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 | ||
ab9a0f19 LJ |
284 | static inline int order_objects(int order, unsigned long size, int reserved) |
285 | { | |
286 | return ((PAGE_SIZE << order) - reserved) / size; | |
287 | } | |
288 | ||
834f3d11 | 289 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 290 | unsigned long size, int reserved) |
834f3d11 CL |
291 | { |
292 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 293 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
294 | }; |
295 | ||
296 | return x; | |
297 | } | |
298 | ||
299 | static inline int oo_order(struct kmem_cache_order_objects x) | |
300 | { | |
210b5c06 | 301 | return x.x >> OO_SHIFT; |
834f3d11 CL |
302 | } |
303 | ||
304 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
305 | { | |
210b5c06 | 306 | return x.x & OO_MASK; |
834f3d11 CL |
307 | } |
308 | ||
41ecc55b CL |
309 | #ifdef CONFIG_SLUB_DEBUG |
310 | /* | |
311 | * Debug settings: | |
312 | */ | |
f0630fff CL |
313 | #ifdef CONFIG_SLUB_DEBUG_ON |
314 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
315 | #else | |
41ecc55b | 316 | static int slub_debug; |
f0630fff | 317 | #endif |
41ecc55b CL |
318 | |
319 | static char *slub_debug_slabs; | |
fa5ec8a1 | 320 | static int disable_higher_order_debug; |
41ecc55b | 321 | |
81819f0f CL |
322 | /* |
323 | * Object debugging | |
324 | */ | |
325 | static void print_section(char *text, u8 *addr, unsigned int length) | |
326 | { | |
327 | int i, offset; | |
328 | int newline = 1; | |
329 | char ascii[17]; | |
330 | ||
331 | ascii[16] = 0; | |
332 | ||
333 | for (i = 0; i < length; i++) { | |
334 | if (newline) { | |
24922684 | 335 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
336 | newline = 0; |
337 | } | |
06428780 | 338 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
339 | offset = i % 16; |
340 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
341 | if (offset == 15) { | |
06428780 | 342 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
343 | newline = 1; |
344 | } | |
345 | } | |
346 | if (!newline) { | |
347 | i %= 16; | |
348 | while (i < 16) { | |
06428780 | 349 | printk(KERN_CONT " "); |
81819f0f CL |
350 | ascii[i] = ' '; |
351 | i++; | |
352 | } | |
06428780 | 353 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
354 | } |
355 | } | |
356 | ||
81819f0f CL |
357 | static struct track *get_track(struct kmem_cache *s, void *object, |
358 | enum track_item alloc) | |
359 | { | |
360 | struct track *p; | |
361 | ||
362 | if (s->offset) | |
363 | p = object + s->offset + sizeof(void *); | |
364 | else | |
365 | p = object + s->inuse; | |
366 | ||
367 | return p + alloc; | |
368 | } | |
369 | ||
370 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 371 | enum track_item alloc, unsigned long addr) |
81819f0f | 372 | { |
1a00df4a | 373 | struct track *p = get_track(s, object, alloc); |
81819f0f | 374 | |
81819f0f CL |
375 | if (addr) { |
376 | p->addr = addr; | |
377 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 378 | p->pid = current->pid; |
81819f0f CL |
379 | p->when = jiffies; |
380 | } else | |
381 | memset(p, 0, sizeof(struct track)); | |
382 | } | |
383 | ||
81819f0f CL |
384 | static void init_tracking(struct kmem_cache *s, void *object) |
385 | { | |
24922684 CL |
386 | if (!(s->flags & SLAB_STORE_USER)) |
387 | return; | |
388 | ||
ce71e27c EGM |
389 | set_track(s, object, TRACK_FREE, 0UL); |
390 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
391 | } |
392 | ||
393 | static void print_track(const char *s, struct track *t) | |
394 | { | |
395 | if (!t->addr) | |
396 | return; | |
397 | ||
7daf705f | 398 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 399 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
400 | } |
401 | ||
402 | static void print_tracking(struct kmem_cache *s, void *object) | |
403 | { | |
404 | if (!(s->flags & SLAB_STORE_USER)) | |
405 | return; | |
406 | ||
407 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
408 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
409 | } | |
410 | ||
411 | static void print_page_info(struct page *page) | |
412 | { | |
39b26464 CL |
413 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
414 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
415 | |
416 | } | |
417 | ||
418 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
419 | { | |
420 | va_list args; | |
421 | char buf[100]; | |
422 | ||
423 | va_start(args, fmt); | |
424 | vsnprintf(buf, sizeof(buf), fmt, args); | |
425 | va_end(args); | |
426 | printk(KERN_ERR "========================================" | |
427 | "=====================================\n"); | |
428 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
429 | printk(KERN_ERR "----------------------------------------" | |
430 | "-------------------------------------\n\n"); | |
81819f0f CL |
431 | } |
432 | ||
24922684 CL |
433 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
434 | { | |
435 | va_list args; | |
436 | char buf[100]; | |
437 | ||
438 | va_start(args, fmt); | |
439 | vsnprintf(buf, sizeof(buf), fmt, args); | |
440 | va_end(args); | |
441 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
442 | } | |
443 | ||
444 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
445 | { |
446 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 447 | u8 *addr = page_address(page); |
24922684 CL |
448 | |
449 | print_tracking(s, p); | |
450 | ||
451 | print_page_info(page); | |
452 | ||
453 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
454 | p, p - addr, get_freepointer(s, p)); | |
455 | ||
456 | if (p > addr + 16) | |
457 | print_section("Bytes b4", p - 16, 16); | |
458 | ||
0ebd652b | 459 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
460 | |
461 | if (s->flags & SLAB_RED_ZONE) | |
462 | print_section("Redzone", p + s->objsize, | |
463 | s->inuse - s->objsize); | |
464 | ||
81819f0f CL |
465 | if (s->offset) |
466 | off = s->offset + sizeof(void *); | |
467 | else | |
468 | off = s->inuse; | |
469 | ||
24922684 | 470 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 471 | off += 2 * sizeof(struct track); |
81819f0f CL |
472 | |
473 | if (off != s->size) | |
474 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
475 | print_section("Padding", p + off, s->size - off); |
476 | ||
477 | dump_stack(); | |
81819f0f CL |
478 | } |
479 | ||
480 | static void object_err(struct kmem_cache *s, struct page *page, | |
481 | u8 *object, char *reason) | |
482 | { | |
3dc50637 | 483 | slab_bug(s, "%s", reason); |
24922684 | 484 | print_trailer(s, page, object); |
81819f0f CL |
485 | } |
486 | ||
24922684 | 487 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
488 | { |
489 | va_list args; | |
490 | char buf[100]; | |
491 | ||
24922684 CL |
492 | va_start(args, fmt); |
493 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 494 | va_end(args); |
3dc50637 | 495 | slab_bug(s, "%s", buf); |
24922684 | 496 | print_page_info(page); |
81819f0f CL |
497 | dump_stack(); |
498 | } | |
499 | ||
f7cb1933 | 500 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
501 | { |
502 | u8 *p = object; | |
503 | ||
504 | if (s->flags & __OBJECT_POISON) { | |
505 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 506 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
507 | } |
508 | ||
509 | if (s->flags & SLAB_RED_ZONE) | |
f7cb1933 | 510 | memset(p + s->objsize, val, s->inuse - s->objsize); |
81819f0f CL |
511 | } |
512 | ||
24922684 | 513 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
514 | { |
515 | while (bytes) { | |
516 | if (*start != (u8)value) | |
24922684 | 517 | return start; |
81819f0f CL |
518 | start++; |
519 | bytes--; | |
520 | } | |
24922684 CL |
521 | return NULL; |
522 | } | |
523 | ||
524 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
525 | void *from, void *to) | |
526 | { | |
527 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
528 | memset(from, data, to - from); | |
529 | } | |
530 | ||
531 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
532 | u8 *object, char *what, | |
06428780 | 533 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
534 | { |
535 | u8 *fault; | |
536 | u8 *end; | |
537 | ||
538 | fault = check_bytes(start, value, bytes); | |
539 | if (!fault) | |
540 | return 1; | |
541 | ||
542 | end = start + bytes; | |
543 | while (end > fault && end[-1] == value) | |
544 | end--; | |
545 | ||
546 | slab_bug(s, "%s overwritten", what); | |
547 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
548 | fault, end - 1, fault[0], value); | |
549 | print_trailer(s, page, object); | |
550 | ||
551 | restore_bytes(s, what, value, fault, end); | |
552 | return 0; | |
81819f0f CL |
553 | } |
554 | ||
81819f0f CL |
555 | /* |
556 | * Object layout: | |
557 | * | |
558 | * object address | |
559 | * Bytes of the object to be managed. | |
560 | * If the freepointer may overlay the object then the free | |
561 | * pointer is the first word of the object. | |
672bba3a | 562 | * |
81819f0f CL |
563 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
564 | * 0xa5 (POISON_END) | |
565 | * | |
566 | * object + s->objsize | |
567 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
568 | * Padding is extended by another word if Redzoning is enabled and |
569 | * objsize == inuse. | |
570 | * | |
81819f0f CL |
571 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
572 | * 0xcc (RED_ACTIVE) for objects in use. | |
573 | * | |
574 | * object + s->inuse | |
672bba3a CL |
575 | * Meta data starts here. |
576 | * | |
81819f0f CL |
577 | * A. Free pointer (if we cannot overwrite object on free) |
578 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 579 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 580 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
581 | * before the word boundary. |
582 | * | |
583 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
584 | * |
585 | * object + s->size | |
672bba3a | 586 | * Nothing is used beyond s->size. |
81819f0f | 587 | * |
672bba3a CL |
588 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
589 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
590 | * may be used with merged slabcaches. |
591 | */ | |
592 | ||
81819f0f CL |
593 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
594 | { | |
595 | unsigned long off = s->inuse; /* The end of info */ | |
596 | ||
597 | if (s->offset) | |
598 | /* Freepointer is placed after the object. */ | |
599 | off += sizeof(void *); | |
600 | ||
601 | if (s->flags & SLAB_STORE_USER) | |
602 | /* We also have user information there */ | |
603 | off += 2 * sizeof(struct track); | |
604 | ||
605 | if (s->size == off) | |
606 | return 1; | |
607 | ||
24922684 CL |
608 | return check_bytes_and_report(s, page, p, "Object padding", |
609 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
610 | } |
611 | ||
39b26464 | 612 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
613 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
614 | { | |
24922684 CL |
615 | u8 *start; |
616 | u8 *fault; | |
617 | u8 *end; | |
618 | int length; | |
619 | int remainder; | |
81819f0f CL |
620 | |
621 | if (!(s->flags & SLAB_POISON)) | |
622 | return 1; | |
623 | ||
a973e9dd | 624 | start = page_address(page); |
ab9a0f19 | 625 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
626 | end = start + length; |
627 | remainder = length % s->size; | |
81819f0f CL |
628 | if (!remainder) |
629 | return 1; | |
630 | ||
39b26464 | 631 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
632 | if (!fault) |
633 | return 1; | |
634 | while (end > fault && end[-1] == POISON_INUSE) | |
635 | end--; | |
636 | ||
637 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 638 | print_section("Padding", end - remainder, remainder); |
24922684 | 639 | |
8a3d271d | 640 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 641 | return 0; |
81819f0f CL |
642 | } |
643 | ||
644 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 645 | void *object, u8 val) |
81819f0f CL |
646 | { |
647 | u8 *p = object; | |
648 | u8 *endobject = object + s->objsize; | |
649 | ||
650 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 651 | if (!check_bytes_and_report(s, page, object, "Redzone", |
f7cb1933 | 652 | endobject, val, s->inuse - s->objsize)) |
81819f0f | 653 | return 0; |
81819f0f | 654 | } else { |
3adbefee IM |
655 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
656 | check_bytes_and_report(s, page, p, "Alignment padding", | |
657 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
658 | } | |
81819f0f CL |
659 | } |
660 | ||
661 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 662 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 CL |
663 | (!check_bytes_and_report(s, page, p, "Poison", p, |
664 | POISON_FREE, s->objsize - 1) || | |
665 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 666 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 667 | return 0; |
81819f0f CL |
668 | /* |
669 | * check_pad_bytes cleans up on its own. | |
670 | */ | |
671 | check_pad_bytes(s, page, p); | |
672 | } | |
673 | ||
f7cb1933 | 674 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
675 | /* |
676 | * Object and freepointer overlap. Cannot check | |
677 | * freepointer while object is allocated. | |
678 | */ | |
679 | return 1; | |
680 | ||
681 | /* Check free pointer validity */ | |
682 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
683 | object_err(s, page, p, "Freepointer corrupt"); | |
684 | /* | |
9f6c708e | 685 | * No choice but to zap it and thus lose the remainder |
81819f0f | 686 | * of the free objects in this slab. May cause |
672bba3a | 687 | * another error because the object count is now wrong. |
81819f0f | 688 | */ |
a973e9dd | 689 | set_freepointer(s, p, NULL); |
81819f0f CL |
690 | return 0; |
691 | } | |
692 | return 1; | |
693 | } | |
694 | ||
695 | static int check_slab(struct kmem_cache *s, struct page *page) | |
696 | { | |
39b26464 CL |
697 | int maxobj; |
698 | ||
81819f0f CL |
699 | VM_BUG_ON(!irqs_disabled()); |
700 | ||
701 | if (!PageSlab(page)) { | |
24922684 | 702 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
703 | return 0; |
704 | } | |
39b26464 | 705 | |
ab9a0f19 | 706 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
707 | if (page->objects > maxobj) { |
708 | slab_err(s, page, "objects %u > max %u", | |
709 | s->name, page->objects, maxobj); | |
710 | return 0; | |
711 | } | |
712 | if (page->inuse > page->objects) { | |
24922684 | 713 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 714 | s->name, page->inuse, page->objects); |
81819f0f CL |
715 | return 0; |
716 | } | |
717 | /* Slab_pad_check fixes things up after itself */ | |
718 | slab_pad_check(s, page); | |
719 | return 1; | |
720 | } | |
721 | ||
722 | /* | |
672bba3a CL |
723 | * Determine if a certain object on a page is on the freelist. Must hold the |
724 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
725 | */ |
726 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
727 | { | |
728 | int nr = 0; | |
729 | void *fp = page->freelist; | |
730 | void *object = NULL; | |
224a88be | 731 | unsigned long max_objects; |
81819f0f | 732 | |
39b26464 | 733 | while (fp && nr <= page->objects) { |
81819f0f CL |
734 | if (fp == search) |
735 | return 1; | |
736 | if (!check_valid_pointer(s, page, fp)) { | |
737 | if (object) { | |
738 | object_err(s, page, object, | |
739 | "Freechain corrupt"); | |
a973e9dd | 740 | set_freepointer(s, object, NULL); |
81819f0f CL |
741 | break; |
742 | } else { | |
24922684 | 743 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 744 | page->freelist = NULL; |
39b26464 | 745 | page->inuse = page->objects; |
24922684 | 746 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
747 | return 0; |
748 | } | |
749 | break; | |
750 | } | |
751 | object = fp; | |
752 | fp = get_freepointer(s, object); | |
753 | nr++; | |
754 | } | |
755 | ||
ab9a0f19 | 756 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
757 | if (max_objects > MAX_OBJS_PER_PAGE) |
758 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
759 | |
760 | if (page->objects != max_objects) { | |
761 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
762 | "should be %d", page->objects, max_objects); | |
763 | page->objects = max_objects; | |
764 | slab_fix(s, "Number of objects adjusted."); | |
765 | } | |
39b26464 | 766 | if (page->inuse != page->objects - nr) { |
70d71228 | 767 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
768 | "counted were %d", page->inuse, page->objects - nr); |
769 | page->inuse = page->objects - nr; | |
24922684 | 770 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
771 | } |
772 | return search == NULL; | |
773 | } | |
774 | ||
0121c619 CL |
775 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
776 | int alloc) | |
3ec09742 CL |
777 | { |
778 | if (s->flags & SLAB_TRACE) { | |
779 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
780 | s->name, | |
781 | alloc ? "alloc" : "free", | |
782 | object, page->inuse, | |
783 | page->freelist); | |
784 | ||
785 | if (!alloc) | |
786 | print_section("Object", (void *)object, s->objsize); | |
787 | ||
788 | dump_stack(); | |
789 | } | |
790 | } | |
791 | ||
c016b0bd CL |
792 | /* |
793 | * Hooks for other subsystems that check memory allocations. In a typical | |
794 | * production configuration these hooks all should produce no code at all. | |
795 | */ | |
796 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
797 | { | |
c1d50836 | 798 | flags &= gfp_allowed_mask; |
c016b0bd CL |
799 | lockdep_trace_alloc(flags); |
800 | might_sleep_if(flags & __GFP_WAIT); | |
801 | ||
802 | return should_failslab(s->objsize, flags, s->flags); | |
803 | } | |
804 | ||
805 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
806 | { | |
c1d50836 | 807 | flags &= gfp_allowed_mask; |
c016b0bd CL |
808 | kmemcheck_slab_alloc(s, flags, object, s->objsize); |
809 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); | |
810 | } | |
811 | ||
812 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
813 | { | |
814 | kmemleak_free_recursive(x, s->flags); | |
815 | } | |
816 | ||
817 | static inline void slab_free_hook_irq(struct kmem_cache *s, void *object) | |
818 | { | |
819 | kmemcheck_slab_free(s, object, s->objsize); | |
820 | debug_check_no_locks_freed(object, s->objsize); | |
821 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
822 | debug_check_no_obj_freed(object, s->objsize); | |
823 | } | |
824 | ||
643b1138 | 825 | /* |
672bba3a | 826 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 827 | */ |
e95eed57 | 828 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 829 | { |
643b1138 CL |
830 | spin_lock(&n->list_lock); |
831 | list_add(&page->lru, &n->full); | |
832 | spin_unlock(&n->list_lock); | |
833 | } | |
834 | ||
835 | static void remove_full(struct kmem_cache *s, struct page *page) | |
836 | { | |
837 | struct kmem_cache_node *n; | |
838 | ||
839 | if (!(s->flags & SLAB_STORE_USER)) | |
840 | return; | |
841 | ||
842 | n = get_node(s, page_to_nid(page)); | |
843 | ||
844 | spin_lock(&n->list_lock); | |
845 | list_del(&page->lru); | |
846 | spin_unlock(&n->list_lock); | |
847 | } | |
848 | ||
0f389ec6 CL |
849 | /* Tracking of the number of slabs for debugging purposes */ |
850 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
851 | { | |
852 | struct kmem_cache_node *n = get_node(s, node); | |
853 | ||
854 | return atomic_long_read(&n->nr_slabs); | |
855 | } | |
856 | ||
26c02cf0 AB |
857 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
858 | { | |
859 | return atomic_long_read(&n->nr_slabs); | |
860 | } | |
861 | ||
205ab99d | 862 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
863 | { |
864 | struct kmem_cache_node *n = get_node(s, node); | |
865 | ||
866 | /* | |
867 | * May be called early in order to allocate a slab for the | |
868 | * kmem_cache_node structure. Solve the chicken-egg | |
869 | * dilemma by deferring the increment of the count during | |
870 | * bootstrap (see early_kmem_cache_node_alloc). | |
871 | */ | |
7340cc84 | 872 | if (n) { |
0f389ec6 | 873 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
874 | atomic_long_add(objects, &n->total_objects); |
875 | } | |
0f389ec6 | 876 | } |
205ab99d | 877 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
878 | { |
879 | struct kmem_cache_node *n = get_node(s, node); | |
880 | ||
881 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 882 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
883 | } |
884 | ||
885 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
886 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
887 | void *object) | |
888 | { | |
889 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
890 | return; | |
891 | ||
f7cb1933 | 892 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
893 | init_tracking(s, object); |
894 | } | |
895 | ||
1537066c | 896 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 897 | void *object, unsigned long addr) |
81819f0f CL |
898 | { |
899 | if (!check_slab(s, page)) | |
900 | goto bad; | |
901 | ||
d692ef6d | 902 | if (!on_freelist(s, page, object)) { |
24922684 | 903 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 904 | goto bad; |
81819f0f CL |
905 | } |
906 | ||
907 | if (!check_valid_pointer(s, page, object)) { | |
908 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 909 | goto bad; |
81819f0f CL |
910 | } |
911 | ||
f7cb1933 | 912 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 913 | goto bad; |
81819f0f | 914 | |
3ec09742 CL |
915 | /* Success perform special debug activities for allocs */ |
916 | if (s->flags & SLAB_STORE_USER) | |
917 | set_track(s, object, TRACK_ALLOC, addr); | |
918 | trace(s, page, object, 1); | |
f7cb1933 | 919 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 920 | return 1; |
3ec09742 | 921 | |
81819f0f CL |
922 | bad: |
923 | if (PageSlab(page)) { | |
924 | /* | |
925 | * If this is a slab page then lets do the best we can | |
926 | * to avoid issues in the future. Marking all objects | |
672bba3a | 927 | * as used avoids touching the remaining objects. |
81819f0f | 928 | */ |
24922684 | 929 | slab_fix(s, "Marking all objects used"); |
39b26464 | 930 | page->inuse = page->objects; |
a973e9dd | 931 | page->freelist = NULL; |
81819f0f CL |
932 | } |
933 | return 0; | |
934 | } | |
935 | ||
1537066c CL |
936 | static noinline int free_debug_processing(struct kmem_cache *s, |
937 | struct page *page, void *object, unsigned long addr) | |
81819f0f CL |
938 | { |
939 | if (!check_slab(s, page)) | |
940 | goto fail; | |
941 | ||
942 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 943 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
944 | goto fail; |
945 | } | |
946 | ||
947 | if (on_freelist(s, page, object)) { | |
24922684 | 948 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
949 | goto fail; |
950 | } | |
951 | ||
f7cb1933 | 952 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
81819f0f CL |
953 | return 0; |
954 | ||
955 | if (unlikely(s != page->slab)) { | |
3adbefee | 956 | if (!PageSlab(page)) { |
70d71228 CL |
957 | slab_err(s, page, "Attempt to free object(0x%p) " |
958 | "outside of slab", object); | |
3adbefee | 959 | } else if (!page->slab) { |
81819f0f | 960 | printk(KERN_ERR |
70d71228 | 961 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 962 | object); |
70d71228 | 963 | dump_stack(); |
06428780 | 964 | } else |
24922684 CL |
965 | object_err(s, page, object, |
966 | "page slab pointer corrupt."); | |
81819f0f CL |
967 | goto fail; |
968 | } | |
3ec09742 CL |
969 | |
970 | /* Special debug activities for freeing objects */ | |
8a38082d | 971 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
972 | remove_full(s, page); |
973 | if (s->flags & SLAB_STORE_USER) | |
974 | set_track(s, object, TRACK_FREE, addr); | |
975 | trace(s, page, object, 0); | |
f7cb1933 | 976 | init_object(s, object, SLUB_RED_INACTIVE); |
81819f0f | 977 | return 1; |
3ec09742 | 978 | |
81819f0f | 979 | fail: |
24922684 | 980 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
981 | return 0; |
982 | } | |
983 | ||
41ecc55b CL |
984 | static int __init setup_slub_debug(char *str) |
985 | { | |
f0630fff CL |
986 | slub_debug = DEBUG_DEFAULT_FLAGS; |
987 | if (*str++ != '=' || !*str) | |
988 | /* | |
989 | * No options specified. Switch on full debugging. | |
990 | */ | |
991 | goto out; | |
992 | ||
993 | if (*str == ',') | |
994 | /* | |
995 | * No options but restriction on slabs. This means full | |
996 | * debugging for slabs matching a pattern. | |
997 | */ | |
998 | goto check_slabs; | |
999 | ||
fa5ec8a1 DR |
1000 | if (tolower(*str) == 'o') { |
1001 | /* | |
1002 | * Avoid enabling debugging on caches if its minimum order | |
1003 | * would increase as a result. | |
1004 | */ | |
1005 | disable_higher_order_debug = 1; | |
1006 | goto out; | |
1007 | } | |
1008 | ||
f0630fff CL |
1009 | slub_debug = 0; |
1010 | if (*str == '-') | |
1011 | /* | |
1012 | * Switch off all debugging measures. | |
1013 | */ | |
1014 | goto out; | |
1015 | ||
1016 | /* | |
1017 | * Determine which debug features should be switched on | |
1018 | */ | |
06428780 | 1019 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1020 | switch (tolower(*str)) { |
1021 | case 'f': | |
1022 | slub_debug |= SLAB_DEBUG_FREE; | |
1023 | break; | |
1024 | case 'z': | |
1025 | slub_debug |= SLAB_RED_ZONE; | |
1026 | break; | |
1027 | case 'p': | |
1028 | slub_debug |= SLAB_POISON; | |
1029 | break; | |
1030 | case 'u': | |
1031 | slub_debug |= SLAB_STORE_USER; | |
1032 | break; | |
1033 | case 't': | |
1034 | slub_debug |= SLAB_TRACE; | |
1035 | break; | |
4c13dd3b DM |
1036 | case 'a': |
1037 | slub_debug |= SLAB_FAILSLAB; | |
1038 | break; | |
f0630fff CL |
1039 | default: |
1040 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1041 | "unknown. skipped\n", *str); |
f0630fff | 1042 | } |
41ecc55b CL |
1043 | } |
1044 | ||
f0630fff | 1045 | check_slabs: |
41ecc55b CL |
1046 | if (*str == ',') |
1047 | slub_debug_slabs = str + 1; | |
f0630fff | 1048 | out: |
41ecc55b CL |
1049 | return 1; |
1050 | } | |
1051 | ||
1052 | __setup("slub_debug", setup_slub_debug); | |
1053 | ||
ba0268a8 CL |
1054 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1055 | unsigned long flags, const char *name, | |
51cc5068 | 1056 | void (*ctor)(void *)) |
41ecc55b CL |
1057 | { |
1058 | /* | |
e153362a | 1059 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1060 | */ |
e153362a | 1061 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1062 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1063 | flags |= slub_debug; | |
ba0268a8 CL |
1064 | |
1065 | return flags; | |
41ecc55b CL |
1066 | } |
1067 | #else | |
3ec09742 CL |
1068 | static inline void setup_object_debug(struct kmem_cache *s, |
1069 | struct page *page, void *object) {} | |
41ecc55b | 1070 | |
3ec09742 | 1071 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1072 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1073 | |
3ec09742 | 1074 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1075 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1076 | |
41ecc55b CL |
1077 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1078 | { return 1; } | |
1079 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1080 | void *object, u8 val) { return 1; } |
3ec09742 | 1081 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1082 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1083 | unsigned long flags, const char *name, | |
51cc5068 | 1084 | void (*ctor)(void *)) |
ba0268a8 CL |
1085 | { |
1086 | return flags; | |
1087 | } | |
41ecc55b | 1088 | #define slub_debug 0 |
0f389ec6 | 1089 | |
fdaa45e9 IM |
1090 | #define disable_higher_order_debug 0 |
1091 | ||
0f389ec6 CL |
1092 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1093 | { return 0; } | |
26c02cf0 AB |
1094 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1095 | { return 0; } | |
205ab99d CL |
1096 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1097 | int objects) {} | |
1098 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1099 | int objects) {} | |
7d550c56 CL |
1100 | |
1101 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1102 | { return 0; } | |
1103 | ||
1104 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1105 | void *object) {} | |
1106 | ||
1107 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1108 | ||
1109 | static inline void slab_free_hook_irq(struct kmem_cache *s, | |
1110 | void *object) {} | |
1111 | ||
ab4d5ed5 | 1112 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1113 | |
81819f0f CL |
1114 | /* |
1115 | * Slab allocation and freeing | |
1116 | */ | |
65c3376a CL |
1117 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1118 | struct kmem_cache_order_objects oo) | |
1119 | { | |
1120 | int order = oo_order(oo); | |
1121 | ||
b1eeab67 VN |
1122 | flags |= __GFP_NOTRACK; |
1123 | ||
2154a336 | 1124 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1125 | return alloc_pages(flags, order); |
1126 | else | |
6b65aaf3 | 1127 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1128 | } |
1129 | ||
81819f0f CL |
1130 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1131 | { | |
06428780 | 1132 | struct page *page; |
834f3d11 | 1133 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1134 | gfp_t alloc_gfp; |
81819f0f | 1135 | |
b7a49f0d | 1136 | flags |= s->allocflags; |
e12ba74d | 1137 | |
ba52270d PE |
1138 | /* |
1139 | * Let the initial higher-order allocation fail under memory pressure | |
1140 | * so we fall-back to the minimum order allocation. | |
1141 | */ | |
1142 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1143 | ||
1144 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1145 | if (unlikely(!page)) { |
1146 | oo = s->min; | |
1147 | /* | |
1148 | * Allocation may have failed due to fragmentation. | |
1149 | * Try a lower order alloc if possible | |
1150 | */ | |
1151 | page = alloc_slab_page(flags, node, oo); | |
1152 | if (!page) | |
1153 | return NULL; | |
81819f0f | 1154 | |
84e554e6 | 1155 | stat(s, ORDER_FALLBACK); |
65c3376a | 1156 | } |
5a896d9e VN |
1157 | |
1158 | if (kmemcheck_enabled | |
5086c389 | 1159 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1160 | int pages = 1 << oo_order(oo); |
1161 | ||
1162 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1163 | ||
1164 | /* | |
1165 | * Objects from caches that have a constructor don't get | |
1166 | * cleared when they're allocated, so we need to do it here. | |
1167 | */ | |
1168 | if (s->ctor) | |
1169 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1170 | else | |
1171 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1172 | } |
1173 | ||
834f3d11 | 1174 | page->objects = oo_objects(oo); |
81819f0f CL |
1175 | mod_zone_page_state(page_zone(page), |
1176 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1177 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1178 | 1 << oo_order(oo)); |
81819f0f CL |
1179 | |
1180 | return page; | |
1181 | } | |
1182 | ||
1183 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1184 | void *object) | |
1185 | { | |
3ec09742 | 1186 | setup_object_debug(s, page, object); |
4f104934 | 1187 | if (unlikely(s->ctor)) |
51cc5068 | 1188 | s->ctor(object); |
81819f0f CL |
1189 | } |
1190 | ||
1191 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1192 | { | |
1193 | struct page *page; | |
81819f0f | 1194 | void *start; |
81819f0f CL |
1195 | void *last; |
1196 | void *p; | |
1197 | ||
6cb06229 | 1198 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1199 | |
6cb06229 CL |
1200 | page = allocate_slab(s, |
1201 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1202 | if (!page) |
1203 | goto out; | |
1204 | ||
205ab99d | 1205 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1206 | page->slab = s; |
1207 | page->flags |= 1 << PG_slab; | |
81819f0f CL |
1208 | |
1209 | start = page_address(page); | |
81819f0f CL |
1210 | |
1211 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1212 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1213 | |
1214 | last = start; | |
224a88be | 1215 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1216 | setup_object(s, page, last); |
1217 | set_freepointer(s, last, p); | |
1218 | last = p; | |
1219 | } | |
1220 | setup_object(s, page, last); | |
a973e9dd | 1221 | set_freepointer(s, last, NULL); |
81819f0f CL |
1222 | |
1223 | page->freelist = start; | |
1224 | page->inuse = 0; | |
1225 | out: | |
81819f0f CL |
1226 | return page; |
1227 | } | |
1228 | ||
1229 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1230 | { | |
834f3d11 CL |
1231 | int order = compound_order(page); |
1232 | int pages = 1 << order; | |
81819f0f | 1233 | |
af537b0a | 1234 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1235 | void *p; |
1236 | ||
1237 | slab_pad_check(s, page); | |
224a88be CL |
1238 | for_each_object(p, s, page_address(page), |
1239 | page->objects) | |
f7cb1933 | 1240 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1241 | } |
1242 | ||
b1eeab67 | 1243 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1244 | |
81819f0f CL |
1245 | mod_zone_page_state(page_zone(page), |
1246 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1247 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1248 | -pages); |
81819f0f | 1249 | |
49bd5221 CL |
1250 | __ClearPageSlab(page); |
1251 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1252 | if (current->reclaim_state) |
1253 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1254 | __free_pages(page, order); |
81819f0f CL |
1255 | } |
1256 | ||
1257 | static void rcu_free_slab(struct rcu_head *h) | |
1258 | { | |
1259 | struct page *page; | |
1260 | ||
1261 | page = container_of((struct list_head *)h, struct page, lru); | |
1262 | __free_slab(page->slab, page); | |
1263 | } | |
1264 | ||
1265 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1266 | { | |
1267 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1268 | /* | |
1269 | * RCU free overloads the RCU head over the LRU | |
1270 | */ | |
1271 | struct rcu_head *head = (void *)&page->lru; | |
1272 | ||
1273 | call_rcu(head, rcu_free_slab); | |
1274 | } else | |
1275 | __free_slab(s, page); | |
1276 | } | |
1277 | ||
1278 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1279 | { | |
205ab99d | 1280 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1281 | free_slab(s, page); |
1282 | } | |
1283 | ||
1284 | /* | |
1285 | * Per slab locking using the pagelock | |
1286 | */ | |
1287 | static __always_inline void slab_lock(struct page *page) | |
1288 | { | |
1289 | bit_spin_lock(PG_locked, &page->flags); | |
1290 | } | |
1291 | ||
1292 | static __always_inline void slab_unlock(struct page *page) | |
1293 | { | |
a76d3546 | 1294 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1295 | } |
1296 | ||
1297 | static __always_inline int slab_trylock(struct page *page) | |
1298 | { | |
1299 | int rc = 1; | |
1300 | ||
1301 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1302 | return rc; | |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * Management of partially allocated slabs | |
1307 | */ | |
7c2e132c CL |
1308 | static void add_partial(struct kmem_cache_node *n, |
1309 | struct page *page, int tail) | |
81819f0f | 1310 | { |
e95eed57 CL |
1311 | spin_lock(&n->list_lock); |
1312 | n->nr_partial++; | |
7c2e132c CL |
1313 | if (tail) |
1314 | list_add_tail(&page->lru, &n->partial); | |
1315 | else | |
1316 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1317 | spin_unlock(&n->list_lock); |
1318 | } | |
1319 | ||
62e346a8 CL |
1320 | static inline void __remove_partial(struct kmem_cache_node *n, |
1321 | struct page *page) | |
1322 | { | |
1323 | list_del(&page->lru); | |
1324 | n->nr_partial--; | |
1325 | } | |
1326 | ||
0121c619 | 1327 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1328 | { |
1329 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1330 | ||
1331 | spin_lock(&n->list_lock); | |
62e346a8 | 1332 | __remove_partial(n, page); |
81819f0f CL |
1333 | spin_unlock(&n->list_lock); |
1334 | } | |
1335 | ||
1336 | /* | |
672bba3a | 1337 | * Lock slab and remove from the partial list. |
81819f0f | 1338 | * |
672bba3a | 1339 | * Must hold list_lock. |
81819f0f | 1340 | */ |
0121c619 CL |
1341 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1342 | struct page *page) | |
81819f0f CL |
1343 | { |
1344 | if (slab_trylock(page)) { | |
62e346a8 | 1345 | __remove_partial(n, page); |
8a38082d | 1346 | __SetPageSlubFrozen(page); |
81819f0f CL |
1347 | return 1; |
1348 | } | |
1349 | return 0; | |
1350 | } | |
1351 | ||
1352 | /* | |
672bba3a | 1353 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1354 | */ |
1355 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1356 | { | |
1357 | struct page *page; | |
1358 | ||
1359 | /* | |
1360 | * Racy check. If we mistakenly see no partial slabs then we | |
1361 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1362 | * partial slab and there is none available then get_partials() |
1363 | * will return NULL. | |
81819f0f CL |
1364 | */ |
1365 | if (!n || !n->nr_partial) | |
1366 | return NULL; | |
1367 | ||
1368 | spin_lock(&n->list_lock); | |
1369 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1370 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1371 | goto out; |
1372 | page = NULL; | |
1373 | out: | |
1374 | spin_unlock(&n->list_lock); | |
1375 | return page; | |
1376 | } | |
1377 | ||
1378 | /* | |
672bba3a | 1379 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1380 | */ |
1381 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1382 | { | |
1383 | #ifdef CONFIG_NUMA | |
1384 | struct zonelist *zonelist; | |
dd1a239f | 1385 | struct zoneref *z; |
54a6eb5c MG |
1386 | struct zone *zone; |
1387 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1388 | struct page *page; |
1389 | ||
1390 | /* | |
672bba3a CL |
1391 | * The defrag ratio allows a configuration of the tradeoffs between |
1392 | * inter node defragmentation and node local allocations. A lower | |
1393 | * defrag_ratio increases the tendency to do local allocations | |
1394 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1395 | * |
672bba3a CL |
1396 | * If the defrag_ratio is set to 0 then kmalloc() always |
1397 | * returns node local objects. If the ratio is higher then kmalloc() | |
1398 | * may return off node objects because partial slabs are obtained | |
1399 | * from other nodes and filled up. | |
81819f0f | 1400 | * |
6446faa2 | 1401 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1402 | * defrag_ratio = 1000) then every (well almost) allocation will |
1403 | * first attempt to defrag slab caches on other nodes. This means | |
1404 | * scanning over all nodes to look for partial slabs which may be | |
1405 | * expensive if we do it every time we are trying to find a slab | |
1406 | * with available objects. | |
81819f0f | 1407 | */ |
9824601e CL |
1408 | if (!s->remote_node_defrag_ratio || |
1409 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1410 | return NULL; |
1411 | ||
c0ff7453 | 1412 | get_mems_allowed(); |
0e88460d | 1413 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1414 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1415 | struct kmem_cache_node *n; |
1416 | ||
54a6eb5c | 1417 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1418 | |
54a6eb5c | 1419 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1420 | n->nr_partial > s->min_partial) { |
81819f0f | 1421 | page = get_partial_node(n); |
c0ff7453 MX |
1422 | if (page) { |
1423 | put_mems_allowed(); | |
81819f0f | 1424 | return page; |
c0ff7453 | 1425 | } |
81819f0f CL |
1426 | } |
1427 | } | |
c0ff7453 | 1428 | put_mems_allowed(); |
81819f0f CL |
1429 | #endif |
1430 | return NULL; | |
1431 | } | |
1432 | ||
1433 | /* | |
1434 | * Get a partial page, lock it and return it. | |
1435 | */ | |
1436 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1437 | { | |
1438 | struct page *page; | |
2154a336 | 1439 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f CL |
1440 | |
1441 | page = get_partial_node(get_node(s, searchnode)); | |
bc6488e9 | 1442 | if (page || node != -1) |
81819f0f CL |
1443 | return page; |
1444 | ||
1445 | return get_any_partial(s, flags); | |
1446 | } | |
1447 | ||
1448 | /* | |
1449 | * Move a page back to the lists. | |
1450 | * | |
1451 | * Must be called with the slab lock held. | |
1452 | * | |
1453 | * On exit the slab lock will have been dropped. | |
1454 | */ | |
7c2e132c | 1455 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
3478973d | 1456 | __releases(bitlock) |
81819f0f | 1457 | { |
e95eed57 CL |
1458 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1459 | ||
8a38082d | 1460 | __ClearPageSlubFrozen(page); |
81819f0f | 1461 | if (page->inuse) { |
e95eed57 | 1462 | |
a973e9dd | 1463 | if (page->freelist) { |
7c2e132c | 1464 | add_partial(n, page, tail); |
84e554e6 | 1465 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
8ff12cfc | 1466 | } else { |
84e554e6 | 1467 | stat(s, DEACTIVATE_FULL); |
af537b0a | 1468 | if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER)) |
8ff12cfc CL |
1469 | add_full(n, page); |
1470 | } | |
81819f0f CL |
1471 | slab_unlock(page); |
1472 | } else { | |
84e554e6 | 1473 | stat(s, DEACTIVATE_EMPTY); |
3b89d7d8 | 1474 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1475 | /* |
672bba3a CL |
1476 | * Adding an empty slab to the partial slabs in order |
1477 | * to avoid page allocator overhead. This slab needs | |
1478 | * to come after the other slabs with objects in | |
6446faa2 CL |
1479 | * so that the others get filled first. That way the |
1480 | * size of the partial list stays small. | |
1481 | * | |
0121c619 CL |
1482 | * kmem_cache_shrink can reclaim any empty slabs from |
1483 | * the partial list. | |
e95eed57 | 1484 | */ |
7c2e132c | 1485 | add_partial(n, page, 1); |
e95eed57 CL |
1486 | slab_unlock(page); |
1487 | } else { | |
1488 | slab_unlock(page); | |
84e554e6 | 1489 | stat(s, FREE_SLAB); |
e95eed57 CL |
1490 | discard_slab(s, page); |
1491 | } | |
81819f0f CL |
1492 | } |
1493 | } | |
1494 | ||
1495 | /* | |
1496 | * Remove the cpu slab | |
1497 | */ | |
dfb4f096 | 1498 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
3478973d | 1499 | __releases(bitlock) |
81819f0f | 1500 | { |
dfb4f096 | 1501 | struct page *page = c->page; |
7c2e132c | 1502 | int tail = 1; |
8ff12cfc | 1503 | |
b773ad73 | 1504 | if (page->freelist) |
84e554e6 | 1505 | stat(s, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1506 | /* |
6446faa2 | 1507 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1508 | * because both freelists are empty. So this is unlikely |
1509 | * to occur. | |
1510 | */ | |
a973e9dd | 1511 | while (unlikely(c->freelist)) { |
894b8788 CL |
1512 | void **object; |
1513 | ||
7c2e132c CL |
1514 | tail = 0; /* Hot objects. Put the slab first */ |
1515 | ||
894b8788 | 1516 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1517 | object = c->freelist; |
ff12059e | 1518 | c->freelist = get_freepointer(s, c->freelist); |
894b8788 CL |
1519 | |
1520 | /* And put onto the regular freelist */ | |
ff12059e | 1521 | set_freepointer(s, object, page->freelist); |
894b8788 CL |
1522 | page->freelist = object; |
1523 | page->inuse--; | |
1524 | } | |
dfb4f096 | 1525 | c->page = NULL; |
7c2e132c | 1526 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1527 | } |
1528 | ||
dfb4f096 | 1529 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1530 | { |
84e554e6 | 1531 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 CL |
1532 | slab_lock(c->page); |
1533 | deactivate_slab(s, c); | |
81819f0f CL |
1534 | } |
1535 | ||
1536 | /* | |
1537 | * Flush cpu slab. | |
6446faa2 | 1538 | * |
81819f0f CL |
1539 | * Called from IPI handler with interrupts disabled. |
1540 | */ | |
0c710013 | 1541 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1542 | { |
9dfc6e68 | 1543 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1544 | |
dfb4f096 CL |
1545 | if (likely(c && c->page)) |
1546 | flush_slab(s, c); | |
81819f0f CL |
1547 | } |
1548 | ||
1549 | static void flush_cpu_slab(void *d) | |
1550 | { | |
1551 | struct kmem_cache *s = d; | |
81819f0f | 1552 | |
dfb4f096 | 1553 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1554 | } |
1555 | ||
1556 | static void flush_all(struct kmem_cache *s) | |
1557 | { | |
15c8b6c1 | 1558 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1559 | } |
1560 | ||
dfb4f096 CL |
1561 | /* |
1562 | * Check if the objects in a per cpu structure fit numa | |
1563 | * locality expectations. | |
1564 | */ | |
1565 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1566 | { | |
1567 | #ifdef CONFIG_NUMA | |
2154a336 | 1568 | if (node != NUMA_NO_NODE && c->node != node) |
dfb4f096 CL |
1569 | return 0; |
1570 | #endif | |
1571 | return 1; | |
1572 | } | |
1573 | ||
781b2ba6 PE |
1574 | static int count_free(struct page *page) |
1575 | { | |
1576 | return page->objects - page->inuse; | |
1577 | } | |
1578 | ||
1579 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1580 | int (*get_count)(struct page *)) | |
1581 | { | |
1582 | unsigned long flags; | |
1583 | unsigned long x = 0; | |
1584 | struct page *page; | |
1585 | ||
1586 | spin_lock_irqsave(&n->list_lock, flags); | |
1587 | list_for_each_entry(page, &n->partial, lru) | |
1588 | x += get_count(page); | |
1589 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1590 | return x; | |
1591 | } | |
1592 | ||
26c02cf0 AB |
1593 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1594 | { | |
1595 | #ifdef CONFIG_SLUB_DEBUG | |
1596 | return atomic_long_read(&n->total_objects); | |
1597 | #else | |
1598 | return 0; | |
1599 | #endif | |
1600 | } | |
1601 | ||
781b2ba6 PE |
1602 | static noinline void |
1603 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1604 | { | |
1605 | int node; | |
1606 | ||
1607 | printk(KERN_WARNING | |
1608 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1609 | nid, gfpflags); | |
1610 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1611 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1612 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1613 | ||
fa5ec8a1 DR |
1614 | if (oo_order(s->min) > get_order(s->objsize)) |
1615 | printk(KERN_WARNING " %s debugging increased min order, use " | |
1616 | "slub_debug=O to disable.\n", s->name); | |
1617 | ||
781b2ba6 PE |
1618 | for_each_online_node(node) { |
1619 | struct kmem_cache_node *n = get_node(s, node); | |
1620 | unsigned long nr_slabs; | |
1621 | unsigned long nr_objs; | |
1622 | unsigned long nr_free; | |
1623 | ||
1624 | if (!n) | |
1625 | continue; | |
1626 | ||
26c02cf0 AB |
1627 | nr_free = count_partial(n, count_free); |
1628 | nr_slabs = node_nr_slabs(n); | |
1629 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1630 | |
1631 | printk(KERN_WARNING | |
1632 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1633 | node, nr_slabs, nr_objs, nr_free); | |
1634 | } | |
1635 | } | |
1636 | ||
81819f0f | 1637 | /* |
894b8788 CL |
1638 | * Slow path. The lockless freelist is empty or we need to perform |
1639 | * debugging duties. | |
1640 | * | |
1641 | * Interrupts are disabled. | |
81819f0f | 1642 | * |
894b8788 CL |
1643 | * Processing is still very fast if new objects have been freed to the |
1644 | * regular freelist. In that case we simply take over the regular freelist | |
1645 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1646 | * |
894b8788 CL |
1647 | * If that is not working then we fall back to the partial lists. We take the |
1648 | * first element of the freelist as the object to allocate now and move the | |
1649 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1650 | * |
894b8788 | 1651 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1652 | * we need to allocate a new slab. This is the slowest path since it involves |
1653 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1654 | */ |
ce71e27c EGM |
1655 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1656 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1657 | { |
81819f0f | 1658 | void **object; |
dfb4f096 | 1659 | struct page *new; |
81819f0f | 1660 | |
e72e9c23 LT |
1661 | /* We handle __GFP_ZERO in the caller */ |
1662 | gfpflags &= ~__GFP_ZERO; | |
1663 | ||
dfb4f096 | 1664 | if (!c->page) |
81819f0f CL |
1665 | goto new_slab; |
1666 | ||
dfb4f096 CL |
1667 | slab_lock(c->page); |
1668 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1669 | goto another_slab; |
6446faa2 | 1670 | |
84e554e6 | 1671 | stat(s, ALLOC_REFILL); |
6446faa2 | 1672 | |
894b8788 | 1673 | load_freelist: |
dfb4f096 | 1674 | object = c->page->freelist; |
a973e9dd | 1675 | if (unlikely(!object)) |
81819f0f | 1676 | goto another_slab; |
af537b0a | 1677 | if (kmem_cache_debug(s)) |
81819f0f CL |
1678 | goto debug; |
1679 | ||
ff12059e | 1680 | c->freelist = get_freepointer(s, object); |
39b26464 | 1681 | c->page->inuse = c->page->objects; |
a973e9dd | 1682 | c->page->freelist = NULL; |
dfb4f096 | 1683 | c->node = page_to_nid(c->page); |
1f84260c | 1684 | unlock_out: |
dfb4f096 | 1685 | slab_unlock(c->page); |
84e554e6 | 1686 | stat(s, ALLOC_SLOWPATH); |
81819f0f CL |
1687 | return object; |
1688 | ||
1689 | another_slab: | |
dfb4f096 | 1690 | deactivate_slab(s, c); |
81819f0f CL |
1691 | |
1692 | new_slab: | |
dfb4f096 CL |
1693 | new = get_partial(s, gfpflags, node); |
1694 | if (new) { | |
1695 | c->page = new; | |
84e554e6 | 1696 | stat(s, ALLOC_FROM_PARTIAL); |
894b8788 | 1697 | goto load_freelist; |
81819f0f CL |
1698 | } |
1699 | ||
c1d50836 | 1700 | gfpflags &= gfp_allowed_mask; |
b811c202 CL |
1701 | if (gfpflags & __GFP_WAIT) |
1702 | local_irq_enable(); | |
1703 | ||
dfb4f096 | 1704 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1705 | |
1706 | if (gfpflags & __GFP_WAIT) | |
1707 | local_irq_disable(); | |
1708 | ||
dfb4f096 | 1709 | if (new) { |
9dfc6e68 | 1710 | c = __this_cpu_ptr(s->cpu_slab); |
84e554e6 | 1711 | stat(s, ALLOC_SLAB); |
05aa3450 | 1712 | if (c->page) |
dfb4f096 | 1713 | flush_slab(s, c); |
dfb4f096 | 1714 | slab_lock(new); |
8a38082d | 1715 | __SetPageSlubFrozen(new); |
dfb4f096 | 1716 | c->page = new; |
4b6f0750 | 1717 | goto load_freelist; |
81819f0f | 1718 | } |
95f85989 PE |
1719 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
1720 | slab_out_of_memory(s, gfpflags, node); | |
71c7a06f | 1721 | return NULL; |
81819f0f | 1722 | debug: |
dfb4f096 | 1723 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1724 | goto another_slab; |
894b8788 | 1725 | |
dfb4f096 | 1726 | c->page->inuse++; |
ff12059e | 1727 | c->page->freelist = get_freepointer(s, object); |
15b7c514 | 1728 | c->node = NUMA_NO_NODE; |
1f84260c | 1729 | goto unlock_out; |
894b8788 CL |
1730 | } |
1731 | ||
1732 | /* | |
1733 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1734 | * have the fastpath folded into their functions. So no function call | |
1735 | * overhead for requests that can be satisfied on the fastpath. | |
1736 | * | |
1737 | * The fastpath works by first checking if the lockless freelist can be used. | |
1738 | * If not then __slab_alloc is called for slow processing. | |
1739 | * | |
1740 | * Otherwise we can simply pick the next object from the lockless free list. | |
1741 | */ | |
06428780 | 1742 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1743 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1744 | { |
894b8788 | 1745 | void **object; |
dfb4f096 | 1746 | struct kmem_cache_cpu *c; |
1f84260c CL |
1747 | unsigned long flags; |
1748 | ||
c016b0bd | 1749 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 1750 | return NULL; |
1f84260c | 1751 | |
894b8788 | 1752 | local_irq_save(flags); |
9dfc6e68 CL |
1753 | c = __this_cpu_ptr(s->cpu_slab); |
1754 | object = c->freelist; | |
9dfc6e68 | 1755 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 1756 | |
dfb4f096 | 1757 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1758 | |
1759 | else { | |
ff12059e | 1760 | c->freelist = get_freepointer(s, object); |
84e554e6 | 1761 | stat(s, ALLOC_FASTPATH); |
894b8788 CL |
1762 | } |
1763 | local_irq_restore(flags); | |
d07dbea4 | 1764 | |
74e2134f | 1765 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 1766 | memset(object, 0, s->objsize); |
d07dbea4 | 1767 | |
c016b0bd | 1768 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 1769 | |
894b8788 | 1770 | return object; |
81819f0f CL |
1771 | } |
1772 | ||
1773 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1774 | { | |
2154a336 | 1775 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 1776 | |
ca2b84cb | 1777 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1778 | |
1779 | return ret; | |
81819f0f CL |
1780 | } |
1781 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1782 | ||
0f24f128 | 1783 | #ifdef CONFIG_TRACING |
4a92379b RK |
1784 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
1785 | { | |
1786 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
1787 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
1788 | return ret; | |
1789 | } | |
1790 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
1791 | ||
1792 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 1793 | { |
4a92379b RK |
1794 | void *ret = kmalloc_order(size, flags, order); |
1795 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1796 | return ret; | |
5b882be4 | 1797 | } |
4a92379b | 1798 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
1799 | #endif |
1800 | ||
81819f0f CL |
1801 | #ifdef CONFIG_NUMA |
1802 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1803 | { | |
5b882be4 EGM |
1804 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1805 | ||
ca2b84cb EGM |
1806 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
1807 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
1808 | |
1809 | return ret; | |
81819f0f CL |
1810 | } |
1811 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 1812 | |
0f24f128 | 1813 | #ifdef CONFIG_TRACING |
4a92379b | 1814 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 1815 | gfp_t gfpflags, |
4a92379b | 1816 | int node, size_t size) |
5b882be4 | 1817 | { |
4a92379b RK |
1818 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1819 | ||
1820 | trace_kmalloc_node(_RET_IP_, ret, | |
1821 | size, s->size, gfpflags, node); | |
1822 | return ret; | |
5b882be4 | 1823 | } |
4a92379b | 1824 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 1825 | #endif |
5d1f57e4 | 1826 | #endif |
5b882be4 | 1827 | |
81819f0f | 1828 | /* |
894b8788 CL |
1829 | * Slow patch handling. This may still be called frequently since objects |
1830 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1831 | * |
894b8788 CL |
1832 | * So we still attempt to reduce cache line usage. Just take the slab |
1833 | * lock and free the item. If there is no additional partial page | |
1834 | * handling required then we can return immediately. | |
81819f0f | 1835 | */ |
894b8788 | 1836 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 1837 | void *x, unsigned long addr) |
81819f0f CL |
1838 | { |
1839 | void *prior; | |
1840 | void **object = (void *)x; | |
81819f0f | 1841 | |
84e554e6 | 1842 | stat(s, FREE_SLOWPATH); |
81819f0f CL |
1843 | slab_lock(page); |
1844 | ||
af537b0a | 1845 | if (kmem_cache_debug(s)) |
81819f0f | 1846 | goto debug; |
6446faa2 | 1847 | |
81819f0f | 1848 | checks_ok: |
ff12059e CL |
1849 | prior = page->freelist; |
1850 | set_freepointer(s, object, prior); | |
81819f0f CL |
1851 | page->freelist = object; |
1852 | page->inuse--; | |
1853 | ||
8a38082d | 1854 | if (unlikely(PageSlubFrozen(page))) { |
84e554e6 | 1855 | stat(s, FREE_FROZEN); |
81819f0f | 1856 | goto out_unlock; |
8ff12cfc | 1857 | } |
81819f0f CL |
1858 | |
1859 | if (unlikely(!page->inuse)) | |
1860 | goto slab_empty; | |
1861 | ||
1862 | /* | |
6446faa2 | 1863 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1864 | * then add it. |
1865 | */ | |
a973e9dd | 1866 | if (unlikely(!prior)) { |
7c2e132c | 1867 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
84e554e6 | 1868 | stat(s, FREE_ADD_PARTIAL); |
8ff12cfc | 1869 | } |
81819f0f CL |
1870 | |
1871 | out_unlock: | |
1872 | slab_unlock(page); | |
81819f0f CL |
1873 | return; |
1874 | ||
1875 | slab_empty: | |
a973e9dd | 1876 | if (prior) { |
81819f0f | 1877 | /* |
672bba3a | 1878 | * Slab still on the partial list. |
81819f0f CL |
1879 | */ |
1880 | remove_partial(s, page); | |
84e554e6 | 1881 | stat(s, FREE_REMOVE_PARTIAL); |
8ff12cfc | 1882 | } |
81819f0f | 1883 | slab_unlock(page); |
84e554e6 | 1884 | stat(s, FREE_SLAB); |
81819f0f | 1885 | discard_slab(s, page); |
81819f0f CL |
1886 | return; |
1887 | ||
1888 | debug: | |
3ec09742 | 1889 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1890 | goto out_unlock; |
77c5e2d0 | 1891 | goto checks_ok; |
81819f0f CL |
1892 | } |
1893 | ||
894b8788 CL |
1894 | /* |
1895 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1896 | * can perform fastpath freeing without additional function calls. | |
1897 | * | |
1898 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1899 | * of this processor. This typically the case if we have just allocated | |
1900 | * the item before. | |
1901 | * | |
1902 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1903 | * with all sorts of special processing. | |
1904 | */ | |
06428780 | 1905 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1906 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1907 | { |
1908 | void **object = (void *)x; | |
dfb4f096 | 1909 | struct kmem_cache_cpu *c; |
1f84260c CL |
1910 | unsigned long flags; |
1911 | ||
c016b0bd CL |
1912 | slab_free_hook(s, x); |
1913 | ||
894b8788 | 1914 | local_irq_save(flags); |
9dfc6e68 | 1915 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd CL |
1916 | |
1917 | slab_free_hook_irq(s, x); | |
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 | 1995 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 1996 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
1997 | { |
1998 | int order; | |
1999 | int rem; | |
6300ea75 | 2000 | int min_order = slub_min_order; |
81819f0f | 2001 | |
ab9a0f19 | 2002 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 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 | |
ab9a0f19 | 2011 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2012 | continue; |
2013 | ||
ab9a0f19 | 2014 | rem = (slab_size - reserved) % size; |
81819f0f | 2015 | |
5e6d444e | 2016 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2017 | break; |
2018 | ||
2019 | } | |
672bba3a | 2020 | |
81819f0f CL |
2021 | return order; |
2022 | } | |
2023 | ||
ab9a0f19 | 2024 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
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); | |
ab9a0f19 | 2042 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
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, | |
ab9a0f19 | 2049 | slub_max_order, fraction, reserved); |
5e6d444e CL |
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 | */ | |
ab9a0f19 | 2061 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2062 | if (order <= slub_max_order) |
2063 | return order; | |
2064 | ||
2065 | /* | |
2066 | * Doh this slab cannot be placed using slub_max_order. | |
2067 | */ | |
ab9a0f19 | 2068 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
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 | |
ab9a0f19 | 2319 | order = calculate_order(size, s->reserved); |
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 | */ | |
ab9a0f19 LJ |
2337 | s->oo = oo_make(order, size, s->reserved); |
2338 | s->min = oo_make(get_order(size), size, s->reserved); | |
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); |
ab9a0f19 | 2357 | s->reserved = 0; |
81819f0f | 2358 | |
06b285dc | 2359 | if (!calculate_sizes(s, -1)) |
81819f0f | 2360 | goto error; |
3de47213 DR |
2361 | if (disable_higher_order_debug) { |
2362 | /* | |
2363 | * Disable debugging flags that store metadata if the min slab | |
2364 | * order increased. | |
2365 | */ | |
2366 | if (get_order(s->size) > get_order(s->objsize)) { | |
2367 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2368 | s->offset = 0; | |
2369 | if (!calculate_sizes(s, -1)) | |
2370 | goto error; | |
2371 | } | |
2372 | } | |
81819f0f | 2373 | |
3b89d7d8 DR |
2374 | /* |
2375 | * The larger the object size is, the more pages we want on the partial | |
2376 | * list to avoid pounding the page allocator excessively. | |
2377 | */ | |
c0bdb232 | 2378 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2379 | s->refcount = 1; |
2380 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2381 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2382 | #endif |
55136592 | 2383 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 2384 | goto error; |
81819f0f | 2385 | |
55136592 | 2386 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 2387 | return 1; |
ff12059e | 2388 | |
4c93c355 | 2389 | free_kmem_cache_nodes(s); |
81819f0f CL |
2390 | error: |
2391 | if (flags & SLAB_PANIC) | |
2392 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2393 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2394 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2395 | s->offset, flags); |
2396 | return 0; | |
2397 | } | |
81819f0f | 2398 | |
81819f0f CL |
2399 | /* |
2400 | * Determine the size of a slab object | |
2401 | */ | |
2402 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2403 | { | |
2404 | return s->objsize; | |
2405 | } | |
2406 | EXPORT_SYMBOL(kmem_cache_size); | |
2407 | ||
2408 | const char *kmem_cache_name(struct kmem_cache *s) | |
2409 | { | |
2410 | return s->name; | |
2411 | } | |
2412 | EXPORT_SYMBOL(kmem_cache_name); | |
2413 | ||
33b12c38 CL |
2414 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2415 | const char *text) | |
2416 | { | |
2417 | #ifdef CONFIG_SLUB_DEBUG | |
2418 | void *addr = page_address(page); | |
2419 | void *p; | |
a5dd5c11 NK |
2420 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
2421 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
2422 | if (!map) |
2423 | return; | |
33b12c38 CL |
2424 | slab_err(s, page, "%s", text); |
2425 | slab_lock(page); | |
2426 | for_each_free_object(p, s, page->freelist) | |
2427 | set_bit(slab_index(p, s, addr), map); | |
2428 | ||
2429 | for_each_object(p, s, addr, page->objects) { | |
2430 | ||
2431 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2432 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2433 | p, p - addr); | |
2434 | print_tracking(s, p); | |
2435 | } | |
2436 | } | |
2437 | slab_unlock(page); | |
bbd7d57b | 2438 | kfree(map); |
33b12c38 CL |
2439 | #endif |
2440 | } | |
2441 | ||
81819f0f | 2442 | /* |
599870b1 | 2443 | * Attempt to free all partial slabs on a node. |
81819f0f | 2444 | */ |
599870b1 | 2445 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2446 | { |
81819f0f CL |
2447 | unsigned long flags; |
2448 | struct page *page, *h; | |
2449 | ||
2450 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2451 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 2452 | if (!page->inuse) { |
62e346a8 | 2453 | __remove_partial(n, page); |
81819f0f | 2454 | discard_slab(s, page); |
33b12c38 CL |
2455 | } else { |
2456 | list_slab_objects(s, page, | |
2457 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2458 | } |
33b12c38 | 2459 | } |
81819f0f | 2460 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2461 | } |
2462 | ||
2463 | /* | |
672bba3a | 2464 | * Release all resources used by a slab cache. |
81819f0f | 2465 | */ |
0c710013 | 2466 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2467 | { |
2468 | int node; | |
2469 | ||
2470 | flush_all(s); | |
9dfc6e68 | 2471 | free_percpu(s->cpu_slab); |
81819f0f | 2472 | /* Attempt to free all objects */ |
f64dc58c | 2473 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2474 | struct kmem_cache_node *n = get_node(s, node); |
2475 | ||
599870b1 CL |
2476 | free_partial(s, n); |
2477 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2478 | return 1; |
2479 | } | |
2480 | free_kmem_cache_nodes(s); | |
2481 | return 0; | |
2482 | } | |
2483 | ||
2484 | /* | |
2485 | * Close a cache and release the kmem_cache structure | |
2486 | * (must be used for caches created using kmem_cache_create) | |
2487 | */ | |
2488 | void kmem_cache_destroy(struct kmem_cache *s) | |
2489 | { | |
2490 | down_write(&slub_lock); | |
2491 | s->refcount--; | |
2492 | if (!s->refcount) { | |
2493 | list_del(&s->list); | |
d629d819 PE |
2494 | if (kmem_cache_close(s)) { |
2495 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2496 | "still has objects.\n", s->name, __func__); | |
2497 | dump_stack(); | |
2498 | } | |
d76b1590 ED |
2499 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2500 | rcu_barrier(); | |
81819f0f | 2501 | sysfs_slab_remove(s); |
2bce6485 CL |
2502 | } |
2503 | up_write(&slub_lock); | |
81819f0f CL |
2504 | } |
2505 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2506 | ||
2507 | /******************************************************************** | |
2508 | * Kmalloc subsystem | |
2509 | *******************************************************************/ | |
2510 | ||
51df1142 | 2511 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
2512 | EXPORT_SYMBOL(kmalloc_caches); |
2513 | ||
51df1142 CL |
2514 | static struct kmem_cache *kmem_cache; |
2515 | ||
55136592 | 2516 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 2517 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
2518 | #endif |
2519 | ||
81819f0f CL |
2520 | static int __init setup_slub_min_order(char *str) |
2521 | { | |
06428780 | 2522 | get_option(&str, &slub_min_order); |
81819f0f CL |
2523 | |
2524 | return 1; | |
2525 | } | |
2526 | ||
2527 | __setup("slub_min_order=", setup_slub_min_order); | |
2528 | ||
2529 | static int __init setup_slub_max_order(char *str) | |
2530 | { | |
06428780 | 2531 | get_option(&str, &slub_max_order); |
818cf590 | 2532 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2533 | |
2534 | return 1; | |
2535 | } | |
2536 | ||
2537 | __setup("slub_max_order=", setup_slub_max_order); | |
2538 | ||
2539 | static int __init setup_slub_min_objects(char *str) | |
2540 | { | |
06428780 | 2541 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2542 | |
2543 | return 1; | |
2544 | } | |
2545 | ||
2546 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2547 | ||
2548 | static int __init setup_slub_nomerge(char *str) | |
2549 | { | |
2550 | slub_nomerge = 1; | |
2551 | return 1; | |
2552 | } | |
2553 | ||
2554 | __setup("slub_nomerge", setup_slub_nomerge); | |
2555 | ||
51df1142 CL |
2556 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
2557 | int size, unsigned int flags) | |
81819f0f | 2558 | { |
51df1142 CL |
2559 | struct kmem_cache *s; |
2560 | ||
2561 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
2562 | ||
83b519e8 PE |
2563 | /* |
2564 | * This function is called with IRQs disabled during early-boot on | |
2565 | * single CPU so there's no need to take slub_lock here. | |
2566 | */ | |
55136592 | 2567 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2568 | flags, NULL)) |
81819f0f CL |
2569 | goto panic; |
2570 | ||
2571 | list_add(&s->list, &slab_caches); | |
51df1142 | 2572 | return s; |
81819f0f CL |
2573 | |
2574 | panic: | |
2575 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 2576 | return NULL; |
81819f0f CL |
2577 | } |
2578 | ||
f1b26339 CL |
2579 | /* |
2580 | * Conversion table for small slabs sizes / 8 to the index in the | |
2581 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2582 | * of two cache sizes there. The size of larger slabs can be determined using | |
2583 | * fls. | |
2584 | */ | |
2585 | static s8 size_index[24] = { | |
2586 | 3, /* 8 */ | |
2587 | 4, /* 16 */ | |
2588 | 5, /* 24 */ | |
2589 | 5, /* 32 */ | |
2590 | 6, /* 40 */ | |
2591 | 6, /* 48 */ | |
2592 | 6, /* 56 */ | |
2593 | 6, /* 64 */ | |
2594 | 1, /* 72 */ | |
2595 | 1, /* 80 */ | |
2596 | 1, /* 88 */ | |
2597 | 1, /* 96 */ | |
2598 | 7, /* 104 */ | |
2599 | 7, /* 112 */ | |
2600 | 7, /* 120 */ | |
2601 | 7, /* 128 */ | |
2602 | 2, /* 136 */ | |
2603 | 2, /* 144 */ | |
2604 | 2, /* 152 */ | |
2605 | 2, /* 160 */ | |
2606 | 2, /* 168 */ | |
2607 | 2, /* 176 */ | |
2608 | 2, /* 184 */ | |
2609 | 2 /* 192 */ | |
2610 | }; | |
2611 | ||
acdfcd04 AK |
2612 | static inline int size_index_elem(size_t bytes) |
2613 | { | |
2614 | return (bytes - 1) / 8; | |
2615 | } | |
2616 | ||
81819f0f CL |
2617 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2618 | { | |
f1b26339 | 2619 | int index; |
81819f0f | 2620 | |
f1b26339 CL |
2621 | if (size <= 192) { |
2622 | if (!size) | |
2623 | return ZERO_SIZE_PTR; | |
81819f0f | 2624 | |
acdfcd04 | 2625 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 2626 | } else |
f1b26339 | 2627 | index = fls(size - 1); |
81819f0f CL |
2628 | |
2629 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2630 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 2631 | return kmalloc_dma_caches[index]; |
f1b26339 | 2632 | |
81819f0f | 2633 | #endif |
51df1142 | 2634 | return kmalloc_caches[index]; |
81819f0f CL |
2635 | } |
2636 | ||
2637 | void *__kmalloc(size_t size, gfp_t flags) | |
2638 | { | |
aadb4bc4 | 2639 | struct kmem_cache *s; |
5b882be4 | 2640 | void *ret; |
81819f0f | 2641 | |
ffadd4d0 | 2642 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2643 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2644 | |
2645 | s = get_slab(size, flags); | |
2646 | ||
2647 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2648 | return s; |
2649 | ||
2154a336 | 2650 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2651 | |
ca2b84cb | 2652 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2653 | |
2654 | return ret; | |
81819f0f CL |
2655 | } |
2656 | EXPORT_SYMBOL(__kmalloc); | |
2657 | ||
5d1f57e4 | 2658 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
2659 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2660 | { | |
b1eeab67 | 2661 | struct page *page; |
e4f7c0b4 | 2662 | void *ptr = NULL; |
f619cfe1 | 2663 | |
b1eeab67 VN |
2664 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2665 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 2666 | if (page) |
e4f7c0b4 CM |
2667 | ptr = page_address(page); |
2668 | ||
2669 | kmemleak_alloc(ptr, size, 1, flags); | |
2670 | return ptr; | |
f619cfe1 CL |
2671 | } |
2672 | ||
81819f0f CL |
2673 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
2674 | { | |
aadb4bc4 | 2675 | struct kmem_cache *s; |
5b882be4 | 2676 | void *ret; |
81819f0f | 2677 | |
057685cf | 2678 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2679 | ret = kmalloc_large_node(size, flags, node); |
2680 | ||
ca2b84cb EGM |
2681 | trace_kmalloc_node(_RET_IP_, ret, |
2682 | size, PAGE_SIZE << get_order(size), | |
2683 | flags, node); | |
5b882be4 EGM |
2684 | |
2685 | return ret; | |
2686 | } | |
aadb4bc4 CL |
2687 | |
2688 | s = get_slab(size, flags); | |
2689 | ||
2690 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2691 | return s; |
2692 | ||
5b882be4 EGM |
2693 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2694 | ||
ca2b84cb | 2695 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2696 | |
2697 | return ret; | |
81819f0f CL |
2698 | } |
2699 | EXPORT_SYMBOL(__kmalloc_node); | |
2700 | #endif | |
2701 | ||
2702 | size_t ksize(const void *object) | |
2703 | { | |
272c1d21 | 2704 | struct page *page; |
81819f0f CL |
2705 | struct kmem_cache *s; |
2706 | ||
ef8b4520 | 2707 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2708 | return 0; |
2709 | ||
294a80a8 | 2710 | page = virt_to_head_page(object); |
294a80a8 | 2711 | |
76994412 PE |
2712 | if (unlikely(!PageSlab(page))) { |
2713 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2714 | return PAGE_SIZE << compound_order(page); |
76994412 | 2715 | } |
81819f0f | 2716 | s = page->slab; |
81819f0f | 2717 | |
ae20bfda | 2718 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2719 | /* |
2720 | * Debugging requires use of the padding between object | |
2721 | * and whatever may come after it. | |
2722 | */ | |
2723 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2724 | return s->objsize; | |
2725 | ||
ae20bfda | 2726 | #endif |
81819f0f CL |
2727 | /* |
2728 | * If we have the need to store the freelist pointer | |
2729 | * back there or track user information then we can | |
2730 | * only use the space before that information. | |
2731 | */ | |
2732 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2733 | return s->inuse; | |
81819f0f CL |
2734 | /* |
2735 | * Else we can use all the padding etc for the allocation | |
2736 | */ | |
2737 | return s->size; | |
2738 | } | |
b1aabecd | 2739 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2740 | |
2741 | void kfree(const void *x) | |
2742 | { | |
81819f0f | 2743 | struct page *page; |
5bb983b0 | 2744 | void *object = (void *)x; |
81819f0f | 2745 | |
2121db74 PE |
2746 | trace_kfree(_RET_IP_, x); |
2747 | ||
2408c550 | 2748 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2749 | return; |
2750 | ||
b49af68f | 2751 | page = virt_to_head_page(x); |
aadb4bc4 | 2752 | if (unlikely(!PageSlab(page))) { |
0937502a | 2753 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 2754 | kmemleak_free(x); |
aadb4bc4 CL |
2755 | put_page(page); |
2756 | return; | |
2757 | } | |
ce71e27c | 2758 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2759 | } |
2760 | EXPORT_SYMBOL(kfree); | |
2761 | ||
2086d26a | 2762 | /* |
672bba3a CL |
2763 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2764 | * the remaining slabs by the number of items in use. The slabs with the | |
2765 | * most items in use come first. New allocations will then fill those up | |
2766 | * and thus they can be removed from the partial lists. | |
2767 | * | |
2768 | * The slabs with the least items are placed last. This results in them | |
2769 | * being allocated from last increasing the chance that the last objects | |
2770 | * are freed in them. | |
2086d26a CL |
2771 | */ |
2772 | int kmem_cache_shrink(struct kmem_cache *s) | |
2773 | { | |
2774 | int node; | |
2775 | int i; | |
2776 | struct kmem_cache_node *n; | |
2777 | struct page *page; | |
2778 | struct page *t; | |
205ab99d | 2779 | int objects = oo_objects(s->max); |
2086d26a | 2780 | struct list_head *slabs_by_inuse = |
834f3d11 | 2781 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2782 | unsigned long flags; |
2783 | ||
2784 | if (!slabs_by_inuse) | |
2785 | return -ENOMEM; | |
2786 | ||
2787 | flush_all(s); | |
f64dc58c | 2788 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2789 | n = get_node(s, node); |
2790 | ||
2791 | if (!n->nr_partial) | |
2792 | continue; | |
2793 | ||
834f3d11 | 2794 | for (i = 0; i < objects; i++) |
2086d26a CL |
2795 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2796 | ||
2797 | spin_lock_irqsave(&n->list_lock, flags); | |
2798 | ||
2799 | /* | |
672bba3a | 2800 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2801 | * |
672bba3a CL |
2802 | * Note that concurrent frees may occur while we hold the |
2803 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2804 | */ |
2805 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2806 | if (!page->inuse && slab_trylock(page)) { | |
2807 | /* | |
2808 | * Must hold slab lock here because slab_free | |
2809 | * may have freed the last object and be | |
2810 | * waiting to release the slab. | |
2811 | */ | |
62e346a8 | 2812 | __remove_partial(n, page); |
2086d26a CL |
2813 | slab_unlock(page); |
2814 | discard_slab(s, page); | |
2815 | } else { | |
fcda3d89 CL |
2816 | list_move(&page->lru, |
2817 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2818 | } |
2819 | } | |
2820 | ||
2086d26a | 2821 | /* |
672bba3a CL |
2822 | * Rebuild the partial list with the slabs filled up most |
2823 | * first and the least used slabs at the end. | |
2086d26a | 2824 | */ |
834f3d11 | 2825 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2826 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2827 | ||
2086d26a CL |
2828 | spin_unlock_irqrestore(&n->list_lock, flags); |
2829 | } | |
2830 | ||
2831 | kfree(slabs_by_inuse); | |
2832 | return 0; | |
2833 | } | |
2834 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2835 | ||
92a5bbc1 | 2836 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
2837 | static int slab_mem_going_offline_callback(void *arg) |
2838 | { | |
2839 | struct kmem_cache *s; | |
2840 | ||
2841 | down_read(&slub_lock); | |
2842 | list_for_each_entry(s, &slab_caches, list) | |
2843 | kmem_cache_shrink(s); | |
2844 | up_read(&slub_lock); | |
2845 | ||
2846 | return 0; | |
2847 | } | |
2848 | ||
2849 | static void slab_mem_offline_callback(void *arg) | |
2850 | { | |
2851 | struct kmem_cache_node *n; | |
2852 | struct kmem_cache *s; | |
2853 | struct memory_notify *marg = arg; | |
2854 | int offline_node; | |
2855 | ||
2856 | offline_node = marg->status_change_nid; | |
2857 | ||
2858 | /* | |
2859 | * If the node still has available memory. we need kmem_cache_node | |
2860 | * for it yet. | |
2861 | */ | |
2862 | if (offline_node < 0) | |
2863 | return; | |
2864 | ||
2865 | down_read(&slub_lock); | |
2866 | list_for_each_entry(s, &slab_caches, list) { | |
2867 | n = get_node(s, offline_node); | |
2868 | if (n) { | |
2869 | /* | |
2870 | * if n->nr_slabs > 0, slabs still exist on the node | |
2871 | * that is going down. We were unable to free them, | |
c9404c9c | 2872 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
2873 | * callback. So, we must fail. |
2874 | */ | |
0f389ec6 | 2875 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2876 | |
2877 | s->node[offline_node] = NULL; | |
8de66a0c | 2878 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
2879 | } |
2880 | } | |
2881 | up_read(&slub_lock); | |
2882 | } | |
2883 | ||
2884 | static int slab_mem_going_online_callback(void *arg) | |
2885 | { | |
2886 | struct kmem_cache_node *n; | |
2887 | struct kmem_cache *s; | |
2888 | struct memory_notify *marg = arg; | |
2889 | int nid = marg->status_change_nid; | |
2890 | int ret = 0; | |
2891 | ||
2892 | /* | |
2893 | * If the node's memory is already available, then kmem_cache_node is | |
2894 | * already created. Nothing to do. | |
2895 | */ | |
2896 | if (nid < 0) | |
2897 | return 0; | |
2898 | ||
2899 | /* | |
0121c619 | 2900 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
2901 | * allocate a kmem_cache_node structure in order to bring the node |
2902 | * online. | |
2903 | */ | |
2904 | down_read(&slub_lock); | |
2905 | list_for_each_entry(s, &slab_caches, list) { | |
2906 | /* | |
2907 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2908 | * since memory is not yet available from the node that | |
2909 | * is brought up. | |
2910 | */ | |
8de66a0c | 2911 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
2912 | if (!n) { |
2913 | ret = -ENOMEM; | |
2914 | goto out; | |
2915 | } | |
5595cffc | 2916 | init_kmem_cache_node(n, s); |
b9049e23 YG |
2917 | s->node[nid] = n; |
2918 | } | |
2919 | out: | |
2920 | up_read(&slub_lock); | |
2921 | return ret; | |
2922 | } | |
2923 | ||
2924 | static int slab_memory_callback(struct notifier_block *self, | |
2925 | unsigned long action, void *arg) | |
2926 | { | |
2927 | int ret = 0; | |
2928 | ||
2929 | switch (action) { | |
2930 | case MEM_GOING_ONLINE: | |
2931 | ret = slab_mem_going_online_callback(arg); | |
2932 | break; | |
2933 | case MEM_GOING_OFFLINE: | |
2934 | ret = slab_mem_going_offline_callback(arg); | |
2935 | break; | |
2936 | case MEM_OFFLINE: | |
2937 | case MEM_CANCEL_ONLINE: | |
2938 | slab_mem_offline_callback(arg); | |
2939 | break; | |
2940 | case MEM_ONLINE: | |
2941 | case MEM_CANCEL_OFFLINE: | |
2942 | break; | |
2943 | } | |
dc19f9db KH |
2944 | if (ret) |
2945 | ret = notifier_from_errno(ret); | |
2946 | else | |
2947 | ret = NOTIFY_OK; | |
b9049e23 YG |
2948 | return ret; |
2949 | } | |
2950 | ||
2951 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2952 | ||
81819f0f CL |
2953 | /******************************************************************** |
2954 | * Basic setup of slabs | |
2955 | *******************************************************************/ | |
2956 | ||
51df1142 CL |
2957 | /* |
2958 | * Used for early kmem_cache structures that were allocated using | |
2959 | * the page allocator | |
2960 | */ | |
2961 | ||
2962 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
2963 | { | |
2964 | int node; | |
2965 | ||
2966 | list_add(&s->list, &slab_caches); | |
2967 | s->refcount = -1; | |
2968 | ||
2969 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
2970 | struct kmem_cache_node *n = get_node(s, node); | |
2971 | struct page *p; | |
2972 | ||
2973 | if (n) { | |
2974 | list_for_each_entry(p, &n->partial, lru) | |
2975 | p->slab = s; | |
2976 | ||
2977 | #ifdef CONFIG_SLAB_DEBUG | |
2978 | list_for_each_entry(p, &n->full, lru) | |
2979 | p->slab = s; | |
2980 | #endif | |
2981 | } | |
2982 | } | |
2983 | } | |
2984 | ||
81819f0f CL |
2985 | void __init kmem_cache_init(void) |
2986 | { | |
2987 | int i; | |
4b356be0 | 2988 | int caches = 0; |
51df1142 CL |
2989 | struct kmem_cache *temp_kmem_cache; |
2990 | int order; | |
51df1142 CL |
2991 | struct kmem_cache *temp_kmem_cache_node; |
2992 | unsigned long kmalloc_size; | |
2993 | ||
2994 | kmem_size = offsetof(struct kmem_cache, node) + | |
2995 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
2996 | ||
2997 | /* Allocate two kmem_caches from the page allocator */ | |
2998 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
2999 | order = get_order(2 * kmalloc_size); | |
3000 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3001 | ||
81819f0f CL |
3002 | /* |
3003 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3004 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3005 | * kmem_cache_open for slab_state == DOWN. |
3006 | */ | |
51df1142 CL |
3007 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3008 | ||
3009 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3010 | sizeof(struct kmem_cache_node), | |
3011 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3012 | |
0c40ba4f | 3013 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3014 | |
3015 | /* Able to allocate the per node structures */ | |
3016 | slab_state = PARTIAL; | |
3017 | ||
51df1142 CL |
3018 | temp_kmem_cache = kmem_cache; |
3019 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3020 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3021 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3022 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3023 | |
51df1142 CL |
3024 | /* |
3025 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3026 | * kmem_cache_node is separately allocated so no need to | |
3027 | * update any list pointers. | |
3028 | */ | |
3029 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3030 | |
51df1142 CL |
3031 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3032 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3033 | ||
3034 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3035 | ||
3036 | caches++; | |
51df1142 CL |
3037 | kmem_cache_bootstrap_fixup(kmem_cache); |
3038 | caches++; | |
3039 | /* Free temporary boot structure */ | |
3040 | free_pages((unsigned long)temp_kmem_cache, order); | |
3041 | ||
3042 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3043 | |
3044 | /* | |
3045 | * Patch up the size_index table if we have strange large alignment | |
3046 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3047 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3048 | * |
3049 | * Largest permitted alignment is 256 bytes due to the way we | |
3050 | * handle the index determination for the smaller caches. | |
3051 | * | |
3052 | * Make sure that nothing crazy happens if someone starts tinkering | |
3053 | * around with ARCH_KMALLOC_MINALIGN | |
3054 | */ | |
3055 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3056 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3057 | ||
acdfcd04 AK |
3058 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3059 | int elem = size_index_elem(i); | |
3060 | if (elem >= ARRAY_SIZE(size_index)) | |
3061 | break; | |
3062 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3063 | } | |
f1b26339 | 3064 | |
acdfcd04 AK |
3065 | if (KMALLOC_MIN_SIZE == 64) { |
3066 | /* | |
3067 | * The 96 byte size cache is not used if the alignment | |
3068 | * is 64 byte. | |
3069 | */ | |
3070 | for (i = 64 + 8; i <= 96; i += 8) | |
3071 | size_index[size_index_elem(i)] = 7; | |
3072 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3073 | /* |
3074 | * The 192 byte sized cache is not used if the alignment | |
3075 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3076 | * instead. | |
3077 | */ | |
3078 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3079 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3080 | } |
3081 | ||
51df1142 CL |
3082 | /* Caches that are not of the two-to-the-power-of size */ |
3083 | if (KMALLOC_MIN_SIZE <= 32) { | |
3084 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3085 | caches++; | |
3086 | } | |
3087 | ||
3088 | if (KMALLOC_MIN_SIZE <= 64) { | |
3089 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3090 | caches++; | |
3091 | } | |
3092 | ||
3093 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3094 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3095 | caches++; | |
3096 | } | |
3097 | ||
81819f0f CL |
3098 | slab_state = UP; |
3099 | ||
3100 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3101 | if (KMALLOC_MIN_SIZE <= 32) { |
3102 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3103 | BUG_ON(!kmalloc_caches[1]->name); | |
3104 | } | |
3105 | ||
3106 | if (KMALLOC_MIN_SIZE <= 64) { | |
3107 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3108 | BUG_ON(!kmalloc_caches[2]->name); | |
3109 | } | |
3110 | ||
d7278bd7 CL |
3111 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3112 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3113 | ||
3114 | BUG_ON(!s); | |
51df1142 | 3115 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3116 | } |
81819f0f CL |
3117 | |
3118 | #ifdef CONFIG_SMP | |
3119 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3120 | #endif |
81819f0f | 3121 | |
55136592 | 3122 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3123 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3124 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3125 | |
51df1142 | 3126 | if (s && s->size) { |
55136592 CL |
3127 | char *name = kasprintf(GFP_NOWAIT, |
3128 | "dma-kmalloc-%d", s->objsize); | |
3129 | ||
3130 | BUG_ON(!name); | |
51df1142 CL |
3131 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3132 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3133 | } |
3134 | } | |
3135 | #endif | |
3adbefee IM |
3136 | printk(KERN_INFO |
3137 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3138 | " CPUs=%d, Nodes=%d\n", |
3139 | caches, cache_line_size(), | |
81819f0f CL |
3140 | slub_min_order, slub_max_order, slub_min_objects, |
3141 | nr_cpu_ids, nr_node_ids); | |
3142 | } | |
3143 | ||
7e85ee0c PE |
3144 | void __init kmem_cache_init_late(void) |
3145 | { | |
7e85ee0c PE |
3146 | } |
3147 | ||
81819f0f CL |
3148 | /* |
3149 | * Find a mergeable slab cache | |
3150 | */ | |
3151 | static int slab_unmergeable(struct kmem_cache *s) | |
3152 | { | |
3153 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3154 | return 1; | |
3155 | ||
c59def9f | 3156 | if (s->ctor) |
81819f0f CL |
3157 | return 1; |
3158 | ||
8ffa6875 CL |
3159 | /* |
3160 | * We may have set a slab to be unmergeable during bootstrap. | |
3161 | */ | |
3162 | if (s->refcount < 0) | |
3163 | return 1; | |
3164 | ||
81819f0f CL |
3165 | return 0; |
3166 | } | |
3167 | ||
3168 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3169 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3170 | void (*ctor)(void *)) |
81819f0f | 3171 | { |
5b95a4ac | 3172 | struct kmem_cache *s; |
81819f0f CL |
3173 | |
3174 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3175 | return NULL; | |
3176 | ||
c59def9f | 3177 | if (ctor) |
81819f0f CL |
3178 | return NULL; |
3179 | ||
3180 | size = ALIGN(size, sizeof(void *)); | |
3181 | align = calculate_alignment(flags, align, size); | |
3182 | size = ALIGN(size, align); | |
ba0268a8 | 3183 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3184 | |
5b95a4ac | 3185 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3186 | if (slab_unmergeable(s)) |
3187 | continue; | |
3188 | ||
3189 | if (size > s->size) | |
3190 | continue; | |
3191 | ||
ba0268a8 | 3192 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3193 | continue; |
3194 | /* | |
3195 | * Check if alignment is compatible. | |
3196 | * Courtesy of Adrian Drzewiecki | |
3197 | */ | |
06428780 | 3198 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3199 | continue; |
3200 | ||
3201 | if (s->size - size >= sizeof(void *)) | |
3202 | continue; | |
3203 | ||
3204 | return s; | |
3205 | } | |
3206 | return NULL; | |
3207 | } | |
3208 | ||
3209 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3210 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3211 | { |
3212 | struct kmem_cache *s; | |
84c1cf62 | 3213 | char *n; |
81819f0f | 3214 | |
fe1ff49d BH |
3215 | if (WARN_ON(!name)) |
3216 | return NULL; | |
3217 | ||
81819f0f | 3218 | down_write(&slub_lock); |
ba0268a8 | 3219 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3220 | if (s) { |
3221 | s->refcount++; | |
3222 | /* | |
3223 | * Adjust the object sizes so that we clear | |
3224 | * the complete object on kzalloc. | |
3225 | */ | |
3226 | s->objsize = max(s->objsize, (int)size); | |
3227 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3228 | |
7b8f3b66 | 3229 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3230 | s->refcount--; |
81819f0f | 3231 | goto err; |
7b8f3b66 | 3232 | } |
2bce6485 | 3233 | up_write(&slub_lock); |
a0e1d1be CL |
3234 | return s; |
3235 | } | |
6446faa2 | 3236 | |
84c1cf62 PE |
3237 | n = kstrdup(name, GFP_KERNEL); |
3238 | if (!n) | |
3239 | goto err; | |
3240 | ||
a0e1d1be CL |
3241 | s = kmalloc(kmem_size, GFP_KERNEL); |
3242 | if (s) { | |
84c1cf62 | 3243 | if (kmem_cache_open(s, n, |
c59def9f | 3244 | size, align, flags, ctor)) { |
81819f0f | 3245 | list_add(&s->list, &slab_caches); |
7b8f3b66 | 3246 | if (sysfs_slab_add(s)) { |
7b8f3b66 | 3247 | list_del(&s->list); |
84c1cf62 | 3248 | kfree(n); |
7b8f3b66 | 3249 | kfree(s); |
a0e1d1be | 3250 | goto err; |
7b8f3b66 | 3251 | } |
2bce6485 | 3252 | up_write(&slub_lock); |
a0e1d1be CL |
3253 | return s; |
3254 | } | |
84c1cf62 | 3255 | kfree(n); |
a0e1d1be | 3256 | kfree(s); |
81819f0f | 3257 | } |
68cee4f1 | 3258 | err: |
81819f0f | 3259 | up_write(&slub_lock); |
81819f0f | 3260 | |
81819f0f CL |
3261 | if (flags & SLAB_PANIC) |
3262 | panic("Cannot create slabcache %s\n", name); | |
3263 | else | |
3264 | s = NULL; | |
3265 | return s; | |
3266 | } | |
3267 | EXPORT_SYMBOL(kmem_cache_create); | |
3268 | ||
81819f0f | 3269 | #ifdef CONFIG_SMP |
81819f0f | 3270 | /* |
672bba3a CL |
3271 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3272 | * necessary. | |
81819f0f CL |
3273 | */ |
3274 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3275 | unsigned long action, void *hcpu) | |
3276 | { | |
3277 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3278 | struct kmem_cache *s; |
3279 | unsigned long flags; | |
81819f0f CL |
3280 | |
3281 | switch (action) { | |
3282 | case CPU_UP_CANCELED: | |
8bb78442 | 3283 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3284 | case CPU_DEAD: |
8bb78442 | 3285 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3286 | down_read(&slub_lock); |
3287 | list_for_each_entry(s, &slab_caches, list) { | |
3288 | local_irq_save(flags); | |
3289 | __flush_cpu_slab(s, cpu); | |
3290 | local_irq_restore(flags); | |
3291 | } | |
3292 | up_read(&slub_lock); | |
81819f0f CL |
3293 | break; |
3294 | default: | |
3295 | break; | |
3296 | } | |
3297 | return NOTIFY_OK; | |
3298 | } | |
3299 | ||
06428780 | 3300 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3301 | .notifier_call = slab_cpuup_callback |
06428780 | 3302 | }; |
81819f0f CL |
3303 | |
3304 | #endif | |
3305 | ||
ce71e27c | 3306 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3307 | { |
aadb4bc4 | 3308 | struct kmem_cache *s; |
94b528d0 | 3309 | void *ret; |
aadb4bc4 | 3310 | |
ffadd4d0 | 3311 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3312 | return kmalloc_large(size, gfpflags); |
3313 | ||
aadb4bc4 | 3314 | s = get_slab(size, gfpflags); |
81819f0f | 3315 | |
2408c550 | 3316 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3317 | return s; |
81819f0f | 3318 | |
2154a336 | 3319 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 EGM |
3320 | |
3321 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3322 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3323 | |
3324 | return ret; | |
81819f0f CL |
3325 | } |
3326 | ||
5d1f57e4 | 3327 | #ifdef CONFIG_NUMA |
81819f0f | 3328 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3329 | int node, unsigned long caller) |
81819f0f | 3330 | { |
aadb4bc4 | 3331 | struct kmem_cache *s; |
94b528d0 | 3332 | void *ret; |
aadb4bc4 | 3333 | |
d3e14aa3 XF |
3334 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3335 | ret = kmalloc_large_node(size, gfpflags, node); | |
3336 | ||
3337 | trace_kmalloc_node(caller, ret, | |
3338 | size, PAGE_SIZE << get_order(size), | |
3339 | gfpflags, node); | |
3340 | ||
3341 | return ret; | |
3342 | } | |
eada35ef | 3343 | |
aadb4bc4 | 3344 | s = get_slab(size, gfpflags); |
81819f0f | 3345 | |
2408c550 | 3346 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3347 | return s; |
81819f0f | 3348 | |
94b528d0 EGM |
3349 | ret = slab_alloc(s, gfpflags, node, caller); |
3350 | ||
3351 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3352 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3353 | |
3354 | return ret; | |
81819f0f | 3355 | } |
5d1f57e4 | 3356 | #endif |
81819f0f | 3357 | |
ab4d5ed5 | 3358 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3359 | static int count_inuse(struct page *page) |
3360 | { | |
3361 | return page->inuse; | |
3362 | } | |
3363 | ||
3364 | static int count_total(struct page *page) | |
3365 | { | |
3366 | return page->objects; | |
3367 | } | |
ab4d5ed5 | 3368 | #endif |
205ab99d | 3369 | |
ab4d5ed5 | 3370 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3371 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3372 | unsigned long *map) | |
53e15af0 CL |
3373 | { |
3374 | void *p; | |
a973e9dd | 3375 | void *addr = page_address(page); |
53e15af0 CL |
3376 | |
3377 | if (!check_slab(s, page) || | |
3378 | !on_freelist(s, page, NULL)) | |
3379 | return 0; | |
3380 | ||
3381 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3382 | bitmap_zero(map, page->objects); |
53e15af0 | 3383 | |
7656c72b CL |
3384 | for_each_free_object(p, s, page->freelist) { |
3385 | set_bit(slab_index(p, s, addr), map); | |
37d57443 | 3386 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) |
53e15af0 CL |
3387 | return 0; |
3388 | } | |
3389 | ||
224a88be | 3390 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3391 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3392 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3393 | return 0; |
3394 | return 1; | |
3395 | } | |
3396 | ||
434e245d CL |
3397 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3398 | unsigned long *map) | |
53e15af0 CL |
3399 | { |
3400 | if (slab_trylock(page)) { | |
434e245d | 3401 | validate_slab(s, page, map); |
53e15af0 CL |
3402 | slab_unlock(page); |
3403 | } else | |
3404 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3405 | s->name, page); | |
53e15af0 CL |
3406 | } |
3407 | ||
434e245d CL |
3408 | static int validate_slab_node(struct kmem_cache *s, |
3409 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3410 | { |
3411 | unsigned long count = 0; | |
3412 | struct page *page; | |
3413 | unsigned long flags; | |
3414 | ||
3415 | spin_lock_irqsave(&n->list_lock, flags); | |
3416 | ||
3417 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3418 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3419 | count++; |
3420 | } | |
3421 | if (count != n->nr_partial) | |
3422 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3423 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3424 | ||
3425 | if (!(s->flags & SLAB_STORE_USER)) | |
3426 | goto out; | |
3427 | ||
3428 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3429 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3430 | count++; |
3431 | } | |
3432 | if (count != atomic_long_read(&n->nr_slabs)) | |
3433 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3434 | "counter=%ld\n", s->name, count, | |
3435 | atomic_long_read(&n->nr_slabs)); | |
3436 | ||
3437 | out: | |
3438 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3439 | return count; | |
3440 | } | |
3441 | ||
434e245d | 3442 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3443 | { |
3444 | int node; | |
3445 | unsigned long count = 0; | |
205ab99d | 3446 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3447 | sizeof(unsigned long), GFP_KERNEL); |
3448 | ||
3449 | if (!map) | |
3450 | return -ENOMEM; | |
53e15af0 CL |
3451 | |
3452 | flush_all(s); | |
f64dc58c | 3453 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3454 | struct kmem_cache_node *n = get_node(s, node); |
3455 | ||
434e245d | 3456 | count += validate_slab_node(s, n, map); |
53e15af0 | 3457 | } |
434e245d | 3458 | kfree(map); |
53e15af0 CL |
3459 | return count; |
3460 | } | |
88a420e4 | 3461 | /* |
672bba3a | 3462 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3463 | * and freed. |
3464 | */ | |
3465 | ||
3466 | struct location { | |
3467 | unsigned long count; | |
ce71e27c | 3468 | unsigned long addr; |
45edfa58 CL |
3469 | long long sum_time; |
3470 | long min_time; | |
3471 | long max_time; | |
3472 | long min_pid; | |
3473 | long max_pid; | |
174596a0 | 3474 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3475 | nodemask_t nodes; |
88a420e4 CL |
3476 | }; |
3477 | ||
3478 | struct loc_track { | |
3479 | unsigned long max; | |
3480 | unsigned long count; | |
3481 | struct location *loc; | |
3482 | }; | |
3483 | ||
3484 | static void free_loc_track(struct loc_track *t) | |
3485 | { | |
3486 | if (t->max) | |
3487 | free_pages((unsigned long)t->loc, | |
3488 | get_order(sizeof(struct location) * t->max)); | |
3489 | } | |
3490 | ||
68dff6a9 | 3491 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3492 | { |
3493 | struct location *l; | |
3494 | int order; | |
3495 | ||
88a420e4 CL |
3496 | order = get_order(sizeof(struct location) * max); |
3497 | ||
68dff6a9 | 3498 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3499 | if (!l) |
3500 | return 0; | |
3501 | ||
3502 | if (t->count) { | |
3503 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3504 | free_loc_track(t); | |
3505 | } | |
3506 | t->max = max; | |
3507 | t->loc = l; | |
3508 | return 1; | |
3509 | } | |
3510 | ||
3511 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3512 | const struct track *track) |
88a420e4 CL |
3513 | { |
3514 | long start, end, pos; | |
3515 | struct location *l; | |
ce71e27c | 3516 | unsigned long caddr; |
45edfa58 | 3517 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3518 | |
3519 | start = -1; | |
3520 | end = t->count; | |
3521 | ||
3522 | for ( ; ; ) { | |
3523 | pos = start + (end - start + 1) / 2; | |
3524 | ||
3525 | /* | |
3526 | * There is nothing at "end". If we end up there | |
3527 | * we need to add something to before end. | |
3528 | */ | |
3529 | if (pos == end) | |
3530 | break; | |
3531 | ||
3532 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3533 | if (track->addr == caddr) { |
3534 | ||
3535 | l = &t->loc[pos]; | |
3536 | l->count++; | |
3537 | if (track->when) { | |
3538 | l->sum_time += age; | |
3539 | if (age < l->min_time) | |
3540 | l->min_time = age; | |
3541 | if (age > l->max_time) | |
3542 | l->max_time = age; | |
3543 | ||
3544 | if (track->pid < l->min_pid) | |
3545 | l->min_pid = track->pid; | |
3546 | if (track->pid > l->max_pid) | |
3547 | l->max_pid = track->pid; | |
3548 | ||
174596a0 RR |
3549 | cpumask_set_cpu(track->cpu, |
3550 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3551 | } |
3552 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3553 | return 1; |
3554 | } | |
3555 | ||
45edfa58 | 3556 | if (track->addr < caddr) |
88a420e4 CL |
3557 | end = pos; |
3558 | else | |
3559 | start = pos; | |
3560 | } | |
3561 | ||
3562 | /* | |
672bba3a | 3563 | * Not found. Insert new tracking element. |
88a420e4 | 3564 | */ |
68dff6a9 | 3565 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3566 | return 0; |
3567 | ||
3568 | l = t->loc + pos; | |
3569 | if (pos < t->count) | |
3570 | memmove(l + 1, l, | |
3571 | (t->count - pos) * sizeof(struct location)); | |
3572 | t->count++; | |
3573 | l->count = 1; | |
45edfa58 CL |
3574 | l->addr = track->addr; |
3575 | l->sum_time = age; | |
3576 | l->min_time = age; | |
3577 | l->max_time = age; | |
3578 | l->min_pid = track->pid; | |
3579 | l->max_pid = track->pid; | |
174596a0 RR |
3580 | cpumask_clear(to_cpumask(l->cpus)); |
3581 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3582 | nodes_clear(l->nodes); |
3583 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3584 | return 1; |
3585 | } | |
3586 | ||
3587 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 3588 | struct page *page, enum track_item alloc, |
a5dd5c11 | 3589 | unsigned long *map) |
88a420e4 | 3590 | { |
a973e9dd | 3591 | void *addr = page_address(page); |
88a420e4 CL |
3592 | void *p; |
3593 | ||
39b26464 | 3594 | bitmap_zero(map, page->objects); |
7656c72b CL |
3595 | for_each_free_object(p, s, page->freelist) |
3596 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3597 | |
224a88be | 3598 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3599 | if (!test_bit(slab_index(p, s, addr), map)) |
3600 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3601 | } |
3602 | ||
3603 | static int list_locations(struct kmem_cache *s, char *buf, | |
3604 | enum track_item alloc) | |
3605 | { | |
e374d483 | 3606 | int len = 0; |
88a420e4 | 3607 | unsigned long i; |
68dff6a9 | 3608 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 3609 | int node; |
bbd7d57b ED |
3610 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
3611 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 3612 | |
bbd7d57b ED |
3613 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
3614 | GFP_TEMPORARY)) { | |
3615 | kfree(map); | |
68dff6a9 | 3616 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 3617 | } |
88a420e4 CL |
3618 | /* Push back cpu slabs */ |
3619 | flush_all(s); | |
3620 | ||
f64dc58c | 3621 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3622 | struct kmem_cache_node *n = get_node(s, node); |
3623 | unsigned long flags; | |
3624 | struct page *page; | |
3625 | ||
9e86943b | 3626 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3627 | continue; |
3628 | ||
3629 | spin_lock_irqsave(&n->list_lock, flags); | |
3630 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 3631 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 3632 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 3633 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
3634 | spin_unlock_irqrestore(&n->list_lock, flags); |
3635 | } | |
3636 | ||
3637 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3638 | struct location *l = &t.loc[i]; |
88a420e4 | 3639 | |
9c246247 | 3640 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3641 | break; |
e374d483 | 3642 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3643 | |
3644 | if (l->addr) | |
62c70bce | 3645 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 3646 | else |
e374d483 | 3647 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3648 | |
3649 | if (l->sum_time != l->min_time) { | |
e374d483 | 3650 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3651 | l->min_time, |
3652 | (long)div_u64(l->sum_time, l->count), | |
3653 | l->max_time); | |
45edfa58 | 3654 | } else |
e374d483 | 3655 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3656 | l->min_time); |
3657 | ||
3658 | if (l->min_pid != l->max_pid) | |
e374d483 | 3659 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3660 | l->min_pid, l->max_pid); |
3661 | else | |
e374d483 | 3662 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3663 | l->min_pid); |
3664 | ||
174596a0 RR |
3665 | if (num_online_cpus() > 1 && |
3666 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3667 | len < PAGE_SIZE - 60) { |
3668 | len += sprintf(buf + len, " cpus="); | |
3669 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3670 | to_cpumask(l->cpus)); |
45edfa58 CL |
3671 | } |
3672 | ||
62bc62a8 | 3673 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3674 | len < PAGE_SIZE - 60) { |
3675 | len += sprintf(buf + len, " nodes="); | |
3676 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3677 | l->nodes); |
3678 | } | |
3679 | ||
e374d483 | 3680 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3681 | } |
3682 | ||
3683 | free_loc_track(&t); | |
bbd7d57b | 3684 | kfree(map); |
88a420e4 | 3685 | if (!t.count) |
e374d483 HH |
3686 | len += sprintf(buf, "No data\n"); |
3687 | return len; | |
88a420e4 | 3688 | } |
ab4d5ed5 | 3689 | #endif |
88a420e4 | 3690 | |
a5a84755 CL |
3691 | #ifdef SLUB_RESILIENCY_TEST |
3692 | static void resiliency_test(void) | |
3693 | { | |
3694 | u8 *p; | |
3695 | ||
3696 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
3697 | ||
3698 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3699 | printk(KERN_ERR "-----------------------\n"); | |
3700 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3701 | ||
3702 | p = kzalloc(16, GFP_KERNEL); | |
3703 | p[16] = 0x12; | |
3704 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3705 | " 0x12->0x%p\n\n", p + 16); | |
3706 | ||
3707 | validate_slab_cache(kmalloc_caches[4]); | |
3708 | ||
3709 | /* Hmmm... The next two are dangerous */ | |
3710 | p = kzalloc(32, GFP_KERNEL); | |
3711 | p[32 + sizeof(void *)] = 0x34; | |
3712 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3713 | " 0x34 -> -0x%p\n", p); | |
3714 | printk(KERN_ERR | |
3715 | "If allocated object is overwritten then not detectable\n\n"); | |
3716 | ||
3717 | validate_slab_cache(kmalloc_caches[5]); | |
3718 | p = kzalloc(64, GFP_KERNEL); | |
3719 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3720 | *p = 0x56; | |
3721 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3722 | p); | |
3723 | printk(KERN_ERR | |
3724 | "If allocated object is overwritten then not detectable\n\n"); | |
3725 | validate_slab_cache(kmalloc_caches[6]); | |
3726 | ||
3727 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3728 | p = kzalloc(128, GFP_KERNEL); | |
3729 | kfree(p); | |
3730 | *p = 0x78; | |
3731 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3732 | validate_slab_cache(kmalloc_caches[7]); | |
3733 | ||
3734 | p = kzalloc(256, GFP_KERNEL); | |
3735 | kfree(p); | |
3736 | p[50] = 0x9a; | |
3737 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
3738 | p); | |
3739 | validate_slab_cache(kmalloc_caches[8]); | |
3740 | ||
3741 | p = kzalloc(512, GFP_KERNEL); | |
3742 | kfree(p); | |
3743 | p[512] = 0xab; | |
3744 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3745 | validate_slab_cache(kmalloc_caches[9]); | |
3746 | } | |
3747 | #else | |
3748 | #ifdef CONFIG_SYSFS | |
3749 | static void resiliency_test(void) {}; | |
3750 | #endif | |
3751 | #endif | |
3752 | ||
ab4d5ed5 | 3753 | #ifdef CONFIG_SYSFS |
81819f0f | 3754 | enum slab_stat_type { |
205ab99d CL |
3755 | SL_ALL, /* All slabs */ |
3756 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3757 | SL_CPU, /* Only slabs used for cpu caches */ | |
3758 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3759 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3760 | }; |
3761 | ||
205ab99d | 3762 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3763 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3764 | #define SO_CPU (1 << SL_CPU) | |
3765 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3766 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3767 | |
62e5c4b4 CG |
3768 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3769 | char *buf, unsigned long flags) | |
81819f0f CL |
3770 | { |
3771 | unsigned long total = 0; | |
81819f0f CL |
3772 | int node; |
3773 | int x; | |
3774 | unsigned long *nodes; | |
3775 | unsigned long *per_cpu; | |
3776 | ||
3777 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3778 | if (!nodes) |
3779 | return -ENOMEM; | |
81819f0f CL |
3780 | per_cpu = nodes + nr_node_ids; |
3781 | ||
205ab99d CL |
3782 | if (flags & SO_CPU) { |
3783 | int cpu; | |
81819f0f | 3784 | |
205ab99d | 3785 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 3786 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
dfb4f096 | 3787 | |
205ab99d CL |
3788 | if (!c || c->node < 0) |
3789 | continue; | |
3790 | ||
3791 | if (c->page) { | |
3792 | if (flags & SO_TOTAL) | |
3793 | x = c->page->objects; | |
3794 | else if (flags & SO_OBJECTS) | |
3795 | x = c->page->inuse; | |
81819f0f CL |
3796 | else |
3797 | x = 1; | |
205ab99d | 3798 | |
81819f0f | 3799 | total += x; |
205ab99d | 3800 | nodes[c->node] += x; |
81819f0f | 3801 | } |
205ab99d | 3802 | per_cpu[c->node]++; |
81819f0f CL |
3803 | } |
3804 | } | |
3805 | ||
04d94879 | 3806 | lock_memory_hotplug(); |
ab4d5ed5 | 3807 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
3808 | if (flags & SO_ALL) { |
3809 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3810 | struct kmem_cache_node *n = get_node(s, node); | |
3811 | ||
3812 | if (flags & SO_TOTAL) | |
3813 | x = atomic_long_read(&n->total_objects); | |
3814 | else if (flags & SO_OBJECTS) | |
3815 | x = atomic_long_read(&n->total_objects) - | |
3816 | count_partial(n, count_free); | |
81819f0f | 3817 | |
81819f0f | 3818 | else |
205ab99d | 3819 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3820 | total += x; |
3821 | nodes[node] += x; | |
3822 | } | |
3823 | ||
ab4d5ed5 CL |
3824 | } else |
3825 | #endif | |
3826 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
3827 | for_each_node_state(node, N_NORMAL_MEMORY) { |
3828 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3829 | |
205ab99d CL |
3830 | if (flags & SO_TOTAL) |
3831 | x = count_partial(n, count_total); | |
3832 | else if (flags & SO_OBJECTS) | |
3833 | x = count_partial(n, count_inuse); | |
81819f0f | 3834 | else |
205ab99d | 3835 | x = n->nr_partial; |
81819f0f CL |
3836 | total += x; |
3837 | nodes[node] += x; | |
3838 | } | |
3839 | } | |
81819f0f CL |
3840 | x = sprintf(buf, "%lu", total); |
3841 | #ifdef CONFIG_NUMA | |
f64dc58c | 3842 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3843 | if (nodes[node]) |
3844 | x += sprintf(buf + x, " N%d=%lu", | |
3845 | node, nodes[node]); | |
3846 | #endif | |
04d94879 | 3847 | unlock_memory_hotplug(); |
81819f0f CL |
3848 | kfree(nodes); |
3849 | return x + sprintf(buf + x, "\n"); | |
3850 | } | |
3851 | ||
ab4d5ed5 | 3852 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3853 | static int any_slab_objects(struct kmem_cache *s) |
3854 | { | |
3855 | int node; | |
81819f0f | 3856 | |
dfb4f096 | 3857 | for_each_online_node(node) { |
81819f0f CL |
3858 | struct kmem_cache_node *n = get_node(s, node); |
3859 | ||
dfb4f096 CL |
3860 | if (!n) |
3861 | continue; | |
3862 | ||
4ea33e2d | 3863 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3864 | return 1; |
3865 | } | |
3866 | return 0; | |
3867 | } | |
ab4d5ed5 | 3868 | #endif |
81819f0f CL |
3869 | |
3870 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3871 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3872 | ||
3873 | struct slab_attribute { | |
3874 | struct attribute attr; | |
3875 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3876 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3877 | }; | |
3878 | ||
3879 | #define SLAB_ATTR_RO(_name) \ | |
3880 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3881 | ||
3882 | #define SLAB_ATTR(_name) \ | |
3883 | static struct slab_attribute _name##_attr = \ | |
3884 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3885 | ||
81819f0f CL |
3886 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3887 | { | |
3888 | return sprintf(buf, "%d\n", s->size); | |
3889 | } | |
3890 | SLAB_ATTR_RO(slab_size); | |
3891 | ||
3892 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3893 | { | |
3894 | return sprintf(buf, "%d\n", s->align); | |
3895 | } | |
3896 | SLAB_ATTR_RO(align); | |
3897 | ||
3898 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3899 | { | |
3900 | return sprintf(buf, "%d\n", s->objsize); | |
3901 | } | |
3902 | SLAB_ATTR_RO(object_size); | |
3903 | ||
3904 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3905 | { | |
834f3d11 | 3906 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3907 | } |
3908 | SLAB_ATTR_RO(objs_per_slab); | |
3909 | ||
06b285dc CL |
3910 | static ssize_t order_store(struct kmem_cache *s, |
3911 | const char *buf, size_t length) | |
3912 | { | |
0121c619 CL |
3913 | unsigned long order; |
3914 | int err; | |
3915 | ||
3916 | err = strict_strtoul(buf, 10, &order); | |
3917 | if (err) | |
3918 | return err; | |
06b285dc CL |
3919 | |
3920 | if (order > slub_max_order || order < slub_min_order) | |
3921 | return -EINVAL; | |
3922 | ||
3923 | calculate_sizes(s, order); | |
3924 | return length; | |
3925 | } | |
3926 | ||
81819f0f CL |
3927 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
3928 | { | |
834f3d11 | 3929 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 3930 | } |
06b285dc | 3931 | SLAB_ATTR(order); |
81819f0f | 3932 | |
73d342b1 DR |
3933 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
3934 | { | |
3935 | return sprintf(buf, "%lu\n", s->min_partial); | |
3936 | } | |
3937 | ||
3938 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
3939 | size_t length) | |
3940 | { | |
3941 | unsigned long min; | |
3942 | int err; | |
3943 | ||
3944 | err = strict_strtoul(buf, 10, &min); | |
3945 | if (err) | |
3946 | return err; | |
3947 | ||
c0bdb232 | 3948 | set_min_partial(s, min); |
73d342b1 DR |
3949 | return length; |
3950 | } | |
3951 | SLAB_ATTR(min_partial); | |
3952 | ||
81819f0f CL |
3953 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
3954 | { | |
62c70bce JP |
3955 | if (!s->ctor) |
3956 | return 0; | |
3957 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
3958 | } |
3959 | SLAB_ATTR_RO(ctor); | |
3960 | ||
81819f0f CL |
3961 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3962 | { | |
3963 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3964 | } | |
3965 | SLAB_ATTR_RO(aliases); | |
3966 | ||
81819f0f CL |
3967 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
3968 | { | |
d9acf4b7 | 3969 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
3970 | } |
3971 | SLAB_ATTR_RO(partial); | |
3972 | ||
3973 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3974 | { | |
d9acf4b7 | 3975 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
3976 | } |
3977 | SLAB_ATTR_RO(cpu_slabs); | |
3978 | ||
3979 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3980 | { | |
205ab99d | 3981 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
3982 | } |
3983 | SLAB_ATTR_RO(objects); | |
3984 | ||
205ab99d CL |
3985 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
3986 | { | |
3987 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
3988 | } | |
3989 | SLAB_ATTR_RO(objects_partial); | |
3990 | ||
a5a84755 CL |
3991 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
3992 | { | |
3993 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3994 | } | |
3995 | ||
3996 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3997 | const char *buf, size_t length) | |
3998 | { | |
3999 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4000 | if (buf[0] == '1') | |
4001 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4002 | return length; | |
4003 | } | |
4004 | SLAB_ATTR(reclaim_account); | |
4005 | ||
4006 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4007 | { | |
4008 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4009 | } | |
4010 | SLAB_ATTR_RO(hwcache_align); | |
4011 | ||
4012 | #ifdef CONFIG_ZONE_DMA | |
4013 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4014 | { | |
4015 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4016 | } | |
4017 | SLAB_ATTR_RO(cache_dma); | |
4018 | #endif | |
4019 | ||
4020 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4021 | { | |
4022 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4023 | } | |
4024 | SLAB_ATTR_RO(destroy_by_rcu); | |
4025 | ||
ab9a0f19 LJ |
4026 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4027 | { | |
4028 | return sprintf(buf, "%d\n", s->reserved); | |
4029 | } | |
4030 | SLAB_ATTR_RO(reserved); | |
4031 | ||
ab4d5ed5 | 4032 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4033 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4034 | { | |
4035 | return show_slab_objects(s, buf, SO_ALL); | |
4036 | } | |
4037 | SLAB_ATTR_RO(slabs); | |
4038 | ||
205ab99d CL |
4039 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4040 | { | |
4041 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4042 | } | |
4043 | SLAB_ATTR_RO(total_objects); | |
4044 | ||
81819f0f CL |
4045 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4046 | { | |
4047 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4048 | } | |
4049 | ||
4050 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4051 | const char *buf, size_t length) | |
4052 | { | |
4053 | s->flags &= ~SLAB_DEBUG_FREE; | |
4054 | if (buf[0] == '1') | |
4055 | s->flags |= SLAB_DEBUG_FREE; | |
4056 | return length; | |
4057 | } | |
4058 | SLAB_ATTR(sanity_checks); | |
4059 | ||
4060 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4061 | { | |
4062 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4063 | } | |
4064 | ||
4065 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4066 | size_t length) | |
4067 | { | |
4068 | s->flags &= ~SLAB_TRACE; | |
4069 | if (buf[0] == '1') | |
4070 | s->flags |= SLAB_TRACE; | |
4071 | return length; | |
4072 | } | |
4073 | SLAB_ATTR(trace); | |
4074 | ||
81819f0f CL |
4075 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4076 | { | |
4077 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4078 | } | |
4079 | ||
4080 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4081 | const char *buf, size_t length) | |
4082 | { | |
4083 | if (any_slab_objects(s)) | |
4084 | return -EBUSY; | |
4085 | ||
4086 | s->flags &= ~SLAB_RED_ZONE; | |
4087 | if (buf[0] == '1') | |
4088 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4089 | calculate_sizes(s, -1); |
81819f0f CL |
4090 | return length; |
4091 | } | |
4092 | SLAB_ATTR(red_zone); | |
4093 | ||
4094 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4095 | { | |
4096 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4097 | } | |
4098 | ||
4099 | static ssize_t poison_store(struct kmem_cache *s, | |
4100 | const char *buf, size_t length) | |
4101 | { | |
4102 | if (any_slab_objects(s)) | |
4103 | return -EBUSY; | |
4104 | ||
4105 | s->flags &= ~SLAB_POISON; | |
4106 | if (buf[0] == '1') | |
4107 | s->flags |= SLAB_POISON; | |
06b285dc | 4108 | calculate_sizes(s, -1); |
81819f0f CL |
4109 | return length; |
4110 | } | |
4111 | SLAB_ATTR(poison); | |
4112 | ||
4113 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4114 | { | |
4115 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4116 | } | |
4117 | ||
4118 | static ssize_t store_user_store(struct kmem_cache *s, | |
4119 | const char *buf, size_t length) | |
4120 | { | |
4121 | if (any_slab_objects(s)) | |
4122 | return -EBUSY; | |
4123 | ||
4124 | s->flags &= ~SLAB_STORE_USER; | |
4125 | if (buf[0] == '1') | |
4126 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4127 | calculate_sizes(s, -1); |
81819f0f CL |
4128 | return length; |
4129 | } | |
4130 | SLAB_ATTR(store_user); | |
4131 | ||
53e15af0 CL |
4132 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4133 | { | |
4134 | return 0; | |
4135 | } | |
4136 | ||
4137 | static ssize_t validate_store(struct kmem_cache *s, | |
4138 | const char *buf, size_t length) | |
4139 | { | |
434e245d CL |
4140 | int ret = -EINVAL; |
4141 | ||
4142 | if (buf[0] == '1') { | |
4143 | ret = validate_slab_cache(s); | |
4144 | if (ret >= 0) | |
4145 | ret = length; | |
4146 | } | |
4147 | return ret; | |
53e15af0 CL |
4148 | } |
4149 | SLAB_ATTR(validate); | |
a5a84755 CL |
4150 | |
4151 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4152 | { | |
4153 | if (!(s->flags & SLAB_STORE_USER)) | |
4154 | return -ENOSYS; | |
4155 | return list_locations(s, buf, TRACK_ALLOC); | |
4156 | } | |
4157 | SLAB_ATTR_RO(alloc_calls); | |
4158 | ||
4159 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4160 | { | |
4161 | if (!(s->flags & SLAB_STORE_USER)) | |
4162 | return -ENOSYS; | |
4163 | return list_locations(s, buf, TRACK_FREE); | |
4164 | } | |
4165 | SLAB_ATTR_RO(free_calls); | |
4166 | #endif /* CONFIG_SLUB_DEBUG */ | |
4167 | ||
4168 | #ifdef CONFIG_FAILSLAB | |
4169 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4170 | { | |
4171 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4172 | } | |
4173 | ||
4174 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4175 | size_t length) | |
4176 | { | |
4177 | s->flags &= ~SLAB_FAILSLAB; | |
4178 | if (buf[0] == '1') | |
4179 | s->flags |= SLAB_FAILSLAB; | |
4180 | return length; | |
4181 | } | |
4182 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4183 | #endif |
53e15af0 | 4184 | |
2086d26a CL |
4185 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4186 | { | |
4187 | return 0; | |
4188 | } | |
4189 | ||
4190 | static ssize_t shrink_store(struct kmem_cache *s, | |
4191 | const char *buf, size_t length) | |
4192 | { | |
4193 | if (buf[0] == '1') { | |
4194 | int rc = kmem_cache_shrink(s); | |
4195 | ||
4196 | if (rc) | |
4197 | return rc; | |
4198 | } else | |
4199 | return -EINVAL; | |
4200 | return length; | |
4201 | } | |
4202 | SLAB_ATTR(shrink); | |
4203 | ||
81819f0f | 4204 | #ifdef CONFIG_NUMA |
9824601e | 4205 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4206 | { |
9824601e | 4207 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4208 | } |
4209 | ||
9824601e | 4210 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4211 | const char *buf, size_t length) |
4212 | { | |
0121c619 CL |
4213 | unsigned long ratio; |
4214 | int err; | |
4215 | ||
4216 | err = strict_strtoul(buf, 10, &ratio); | |
4217 | if (err) | |
4218 | return err; | |
4219 | ||
e2cb96b7 | 4220 | if (ratio <= 100) |
0121c619 | 4221 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4222 | |
81819f0f CL |
4223 | return length; |
4224 | } | |
9824601e | 4225 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4226 | #endif |
4227 | ||
8ff12cfc | 4228 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4229 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4230 | { | |
4231 | unsigned long sum = 0; | |
4232 | int cpu; | |
4233 | int len; | |
4234 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4235 | ||
4236 | if (!data) | |
4237 | return -ENOMEM; | |
4238 | ||
4239 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4240 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4241 | |
4242 | data[cpu] = x; | |
4243 | sum += x; | |
4244 | } | |
4245 | ||
4246 | len = sprintf(buf, "%lu", sum); | |
4247 | ||
50ef37b9 | 4248 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4249 | for_each_online_cpu(cpu) { |
4250 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4251 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4252 | } |
50ef37b9 | 4253 | #endif |
8ff12cfc CL |
4254 | kfree(data); |
4255 | return len + sprintf(buf + len, "\n"); | |
4256 | } | |
4257 | ||
78eb00cc DR |
4258 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4259 | { | |
4260 | int cpu; | |
4261 | ||
4262 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4263 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4264 | } |
4265 | ||
8ff12cfc CL |
4266 | #define STAT_ATTR(si, text) \ |
4267 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4268 | { \ | |
4269 | return show_stat(s, buf, si); \ | |
4270 | } \ | |
78eb00cc DR |
4271 | static ssize_t text##_store(struct kmem_cache *s, \ |
4272 | const char *buf, size_t length) \ | |
4273 | { \ | |
4274 | if (buf[0] != '0') \ | |
4275 | return -EINVAL; \ | |
4276 | clear_stat(s, si); \ | |
4277 | return length; \ | |
4278 | } \ | |
4279 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4280 | |
4281 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4282 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4283 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4284 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4285 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4286 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4287 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4288 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4289 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4290 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4291 | STAT_ATTR(FREE_SLAB, free_slab); | |
4292 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4293 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4294 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4295 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4296 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4297 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4298 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4299 | #endif |
4300 | ||
06428780 | 4301 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4302 | &slab_size_attr.attr, |
4303 | &object_size_attr.attr, | |
4304 | &objs_per_slab_attr.attr, | |
4305 | &order_attr.attr, | |
73d342b1 | 4306 | &min_partial_attr.attr, |
81819f0f | 4307 | &objects_attr.attr, |
205ab99d | 4308 | &objects_partial_attr.attr, |
81819f0f CL |
4309 | &partial_attr.attr, |
4310 | &cpu_slabs_attr.attr, | |
4311 | &ctor_attr.attr, | |
81819f0f CL |
4312 | &aliases_attr.attr, |
4313 | &align_attr.attr, | |
81819f0f CL |
4314 | &hwcache_align_attr.attr, |
4315 | &reclaim_account_attr.attr, | |
4316 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4317 | &shrink_attr.attr, |
ab9a0f19 | 4318 | &reserved_attr.attr, |
ab4d5ed5 | 4319 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4320 | &total_objects_attr.attr, |
4321 | &slabs_attr.attr, | |
4322 | &sanity_checks_attr.attr, | |
4323 | &trace_attr.attr, | |
81819f0f CL |
4324 | &red_zone_attr.attr, |
4325 | &poison_attr.attr, | |
4326 | &store_user_attr.attr, | |
53e15af0 | 4327 | &validate_attr.attr, |
88a420e4 CL |
4328 | &alloc_calls_attr.attr, |
4329 | &free_calls_attr.attr, | |
ab4d5ed5 | 4330 | #endif |
81819f0f CL |
4331 | #ifdef CONFIG_ZONE_DMA |
4332 | &cache_dma_attr.attr, | |
4333 | #endif | |
4334 | #ifdef CONFIG_NUMA | |
9824601e | 4335 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4336 | #endif |
4337 | #ifdef CONFIG_SLUB_STATS | |
4338 | &alloc_fastpath_attr.attr, | |
4339 | &alloc_slowpath_attr.attr, | |
4340 | &free_fastpath_attr.attr, | |
4341 | &free_slowpath_attr.attr, | |
4342 | &free_frozen_attr.attr, | |
4343 | &free_add_partial_attr.attr, | |
4344 | &free_remove_partial_attr.attr, | |
4345 | &alloc_from_partial_attr.attr, | |
4346 | &alloc_slab_attr.attr, | |
4347 | &alloc_refill_attr.attr, | |
4348 | &free_slab_attr.attr, | |
4349 | &cpuslab_flush_attr.attr, | |
4350 | &deactivate_full_attr.attr, | |
4351 | &deactivate_empty_attr.attr, | |
4352 | &deactivate_to_head_attr.attr, | |
4353 | &deactivate_to_tail_attr.attr, | |
4354 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4355 | &order_fallback_attr.attr, |
81819f0f | 4356 | #endif |
4c13dd3b DM |
4357 | #ifdef CONFIG_FAILSLAB |
4358 | &failslab_attr.attr, | |
4359 | #endif | |
4360 | ||
81819f0f CL |
4361 | NULL |
4362 | }; | |
4363 | ||
4364 | static struct attribute_group slab_attr_group = { | |
4365 | .attrs = slab_attrs, | |
4366 | }; | |
4367 | ||
4368 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4369 | struct attribute *attr, | |
4370 | char *buf) | |
4371 | { | |
4372 | struct slab_attribute *attribute; | |
4373 | struct kmem_cache *s; | |
4374 | int err; | |
4375 | ||
4376 | attribute = to_slab_attr(attr); | |
4377 | s = to_slab(kobj); | |
4378 | ||
4379 | if (!attribute->show) | |
4380 | return -EIO; | |
4381 | ||
4382 | err = attribute->show(s, buf); | |
4383 | ||
4384 | return err; | |
4385 | } | |
4386 | ||
4387 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4388 | struct attribute *attr, | |
4389 | const char *buf, size_t len) | |
4390 | { | |
4391 | struct slab_attribute *attribute; | |
4392 | struct kmem_cache *s; | |
4393 | int err; | |
4394 | ||
4395 | attribute = to_slab_attr(attr); | |
4396 | s = to_slab(kobj); | |
4397 | ||
4398 | if (!attribute->store) | |
4399 | return -EIO; | |
4400 | ||
4401 | err = attribute->store(s, buf, len); | |
4402 | ||
4403 | return err; | |
4404 | } | |
4405 | ||
151c602f CL |
4406 | static void kmem_cache_release(struct kobject *kobj) |
4407 | { | |
4408 | struct kmem_cache *s = to_slab(kobj); | |
4409 | ||
84c1cf62 | 4410 | kfree(s->name); |
151c602f CL |
4411 | kfree(s); |
4412 | } | |
4413 | ||
52cf25d0 | 4414 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
4415 | .show = slab_attr_show, |
4416 | .store = slab_attr_store, | |
4417 | }; | |
4418 | ||
4419 | static struct kobj_type slab_ktype = { | |
4420 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4421 | .release = kmem_cache_release |
81819f0f CL |
4422 | }; |
4423 | ||
4424 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4425 | { | |
4426 | struct kobj_type *ktype = get_ktype(kobj); | |
4427 | ||
4428 | if (ktype == &slab_ktype) | |
4429 | return 1; | |
4430 | return 0; | |
4431 | } | |
4432 | ||
9cd43611 | 4433 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
4434 | .filter = uevent_filter, |
4435 | }; | |
4436 | ||
27c3a314 | 4437 | static struct kset *slab_kset; |
81819f0f CL |
4438 | |
4439 | #define ID_STR_LENGTH 64 | |
4440 | ||
4441 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4442 | * |
4443 | * Format :[flags-]size | |
81819f0f CL |
4444 | */ |
4445 | static char *create_unique_id(struct kmem_cache *s) | |
4446 | { | |
4447 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4448 | char *p = name; | |
4449 | ||
4450 | BUG_ON(!name); | |
4451 | ||
4452 | *p++ = ':'; | |
4453 | /* | |
4454 | * First flags affecting slabcache operations. We will only | |
4455 | * get here for aliasable slabs so we do not need to support | |
4456 | * too many flags. The flags here must cover all flags that | |
4457 | * are matched during merging to guarantee that the id is | |
4458 | * unique. | |
4459 | */ | |
4460 | if (s->flags & SLAB_CACHE_DMA) | |
4461 | *p++ = 'd'; | |
4462 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4463 | *p++ = 'a'; | |
4464 | if (s->flags & SLAB_DEBUG_FREE) | |
4465 | *p++ = 'F'; | |
5a896d9e VN |
4466 | if (!(s->flags & SLAB_NOTRACK)) |
4467 | *p++ = 't'; | |
81819f0f CL |
4468 | if (p != name + 1) |
4469 | *p++ = '-'; | |
4470 | p += sprintf(p, "%07d", s->size); | |
4471 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4472 | return name; | |
4473 | } | |
4474 | ||
4475 | static int sysfs_slab_add(struct kmem_cache *s) | |
4476 | { | |
4477 | int err; | |
4478 | const char *name; | |
4479 | int unmergeable; | |
4480 | ||
4481 | if (slab_state < SYSFS) | |
4482 | /* Defer until later */ | |
4483 | return 0; | |
4484 | ||
4485 | unmergeable = slab_unmergeable(s); | |
4486 | if (unmergeable) { | |
4487 | /* | |
4488 | * Slabcache can never be merged so we can use the name proper. | |
4489 | * This is typically the case for debug situations. In that | |
4490 | * case we can catch duplicate names easily. | |
4491 | */ | |
27c3a314 | 4492 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4493 | name = s->name; |
4494 | } else { | |
4495 | /* | |
4496 | * Create a unique name for the slab as a target | |
4497 | * for the symlinks. | |
4498 | */ | |
4499 | name = create_unique_id(s); | |
4500 | } | |
4501 | ||
27c3a314 | 4502 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4503 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4504 | if (err) { | |
4505 | kobject_put(&s->kobj); | |
81819f0f | 4506 | return err; |
1eada11c | 4507 | } |
81819f0f CL |
4508 | |
4509 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
4510 | if (err) { |
4511 | kobject_del(&s->kobj); | |
4512 | kobject_put(&s->kobj); | |
81819f0f | 4513 | return err; |
5788d8ad | 4514 | } |
81819f0f CL |
4515 | kobject_uevent(&s->kobj, KOBJ_ADD); |
4516 | if (!unmergeable) { | |
4517 | /* Setup first alias */ | |
4518 | sysfs_slab_alias(s, s->name); | |
4519 | kfree(name); | |
4520 | } | |
4521 | return 0; | |
4522 | } | |
4523 | ||
4524 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4525 | { | |
2bce6485 CL |
4526 | if (slab_state < SYSFS) |
4527 | /* | |
4528 | * Sysfs has not been setup yet so no need to remove the | |
4529 | * cache from sysfs. | |
4530 | */ | |
4531 | return; | |
4532 | ||
81819f0f CL |
4533 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
4534 | kobject_del(&s->kobj); | |
151c602f | 4535 | kobject_put(&s->kobj); |
81819f0f CL |
4536 | } |
4537 | ||
4538 | /* | |
4539 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4540 | * available lest we lose that information. |
81819f0f CL |
4541 | */ |
4542 | struct saved_alias { | |
4543 | struct kmem_cache *s; | |
4544 | const char *name; | |
4545 | struct saved_alias *next; | |
4546 | }; | |
4547 | ||
5af328a5 | 4548 | static struct saved_alias *alias_list; |
81819f0f CL |
4549 | |
4550 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4551 | { | |
4552 | struct saved_alias *al; | |
4553 | ||
4554 | if (slab_state == SYSFS) { | |
4555 | /* | |
4556 | * If we have a leftover link then remove it. | |
4557 | */ | |
27c3a314 GKH |
4558 | sysfs_remove_link(&slab_kset->kobj, name); |
4559 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4560 | } |
4561 | ||
4562 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4563 | if (!al) | |
4564 | return -ENOMEM; | |
4565 | ||
4566 | al->s = s; | |
4567 | al->name = name; | |
4568 | al->next = alias_list; | |
4569 | alias_list = al; | |
4570 | return 0; | |
4571 | } | |
4572 | ||
4573 | static int __init slab_sysfs_init(void) | |
4574 | { | |
5b95a4ac | 4575 | struct kmem_cache *s; |
81819f0f CL |
4576 | int err; |
4577 | ||
2bce6485 CL |
4578 | down_write(&slub_lock); |
4579 | ||
0ff21e46 | 4580 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4581 | if (!slab_kset) { |
2bce6485 | 4582 | up_write(&slub_lock); |
81819f0f CL |
4583 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4584 | return -ENOSYS; | |
4585 | } | |
4586 | ||
26a7bd03 CL |
4587 | slab_state = SYSFS; |
4588 | ||
5b95a4ac | 4589 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4590 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4591 | if (err) |
4592 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4593 | " to sysfs\n", s->name); | |
26a7bd03 | 4594 | } |
81819f0f CL |
4595 | |
4596 | while (alias_list) { | |
4597 | struct saved_alias *al = alias_list; | |
4598 | ||
4599 | alias_list = alias_list->next; | |
4600 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4601 | if (err) |
4602 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4603 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4604 | kfree(al); |
4605 | } | |
4606 | ||
2bce6485 | 4607 | up_write(&slub_lock); |
81819f0f CL |
4608 | resiliency_test(); |
4609 | return 0; | |
4610 | } | |
4611 | ||
4612 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 4613 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
4614 | |
4615 | /* | |
4616 | * The /proc/slabinfo ABI | |
4617 | */ | |
158a9624 | 4618 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4619 | static void print_slabinfo_header(struct seq_file *m) |
4620 | { | |
4621 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4622 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4623 | "<objperslab> <pagesperslab>"); | |
4624 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4625 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4626 | seq_putc(m, '\n'); | |
4627 | } | |
4628 | ||
4629 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4630 | { | |
4631 | loff_t n = *pos; | |
4632 | ||
4633 | down_read(&slub_lock); | |
4634 | if (!n) | |
4635 | print_slabinfo_header(m); | |
4636 | ||
4637 | return seq_list_start(&slab_caches, *pos); | |
4638 | } | |
4639 | ||
4640 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4641 | { | |
4642 | return seq_list_next(p, &slab_caches, pos); | |
4643 | } | |
4644 | ||
4645 | static void s_stop(struct seq_file *m, void *p) | |
4646 | { | |
4647 | up_read(&slub_lock); | |
4648 | } | |
4649 | ||
4650 | static int s_show(struct seq_file *m, void *p) | |
4651 | { | |
4652 | unsigned long nr_partials = 0; | |
4653 | unsigned long nr_slabs = 0; | |
4654 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4655 | unsigned long nr_objs = 0; |
4656 | unsigned long nr_free = 0; | |
57ed3eda PE |
4657 | struct kmem_cache *s; |
4658 | int node; | |
4659 | ||
4660 | s = list_entry(p, struct kmem_cache, list); | |
4661 | ||
4662 | for_each_online_node(node) { | |
4663 | struct kmem_cache_node *n = get_node(s, node); | |
4664 | ||
4665 | if (!n) | |
4666 | continue; | |
4667 | ||
4668 | nr_partials += n->nr_partial; | |
4669 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4670 | nr_objs += atomic_long_read(&n->total_objects); |
4671 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4672 | } |
4673 | ||
205ab99d | 4674 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4675 | |
4676 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4677 | nr_objs, s->size, oo_objects(s->oo), |
4678 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4679 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4680 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4681 | 0UL); | |
4682 | seq_putc(m, '\n'); | |
4683 | return 0; | |
4684 | } | |
4685 | ||
7b3c3a50 | 4686 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4687 | .start = s_start, |
4688 | .next = s_next, | |
4689 | .stop = s_stop, | |
4690 | .show = s_show, | |
4691 | }; | |
4692 | ||
7b3c3a50 AD |
4693 | static int slabinfo_open(struct inode *inode, struct file *file) |
4694 | { | |
4695 | return seq_open(file, &slabinfo_op); | |
4696 | } | |
4697 | ||
4698 | static const struct file_operations proc_slabinfo_operations = { | |
4699 | .open = slabinfo_open, | |
4700 | .read = seq_read, | |
4701 | .llseek = seq_lseek, | |
4702 | .release = seq_release, | |
4703 | }; | |
4704 | ||
4705 | static int __init slab_proc_init(void) | |
4706 | { | |
cf5d1131 | 4707 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
4708 | return 0; |
4709 | } | |
4710 | module_init(slab_proc_init); | |
158a9624 | 4711 | #endif /* CONFIG_SLABINFO */ |