2225 | */ | |
06b285dc | 2226 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2227 | { |
2228 | unsigned long flags = s->flags; | |
2229 | unsigned long size = s->objsize; | |
2230 | unsigned long align = s->align; | |
834f3d11 | 2231 | int order; |
81819f0f | 2232 | |
d8b42bf5 CL |
2233 | /* |
2234 | * Round up object size to the next word boundary. We can only | |
2235 | * place the free pointer at word boundaries and this determines | |
2236 | * the possible location of the free pointer. | |
2237 | */ | |
2238 | size = ALIGN(size, sizeof(void *)); | |
2239 | ||
2240 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2241 | /* |
2242 | * Determine if we can poison the object itself. If the user of | |
2243 | * the slab may touch the object after free or before allocation | |
2244 | * then we should never poison the object itself. | |
2245 | */ | |
2246 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2247 | !s->ctor) |
81819f0f CL |
2248 | s->flags |= __OBJECT_POISON; |
2249 | else | |
2250 | s->flags &= ~__OBJECT_POISON; | |
2251 | ||
81819f0f CL |
2252 | |
2253 | /* | |
672bba3a | 2254 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2255 | * end of the object and the free pointer. If not then add an |
672bba3a | 2256 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2257 | */ |
2258 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2259 | size += sizeof(void *); | |
41ecc55b | 2260 | #endif |
81819f0f CL |
2261 | |
2262 | /* | |
672bba3a CL |
2263 | * With that we have determined the number of bytes in actual use |
2264 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2265 | */ |
2266 | s->inuse = size; | |
2267 | ||
2268 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2269 | s->ctor)) { |
81819f0f CL |
2270 | /* |
2271 | * Relocate free pointer after the object if it is not | |
2272 | * permitted to overwrite the first word of the object on | |
2273 | * kmem_cache_free. | |
2274 | * | |
2275 | * This is the case if we do RCU, have a constructor or | |
2276 | * destructor or are poisoning the objects. | |
2277 | */ | |
2278 | s->offset = size; | |
2279 | size += sizeof(void *); | |
2280 | } | |
2281 | ||
c12b3c62 | 2282 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2283 | if (flags & SLAB_STORE_USER) |
2284 | /* | |
2285 | * Need to store information about allocs and frees after | |
2286 | * the object. | |
2287 | */ | |
2288 | size += 2 * sizeof(struct track); | |
2289 | ||
be7b3fbc | 2290 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2291 | /* |
2292 | * Add some empty padding so that we can catch | |
2293 | * overwrites from earlier objects rather than let | |
2294 | * tracking information or the free pointer be | |
0211a9c8 | 2295 | * corrupted if a user writes before the start |
81819f0f CL |
2296 | * of the object. |
2297 | */ | |
2298 | size += sizeof(void *); | |
41ecc55b | 2299 | #endif |
672bba3a | 2300 | |
81819f0f CL |
2301 | /* |
2302 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2303 | * user specified and the dynamic determination of cache line size |
2304 | * on bootup. | |
81819f0f CL |
2305 | */ |
2306 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2307 | s->align = align; |
81819f0f CL |
2308 | |
2309 | /* | |
2310 | * SLUB stores one object immediately after another beginning from | |
2311 | * offset 0. In order to align the objects we have to simply size | |
2312 | * each object to conform to the alignment. | |
2313 | */ | |
2314 | size = ALIGN(size, align); | |
2315 | s->size = size; | |
06b285dc CL |
2316 | if (forced_order >= 0) |
2317 | order = forced_order; | |
2318 | else | |
2319 | order = calculate_order(size); | |
81819f0f | 2320 | |
834f3d11 | 2321 | if (order < 0) |
81819f0f CL |
2322 | return 0; |
2323 | ||
b7a49f0d | 2324 | s->allocflags = 0; |
834f3d11 | 2325 | if (order) |
b7a49f0d CL |
2326 | s->allocflags |= __GFP_COMP; |
2327 | ||
2328 | if (s->flags & SLAB_CACHE_DMA) | |
2329 | s->allocflags |= SLUB_DMA; | |
2330 | ||
2331 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2332 | s->allocflags |= __GFP_RECLAIMABLE; | |
2333 | ||
81819f0f CL |
2334 | /* |
2335 | * Determine the number of objects per slab | |
2336 | */ | |
834f3d11 | 2337 | s->oo = oo_make(order, size); |
65c3376a | 2338 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2339 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2340 | s->max = s->oo; | |
81819f0f | 2341 | |
834f3d11 | 2342 | return !!oo_objects(s->oo); |
81819f0f CL |
2343 | |
2344 | } | |
2345 | ||
55136592 | 2346 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
2347 | const char *name, size_t size, |
2348 | size_t align, unsigned long flags, | |
51cc5068 | 2349 | void (*ctor)(void *)) |
81819f0f CL |
2350 | { |
2351 | memset(s, 0, kmem_size); | |
2352 | s->name = name; | |
2353 | s->ctor = ctor; | |
81819f0f | 2354 | s->objsize = size; |
81819f0f | 2355 | s->align = align; |
ba0268a8 | 2356 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2357 | |
06b285dc | 2358 | if (!calculate_sizes(s, -1)) |
81819f0f | 2359 | goto error; |
3de47213 DR |
2360 | if (disable_higher_order_debug) { |
2361 | /* | |
2362 | * Disable debugging flags that store metadata if the min slab | |
2363 | * order increased. | |
2364 | */ | |
2365 | if (get_order(s->size) > get_order(s->objsize)) { | |
2366 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2367 | s->offset = 0; | |
2368 | if (!calculate_sizes(s, -1)) | |
2369 | goto error; | |
2370 | } | |
2371 | } | |
81819f0f | 2372 | |
3b89d7d8 DR |
2373 | /* |
2374 | * The larger the object size is, the more pages we want on the partial | |
2375 | * list to avoid pounding the page allocator excessively. | |
2376 | */ | |
c0bdb232 | 2377 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2378 | s->refcount = 1; |
2379 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2380 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2381 | #endif |
55136592 | 2382 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 2383 | goto error; |
81819f0f | 2384 | |
55136592 | 2385 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 2386 | return 1; |
ff12059e | 2387 | |
4c93c355 | 2388 | free_kmem_cache_nodes(s); |
81819f0f CL |
2389 | error: |
2390 | if (flags & SLAB_PANIC) | |
2391 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2392 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2393 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2394 | s->offset, flags); |
2395 | return 0; | |
2396 | } | |
81819f0f | 2397 | |
81819f0f CL |
2398 | /* |
2399 | * Determine the size of a slab object | |
2400 | */ | |
2401 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2402 | { | |
2403 | return s->objsize; | |
2404 | } | |
2405 | EXPORT_SYMBOL(kmem_cache_size); | |
2406 | ||
2407 | const char *kmem_cache_name(struct kmem_cache *s) | |
2408 | { | |
2409 | return s->name; | |
2410 | } | |
2411 | EXPORT_SYMBOL(kmem_cache_name); | |
2412 | ||
33b12c38 CL |
2413 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2414 | const char *text) | |
2415 | { | |
2416 | #ifdef CONFIG_SLUB_DEBUG | |
2417 | void *addr = page_address(page); | |
2418 | void *p; | |
a5dd5c11 NK |
2419 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
2420 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
2421 | if (!map) |
2422 | return; | |
33b12c38 CL |
2423 | slab_err(s, page, "%s", text); |
2424 | slab_lock(page); | |
2425 | for_each_free_object(p, s, page->freelist) | |
2426 | set_bit(slab_index(p, s, addr), map); | |
2427 | ||
2428 | for_each_object(p, s, addr, page->objects) { | |
2429 | ||
2430 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2431 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2432 | p, p - addr); | |
2433 | print_tracking(s, p); | |
2434 | } | |
2435 | } | |
2436 | slab_unlock(page); | |
bbd7d57b | 2437 | kfree(map); |
33b12c38 CL |
2438 | #endif |
2439 | } | |
2440 | ||
81819f0f | 2441 | /* |
599870b1 | 2442 | * Attempt to free all partial slabs on a node. |
81819f0f | 2443 | */ |
599870b1 | 2444 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2445 | { |
81819f0f CL |
2446 | unsigned long flags; |
2447 | struct page *page, *h; | |
2448 | ||
2449 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2450 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 2451 | if (!page->inuse) { |
62e346a8 | 2452 | __remove_partial(n, page); |
81819f0f | 2453 | discard_slab(s, page); |
33b12c38 CL |
2454 | } else { |
2455 | list_slab_objects(s, page, | |
2456 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2457 | } |
33b12c38 | 2458 | } |
81819f0f | 2459 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2460 | } |
2461 | ||
2462 | /* | |
672bba3a | 2463 | * Release all resources used by a slab cache. |
81819f0f | 2464 | */ |
0c710013 | 2465 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2466 | { |
2467 | int node; | |
2468 | ||
2469 | flush_all(s); | |
9dfc6e68 | 2470 | free_percpu(s->cpu_slab); |
81819f0f | 2471 | /* Attempt to free all objects */ |
f64dc58c | 2472 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2473 | struct kmem_cache_node *n = get_node(s, node); |
2474 | ||
599870b1 CL |
2475 | free_partial(s, n); |
2476 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2477 | return 1; |
2478 | } | |
2479 | free_kmem_cache_nodes(s); | |
2480 | return 0; | |
2481 | } | |
2482 | ||
2483 | /* | |
2484 | * Close a cache and release the kmem_cache structure | |
2485 | * (must be used for caches created using kmem_cache_create) | |
2486 | */ | |
2487 | void kmem_cache_destroy(struct kmem_cache *s) | |
2488 | { | |
2489 | down_write(&slub_lock); | |
2490 | s->refcount--; | |
2491 | if (!s->refcount) { | |
2492 | list_del(&s->list); | |
d629d819 PE |
2493 | if (kmem_cache_close(s)) { |
2494 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2495 | "still has objects.\n", s->name, __func__); | |
2496 | dump_stack(); | |
2497 | } | |
d76b1590 ED |
2498 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2499 | rcu_barrier(); | |
81819f0f | 2500 | sysfs_slab_remove(s); |
2bce6485 CL |
2501 | } |
2502 | up_write(&slub_lock); | |
81819f0f CL |
2503 | } |
2504 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2505 | ||
2506 | /******************************************************************** | |
2507 | * Kmalloc subsystem | |
2508 | *******************************************************************/ | |
2509 | ||
51df1142 | 2510 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
2511 | EXPORT_SYMBOL(kmalloc_caches); |
2512 | ||
51df1142 CL |
2513 | static struct kmem_cache *kmem_cache; |
2514 | ||
55136592 | 2515 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 2516 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
2517 | #endif |
2518 | ||
81819f0f CL |
2519 | static int __init setup_slub_min_order(char *str) |
2520 | { | |
06428780 | 2521 | get_option(&str, &slub_min_order); |
81819f0f CL |
2522 | |
2523 | return 1; | |
2524 | } | |
2525 | ||
2526 | __setup("slub_min_order=", setup_slub_min_order); | |
2527 | ||
2528 | static int __init setup_slub_max_order(char *str) | |
2529 | { | |
06428780 | 2530 | get_option(&str, &slub_max_order); |
818cf590 | 2531 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2532 | |
2533 | return 1; | |
2534 | } | |
2535 | ||
2536 | __setup("slub_max_order=", setup_slub_max_order); | |
2537 | ||
2538 | static int __init setup_slub_min_objects(char *str) | |
2539 | { | |
06428780 | 2540 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2541 | |
2542 | return 1; | |
2543 | } | |
2544 | ||
2545 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2546 | ||
2547 | static int __init setup_slub_nomerge(char *str) | |
2548 | { | |
2549 | slub_nomerge = 1; | |
2550 | return 1; | |
2551 | } | |
2552 | ||
2553 | __setup("slub_nomerge", setup_slub_nomerge); | |
2554 | ||
51df1142 CL |
2555 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
2556 | int size, unsigned int flags) | |
81819f0f | 2557 | { |
51df1142 CL |
2558 | struct kmem_cache *s; |
2559 | ||
2560 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
2561 | ||
83b519e8 PE |
2562 | /* |
2563 | * This function is called with IRQs disabled during early-boot on | |
2564 | * single CPU so there's no need to take slub_lock here. | |
2565 | */ | |
55136592 | 2566 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2567 | flags, NULL)) |
81819f0f CL |
2568 | goto panic; |
2569 | ||
2570 | list_add(&s->list, &slab_caches); | |
51df1142 | 2571 | return s; |
81819f0f CL |
2572 | |
2573 | panic: | |
2574 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 2575 | return NULL; |
81819f0f CL |
2576 | } |
2577 | ||
f1b26339 CL |
2578 | /* |
2579 | * Conversion table for small slabs sizes / 8 to the index in the | |
2580 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2581 | * of two cache sizes there. The size of larger slabs can be determined using | |
2582 | * fls. | |
2583 | */ | |
2584 | static s8 size_index[24] = { | |
2585 | 3, /* 8 */ | |
2586 | 4, /* 16 */ | |
2587 | 5, /* 24 */ | |
2588 | 5, /* 32 */ | |
2589 | 6, /* 40 */ | |
2590 | 6, /* 48 */ | |
2591 | 6, /* 56 */ | |
2592 | 6, /* 64 */ | |
2593 | 1, /* 72 */ | |
2594 | 1, /* 80 */ | |
2595 | 1, /* 88 */ | |
2596 | 1, /* 96 */ | |
2597 | 7, /* 104 */ | |
2598 | 7, /* 112 */ | |
2599 | 7, /* 120 */ | |
2600 | 7, /* 128 */ | |
2601 | 2, /* 136 */ | |
2602 | 2, /* 144 */ | |
2603 | 2, /* 152 */ | |
2604 | 2, /* 160 */ | |
2605 | 2, /* 168 */ | |
2606 | 2, /* 176 */ | |
2607 | 2, /* 184 */ | |
2608 | 2 /* 192 */ | |
2609 | }; | |
2610 | ||
acdfcd04 AK |
2611 | static inline int size_index_elem(size_t bytes) |
2612 | { | |
2613 | return (bytes - 1) / 8; | |
2614 | } | |
2615 | ||
81819f0f CL |
2616 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2617 | { | |
f1b26339 | 2618 | int index; |
81819f0f | 2619 | |
f1b26339 CL |
2620 | if (size <= 192) { |
2621 | if (!size) | |
2622 | return ZERO_SIZE_PTR; | |
81819f0f | 2623 | |
acdfcd04 | 2624 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 2625 | } else |
f1b26339 | 2626 | index = fls(size - 1); |
81819f0f CL |
2627 | |
2628 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2629 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 2630 | return kmalloc_dma_caches[index]; |
f1b26339 | 2631 | |
81819f0f | 2632 | #endif |
51df1142 | 2633 | return kmalloc_caches[index]; |
81819f0f CL |
2634 | } |
2635 | ||
2636 | void *__kmalloc(size_t size, gfp_t flags) | |
2637 | { | |
aadb4bc4 | 2638 | struct kmem_cache *s; |
5b882be4 | 2639 | void *ret; |
81819f0f | 2640 | |
ffadd4d0 | 2641 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2642 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2643 | |
2644 | s = get_slab(size, flags); | |
2645 | ||
2646 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2647 | return s; |
2648 | ||
2154a336 | 2649 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2650 | |
ca2b84cb | 2651 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2652 | |
2653 | return ret; | |
81819f0f CL |
2654 | } |
2655 | EXPORT_SYMBOL(__kmalloc); | |
2656 | ||
5d1f57e4 | 2657 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
2658 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2659 | { | |
b1eeab67 | 2660 | struct page *page; |
e4f7c0b4 | 2661 | void *ptr = NULL; |
f619cfe1 | 2662 | |
b1eeab67 VN |
2663 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2664 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 2665 | if (page) |
e4f7c0b4 CM |
2666 | ptr = page_address(page); |
2667 | ||
2668 | kmemleak_alloc(ptr, size, 1, flags); | |
2669 | return ptr; | |
f619cfe1 CL |
2670 | } |
2671 | ||
81819f0f CL |
2672 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
2673 | { | |
aadb4bc4 | 2674 | struct kmem_cache *s; |
5b882be4 | 2675 | void *ret; |
81819f0f | 2676 | |
057685cf | 2677 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2678 | ret = kmalloc_large_node(size, flags, node); |
2679 | ||
ca2b84cb EGM |
2680 | trace_kmalloc_node(_RET_IP_, ret, |
2681 | size, PAGE_SIZE << get_order(size), | |
2682 | flags, node); | |
5b882be4 EGM |
2683 | |
2684 | return ret; | |
2685 | } | |
aadb4bc4 CL |
2686 | |
2687 | s = get_slab(size, flags); | |
2688 | ||
2689 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2690 | return s; |
2691 | ||
5b882be4 EGM |
2692 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2693 | ||
ca2b84cb | 2694 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2695 | |
2696 | return ret; | |
81819f0f CL |
2697 | } |
2698 | EXPORT_SYMBOL(__kmalloc_node); | |
2699 | #endif | |
2700 | ||
2701 | size_t ksize(const void *object) | |
2702 | { | |
272c1d21 | 2703 | struct page *page; |
81819f0f CL |
2704 | struct kmem_cache *s; |
2705 | ||
ef8b4520 | 2706 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2707 | return 0; |
2708 | ||
294a80a8 | 2709 | page = virt_to_head_page(object); |
294a80a8 | 2710 | |
76994412 PE |
2711 | if (unlikely(!PageSlab(page))) { |
2712 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2713 | return PAGE_SIZE << compound_order(page); |
76994412 | 2714 | } |
81819f0f | 2715 | s = page->slab; |
81819f0f | 2716 | |
ae20bfda | 2717 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2718 | /* |
2719 | * Debugging requires use of the padding between object | |
2720 | * and whatever may come after it. | |
2721 | */ | |
2722 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2723 | return s->objsize; | |
2724 | ||
ae20bfda | 2725 | #endif |
81819f0f CL |
2726 | /* |
2727 | * If we have the need to store the freelist pointer | |
2728 | * back there or track user information then we can | |
2729 | * only use the space before that information. | |
2730 | */ | |
2731 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2732 | return s->inuse; | |
81819f0f CL |
2733 | /* |
2734 | * Else we can use all the padding etc for the allocation | |
2735 | */ | |
2736 | return s->size; | |
2737 | } | |
b1aabecd | 2738 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2739 | |
2740 | void kfree(const void *x) | |
2741 | { | |
81819f0f | 2742 | struct page *page; |
5bb983b0 | 2743 | void *object = (void *)x; |
81819f0f | 2744 | |
2121db74 PE |
2745 | trace_kfree(_RET_IP_, x); |
2746 | ||
2408c550 | 2747 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2748 | return; |
2749 | ||
b49af68f | 2750 | page = virt_to_head_page(x); |
aadb4bc4 | 2751 | if (unlikely(!PageSlab(page))) { |
0937502a | 2752 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 2753 | kmemleak_free(x); |
aadb4bc4 CL |
2754 | put_page(page); |
2755 | return; | |
2756 | } | |
ce71e27c | 2757 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2758 | } |
2759 | EXPORT_SYMBOL(kfree); | |
2760 | ||
2086d26a | 2761 | /* |
672bba3a CL |
2762 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2763 | * the remaining slabs by the number of items in use. The slabs with the | |
2764 | * most items in use come first. New allocations will then fill those up | |
2765 | * and thus they can be removed from the partial lists. | |
2766 | * | |
2767 | * The slabs with the least items are placed last. This results in them | |
2768 | * being allocated from last increasing the chance that the last objects | |
2769 | * are freed in them. | |
2086d26a CL |
2770 | */ |
2771 | int kmem_cache_shrink(struct kmem_cache *s) | |
2772 | { | |
2773 | int node; | |
2774 | int i; | |
2775 | struct kmem_cache_node *n; | |
2776 | struct page *page; | |
2777 | struct page *t; | |
205ab99d | 2778 | int objects = oo_objects(s->max); |
2086d26a | 2779 | struct list_head *slabs_by_inuse = |
834f3d11 | 2780 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2781 | unsigned long flags; |
2782 | ||
2783 | if (!slabs_by_inuse) | |
2784 | return -ENOMEM; | |
2785 | ||
2786 | flush_all(s); | |
f64dc58c | 2787 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2788 | n = get_node(s, node); |
2789 | ||
2790 | if (!n->nr_partial) | |
2791 | continue; | |
2792 | ||
834f3d11 | 2793 | for (i = 0; i < objects; i++) |
2086d26a CL |
2794 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2795 | ||
2796 | spin_lock_irqsave(&n->list_lock, flags); | |
2797 | ||
2798 | /* | |
672bba3a | 2799 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2800 | * |
672bba3a CL |
2801 | * Note that concurrent frees may occur while we hold the |
2802 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2803 | */ |
2804 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2805 | if (!page->inuse && slab_trylock(page)) { | |
2806 | /* | |
2807 | * Must hold slab lock here because slab_free | |
2808 | * may have freed the last object and be | |
2809 | * waiting to release the slab. | |
2810 | */ | |
62e346a8 | 2811 | __remove_partial(n, page); |
2086d26a CL |
2812 | slab_unlock(page); |
2813 | discard_slab(s, page); | |
2814 | } else { | |
fcda3d89 CL |
2815 | list_move(&page->lru, |
2816 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2817 | } |
2818 | } | |
2819 | ||
2086d26a | 2820 | /* |
672bba3a CL |
2821 | * Rebuild the partial list with the slabs filled up most |
2822 | * first and the least used slabs at the end. | |
2086d26a | 2823 | */ |
834f3d11 | 2824 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2825 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2826 | ||
2086d26a CL |
2827 | spin_unlock_irqrestore(&n->list_lock, flags); |
2828 | } | |
2829 | ||
2830 | kfree(slabs_by_inuse); | |
2831 | return 0; | |
2832 | } | |
2833 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2834 | ||
92a5bbc1 | 2835 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
2836 | static int slab_mem_going_offline_callback(void *arg) |
2837 | { | |
2838 | struct kmem_cache *s; | |
2839 | ||
2840 | down_read(&slub_lock); | |
2841 | list_for_each_entry(s, &slab_caches, list) | |
2842 | kmem_cache_shrink(s); | |
2843 | up_read(&slub_lock); | |
2844 | ||
2845 | return 0; | |
2846 | } | |
2847 | ||
2848 | static void slab_mem_offline_callback(void *arg) | |
2849 | { | |
2850 | struct kmem_cache_node *n; | |
2851 | struct kmem_cache *s; | |
2852 | struct memory_notify *marg = arg; | |
2853 | int offline_node; | |
2854 | ||
2855 | offline_node = marg->status_change_nid; | |
2856 | ||
2857 | /* | |
2858 | * If the node still has available memory. we need kmem_cache_node | |
2859 | * for it yet. | |
2860 | */ | |
2861 | if (offline_node < 0) | |
2862 | return; | |
2863 | ||
2864 | down_read(&slub_lock); | |
2865 | list_for_each_entry(s, &slab_caches, list) { | |
2866 | n = get_node(s, offline_node); | |
2867 | if (n) { | |
2868 | /* | |
2869 | * if n->nr_slabs > 0, slabs still exist on the node | |
2870 | * that is going down. We were unable to free them, | |
c9404c9c | 2871 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
2872 | * callback. So, we must fail. |
2873 | */ | |
0f389ec6 | 2874 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2875 | |
2876 | s->node[offline_node] = NULL; | |
8de66a0c | 2877 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
2878 | } |
2879 | } | |
2880 | up_read(&slub_lock); | |
2881 | } | |
2882 | ||
2883 | static int slab_mem_going_online_callback(void *arg) | |
2884 | { | |
2885 | struct kmem_cache_node *n; | |
2886 | struct kmem_cache *s; | |
2887 | struct memory_notify *marg = arg; | |
2888 | int nid = marg->status_change_nid; | |
2889 | int ret = 0; | |
2890 | ||
2891 | /* | |
2892 | * If the node's memory is already available, then kmem_cache_node is | |
2893 | * already created. Nothing to do. | |
2894 | */ | |
2895 | if (nid < 0) | |
2896 | return 0; | |
2897 | ||
2898 | /* | |
0121c619 | 2899 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
2900 | * allocate a kmem_cache_node structure in order to bring the node |
2901 | * online. | |
2902 | */ | |
2903 | down_read(&slub_lock); | |
2904 | list_for_each_entry(s, &slab_caches, list) { | |
2905 | /* | |
2906 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2907 | * since memory is not yet available from the node that | |
2908 | * is brought up. | |
2909 | */ | |
8de66a0c | 2910 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
2911 | if (!n) { |
2912 | ret = -ENOMEM; | |
2913 | goto out; | |
2914 | } | |
5595cffc | 2915 | init_kmem_cache_node(n, s); |
b9049e23 YG |
2916 | s->node[nid] = n; |
2917 | } | |
2918 | out: | |
2919 | up_read(&slub_lock); | |
2920 | return ret; | |
2921 | } | |
2922 | ||
2923 | static int slab_memory_callback(struct notifier_block *self, | |
2924 | unsigned long action, void *arg) | |
2925 | { | |
2926 | int ret = 0; | |
2927 | ||
2928 | switch (action) { | |
2929 | case MEM_GOING_ONLINE: | |
2930 | ret = slab_mem_going_online_callback(arg); | |
2931 | break; | |
2932 | case MEM_GOING_OFFLINE: | |
2933 | ret = slab_mem_going_offline_callback(arg); | |
2934 | break; | |
2935 | case MEM_OFFLINE: | |
2936 | case MEM_CANCEL_ONLINE: | |
2937 | slab_mem_offline_callback(arg); | |
2938 | break; | |
2939 | case MEM_ONLINE: | |
2940 | case MEM_CANCEL_OFFLINE: | |
2941 | break; | |
2942 | } | |
dc19f9db KH |
2943 | if (ret) |
2944 | ret = notifier_from_errno(ret); | |
2945 | else | |
2946 | ret = NOTIFY_OK; | |
b9049e23 YG |
2947 | return ret; |
2948 | } | |
2949 | ||
2950 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2951 | ||
81819f0f CL |
2952 | /******************************************************************** |
2953 | * Basic setup of slabs | |
2954 | *******************************************************************/ | |
2955 | ||
51df1142 CL |
2956 | /* |
2957 | * Used for early kmem_cache structures that were allocated using | |
2958 | * the page allocator | |
2959 | */ | |
2960 | ||
2961 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
2962 | { | |
2963 | int node; | |
2964 | ||
2965 | list_add(&s->list, &slab_caches); | |
2966 | s->refcount = -1; | |
2967 | ||
2968 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
2969 | struct kmem_cache_node *n = get_node(s, node); | |
2970 | struct page *p; | |
2971 | ||
2972 | if (n) { | |
2973 | list_for_each_entry(p, &n->partial, lru) | |
2974 | p->slab = s; | |
2975 | ||
2976 | #ifdef CONFIG_SLAB_DEBUG | |
2977 | list_for_each_entry(p, &n->full, lru) | |
2978 | p->slab = s; | |
2979 | #endif | |
2980 | } | |
2981 | } | |
2982 | } | |
2983 | ||
81819f0f CL |
2984 | void __init kmem_cache_init(void) |
2985 | { | |
2986 | int i; | |
4b356be0 | 2987 | int caches = 0; |
51df1142 CL |
2988 | struct kmem_cache *temp_kmem_cache; |
2989 | int order; | |
51df1142 CL |
2990 | struct kmem_cache *temp_kmem_cache_node; |
2991 | unsigned long kmalloc_size; | |
2992 | ||
2993 | kmem_size = offsetof(struct kmem_cache, node) + | |
2994 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
2995 | ||
2996 | /* Allocate two kmem_caches from the page allocator */ | |
2997 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
2998 | order = get_order(2 * kmalloc_size); | |
2999 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3000 | ||
81819f0f CL |
3001 | /* |
3002 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3003 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3004 | * kmem_cache_open for slab_state == DOWN. |
3005 | */ | |
51df1142 CL |
3006 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3007 | ||
3008 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3009 | sizeof(struct kmem_cache_node), | |
3010 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3011 | |
0c40ba4f | 3012 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3013 | |
3014 | /* Able to allocate the per node structures */ | |
3015 | slab_state = PARTIAL; | |
3016 | ||
51df1142 CL |
3017 | temp_kmem_cache = kmem_cache; |
3018 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3019 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3020 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3021 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3022 | |
51df1142 CL |
3023 | /* |
3024 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3025 | * kmem_cache_node is separately allocated so no need to | |
3026 | * update any list pointers. | |
3027 | */ | |
3028 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3029 | |
51df1142 CL |
3030 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3031 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3032 | ||
3033 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3034 | ||
3035 | caches++; | |
51df1142 CL |
3036 | kmem_cache_bootstrap_fixup(kmem_cache); |
3037 | caches++; | |
3038 | /* Free temporary boot structure */ | |
3039 | free_pages((unsigned long)temp_kmem_cache, order); | |
3040 | ||
3041 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3042 | |
3043 | /* | |
3044 | * Patch up the size_index table if we have strange large alignment | |
3045 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3046 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3047 | * |
3048 | * Largest permitted alignment is 256 bytes due to the way we | |
3049 | * handle the index determination for the smaller caches. | |
3050 | * | |
3051 | * Make sure that nothing crazy happens if someone starts tinkering | |
3052 | * around with ARCH_KMALLOC_MINALIGN | |
3053 | */ | |
3054 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3055 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3056 | ||
acdfcd04 AK |
3057 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3058 | int elem = size_index_elem(i); | |
3059 | if (elem >= ARRAY_SIZE(size_index)) | |
3060 | break; | |
3061 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3062 | } | |
f1b26339 | 3063 | |
acdfcd04 AK |
3064 | if (KMALLOC_MIN_SIZE == 64) { |
3065 | /* | |
3066 | * The 96 byte size cache is not used if the alignment | |
3067 | * is 64 byte. | |
3068 | */ | |
3069 | for (i = 64 + 8; i <= 96; i += 8) | |
3070 | size_index[size_index_elem(i)] = 7; | |
3071 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3072 | /* |
3073 | * The 192 byte sized cache is not used if the alignment | |
3074 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3075 | * instead. | |
3076 | */ | |
3077 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3078 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3079 | } |
3080 | ||
51df1142 CL |
3081 | /* Caches that are not of the two-to-the-power-of size */ |
3082 | if (KMALLOC_MIN_SIZE <= 32) { | |
3083 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3084 | caches++; | |
3085 | } | |
3086 | ||
3087 | if (KMALLOC_MIN_SIZE <= 64) { | |
3088 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3089 | caches++; | |
3090 | } | |
3091 | ||
3092 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3093 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3094 | caches++; | |
3095 | } | |
3096 | ||
81819f0f CL |
3097 | slab_state = UP; |
3098 | ||
3099 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3100 | if (KMALLOC_MIN_SIZE <= 32) { |
3101 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3102 | BUG_ON(!kmalloc_caches[1]->name); | |
3103 | } | |
3104 | ||
3105 | if (KMALLOC_MIN_SIZE <= 64) { | |
3106 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3107 | BUG_ON(!kmalloc_caches[2]->name); | |
3108 | } | |
3109 | ||
d7278bd7 CL |
3110 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3111 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3112 | ||
3113 | BUG_ON(!s); | |
51df1142 | 3114 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3115 | } |
81819f0f CL |
3116 | |
3117 | #ifdef CONFIG_SMP | |
3118 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3119 | #endif |
81819f0f | 3120 | |
55136592 | 3121 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3122 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3123 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3124 | |
51df1142 | 3125 | if (s && s->size) { |
55136592 CL |
3126 | char *name = kasprintf(GFP_NOWAIT, |
3127 | "dma-kmalloc-%d", s->objsize); | |
3128 | ||
3129 | BUG_ON(!name); | |
51df1142 CL |
3130 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3131 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3132 | } |
3133 | } | |
3134 | #endif | |
3adbefee IM |
3135 | printk(KERN_INFO |
3136 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3137 | " CPUs=%d, Nodes=%d\n", |
3138 | caches, cache_line_size(), | |
81819f0f CL |
3139 | slub_min_order, slub_max_order, slub_min_objects, |
3140 | nr_cpu_ids, nr_node_ids); | |
3141 | } | |
3142 | ||
7e85ee0c PE |
3143 | void __init kmem_cache_init_late(void) |
3144 | { | |
7e85ee0c PE |
3145 | } |
3146 | ||
81819f0f CL |
3147 | /* |
3148 | * Find a mergeable slab cache | |
3149 | */ | |
3150 | static int slab_unmergeable(struct kmem_cache *s) | |
3151 | { | |
3152 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3153 | return 1; | |
3154 | ||
c59def9f | 3155 | if (s->ctor) |
81819f0f CL |
3156 | return 1; |
3157 | ||
8ffa6875 CL |
3158 | /* |
3159 | * We may have set a slab to be unmergeable during bootstrap. | |
3160 | */ | |
3161 | if (s->refcount < 0) | |
3162 | return 1; | |
3163 | ||
81819f0f CL |
3164 | return 0; |
3165 | } | |
3166 | ||
3167 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3168 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3169 | void (*ctor)(void *)) |
81819f0f | 3170 | { |
5b95a4ac | 3171 | struct kmem_cache *s; |
81819f0f CL |
3172 | |
3173 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3174 | return NULL; | |
3175 | ||
c59def9f | 3176 | if (ctor) |
81819f0f CL |
3177 | return NULL; |
3178 | ||
3179 | size = ALIGN(size, sizeof(void *)); | |
3180 | align = calculate_alignment(flags, align, size); | |
3181 | size = ALIGN(size, align); | |
ba0268a8 | 3182 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3183 | |
5b95a4ac | 3184 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3185 | if (slab_unmergeable(s)) |
3186 | continue; | |
3187 | ||
3188 | if (size > s->size) | |
3189 | continue; | |
3190 | ||
ba0268a8 | 3191 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3192 | continue; |
3193 | /* | |
3194 | * Check if alignment is compatible. | |
3195 | * Courtesy of Adrian Drzewiecki | |
3196 | */ | |
06428780 | 3197 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3198 | continue; |
3199 | ||
3200 | if (s->size - size >= sizeof(void *)) | |
3201 | continue; | |
3202 | ||
3203 | return s; | |
3204 | } | |
3205 | return NULL; | |
3206 | } | |
3207 | ||
3208 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3209 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3210 | { |
3211 | struct kmem_cache *s; | |
84c1cf62 | 3212 | char *n; |
81819f0f | 3213 | |
fe1ff49d BH |
3214 | if (WARN_ON(!name)) |
3215 | return NULL; | |
3216 | ||
81819f0f | 3217 | down_write(&slub_lock); |
ba0268a8 | 3218 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3219 | if (s) { |
3220 | s->refcount++; | |
3221 | /* | |
3222 | * Adjust the object sizes so that we clear | |
3223 | * the complete object on kzalloc. | |
3224 | */ | |
3225 | s->objsize = max(s->objsize, (int)size); | |
3226 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3227 | |
7b8f3b66 | 3228 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3229 | s->refcount--; |
81819f0f | 3230 | goto err; |
7b8f3b66 | 3231 | } |
2bce6485 | 3232 | up_write(&slub_lock); |
a0e1d1be CL |
3233 | return s; |
3234 | } | |
6446faa2 | 3235 | |
84c1cf62 PE |
3236 | n = kstrdup(name, GFP_KERNEL); |
3237 | if (!n) | |
3238 | goto err; | |
3239 | ||
a0e1d1be CL |
3240 | s = kmalloc(kmem_size, GFP_KERNEL); |
3241 | if (s) { | |
84c1cf62 | 3242 | if (kmem_cache_open(s, n, |
c59def9f | 3243 | size, align, flags, ctor)) { |
81819f0f | 3244 | list_add(&s->list, &slab_caches); |
7b8f3b66 | 3245 | if (sysfs_slab_add(s)) { |
7b8f3b66 | 3246 | list_del(&s->list); |
84c1cf62 | 3247 | kfree(n); |
7b8f3b66 | 3248 | kfree(s); |
a0e1d1be | 3249 | goto err; |
7b8f3b66 | 3250 | } |
2bce6485 | 3251 | up_write(&slub_lock); |
a0e1d1be CL |
3252 | return s; |
3253 | } | |
84c1cf62 | 3254 | kfree(n); |
a0e1d1be | 3255 | kfree(s); |
81819f0f | 3256 | } |
68cee4f1 | 3257 | err: |
81819f0f | 3258 | up_write(&slub_lock); |
81819f0f | 3259 | |
81819f0f CL |
3260 | if (flags & SLAB_PANIC) |
3261 | panic("Cannot create slabcache %s\n", name); | |
3262 | else | |
3263 | s = NULL; | |
3264 | return s; | |
3265 | } | |
3266 | EXPORT_SYMBOL(kmem_cache_create); | |
3267 | ||
81819f0f | 3268 | #ifdef CONFIG_SMP |
81819f0f | 3269 | /* |
672bba3a CL |
3270 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3271 | * necessary. | |
81819f0f CL |
3272 | */ |
3273 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3274 | unsigned long action, void *hcpu) | |
3275 | { | |
3276 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3277 | struct kmem_cache *s; |
3278 | unsigned long flags; | |
81819f0f CL |
3279 | |
3280 | switch (action) { | |
3281 | case CPU_UP_CANCELED: | |
8bb78442 | 3282 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3283 | case CPU_DEAD: |
8bb78442 | 3284 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3285 | down_read(&slub_lock); |
3286 | list_for_each_entry(s, &slab_caches, list) { | |
3287 | local_irq_save(flags); | |
3288 | __flush_cpu_slab(s, cpu); | |
3289 | local_irq_restore(flags); | |
3290 | } | |
3291 | up_read(&slub_lock); | |
81819f0f CL |
3292 | break; |
3293 | default: | |
3294 | break; | |
3295 | } | |
3296 | return NOTIFY_OK; | |
3297 | } | |
3298 | ||
06428780 | 3299 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3300 | .notifier_call = slab_cpuup_callback |
06428780 | 3301 | }; |
81819f0f CL |
3302 | |
3303 | #endif | |
3304 | ||
ce71e27c | 3305 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3306 | { |
aadb4bc4 | 3307 | struct kmem_cache *s; |
94b528d0 | 3308 | void *ret; |
aadb4bc4 | 3309 | |
ffadd4d0 | 3310 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3311 | return kmalloc_large(size, gfpflags); |
3312 | ||
aadb4bc4 | 3313 | s = get_slab(size, gfpflags); |
81819f0f | 3314 | |
2408c550 | 3315 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3316 | return s; |
81819f0f | 3317 | |
2154a336 | 3318 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 EGM |
3319 | |
3320 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3321 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3322 | |
3323 | return ret; | |
81819f0f CL |
3324 | } |
3325 | ||
5d1f57e4 | 3326 | #ifdef CONFIG_NUMA |
81819f0f | 3327 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3328 | int node, unsigned long caller) |
81819f0f | 3329 | { |
aadb4bc4 | 3330 | struct kmem_cache *s; |
94b528d0 | 3331 | void *ret; |
aadb4bc4 | 3332 | |
d3e14aa3 XF |
3333 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3334 | ret = kmalloc_large_node(size, gfpflags, node); | |
3335 | ||
3336 | trace_kmalloc_node(caller, ret, | |
3337 | size, PAGE_SIZE << get_order(size), | |
3338 | gfpflags, node); | |
3339 | ||
3340 | return ret; | |
3341 | } | |
eada35ef | 3342 | |
aadb4bc4 | 3343 | s = get_slab(size, gfpflags); |
81819f0f | 3344 | |
2408c550 | 3345 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3346 | return s; |
81819f0f | 3347 | |
94b528d0 EGM |
3348 | ret = slab_alloc(s, gfpflags, node, caller); |
3349 | ||
3350 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3351 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3352 | |
3353 | return ret; | |
81819f0f | 3354 | } |
5d1f57e4 | 3355 | #endif |
81819f0f | 3356 | |
ab4d5ed5 | 3357 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3358 | static int count_inuse(struct page *page) |
3359 | { | |
3360 | return page->inuse; | |
3361 | } | |
3362 | ||
3363 | static int count_total(struct page *page) | |
3364 | { | |
3365 | return page->objects; | |
3366 | } | |
ab4d5ed5 | 3367 | #endif |
205ab99d | 3368 | |
ab4d5ed5 | 3369 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3370 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3371 | unsigned long *map) | |
53e15af0 CL |
3372 | { |
3373 | void *p; | |
a973e9dd | 3374 | void *addr = page_address(page); |
53e15af0 CL |
3375 | |
3376 | if (!check_slab(s, page) || | |
3377 | !on_freelist(s, page, NULL)) | |
3378 | return 0; | |
3379 | ||
3380 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3381 | bitmap_zero(map, page->objects); |
53e15af0 | 3382 | |
7656c72b CL |
3383 | for_each_free_object(p, s, page->freelist) { |
3384 | set_bit(slab_index(p, s, addr), map); | |
37d57443 | 3385 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) |
53e15af0 CL |
3386 | return 0; |
3387 | } | |
3388 | ||
224a88be | 3389 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3390 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3391 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3392 | return 0; |
3393 | return 1; | |
3394 | } | |
3395 | ||
434e245d CL |
3396 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3397 | unsigned long *map) | |
53e15af0 CL |
3398 | { |
3399 | if (slab_trylock(page)) { | |
434e245d | 3400 | validate_slab(s, page, map); |
53e15af0 CL |
3401 | slab_unlock(page); |
3402 | } else | |
3403 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3404 | s->name, page); | |
53e15af0 CL |
3405 | } |
3406 | ||
434e245d CL |
3407 | static int validate_slab_node(struct kmem_cache *s, |
3408 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3409 | { |
3410 | unsigned long count = 0; | |
3411 | struct page *page; | |
3412 | unsigned long flags; | |
3413 | ||
3414 | spin_lock_irqsave(&n->list_lock, flags); | |
3415 | ||
3416 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3417 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3418 | count++; |
3419 | } | |
3420 | if (count != n->nr_partial) | |
3421 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3422 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3423 | ||
3424 | if (!(s->flags & SLAB_STORE_USER)) | |
3425 | goto out; | |
3426 | ||
3427 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3428 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3429 | count++; |
3430 | } | |
3431 | if (count != atomic_long_read(&n->nr_slabs)) | |
3432 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3433 | "counter=%ld\n", s->name, count, | |
3434 | atomic_long_read(&n->nr_slabs)); | |
3435 | ||
3436 | out: | |
3437 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3438 | return count; | |
3439 | } | |
3440 | ||
434e245d | 3441 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3442 | { |
3443 | int node; | |
3444 | unsigned long count = 0; | |
205ab99d | 3445 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3446 | sizeof(unsigned long), GFP_KERNEL); |
3447 | ||
3448 | if (!map) | |
3449 | return -ENOMEM; | |
53e15af0 CL |
3450 | |
3451 | flush_all(s); | |
f64dc58c | 3452 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3453 | struct kmem_cache_node *n = get_node(s, node); |
3454 | ||
434e245d | 3455 | count += validate_slab_node(s, n, map); |
53e15af0 | 3456 | } |
434e245d | 3457 | kfree(map); |
53e15af0 CL |
3458 | return count; |
3459 | } | |
88a420e4 | 3460 | /* |
672bba3a | 3461 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3462 | * and freed. |
3463 | */ | |
3464 | ||
3465 | struct location { | |
3466 | unsigned long count; | |
ce71e27c | 3467 | unsigned long addr; |
45edfa58 CL |
3468 | long long sum_time; |
3469 | long min_time; | |
3470 | long max_time; | |
3471 | long min_pid; | |
3472 | long max_pid; | |
174596a0 | 3473 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3474 | nodemask_t nodes; |
88a420e4 CL |
3475 | }; |
3476 | ||
3477 | struct loc_track { | |
3478 | unsigned long max; | |
3479 | unsigned long count; | |
3480 | struct location *loc; | |
3481 | }; | |
3482 | ||
3483 | static void free_loc_track(struct loc_track *t) | |
3484 | { | |
3485 | if (t->max) | |
3486 | free_pages((unsigned long)t->loc, | |
3487 | get_order(sizeof(struct location) * t->max)); | |
3488 | } | |
3489 | ||
68dff6a9 | 3490 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3491 | { |
3492 | struct location *l; | |
3493 | int order; | |
3494 | ||
88a420e4 CL |
3495 | order = get_order(sizeof(struct location) * max); |
3496 | ||
68dff6a9 | 3497 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3498 | if (!l) |
3499 | return 0; | |
3500 | ||
3501 | if (t->count) { | |
3502 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3503 | free_loc_track(t); | |
3504 | } | |
3505 | t->max = max; | |
3506 | t->loc = l; | |
3507 | return 1; | |
3508 | } | |
3509 | ||
3510 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3511 | const struct track *track) |
88a420e4 CL |
3512 | { |
3513 | long start, end, pos; | |
3514 | struct location *l; | |
ce71e27c | 3515 | unsigned long caddr; |
45edfa58 | 3516 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3517 | |
3518 | start = -1; | |
3519 | end = t->count; | |
3520 | ||
3521 | for ( ; ; ) { | |
3522 | pos = start + (end - start + 1) / 2; | |
3523 | ||
3524 | /* | |
3525 | * There is nothing at "end". If we end up there | |
3526 | * we need to add something to before end. | |
3527 | */ | |
3528 | if (pos == end) | |
3529 | break; | |
3530 | ||
3531 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3532 | if (track->addr == caddr) { |
3533 | ||
3534 | l = &t->loc[pos]; | |
3535 | l->count++; | |
3536 | if (track->when) { | |
3537 | l->sum_time += age; | |
3538 | if (age < l->min_time) | |
3539 | l->min_time = age; | |
3540 | if (age > l->max_time) | |
3541 | l->max_time = age; | |
3542 | ||
3543 | if (track->pid < l->min_pid) | |
3544 | l->min_pid = track->pid; | |
3545 | if (track->pid > l->max_pid) | |
3546 | l->max_pid = track->pid; | |
3547 | ||
174596a0 RR |
3548 | cpumask_set_cpu(track->cpu, |
3549 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3550 | } |
3551 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3552 | return 1; |
3553 | } | |
3554 | ||
45edfa58 | 3555 | if (track->addr < caddr) |
88a420e4 CL |
3556 | end = pos; |
3557 | else | |
3558 | start = pos; | |
3559 | } | |
3560 | ||
3561 | /* | |
672bba3a | 3562 | * Not found. Insert new tracking element. |
88a420e4 | 3563 | */ |
68dff6a9 | 3564 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3565 | return 0; |
3566 | ||
3567 | l = t->loc + pos; | |
3568 | if (pos < t->count) | |
3569 | memmove(l + 1, l, | |
3570 | (t->count - pos) * sizeof(struct location)); | |
3571 | t->count++; | |
3572 | l->count = 1; | |
45edfa58 CL |
3573 | l->addr = track->addr; |
3574 | l->sum_time = age; | |
3575 | l->min_time = age; | |
3576 | l->max_time = age; | |
3577 | l->min_pid = track->pid; | |
3578 | l->max_pid = track->pid; | |
174596a0 RR |
3579 | cpumask_clear(to_cpumask(l->cpus)); |
3580 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3581 | nodes_clear(l->nodes); |
3582 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3583 | return 1; |
3584 | } | |
3585 | ||
3586 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 3587 | struct page *page, enum track_item alloc, |
a5dd5c11 | 3588 | unsigned long *map) |
88a420e4 | 3589 | { |
a973e9dd | 3590 | void *addr = page_address(page); |
88a420e4 CL |
3591 | void *p; |
3592 | ||
39b26464 | 3593 | bitmap_zero(map, page->objects); |
7656c72b CL |
3594 | for_each_free_object(p, s, page->freelist) |
3595 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3596 | |
224a88be | 3597 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3598 | if (!test_bit(slab_index(p, s, addr), map)) |
3599 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3600 | } |
3601 | ||
3602 | static int list_locations(struct kmem_cache *s, char *buf, | |
3603 | enum track_item alloc) | |
3604 | { | |
e374d483 | 3605 | int len = 0; |
88a420e4 | 3606 | unsigned long i; |
68dff6a9 | 3607 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 3608 | int node; |
bbd7d57b ED |
3609 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
3610 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 3611 | |
bbd7d57b ED |
3612 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
3613 | GFP_TEMPORARY)) { | |
3614 | kfree(map); | |
68dff6a9 | 3615 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 3616 | } |
88a420e4 CL |
3617 | /* Push back cpu slabs */ |
3618 | flush_all(s); | |
3619 | ||
f64dc58c | 3620 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3621 | struct kmem_cache_node *n = get_node(s, node); |
3622 | unsigned long flags; | |
3623 | struct page *page; | |
3624 | ||
9e86943b | 3625 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3626 | continue; |
3627 | ||
3628 | spin_lock_irqsave(&n->list_lock, flags); | |
3629 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 3630 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 3631 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 3632 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
3633 | spin_unlock_irqrestore(&n->list_lock, flags); |
3634 | } | |
3635 | ||
3636 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3637 | struct location *l = &t.loc[i]; |
88a420e4 | 3638 | |
9c246247 | 3639 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3640 | break; |
e374d483 | 3641 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3642 | |
3643 | if (l->addr) | |
62c70bce | 3644 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 3645 | else |
e374d483 | 3646 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3647 | |
3648 | if (l->sum_time != l->min_time) { | |
e374d483 | 3649 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3650 | l->min_time, |
3651 | (long)div_u64(l->sum_time, l->count), | |
3652 | l->max_time); | |
45edfa58 | 3653 | } else |
e374d483 | 3654 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3655 | l->min_time); |
3656 | ||
3657 | if (l->min_pid != l->max_pid) | |
e374d483 | 3658 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3659 | l->min_pid, l->max_pid); |
3660 | else | |
e374d483 | 3661 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3662 | l->min_pid); |
3663 | ||
174596a0 RR |
3664 | if (num_online_cpus() > 1 && |
3665 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3666 | len < PAGE_SIZE - 60) { |
3667 | len += sprintf(buf + len, " cpus="); | |
3668 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3669 | to_cpumask(l->cpus)); |
45edfa58 CL |
3670 | } |
3671 | ||
62bc62a8 | 3672 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3673 | len < PAGE_SIZE - 60) { |
3674 | len += sprintf(buf + len, " nodes="); | |
3675 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3676 | l->nodes); |
3677 | } | |
3678 | ||
e374d483 | 3679 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3680 | } |
3681 | ||
3682 | free_loc_track(&t); | |
bbd7d57b | 3683 | kfree(map); |
88a420e4 | 3684 | if (!t.count) |
e374d483 HH |
3685 | len += sprintf(buf, "No data\n"); |
3686 | return len; | |
88a420e4 | 3687 | } |
ab4d5ed5 | 3688 | #endif |
88a420e4 | 3689 | |
a5a84755 CL |
3690 | #ifdef SLUB_RESILIENCY_TEST |
3691 | static void resiliency_test(void) | |
3692 | { | |
3693 | u8 *p; | |
3694 | ||
3695 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
3696 | ||
3697 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3698 | printk(KERN_ERR "-----------------------\n"); | |
3699 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3700 | ||
3701 | p = kzalloc(16, GFP_KERNEL); | |
3702 | p[16] = 0x12; | |
3703 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3704 | " 0x12->0x%p\n\n", p + 16); | |
3705 | ||
3706 | validate_slab_cache(kmalloc_caches[4]); | |
3707 | ||
3708 | /* Hmmm... The next two are dangerous */ | |
3709 | p = kzalloc(32, GFP_KERNEL); | |
3710 | p[32 + sizeof(void *)] = 0x34; | |
3711 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3712 | " 0x34 -> -0x%p\n", p); | |
3713 | printk(KERN_ERR | |
3714 | "If allocated object is overwritten then not detectable\n\n"); | |
3715 | ||
3716 | validate_slab_cache(kmalloc_caches[5]); | |
3717 | p = kzalloc(64, GFP_KERNEL); | |
3718 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3719 | *p = 0x56; | |
3720 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3721 | p); | |
3722 | printk(KERN_ERR | |
3723 | "If allocated object is overwritten then not detectable\n\n"); | |
3724 | validate_slab_cache(kmalloc_caches[6]); | |
3725 | ||
3726 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3727 | p = kzalloc(128, GFP_KERNEL); | |
3728 | kfree(p); | |
3729 | *p = 0x78; | |
3730 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3731 | validate_slab_cache(kmalloc_caches[7]); | |
3732 | ||
3733 | p = kzalloc(256, GFP_KERNEL); | |
3734 | kfree(p); | |
3735 | p[50] = 0x9a; | |
3736 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
3737 | p); | |
3738 | validate_slab_cache(kmalloc_caches[8]); | |
3739 | ||
3740 | p = kzalloc(512, GFP_KERNEL); | |
3741 | kfree(p); | |
3742 | p[512] = 0xab; | |
3743 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3744 | validate_slab_cache(kmalloc_caches[9]); | |
3745 | } | |
3746 | #else | |
3747 | #ifdef CONFIG_SYSFS | |
3748 | static void resiliency_test(void) {}; | |
3749 | #endif | |
3750 | #endif | |
3751 | ||
ab4d5ed5 | 3752 | #ifdef CONFIG_SYSFS |
81819f0f | 3753 | enum slab_stat_type { |
205ab99d CL |
3754 | SL_ALL, /* All slabs */ |
3755 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3756 | SL_CPU, /* Only slabs used for cpu caches */ | |
3757 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3758 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3759 | }; |
3760 | ||
205ab99d | 3761 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3762 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3763 | #define SO_CPU (1 << SL_CPU) | |
3764 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3765 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3766 | |
62e5c4b4 CG |
3767 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3768 | char *buf, unsigned long flags) | |
81819f0f CL |
3769 | { |
3770 | unsigned long total = 0; | |
81819f0f CL |
3771 | int node; |
3772 | int x; | |
3773 | unsigned long *nodes; | |
3774 | unsigned long *per_cpu; | |
3775 | ||
3776 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3777 | if (!nodes) |
3778 | return -ENOMEM; | |
81819f0f CL |
3779 | per_cpu = nodes + nr_node_ids; |
3780 | ||
205ab99d CL |
3781 | if (flags & SO_CPU) { |
3782 | int cpu; | |
81819f0f | 3783 | |
205ab99d | 3784 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 3785 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
dfb4f096 | 3786 | |
205ab99d CL |
3787 | if (!c || c->node < 0) |
3788 | continue; | |
3789 | ||
3790 | if (c->page) { | |
3791 | if (flags & SO_TOTAL) | |
3792 | x = c->page->objects; | |
3793 | else if (flags & SO_OBJECTS) | |
3794 | x = c->page->inuse; | |
81819f0f CL |
3795 | else |
3796 | x = 1; | |
205ab99d | 3797 | |
81819f0f | 3798 | total += x; |
205ab99d | 3799 | nodes[c->node] += x; |
81819f0f | 3800 | } |
205ab99d | 3801 | per_cpu[c->node]++; |
81819f0f CL |
3802 | } |
3803 | } | |
3804 | ||
04d94879 | 3805 | lock_memory_hotplug(); |
ab4d5ed5 | 3806 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
3807 | if (flags & SO_ALL) { |
3808 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3809 | struct kmem_cache_node *n = get_node(s, node); | |
3810 | ||
3811 | if (flags & SO_TOTAL) | |
3812 | x = atomic_long_read(&n->total_objects); | |
3813 | else if (flags & SO_OBJECTS) | |
3814 | x = atomic_long_read(&n->total_objects) - | |
3815 | count_partial(n, count_free); | |
81819f0f | 3816 | |
81819f0f | 3817 | else |
205ab99d | 3818 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3819 | total += x; |
3820 | nodes[node] += x; | |
3821 | } | |
3822 | ||
ab4d5ed5 CL |
3823 | } else |
3824 | #endif | |
3825 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
3826 | for_each_node_state(node, N_NORMAL_MEMORY) { |
3827 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3828 | |
205ab99d CL |
3829 | if (flags & SO_TOTAL) |
3830 | x = count_partial(n, count_total); | |
3831 | else if (flags & SO_OBJECTS) | |
3832 | x = count_partial(n, count_inuse); | |
81819f0f | 3833 | else |
205ab99d | 3834 | x = n->nr_partial; |
81819f0f CL |
3835 | total += x; |
3836 | nodes[node] += x; | |
3837 | } | |
3838 | } | |
81819f0f CL |
3839 | x = sprintf(buf, "%lu", total); |
3840 | #ifdef CONFIG_NUMA | |
f64dc58c | 3841 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3842 | if (nodes[node]) |
3843 | x += sprintf(buf + x, " N%d=%lu", | |
3844 | node, nodes[node]); | |
3845 | #endif | |
04d94879 | 3846 | unlock_memory_hotplug(); |
81819f0f CL |
3847 | kfree(nodes); |
3848 | return x + sprintf(buf + x, "\n"); | |
3849 | } | |
3850 | ||
ab4d5ed5 | 3851 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3852 | static int any_slab_objects(struct kmem_cache *s) |
3853 | { | |
3854 | int node; | |
81819f0f | 3855 | |
dfb4f096 | 3856 | for_each_online_node(node) { |
81819f0f CL |
3857 | struct kmem_cache_node *n = get_node(s, node); |
3858 | ||
dfb4f096 CL |
3859 | if (!n) |
3860 | continue; | |
3861 | ||
4ea33e2d | 3862 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3863 | return 1; |
3864 | } | |
3865 | return 0; | |
3866 | } | |
ab4d5ed5 | 3867 | #endif |
81819f0f CL |
3868 | |
3869 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3870 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3871 | ||
3872 | struct slab_attribute { | |
3873 | struct attribute attr; | |
3874 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3875 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3876 | }; | |
3877 | ||
3878 | #define SLAB_ATTR_RO(_name) \ | |
3879 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3880 | ||
3881 | #define SLAB_ATTR(_name) \ | |
3882 | static struct slab_attribute _name##_attr = \ | |
3883 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3884 | ||
81819f0f CL |
3885 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3886 | { | |
3887 | return sprintf(buf, "%d\n", s->size); | |
3888 | } | |
3889 | SLAB_ATTR_RO(slab_size); | |
3890 | ||
3891 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3892 | { | |
3893 | return sprintf(buf, "%d\n", s->align); | |
3894 | } | |
3895 | SLAB_ATTR_RO(align); | |
3896 | ||
3897 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3898 | { | |
3899 | return sprintf(buf, "%d\n", s->objsize); | |
3900 | } | |
3901 | SLAB_ATTR_RO(object_size); | |
3902 | ||
3903 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3904 | { | |
834f3d11 | 3905 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3906 | } |
3907 | SLAB_ATTR_RO(objs_per_slab); | |
3908 | ||
06b285dc CL |
3909 | static ssize_t order_store(struct kmem_cache *s, |
3910 | const char *buf, size_t length) | |
3911 | { | |
0121c619 CL |
3912 | unsigned long order; |
3913 | int err; | |
3914 | ||
3915 | err = strict_strtoul(buf, 10, &order); | |
3916 | if (err) | |
3917 | return err; | |
06b285dc CL |
3918 | |
3919 | if (order > slub_max_order || order < slub_min_order) | |
3920 | return -EINVAL; | |
3921 | ||
3922 | calculate_sizes(s, order); | |
3923 | return length; | |
3924 | } | |
3925 | ||
81819f0f CL |
3926 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
3927 | { | |
834f3d11 | 3928 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 3929 | } |
06b285dc | 3930 | SLAB_ATTR(order); |
81819f0f | 3931 | |
73d342b1 DR |
3932 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
3933 | { | |
3934 | return sprintf(buf, "%lu\n", s->min_partial); | |
3935 | } | |
3936 | ||
3937 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
3938 | size_t length) | |
3939 | { | |
3940 | unsigned long min; | |
3941 | int err; | |
3942 | ||
3943 | err = strict_strtoul(buf, 10, &min); | |
3944 | if (err) | |
3945 | return err; | |
3946 | ||
c0bdb232 | 3947 | set_min_partial(s, min); |
73d342b1 DR |
3948 | return length; |
3949 | } | |
3950 | SLAB_ATTR(min_partial); | |
3951 | ||
81819f0f CL |
3952 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
3953 | { | |
62c70bce JP |
3954 | if (!s->ctor) |
3955 | return 0; | |
3956 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
3957 | } |
3958 | SLAB_ATTR_RO(ctor); | |
3959 | ||
81819f0f CL |
3960 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3961 | { | |
3962 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3963 | } | |
3964 | SLAB_ATTR_RO(aliases); | |
3965 | ||
81819f0f CL |
3966 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
3967 | { | |
d9acf4b7 | 3968 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
3969 | } |
3970 | SLAB_ATTR_RO(partial); | |
3971 | ||
3972 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3973 | { | |
d9acf4b7 | 3974 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
3975 | } |
3976 | SLAB_ATTR_RO(cpu_slabs); | |
3977 | ||
3978 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3979 | { | |
205ab99d | 3980 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
3981 | } |
3982 | SLAB_ATTR_RO(objects); | |
3983 | ||
205ab99d CL |
3984 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
3985 | { | |
3986 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
3987 | } | |
3988 | SLAB_ATTR_RO(objects_partial); | |
3989 | ||
a5a84755 CL |
3990 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
3991 | { | |
3992 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3993 | } | |
3994 | ||
3995 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3996 | const char *buf, size_t length) | |
3997 | { | |
3998 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3999 | if (buf[0] == '1') | |
4000 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4001 | return length; | |
4002 | } | |
4003 | SLAB_ATTR(reclaim_account); | |
4004 | ||
4005 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4006 | { | |
4007 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4008 | } | |
4009 | SLAB_ATTR_RO(hwcache_align); | |
4010 | ||
4011 | #ifdef CONFIG_ZONE_DMA | |
4012 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4013 | { | |
4014 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4015 | } | |
4016 | SLAB_ATTR_RO(cache_dma); | |
4017 | #endif | |
4018 | ||
4019 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4020 | { | |
4021 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4022 | } | |
4023 | SLAB_ATTR_RO(destroy_by_rcu); | |
4024 | ||
ab4d5ed5 | 4025 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4026 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4027 | { | |
4028 | return show_slab_objects(s, buf, SO_ALL); | |
4029 | } | |
4030 | SLAB_ATTR_RO(slabs); | |
4031 | ||
205ab99d CL |
4032 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4033 | { | |
4034 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4035 | } | |
4036 | SLAB_ATTR_RO(total_objects); | |
4037 | ||
81819f0f CL |
4038 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4039 | { | |
4040 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4041 | } | |
4042 | ||
4043 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4044 | const char *buf, size_t length) | |
4045 | { | |
4046 | s->flags &= ~SLAB_DEBUG_FREE; | |
4047 | if (buf[0] == '1') | |
4048 | s->flags |= SLAB_DEBUG_FREE; | |
4049 | return length; | |
4050 | } | |
4051 | SLAB_ATTR(sanity_checks); | |
4052 | ||
4053 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4054 | { | |
4055 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4056 | } | |
4057 | ||
4058 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4059 | size_t length) | |
4060 | { | |
4061 | s->flags &= ~SLAB_TRACE; | |
4062 | if (buf[0] == '1') | |
4063 | s->flags |= SLAB_TRACE; | |
4064 | return length; | |
4065 | } | |
4066 | SLAB_ATTR(trace); | |
4067 | ||
81819f0f CL |
4068 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4069 | { | |
4070 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4071 | } | |
4072 | ||
4073 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4074 | const char *buf, size_t length) | |
4075 | { | |
4076 | if (any_slab_objects(s)) | |
4077 | return -EBUSY; | |
4078 | ||
4079 | s->flags &= ~SLAB_RED_ZONE; | |
4080 | if (buf[0] == '1') | |
4081 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4082 | calculate_sizes(s, -1); |
81819f0f CL |
4083 | return length; |
4084 | } | |
4085 | SLAB_ATTR(red_zone); | |
4086 | ||
4087 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4088 | { | |
4089 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4090 | } | |
4091 | ||
4092 | static ssize_t poison_store(struct kmem_cache *s, | |
4093 | const char *buf, size_t length) | |
4094 | { | |
4095 | if (any_slab_objects(s)) | |
4096 | return -EBUSY; | |
4097 | ||
4098 | s->flags &= ~SLAB_POISON; | |
4099 | if (buf[0] == '1') | |
4100 | s->flags |= SLAB_POISON; | |
06b285dc | 4101 | calculate_sizes(s, -1); |
81819f0f CL |
4102 | return length; |
4103 | } | |
4104 | SLAB_ATTR(poison); | |
4105 | ||
4106 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4107 | { | |
4108 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4109 | } | |
4110 | ||
4111 | static ssize_t store_user_store(struct kmem_cache *s, | |
4112 | const char *buf, size_t length) | |
4113 | { | |
4114 | if (any_slab_objects(s)) | |
4115 | return -EBUSY; | |
4116 | ||
4117 | s->flags &= ~SLAB_STORE_USER; | |
4118 | if (buf[0] == '1') | |
4119 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4120 | calculate_sizes(s, -1); |
81819f0f CL |
4121 | return length; |
4122 | } | |
4123 | SLAB_ATTR(store_user); | |
4124 | ||
53e15af0 CL |
4125 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4126 | { | |
4127 | return 0; | |
4128 | } | |
4129 | ||
4130 | static ssize_t validate_store(struct kmem_cache *s, | |
4131 | const char *buf, size_t length) | |
4132 | { | |
434e245d CL |
4133 | int ret = -EINVAL; |
4134 | ||
4135 | if (buf[0] == '1') { | |
4136 | ret = validate_slab_cache(s); | |
4137 | if (ret >= 0) | |
4138 | ret = length; | |
4139 | } | |
4140 | return ret; | |
53e15af0 CL |
4141 | } |
4142 | SLAB_ATTR(validate); | |
a5a84755 CL |
4143 | |
4144 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4145 | { | |
4146 | if (!(s->flags & SLAB_STORE_USER)) | |
4147 | return -ENOSYS; | |
4148 | return list_locations(s, buf, TRACK_ALLOC); | |
4149 | } | |
4150 | SLAB_ATTR_RO(alloc_calls); | |
4151 | ||
4152 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4153 | { | |
4154 | if (!(s->flags & SLAB_STORE_USER)) | |
4155 | return -ENOSYS; | |
4156 | return list_locations(s, buf, TRACK_FREE); | |
4157 | } | |
4158 | SLAB_ATTR_RO(free_calls); | |
4159 | #endif /* CONFIG_SLUB_DEBUG */ | |
4160 | ||
4161 | #ifdef CONFIG_FAILSLAB | |
4162 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4163 | { | |
4164 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4165 | } | |
4166 | ||
4167 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4168 | size_t length) | |
4169 | { | |
4170 | s->flags &= ~SLAB_FAILSLAB; | |
4171 | if (buf[0] == '1') | |
4172 | s->flags |= SLAB_FAILSLAB; | |
4173 | return length; | |
4174 | } | |
4175 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4176 | #endif |
53e15af0 | 4177 | |
2086d26a CL |
4178 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4179 | { | |
4180 | return 0; | |
4181 | } | |
4182 | ||
4183 | static ssize_t shrink_store(struct kmem_cache *s, | |
4184 | const char *buf, size_t length) | |
4185 | { | |
4186 | if (buf[0] == '1') { | |
4187 | int rc = kmem_cache_shrink(s); | |
4188 | ||
4189 | if (rc) | |
4190 | return rc; | |
4191 | } else | |
4192 | return -EINVAL; | |
4193 | return length; | |
4194 | } | |
4195 | SLAB_ATTR(shrink); | |
4196 | ||
81819f0f | 4197 | #ifdef CONFIG_NUMA |
9824601e | 4198 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4199 | { |
9824601e | 4200 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4201 | } |
4202 | ||
9824601e | 4203 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4204 | const char *buf, size_t length) |
4205 | { | |
0121c619 CL |
4206 | unsigned long ratio; |
4207 | int err; | |
4208 | ||
4209 | err = strict_strtoul(buf, 10, &ratio); | |
4210 | if (err) | |
4211 | return err; | |
4212 | ||
e2cb96b7 | 4213 | if (ratio <= 100) |
0121c619 | 4214 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4215 | |
81819f0f CL |
4216 | return length; |
4217 | } | |
9824601e | 4218 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4219 | #endif |
4220 | ||
8ff12cfc | 4221 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4222 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4223 | { | |
4224 | unsigned long sum = 0; | |
4225 | int cpu; | |
4226 | int len; | |
4227 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4228 | ||
4229 | if (!data) | |
4230 | return -ENOMEM; | |
4231 | ||
4232 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4233 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4234 | |
4235 | data[cpu] = x; | |
4236 | sum += x; | |
4237 | } | |
4238 | ||
4239 | len = sprintf(buf, "%lu", sum); | |
4240 | ||
50ef37b9 | 4241 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4242 | for_each_online_cpu(cpu) { |
4243 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4244 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4245 | } |
50ef37b9 | 4246 | #endif |
8ff12cfc CL |
4247 | kfree(data); |
4248 | return len + sprintf(buf + len, "\n"); | |
4249 | } | |
4250 | ||
78eb00cc DR |
4251 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4252 | { | |
4253 | int cpu; | |
4254 | ||
4255 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4256 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4257 | } |
4258 | ||
8ff12cfc CL |
4259 | #define STAT_ATTR(si, text) \ |
4260 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4261 | { \ | |
4262 | return show_stat(s, buf, si); \ | |
4263 | } \ | |
78eb00cc DR |
4264 | static ssize_t text##_store(struct kmem_cache *s, \ |
4265 | const char *buf, size_t length) \ | |
4266 | { \ | |
4267 | if (buf[0] != '0') \ | |
4268 | return -EINVAL; \ | |
4269 | clear_stat(s, si); \ | |
4270 | return length; \ | |
4271 | } \ | |
4272 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4273 | |
4274 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4275 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4276 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4277 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4278 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4279 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4280 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4281 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4282 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4283 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4284 | STAT_ATTR(FREE_SLAB, free_slab); | |
4285 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4286 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4287 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4288 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4289 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4290 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4291 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4292 | #endif |
4293 | ||
06428780 | 4294 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4295 | &slab_size_attr.attr, |
4296 | &object_size_attr.attr, | |
4297 | &objs_per_slab_attr.attr, | |
4298 | &order_attr.attr, | |
73d342b1 | 4299 | &min_partial_attr.attr, |
81819f0f | 4300 | &objects_attr.attr, |
205ab99d | 4301 | &objects_partial_attr.attr, |
81819f0f CL |
4302 | &partial_attr.attr, |
4303 | &cpu_slabs_attr.attr, | |
4304 | &ctor_attr.attr, | |
81819f0f CL |
4305 | &aliases_attr.attr, |
4306 | &align_attr.attr, | |
81819f0f CL |
4307 | &hwcache_align_attr.attr, |
4308 | &reclaim_account_attr.attr, | |
4309 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4310 | &shrink_attr.attr, |
ab4d5ed5 | 4311 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4312 | &total_objects_attr.attr, |
4313 | &slabs_attr.attr, | |
4314 | &sanity_checks_attr.attr, | |
4315 | &trace_attr.attr, | |
81819f0f CL |
4316 | &red_zone_attr.attr, |
4317 | &poison_attr.attr, | |
4318 | &store_user_attr.attr, | |
53e15af0 | 4319 | &validate_attr.attr, |
88a420e4 CL |
4320 | &alloc_calls_attr.attr, |
4321 | &free_calls_attr.attr, | |
ab4d5ed5 | 4322 | #endif |
81819f0f CL |
4323 | #ifdef CONFIG_ZONE_DMA |
4324 | &cache_dma_attr.attr, | |
4325 | #endif | |
4326 | #ifdef CONFIG_NUMA | |
9824601e | 4327 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4328 | #endif |
4329 | #ifdef CONFIG_SLUB_STATS | |
4330 | &alloc_fastpath_attr.attr, | |
4331 | &alloc_slowpath_attr.attr, | |
4332 | &free_fastpath_attr.attr, | |
4333 | &free_slowpath_attr.attr, | |
4334 | &free_frozen_attr.attr, | |
4335 | &free_add_partial_attr.attr, | |
4336 | &free_remove_partial_attr.attr, | |
4337 | &alloc_from_partial_attr.attr, | |
4338 | &alloc_slab_attr.attr, | |
4339 | &alloc_refill_attr.attr, | |
4340 | &free_slab_attr.attr, | |
4341 | &cpuslab_flush_attr.attr, | |
4342 | &deactivate_full_attr.attr, | |
4343 | &deactivate_empty_attr.attr, | |
4344 | &deactivate_to_head_attr.attr, | |
4345 | &deactivate_to_tail_attr.attr, | |
4346 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4347 | &order_fallback_attr.attr, |
81819f0f | 4348 | #endif |
4c13dd3b DM |
4349 | #ifdef CONFIG_FAILSLAB |
4350 | &failslab_attr.attr, | |
4351 | #endif | |
4352 | ||
81819f0f CL |
4353 | NULL |
4354 | }; | |
4355 | ||
4356 | static struct attribute_group slab_attr_group = { | |
4357 | .attrs = slab_attrs, | |
4358 | }; | |
4359 | ||
4360 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4361 | struct attribute *attr, | |
4362 | char *buf) | |
4363 | { | |
4364 | struct slab_attribute *attribute; | |
4365 | struct kmem_cache *s; | |
4366 | int err; | |
4367 | ||
4368 | attribute = to_slab_attr(attr); | |
4369 | s = to_slab(kobj); | |
4370 | ||
4371 | if (!attribute->show) | |
4372 | return -EIO; | |
4373 | ||
4374 | err = attribute->show(s, buf); | |
4375 | ||
4376 | return err; | |
4377 | } | |
4378 | ||
4379 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4380 | struct attribute *attr, | |
4381 | const char *buf, size_t len) | |
4382 | { | |
4383 | struct slab_attribute *attribute; | |
4384 | struct kmem_cache *s; | |
4385 | int err; | |
4386 | ||
4387 | attribute = to_slab_attr(attr); | |
4388 | s = to_slab(kobj); | |
4389 | ||
4390 | if (!attribute->store) | |
4391 | return -EIO; | |
4392 | ||
4393 | err = attribute->store(s, buf, len); | |
4394 | ||
4395 | return err; | |
4396 | } | |
4397 | ||
151c602f CL |
4398 | static void kmem_cache_release(struct kobject *kobj) |
4399 | { | |
4400 | struct kmem_cache *s = to_slab(kobj); | |
4401 | ||
84c1cf62 | 4402 | kfree(s->name); |
151c602f CL |
4403 | kfree(s); |
4404 | } | |
4405 | ||
52cf25d0 | 4406 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
4407 | .show = slab_attr_show, |
4408 | .store = slab_attr_store, | |
4409 | }; | |
4410 | ||
4411 | static struct kobj_type slab_ktype = { | |
4412 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4413 | .release = kmem_cache_release |
81819f0f CL |
4414 | }; |
4415 | ||
4416 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4417 | { | |
4418 | struct kobj_type *ktype = get_ktype(kobj); | |
4419 | ||
4420 | if (ktype == &slab_ktype) | |
4421 | return 1; | |
4422 | return 0; | |
4423 | } | |
4424 | ||
9cd43611 | 4425 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
4426 | .filter = uevent_filter, |
4427 | }; | |
4428 | ||
27c3a314 | 4429 | static struct kset *slab_kset; |
81819f0f CL |
4430 | |
4431 | #define ID_STR_LENGTH 64 | |
4432 | ||
4433 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4434 | * |
4435 | * Format :[flags-]size | |
81819f0f CL |
4436 | */ |
4437 | static char *create_unique_id(struct kmem_cache *s) | |
4438 | { | |
4439 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4440 | char *p = name; | |
4441 | ||
4442 | BUG_ON(!name); | |
4443 | ||
4444 | *p++ = ':'; | |
4445 | /* | |
4446 | * First flags affecting slabcache operations. We will only | |
4447 | * get here for aliasable slabs so we do not need to support | |
4448 | * too many flags. The flags here must cover all flags that | |
4449 | * are matched during merging to guarantee that the id is | |
4450 | * unique. | |
4451 | */ | |
4452 | if (s->flags & SLAB_CACHE_DMA) | |
4453 | *p++ = 'd'; | |
4454 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4455 | *p++ = 'a'; | |
4456 | if (s->flags & SLAB_DEBUG_FREE) | |
4457 | *p++ = 'F'; | |
5a896d9e VN |
4458 | if (!(s->flags & SLAB_NOTRACK)) |
4459 | *p++ = 't'; | |
81819f0f CL |
4460 | if (p != name + 1) |
4461 | *p++ = '-'; | |
4462 | p += sprintf(p, "%07d", s->size); | |
4463 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4464 | return name; | |
4465 | } | |
4466 | ||
4467 | static int sysfs_slab_add(struct kmem_cache *s) | |
4468 | { | |
4469 | int err; | |
4470 | const char *name; | |
4471 | int unmergeable; | |
4472 | ||
4473 | if (slab_state < SYSFS) | |
4474 | /* Defer until later */ | |
4475 | return 0; | |
4476 | ||
4477 | unmergeable = slab_unmergeable(s); | |
4478 | if (unmergeable) { | |
4479 | /* | |
4480 | * Slabcache can never be merged so we can use the name proper. | |
4481 | * This is typically the case for debug situations. In that | |
4482 | * case we can catch duplicate names easily. | |
4483 | */ | |
27c3a314 | 4484 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4485 | name = s->name; |
4486 | } else { | |
4487 | /* | |
4488 | * Create a unique name for the slab as a target | |
4489 | * for the symlinks. | |
4490 | */ | |
4491 | name = create_unique_id(s); | |
4492 | } | |
4493 | ||
27c3a314 | 4494 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4495 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4496 | if (err) { | |
4497 | kobject_put(&s->kobj); | |
81819f0f | 4498 | return err; |
1eada11c | 4499 | } |
81819f0f CL |
4500 | |
4501 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
4502 | if (err) { |
4503 | kobject_del(&s->kobj); | |
4504 | kobject_put(&s->kobj); | |
81819f0f | 4505 | return err; |
5788d8ad | 4506 | } |
81819f0f CL |
4507 | kobject_uevent(&s->kobj, KOBJ_ADD); |
4508 | if (!unmergeable) { | |
4509 | /* Setup first alias */ | |
4510 | sysfs_slab_alias(s, s->name); | |
4511 | kfree(name); | |
4512 | } | |
4513 | return 0; | |
4514 | } | |
4515 | ||
4516 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4517 | { | |
2bce6485 CL |
4518 | if (slab_state < SYSFS) |
4519 | /* | |
4520 | * Sysfs has not been setup yet so no need to remove the | |
4521 | * cache from sysfs. | |
4522 | */ | |
4523 | return; | |
4524 | ||
81819f0f CL |
4525 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
4526 | kobject_del(&s->kobj); | |
151c602f | 4527 | kobject_put(&s->kobj); |
81819f0f CL |
4528 | } |
4529 | ||
4530 | /* | |
4531 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4532 | * available lest we lose that information. |
81819f0f CL |
4533 | */ |
4534 | struct saved_alias { | |
4535 | struct kmem_cache *s; | |
4536 | const char *name; | |
4537 | struct saved_alias *next; | |
4538 | }; | |
4539 | ||
5af328a5 | 4540 | static struct saved_alias *alias_list; |
81819f0f CL |
4541 | |
4542 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4543 | { | |
4544 | struct saved_alias *al; | |
4545 | ||
4546 | if (slab_state == SYSFS) { | |
4547 | /* | |
4548 | * If we have a leftover link then remove it. | |
4549 | */ | |
27c3a314 GKH |
4550 | sysfs_remove_link(&slab_kset->kobj, name); |
4551 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4552 | } |
4553 | ||
4554 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4555 | if (!al) | |
4556 | return -ENOMEM; | |
4557 | ||
4558 | al->s = s; | |
4559 | al->name = name; | |
4560 | al->next = alias_list; | |
4561 | alias_list = al; | |
4562 | return 0; | |
4563 | } | |
4564 | ||
4565 | static int __init slab_sysfs_init(void) | |
4566 | { | |
5b95a4ac | 4567 | struct kmem_cache *s; |
81819f0f CL |
4568 | int err; |
4569 | ||
2bce6485 CL |
4570 | down_write(&slub_lock); |
4571 | ||
0ff21e46 | 4572 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4573 | if (!slab_kset) { |
2bce6485 | 4574 | up_write(&slub_lock); |
81819f0f CL |
4575 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4576 | return -ENOSYS; | |
4577 | } | |
4578 | ||
26a7bd03 CL |
4579 | slab_state = SYSFS; |
4580 | ||
5b95a4ac | 4581 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4582 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4583 | if (err) |
4584 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4585 | " to sysfs\n", s->name); | |
26a7bd03 | 4586 | } |
81819f0f CL |
4587 | |
4588 | while (alias_list) { | |
4589 | struct saved_alias *al = alias_list; | |
4590 | ||
4591 | alias_list = alias_list->next; | |
4592 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4593 | if (err) |
4594 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4595 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4596 | kfree(al); |
4597 | } | |
4598 | ||
2bce6485 | 4599 | up_write(&slub_lock); |
81819f0f CL |
4600 | resiliency_test(); |
4601 | return 0; | |
4602 | } | |
4603 | ||
4604 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 4605 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
4606 | |
4607 | /* | |
4608 | * The /proc/slabinfo ABI | |
4609 | */ | |
158a9624 | 4610 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4611 | static void print_slabinfo_header(struct seq_file *m) |
4612 | { | |
4613 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4614 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4615 | "<objperslab> <pagesperslab>"); | |
4616 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4617 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4618 | seq_putc(m, '\n'); | |
4619 | } | |
4620 | ||
4621 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4622 | { | |
4623 | loff_t n = *pos; | |
4624 | ||
4625 | down_read(&slub_lock); | |
4626 | if (!n) | |
4627 | print_slabinfo_header(m); | |
4628 | ||
4629 | return seq_list_start(&slab_caches, *pos); | |
4630 | } | |
4631 | ||
4632 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4633 | { | |
4634 | return seq_list_next(p, &slab_caches, pos); | |
4635 | } | |
4636 | ||
4637 | static void s_stop(struct seq_file *m, void *p) | |
4638 | { | |
4639 | up_read(&slub_lock); | |
4640 | } | |
4641 | ||
4642 | static int s_show(struct seq_file *m, void *p) | |
4643 | { | |
4644 | unsigned long nr_partials = 0; | |
4645 | unsigned long nr_slabs = 0; | |
4646 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4647 | unsigned long nr_objs = 0; |
4648 | unsigned long nr_free = 0; | |
57ed3eda PE |
4649 | struct kmem_cache *s; |
4650 | int node; | |
4651 | ||
4652 | s = list_entry(p, struct kmem_cache, list); | |
4653 | ||
4654 | for_each_online_node(node) { | |
4655 | struct kmem_cache_node *n = get_node(s, node); | |
4656 | ||
4657 | if (!n) | |
4658 | continue; | |
4659 | ||
4660 | nr_partials += n->nr_partial; | |
4661 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4662 | nr_objs += atomic_long_read(&n->total_objects); |
4663 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4664 | } |
4665 | ||
205ab99d | 4666 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4667 | |
4668 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4669 | nr_objs, s->size, oo_objects(s->oo), |
4670 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4671 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4672 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4673 | 0UL); | |
4674 | seq_putc(m, '\n'); | |
4675 | return 0; | |
4676 | } | |
4677 | ||
7b3c3a50 | 4678 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4679 | .start = s_start, |
4680 | .next = s_next, | |
4681 | .stop = s_stop, | |
4682 | .show = s_show, | |
4683 | }; | |
4684 | ||
7b3c3a50 AD |
4685 | static int slabinfo_open(struct inode *inode, struct file *file) |
4686 | { | |
4687 | return seq_open(file, &slabinfo_op); | |
4688 | } | |
4689 | ||
4690 | static const struct file_operations proc_slabinfo_operations = { | |
4691 | .open = slabinfo_open, | |
4692 | .read = seq_read, | |
4693 | .llseek = seq_lseek, | |
4694 | .release = seq_release, | |
4695 | }; | |
4696 | ||
4697 | static int __init slab_proc_init(void) | |
4698 | { | |
cf5d1131 | 4699 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
4700 | return 0; |
4701 | } | |
4702 | module_init(slab_proc_init); | |
158a9624 | 4703 | #endif /* CONFIG_SLABINFO */ |