]>
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 | |
25985edc | 67 | * a partial slab. A new slab has no one operating on it and thus there is |
81819f0f CL |
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) | |
50d5c41c | 169 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 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 | ||
4fdccdfb | 220 | static inline void stat(const 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 | ||
1393d9a1 CL |
264 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
265 | { | |
266 | void *p; | |
267 | ||
268 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
269 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
270 | #else | |
271 | p = get_freepointer(s, object); | |
272 | #endif | |
273 | return p; | |
274 | } | |
275 | ||
7656c72b CL |
276 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
277 | { | |
278 | *(void **)(object + s->offset) = fp; | |
279 | } | |
280 | ||
281 | /* Loop over all objects in a slab */ | |
224a88be CL |
282 | #define for_each_object(__p, __s, __addr, __objects) \ |
283 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
284 | __p += (__s)->size) |
285 | ||
7656c72b CL |
286 | /* Determine object index from a given position */ |
287 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
288 | { | |
289 | return (p - addr) / s->size; | |
290 | } | |
291 | ||
d71f606f MK |
292 | static inline size_t slab_ksize(const struct kmem_cache *s) |
293 | { | |
294 | #ifdef CONFIG_SLUB_DEBUG | |
295 | /* | |
296 | * Debugging requires use of the padding between object | |
297 | * and whatever may come after it. | |
298 | */ | |
299 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
300 | return s->objsize; | |
301 | ||
302 | #endif | |
303 | /* | |
304 | * If we have the need to store the freelist pointer | |
305 | * back there or track user information then we can | |
306 | * only use the space before that information. | |
307 | */ | |
308 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
309 | return s->inuse; | |
310 | /* | |
311 | * Else we can use all the padding etc for the allocation | |
312 | */ | |
313 | return s->size; | |
314 | } | |
315 | ||
ab9a0f19 LJ |
316 | static inline int order_objects(int order, unsigned long size, int reserved) |
317 | { | |
318 | return ((PAGE_SIZE << order) - reserved) / size; | |
319 | } | |
320 | ||
834f3d11 | 321 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 322 | unsigned long size, int reserved) |
834f3d11 CL |
323 | { |
324 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 325 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
326 | }; |
327 | ||
328 | return x; | |
329 | } | |
330 | ||
331 | static inline int oo_order(struct kmem_cache_order_objects x) | |
332 | { | |
210b5c06 | 333 | return x.x >> OO_SHIFT; |
834f3d11 CL |
334 | } |
335 | ||
336 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
337 | { | |
210b5c06 | 338 | return x.x & OO_MASK; |
834f3d11 CL |
339 | } |
340 | ||
41ecc55b | 341 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
342 | /* |
343 | * Determine a map of object in use on a page. | |
344 | * | |
345 | * Slab lock or node listlock must be held to guarantee that the page does | |
346 | * not vanish from under us. | |
347 | */ | |
348 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
349 | { | |
350 | void *p; | |
351 | void *addr = page_address(page); | |
352 | ||
353 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
354 | set_bit(slab_index(p, s, addr), map); | |
355 | } | |
356 | ||
41ecc55b CL |
357 | /* |
358 | * Debug settings: | |
359 | */ | |
f0630fff CL |
360 | #ifdef CONFIG_SLUB_DEBUG_ON |
361 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
362 | #else | |
41ecc55b | 363 | static int slub_debug; |
f0630fff | 364 | #endif |
41ecc55b CL |
365 | |
366 | static char *slub_debug_slabs; | |
fa5ec8a1 | 367 | static int disable_higher_order_debug; |
41ecc55b | 368 | |
81819f0f CL |
369 | /* |
370 | * Object debugging | |
371 | */ | |
372 | static void print_section(char *text, u8 *addr, unsigned int length) | |
373 | { | |
374 | int i, offset; | |
375 | int newline = 1; | |
376 | char ascii[17]; | |
377 | ||
378 | ascii[16] = 0; | |
379 | ||
380 | for (i = 0; i < length; i++) { | |
381 | if (newline) { | |
24922684 | 382 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
383 | newline = 0; |
384 | } | |
06428780 | 385 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
386 | offset = i % 16; |
387 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
388 | if (offset == 15) { | |
06428780 | 389 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
390 | newline = 1; |
391 | } | |
392 | } | |
393 | if (!newline) { | |
394 | i %= 16; | |
395 | while (i < 16) { | |
06428780 | 396 | printk(KERN_CONT " "); |
81819f0f CL |
397 | ascii[i] = ' '; |
398 | i++; | |
399 | } | |
06428780 | 400 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
401 | } |
402 | } | |
403 | ||
81819f0f CL |
404 | static struct track *get_track(struct kmem_cache *s, void *object, |
405 | enum track_item alloc) | |
406 | { | |
407 | struct track *p; | |
408 | ||
409 | if (s->offset) | |
410 | p = object + s->offset + sizeof(void *); | |
411 | else | |
412 | p = object + s->inuse; | |
413 | ||
414 | return p + alloc; | |
415 | } | |
416 | ||
417 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 418 | enum track_item alloc, unsigned long addr) |
81819f0f | 419 | { |
1a00df4a | 420 | struct track *p = get_track(s, object, alloc); |
81819f0f | 421 | |
81819f0f CL |
422 | if (addr) { |
423 | p->addr = addr; | |
424 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 425 | p->pid = current->pid; |
81819f0f CL |
426 | p->when = jiffies; |
427 | } else | |
428 | memset(p, 0, sizeof(struct track)); | |
429 | } | |
430 | ||
81819f0f CL |
431 | static void init_tracking(struct kmem_cache *s, void *object) |
432 | { | |
24922684 CL |
433 | if (!(s->flags & SLAB_STORE_USER)) |
434 | return; | |
435 | ||
ce71e27c EGM |
436 | set_track(s, object, TRACK_FREE, 0UL); |
437 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
438 | } |
439 | ||
440 | static void print_track(const char *s, struct track *t) | |
441 | { | |
442 | if (!t->addr) | |
443 | return; | |
444 | ||
7daf705f | 445 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 446 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
447 | } |
448 | ||
449 | static void print_tracking(struct kmem_cache *s, void *object) | |
450 | { | |
451 | if (!(s->flags & SLAB_STORE_USER)) | |
452 | return; | |
453 | ||
454 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
455 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
456 | } | |
457 | ||
458 | static void print_page_info(struct page *page) | |
459 | { | |
39b26464 CL |
460 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
461 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
462 | |
463 | } | |
464 | ||
465 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
466 | { | |
467 | va_list args; | |
468 | char buf[100]; | |
469 | ||
470 | va_start(args, fmt); | |
471 | vsnprintf(buf, sizeof(buf), fmt, args); | |
472 | va_end(args); | |
473 | printk(KERN_ERR "========================================" | |
474 | "=====================================\n"); | |
475 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
476 | printk(KERN_ERR "----------------------------------------" | |
477 | "-------------------------------------\n\n"); | |
81819f0f CL |
478 | } |
479 | ||
24922684 CL |
480 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
481 | { | |
482 | va_list args; | |
483 | char buf[100]; | |
484 | ||
485 | va_start(args, fmt); | |
486 | vsnprintf(buf, sizeof(buf), fmt, args); | |
487 | va_end(args); | |
488 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
489 | } | |
490 | ||
491 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
492 | { |
493 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 494 | u8 *addr = page_address(page); |
24922684 CL |
495 | |
496 | print_tracking(s, p); | |
497 | ||
498 | print_page_info(page); | |
499 | ||
500 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
501 | p, p - addr, get_freepointer(s, p)); | |
502 | ||
503 | if (p > addr + 16) | |
504 | print_section("Bytes b4", p - 16, 16); | |
505 | ||
0ebd652b | 506 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
507 | |
508 | if (s->flags & SLAB_RED_ZONE) | |
509 | print_section("Redzone", p + s->objsize, | |
510 | s->inuse - s->objsize); | |
511 | ||
81819f0f CL |
512 | if (s->offset) |
513 | off = s->offset + sizeof(void *); | |
514 | else | |
515 | off = s->inuse; | |
516 | ||
24922684 | 517 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 518 | off += 2 * sizeof(struct track); |
81819f0f CL |
519 | |
520 | if (off != s->size) | |
521 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
522 | print_section("Padding", p + off, s->size - off); |
523 | ||
524 | dump_stack(); | |
81819f0f CL |
525 | } |
526 | ||
527 | static void object_err(struct kmem_cache *s, struct page *page, | |
528 | u8 *object, char *reason) | |
529 | { | |
3dc50637 | 530 | slab_bug(s, "%s", reason); |
24922684 | 531 | print_trailer(s, page, object); |
81819f0f CL |
532 | } |
533 | ||
24922684 | 534 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
535 | { |
536 | va_list args; | |
537 | char buf[100]; | |
538 | ||
24922684 CL |
539 | va_start(args, fmt); |
540 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 541 | va_end(args); |
3dc50637 | 542 | slab_bug(s, "%s", buf); |
24922684 | 543 | print_page_info(page); |
81819f0f CL |
544 | dump_stack(); |
545 | } | |
546 | ||
f7cb1933 | 547 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
548 | { |
549 | u8 *p = object; | |
550 | ||
551 | if (s->flags & __OBJECT_POISON) { | |
552 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 553 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
554 | } |
555 | ||
556 | if (s->flags & SLAB_RED_ZONE) | |
f7cb1933 | 557 | memset(p + s->objsize, val, s->inuse - s->objsize); |
81819f0f CL |
558 | } |
559 | ||
24922684 | 560 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
561 | { |
562 | while (bytes) { | |
563 | if (*start != (u8)value) | |
24922684 | 564 | return start; |
81819f0f CL |
565 | start++; |
566 | bytes--; | |
567 | } | |
24922684 CL |
568 | return NULL; |
569 | } | |
570 | ||
571 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
572 | void *from, void *to) | |
573 | { | |
574 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
575 | memset(from, data, to - from); | |
576 | } | |
577 | ||
578 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
579 | u8 *object, char *what, | |
06428780 | 580 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
581 | { |
582 | u8 *fault; | |
583 | u8 *end; | |
584 | ||
585 | fault = check_bytes(start, value, bytes); | |
586 | if (!fault) | |
587 | return 1; | |
588 | ||
589 | end = start + bytes; | |
590 | while (end > fault && end[-1] == value) | |
591 | end--; | |
592 | ||
593 | slab_bug(s, "%s overwritten", what); | |
594 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
595 | fault, end - 1, fault[0], value); | |
596 | print_trailer(s, page, object); | |
597 | ||
598 | restore_bytes(s, what, value, fault, end); | |
599 | return 0; | |
81819f0f CL |
600 | } |
601 | ||
81819f0f CL |
602 | /* |
603 | * Object layout: | |
604 | * | |
605 | * object address | |
606 | * Bytes of the object to be managed. | |
607 | * If the freepointer may overlay the object then the free | |
608 | * pointer is the first word of the object. | |
672bba3a | 609 | * |
81819f0f CL |
610 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
611 | * 0xa5 (POISON_END) | |
612 | * | |
613 | * object + s->objsize | |
614 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
615 | * Padding is extended by another word if Redzoning is enabled and |
616 | * objsize == inuse. | |
617 | * | |
81819f0f CL |
618 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
619 | * 0xcc (RED_ACTIVE) for objects in use. | |
620 | * | |
621 | * object + s->inuse | |
672bba3a CL |
622 | * Meta data starts here. |
623 | * | |
81819f0f CL |
624 | * A. Free pointer (if we cannot overwrite object on free) |
625 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 626 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 627 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
628 | * before the word boundary. |
629 | * | |
630 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
631 | * |
632 | * object + s->size | |
672bba3a | 633 | * Nothing is used beyond s->size. |
81819f0f | 634 | * |
672bba3a CL |
635 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
636 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
637 | * may be used with merged slabcaches. |
638 | */ | |
639 | ||
81819f0f CL |
640 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
641 | { | |
642 | unsigned long off = s->inuse; /* The end of info */ | |
643 | ||
644 | if (s->offset) | |
645 | /* Freepointer is placed after the object. */ | |
646 | off += sizeof(void *); | |
647 | ||
648 | if (s->flags & SLAB_STORE_USER) | |
649 | /* We also have user information there */ | |
650 | off += 2 * sizeof(struct track); | |
651 | ||
652 | if (s->size == off) | |
653 | return 1; | |
654 | ||
24922684 CL |
655 | return check_bytes_and_report(s, page, p, "Object padding", |
656 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
657 | } |
658 | ||
39b26464 | 659 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
660 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
661 | { | |
24922684 CL |
662 | u8 *start; |
663 | u8 *fault; | |
664 | u8 *end; | |
665 | int length; | |
666 | int remainder; | |
81819f0f CL |
667 | |
668 | if (!(s->flags & SLAB_POISON)) | |
669 | return 1; | |
670 | ||
a973e9dd | 671 | start = page_address(page); |
ab9a0f19 | 672 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
673 | end = start + length; |
674 | remainder = length % s->size; | |
81819f0f CL |
675 | if (!remainder) |
676 | return 1; | |
677 | ||
39b26464 | 678 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
679 | if (!fault) |
680 | return 1; | |
681 | while (end > fault && end[-1] == POISON_INUSE) | |
682 | end--; | |
683 | ||
684 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 685 | print_section("Padding", end - remainder, remainder); |
24922684 | 686 | |
8a3d271d | 687 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 688 | return 0; |
81819f0f CL |
689 | } |
690 | ||
691 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 692 | void *object, u8 val) |
81819f0f CL |
693 | { |
694 | u8 *p = object; | |
695 | u8 *endobject = object + s->objsize; | |
696 | ||
697 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 698 | if (!check_bytes_and_report(s, page, object, "Redzone", |
f7cb1933 | 699 | endobject, val, s->inuse - s->objsize)) |
81819f0f | 700 | return 0; |
81819f0f | 701 | } else { |
3adbefee IM |
702 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
703 | check_bytes_and_report(s, page, p, "Alignment padding", | |
704 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
705 | } | |
81819f0f CL |
706 | } |
707 | ||
708 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 709 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 CL |
710 | (!check_bytes_and_report(s, page, p, "Poison", p, |
711 | POISON_FREE, s->objsize - 1) || | |
712 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 713 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 714 | return 0; |
81819f0f CL |
715 | /* |
716 | * check_pad_bytes cleans up on its own. | |
717 | */ | |
718 | check_pad_bytes(s, page, p); | |
719 | } | |
720 | ||
f7cb1933 | 721 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
722 | /* |
723 | * Object and freepointer overlap. Cannot check | |
724 | * freepointer while object is allocated. | |
725 | */ | |
726 | return 1; | |
727 | ||
728 | /* Check free pointer validity */ | |
729 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
730 | object_err(s, page, p, "Freepointer corrupt"); | |
731 | /* | |
9f6c708e | 732 | * No choice but to zap it and thus lose the remainder |
81819f0f | 733 | * of the free objects in this slab. May cause |
672bba3a | 734 | * another error because the object count is now wrong. |
81819f0f | 735 | */ |
a973e9dd | 736 | set_freepointer(s, p, NULL); |
81819f0f CL |
737 | return 0; |
738 | } | |
739 | return 1; | |
740 | } | |
741 | ||
742 | static int check_slab(struct kmem_cache *s, struct page *page) | |
743 | { | |
39b26464 CL |
744 | int maxobj; |
745 | ||
81819f0f CL |
746 | VM_BUG_ON(!irqs_disabled()); |
747 | ||
748 | if (!PageSlab(page)) { | |
24922684 | 749 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
750 | return 0; |
751 | } | |
39b26464 | 752 | |
ab9a0f19 | 753 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
754 | if (page->objects > maxobj) { |
755 | slab_err(s, page, "objects %u > max %u", | |
756 | s->name, page->objects, maxobj); | |
757 | return 0; | |
758 | } | |
759 | if (page->inuse > page->objects) { | |
24922684 | 760 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 761 | s->name, page->inuse, page->objects); |
81819f0f CL |
762 | return 0; |
763 | } | |
764 | /* Slab_pad_check fixes things up after itself */ | |
765 | slab_pad_check(s, page); | |
766 | return 1; | |
767 | } | |
768 | ||
769 | /* | |
672bba3a CL |
770 | * Determine if a certain object on a page is on the freelist. Must hold the |
771 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
772 | */ |
773 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
774 | { | |
775 | int nr = 0; | |
776 | void *fp = page->freelist; | |
777 | void *object = NULL; | |
224a88be | 778 | unsigned long max_objects; |
81819f0f | 779 | |
39b26464 | 780 | while (fp && nr <= page->objects) { |
81819f0f CL |
781 | if (fp == search) |
782 | return 1; | |
783 | if (!check_valid_pointer(s, page, fp)) { | |
784 | if (object) { | |
785 | object_err(s, page, object, | |
786 | "Freechain corrupt"); | |
a973e9dd | 787 | set_freepointer(s, object, NULL); |
81819f0f CL |
788 | break; |
789 | } else { | |
24922684 | 790 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 791 | page->freelist = NULL; |
39b26464 | 792 | page->inuse = page->objects; |
24922684 | 793 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
794 | return 0; |
795 | } | |
796 | break; | |
797 | } | |
798 | object = fp; | |
799 | fp = get_freepointer(s, object); | |
800 | nr++; | |
801 | } | |
802 | ||
ab9a0f19 | 803 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
804 | if (max_objects > MAX_OBJS_PER_PAGE) |
805 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
806 | |
807 | if (page->objects != max_objects) { | |
808 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
809 | "should be %d", page->objects, max_objects); | |
810 | page->objects = max_objects; | |
811 | slab_fix(s, "Number of objects adjusted."); | |
812 | } | |
39b26464 | 813 | if (page->inuse != page->objects - nr) { |
70d71228 | 814 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
815 | "counted were %d", page->inuse, page->objects - nr); |
816 | page->inuse = page->objects - nr; | |
24922684 | 817 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
818 | } |
819 | return search == NULL; | |
820 | } | |
821 | ||
0121c619 CL |
822 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
823 | int alloc) | |
3ec09742 CL |
824 | { |
825 | if (s->flags & SLAB_TRACE) { | |
826 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
827 | s->name, | |
828 | alloc ? "alloc" : "free", | |
829 | object, page->inuse, | |
830 | page->freelist); | |
831 | ||
832 | if (!alloc) | |
833 | print_section("Object", (void *)object, s->objsize); | |
834 | ||
835 | dump_stack(); | |
836 | } | |
837 | } | |
838 | ||
c016b0bd CL |
839 | /* |
840 | * Hooks for other subsystems that check memory allocations. In a typical | |
841 | * production configuration these hooks all should produce no code at all. | |
842 | */ | |
843 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
844 | { | |
c1d50836 | 845 | flags &= gfp_allowed_mask; |
c016b0bd CL |
846 | lockdep_trace_alloc(flags); |
847 | might_sleep_if(flags & __GFP_WAIT); | |
848 | ||
849 | return should_failslab(s->objsize, flags, s->flags); | |
850 | } | |
851 | ||
852 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
853 | { | |
c1d50836 | 854 | flags &= gfp_allowed_mask; |
b3d41885 | 855 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
c016b0bd CL |
856 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); |
857 | } | |
858 | ||
859 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
860 | { | |
861 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 862 | |
d3f661d6 CL |
863 | /* |
864 | * Trouble is that we may no longer disable interupts in the fast path | |
865 | * So in order to make the debug calls that expect irqs to be | |
866 | * disabled we need to disable interrupts temporarily. | |
867 | */ | |
868 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
869 | { | |
870 | unsigned long flags; | |
871 | ||
872 | local_irq_save(flags); | |
873 | kmemcheck_slab_free(s, x, s->objsize); | |
874 | debug_check_no_locks_freed(x, s->objsize); | |
d3f661d6 CL |
875 | local_irq_restore(flags); |
876 | } | |
877 | #endif | |
f9b615de TG |
878 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
879 | debug_check_no_obj_freed(x, s->objsize); | |
c016b0bd CL |
880 | } |
881 | ||
643b1138 | 882 | /* |
672bba3a | 883 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 884 | */ |
e95eed57 | 885 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 886 | { |
643b1138 CL |
887 | spin_lock(&n->list_lock); |
888 | list_add(&page->lru, &n->full); | |
889 | spin_unlock(&n->list_lock); | |
890 | } | |
891 | ||
892 | static void remove_full(struct kmem_cache *s, struct page *page) | |
893 | { | |
894 | struct kmem_cache_node *n; | |
895 | ||
896 | if (!(s->flags & SLAB_STORE_USER)) | |
897 | return; | |
898 | ||
899 | n = get_node(s, page_to_nid(page)); | |
900 | ||
901 | spin_lock(&n->list_lock); | |
902 | list_del(&page->lru); | |
903 | spin_unlock(&n->list_lock); | |
904 | } | |
905 | ||
0f389ec6 CL |
906 | /* Tracking of the number of slabs for debugging purposes */ |
907 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
908 | { | |
909 | struct kmem_cache_node *n = get_node(s, node); | |
910 | ||
911 | return atomic_long_read(&n->nr_slabs); | |
912 | } | |
913 | ||
26c02cf0 AB |
914 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
915 | { | |
916 | return atomic_long_read(&n->nr_slabs); | |
917 | } | |
918 | ||
205ab99d | 919 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
920 | { |
921 | struct kmem_cache_node *n = get_node(s, node); | |
922 | ||
923 | /* | |
924 | * May be called early in order to allocate a slab for the | |
925 | * kmem_cache_node structure. Solve the chicken-egg | |
926 | * dilemma by deferring the increment of the count during | |
927 | * bootstrap (see early_kmem_cache_node_alloc). | |
928 | */ | |
7340cc84 | 929 | if (n) { |
0f389ec6 | 930 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
931 | atomic_long_add(objects, &n->total_objects); |
932 | } | |
0f389ec6 | 933 | } |
205ab99d | 934 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
935 | { |
936 | struct kmem_cache_node *n = get_node(s, node); | |
937 | ||
938 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 939 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
940 | } |
941 | ||
942 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
943 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
944 | void *object) | |
945 | { | |
946 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
947 | return; | |
948 | ||
f7cb1933 | 949 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
950 | init_tracking(s, object); |
951 | } | |
952 | ||
1537066c | 953 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 954 | void *object, unsigned long addr) |
81819f0f CL |
955 | { |
956 | if (!check_slab(s, page)) | |
957 | goto bad; | |
958 | ||
d692ef6d | 959 | if (!on_freelist(s, page, object)) { |
24922684 | 960 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 961 | goto bad; |
81819f0f CL |
962 | } |
963 | ||
964 | if (!check_valid_pointer(s, page, object)) { | |
965 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 966 | goto bad; |
81819f0f CL |
967 | } |
968 | ||
f7cb1933 | 969 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 970 | goto bad; |
81819f0f | 971 | |
3ec09742 CL |
972 | /* Success perform special debug activities for allocs */ |
973 | if (s->flags & SLAB_STORE_USER) | |
974 | set_track(s, object, TRACK_ALLOC, addr); | |
975 | trace(s, page, object, 1); | |
f7cb1933 | 976 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 977 | return 1; |
3ec09742 | 978 | |
81819f0f CL |
979 | bad: |
980 | if (PageSlab(page)) { | |
981 | /* | |
982 | * If this is a slab page then lets do the best we can | |
983 | * to avoid issues in the future. Marking all objects | |
672bba3a | 984 | * as used avoids touching the remaining objects. |
81819f0f | 985 | */ |
24922684 | 986 | slab_fix(s, "Marking all objects used"); |
39b26464 | 987 | page->inuse = page->objects; |
a973e9dd | 988 | page->freelist = NULL; |
81819f0f CL |
989 | } |
990 | return 0; | |
991 | } | |
992 | ||
1537066c CL |
993 | static noinline int free_debug_processing(struct kmem_cache *s, |
994 | struct page *page, void *object, unsigned long addr) | |
81819f0f CL |
995 | { |
996 | if (!check_slab(s, page)) | |
997 | goto fail; | |
998 | ||
999 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1000 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1001 | goto fail; |
1002 | } | |
1003 | ||
1004 | if (on_freelist(s, page, object)) { | |
24922684 | 1005 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1006 | goto fail; |
1007 | } | |
1008 | ||
f7cb1933 | 1009 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
81819f0f CL |
1010 | return 0; |
1011 | ||
1012 | if (unlikely(s != page->slab)) { | |
3adbefee | 1013 | if (!PageSlab(page)) { |
70d71228 CL |
1014 | slab_err(s, page, "Attempt to free object(0x%p) " |
1015 | "outside of slab", object); | |
3adbefee | 1016 | } else if (!page->slab) { |
81819f0f | 1017 | printk(KERN_ERR |
70d71228 | 1018 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1019 | object); |
70d71228 | 1020 | dump_stack(); |
06428780 | 1021 | } else |
24922684 CL |
1022 | object_err(s, page, object, |
1023 | "page slab pointer corrupt."); | |
81819f0f CL |
1024 | goto fail; |
1025 | } | |
3ec09742 CL |
1026 | |
1027 | /* Special debug activities for freeing objects */ | |
50d5c41c | 1028 | if (!page->frozen && !page->freelist) |
3ec09742 CL |
1029 | remove_full(s, page); |
1030 | if (s->flags & SLAB_STORE_USER) | |
1031 | set_track(s, object, TRACK_FREE, addr); | |
1032 | trace(s, page, object, 0); | |
f7cb1933 | 1033 | init_object(s, object, SLUB_RED_INACTIVE); |
81819f0f | 1034 | return 1; |
3ec09742 | 1035 | |
81819f0f | 1036 | fail: |
24922684 | 1037 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
1038 | return 0; |
1039 | } | |
1040 | ||
41ecc55b CL |
1041 | static int __init setup_slub_debug(char *str) |
1042 | { | |
f0630fff CL |
1043 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1044 | if (*str++ != '=' || !*str) | |
1045 | /* | |
1046 | * No options specified. Switch on full debugging. | |
1047 | */ | |
1048 | goto out; | |
1049 | ||
1050 | if (*str == ',') | |
1051 | /* | |
1052 | * No options but restriction on slabs. This means full | |
1053 | * debugging for slabs matching a pattern. | |
1054 | */ | |
1055 | goto check_slabs; | |
1056 | ||
fa5ec8a1 DR |
1057 | if (tolower(*str) == 'o') { |
1058 | /* | |
1059 | * Avoid enabling debugging on caches if its minimum order | |
1060 | * would increase as a result. | |
1061 | */ | |
1062 | disable_higher_order_debug = 1; | |
1063 | goto out; | |
1064 | } | |
1065 | ||
f0630fff CL |
1066 | slub_debug = 0; |
1067 | if (*str == '-') | |
1068 | /* | |
1069 | * Switch off all debugging measures. | |
1070 | */ | |
1071 | goto out; | |
1072 | ||
1073 | /* | |
1074 | * Determine which debug features should be switched on | |
1075 | */ | |
06428780 | 1076 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1077 | switch (tolower(*str)) { |
1078 | case 'f': | |
1079 | slub_debug |= SLAB_DEBUG_FREE; | |
1080 | break; | |
1081 | case 'z': | |
1082 | slub_debug |= SLAB_RED_ZONE; | |
1083 | break; | |
1084 | case 'p': | |
1085 | slub_debug |= SLAB_POISON; | |
1086 | break; | |
1087 | case 'u': | |
1088 | slub_debug |= SLAB_STORE_USER; | |
1089 | break; | |
1090 | case 't': | |
1091 | slub_debug |= SLAB_TRACE; | |
1092 | break; | |
4c13dd3b DM |
1093 | case 'a': |
1094 | slub_debug |= SLAB_FAILSLAB; | |
1095 | break; | |
f0630fff CL |
1096 | default: |
1097 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1098 | "unknown. skipped\n", *str); |
f0630fff | 1099 | } |
41ecc55b CL |
1100 | } |
1101 | ||
f0630fff | 1102 | check_slabs: |
41ecc55b CL |
1103 | if (*str == ',') |
1104 | slub_debug_slabs = str + 1; | |
f0630fff | 1105 | out: |
41ecc55b CL |
1106 | return 1; |
1107 | } | |
1108 | ||
1109 | __setup("slub_debug", setup_slub_debug); | |
1110 | ||
ba0268a8 CL |
1111 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1112 | unsigned long flags, const char *name, | |
51cc5068 | 1113 | void (*ctor)(void *)) |
41ecc55b CL |
1114 | { |
1115 | /* | |
e153362a | 1116 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1117 | */ |
e153362a | 1118 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1119 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1120 | flags |= slub_debug; | |
ba0268a8 CL |
1121 | |
1122 | return flags; | |
41ecc55b CL |
1123 | } |
1124 | #else | |
3ec09742 CL |
1125 | static inline void setup_object_debug(struct kmem_cache *s, |
1126 | struct page *page, void *object) {} | |
41ecc55b | 1127 | |
3ec09742 | 1128 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1129 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1130 | |
3ec09742 | 1131 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1132 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1133 | |
41ecc55b CL |
1134 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1135 | { return 1; } | |
1136 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1137 | void *object, u8 val) { return 1; } |
3ec09742 | 1138 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1139 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1140 | unsigned long flags, const char *name, | |
51cc5068 | 1141 | void (*ctor)(void *)) |
ba0268a8 CL |
1142 | { |
1143 | return flags; | |
1144 | } | |
41ecc55b | 1145 | #define slub_debug 0 |
0f389ec6 | 1146 | |
fdaa45e9 IM |
1147 | #define disable_higher_order_debug 0 |
1148 | ||
0f389ec6 CL |
1149 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1150 | { return 0; } | |
26c02cf0 AB |
1151 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1152 | { return 0; } | |
205ab99d CL |
1153 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1154 | int objects) {} | |
1155 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1156 | int objects) {} | |
7d550c56 CL |
1157 | |
1158 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1159 | { return 0; } | |
1160 | ||
1161 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1162 | void *object) {} | |
1163 | ||
1164 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1165 | ||
ab4d5ed5 | 1166 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1167 | |
81819f0f CL |
1168 | /* |
1169 | * Slab allocation and freeing | |
1170 | */ | |
65c3376a CL |
1171 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1172 | struct kmem_cache_order_objects oo) | |
1173 | { | |
1174 | int order = oo_order(oo); | |
1175 | ||
b1eeab67 VN |
1176 | flags |= __GFP_NOTRACK; |
1177 | ||
2154a336 | 1178 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1179 | return alloc_pages(flags, order); |
1180 | else | |
6b65aaf3 | 1181 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1182 | } |
1183 | ||
81819f0f CL |
1184 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1185 | { | |
06428780 | 1186 | struct page *page; |
834f3d11 | 1187 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1188 | gfp_t alloc_gfp; |
81819f0f | 1189 | |
7e0528da CL |
1190 | flags &= gfp_allowed_mask; |
1191 | ||
1192 | if (flags & __GFP_WAIT) | |
1193 | local_irq_enable(); | |
1194 | ||
b7a49f0d | 1195 | flags |= s->allocflags; |
e12ba74d | 1196 | |
ba52270d PE |
1197 | /* |
1198 | * Let the initial higher-order allocation fail under memory pressure | |
1199 | * so we fall-back to the minimum order allocation. | |
1200 | */ | |
1201 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1202 | ||
1203 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1204 | if (unlikely(!page)) { |
1205 | oo = s->min; | |
1206 | /* | |
1207 | * Allocation may have failed due to fragmentation. | |
1208 | * Try a lower order alloc if possible | |
1209 | */ | |
1210 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1211 | |
7e0528da CL |
1212 | if (page) |
1213 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1214 | } |
5a896d9e | 1215 | |
7e0528da CL |
1216 | if (flags & __GFP_WAIT) |
1217 | local_irq_disable(); | |
1218 | ||
1219 | if (!page) | |
1220 | return NULL; | |
1221 | ||
5a896d9e | 1222 | if (kmemcheck_enabled |
5086c389 | 1223 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1224 | int pages = 1 << oo_order(oo); |
1225 | ||
1226 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1227 | ||
1228 | /* | |
1229 | * Objects from caches that have a constructor don't get | |
1230 | * cleared when they're allocated, so we need to do it here. | |
1231 | */ | |
1232 | if (s->ctor) | |
1233 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1234 | else | |
1235 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1236 | } |
1237 | ||
834f3d11 | 1238 | page->objects = oo_objects(oo); |
81819f0f CL |
1239 | mod_zone_page_state(page_zone(page), |
1240 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1241 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1242 | 1 << oo_order(oo)); |
81819f0f CL |
1243 | |
1244 | return page; | |
1245 | } | |
1246 | ||
1247 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1248 | void *object) | |
1249 | { | |
3ec09742 | 1250 | setup_object_debug(s, page, object); |
4f104934 | 1251 | if (unlikely(s->ctor)) |
51cc5068 | 1252 | s->ctor(object); |
81819f0f CL |
1253 | } |
1254 | ||
1255 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1256 | { | |
1257 | struct page *page; | |
81819f0f | 1258 | void *start; |
81819f0f CL |
1259 | void *last; |
1260 | void *p; | |
1261 | ||
6cb06229 | 1262 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1263 | |
6cb06229 CL |
1264 | page = allocate_slab(s, |
1265 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1266 | if (!page) |
1267 | goto out; | |
1268 | ||
205ab99d | 1269 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1270 | page->slab = s; |
1271 | page->flags |= 1 << PG_slab; | |
81819f0f CL |
1272 | |
1273 | start = page_address(page); | |
81819f0f CL |
1274 | |
1275 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1276 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1277 | |
1278 | last = start; | |
224a88be | 1279 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1280 | setup_object(s, page, last); |
1281 | set_freepointer(s, last, p); | |
1282 | last = p; | |
1283 | } | |
1284 | setup_object(s, page, last); | |
a973e9dd | 1285 | set_freepointer(s, last, NULL); |
81819f0f CL |
1286 | |
1287 | page->freelist = start; | |
1288 | page->inuse = 0; | |
8cb0a506 | 1289 | page->frozen = 1; |
81819f0f | 1290 | out: |
81819f0f CL |
1291 | return page; |
1292 | } | |
1293 | ||
1294 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1295 | { | |
834f3d11 CL |
1296 | int order = compound_order(page); |
1297 | int pages = 1 << order; | |
81819f0f | 1298 | |
af537b0a | 1299 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1300 | void *p; |
1301 | ||
1302 | slab_pad_check(s, page); | |
224a88be CL |
1303 | for_each_object(p, s, page_address(page), |
1304 | page->objects) | |
f7cb1933 | 1305 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1306 | } |
1307 | ||
b1eeab67 | 1308 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1309 | |
81819f0f CL |
1310 | mod_zone_page_state(page_zone(page), |
1311 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1312 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1313 | -pages); |
81819f0f | 1314 | |
49bd5221 CL |
1315 | __ClearPageSlab(page); |
1316 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1317 | if (current->reclaim_state) |
1318 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1319 | __free_pages(page, order); |
81819f0f CL |
1320 | } |
1321 | ||
da9a638c LJ |
1322 | #define need_reserve_slab_rcu \ |
1323 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1324 | ||
81819f0f CL |
1325 | static void rcu_free_slab(struct rcu_head *h) |
1326 | { | |
1327 | struct page *page; | |
1328 | ||
da9a638c LJ |
1329 | if (need_reserve_slab_rcu) |
1330 | page = virt_to_head_page(h); | |
1331 | else | |
1332 | page = container_of((struct list_head *)h, struct page, lru); | |
1333 | ||
81819f0f CL |
1334 | __free_slab(page->slab, page); |
1335 | } | |
1336 | ||
1337 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1338 | { | |
1339 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1340 | struct rcu_head *head; |
1341 | ||
1342 | if (need_reserve_slab_rcu) { | |
1343 | int order = compound_order(page); | |
1344 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1345 | ||
1346 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1347 | head = page_address(page) + offset; | |
1348 | } else { | |
1349 | /* | |
1350 | * RCU free overloads the RCU head over the LRU | |
1351 | */ | |
1352 | head = (void *)&page->lru; | |
1353 | } | |
81819f0f CL |
1354 | |
1355 | call_rcu(head, rcu_free_slab); | |
1356 | } else | |
1357 | __free_slab(s, page); | |
1358 | } | |
1359 | ||
1360 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1361 | { | |
205ab99d | 1362 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1363 | free_slab(s, page); |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * Per slab locking using the pagelock | |
1368 | */ | |
1369 | static __always_inline void slab_lock(struct page *page) | |
1370 | { | |
1371 | bit_spin_lock(PG_locked, &page->flags); | |
1372 | } | |
1373 | ||
1374 | static __always_inline void slab_unlock(struct page *page) | |
1375 | { | |
a76d3546 | 1376 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1377 | } |
1378 | ||
1379 | static __always_inline int slab_trylock(struct page *page) | |
1380 | { | |
1381 | int rc = 1; | |
1382 | ||
1383 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1384 | return rc; | |
1385 | } | |
1386 | ||
1387 | /* | |
1388 | * Management of partially allocated slabs | |
1389 | */ | |
7c2e132c CL |
1390 | static void add_partial(struct kmem_cache_node *n, |
1391 | struct page *page, int tail) | |
81819f0f | 1392 | { |
e95eed57 CL |
1393 | spin_lock(&n->list_lock); |
1394 | n->nr_partial++; | |
7c2e132c CL |
1395 | if (tail) |
1396 | list_add_tail(&page->lru, &n->partial); | |
1397 | else | |
1398 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1399 | spin_unlock(&n->list_lock); |
1400 | } | |
1401 | ||
62e346a8 CL |
1402 | static inline void __remove_partial(struct kmem_cache_node *n, |
1403 | struct page *page) | |
1404 | { | |
1405 | list_del(&page->lru); | |
1406 | n->nr_partial--; | |
1407 | } | |
1408 | ||
0121c619 | 1409 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1410 | { |
1411 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1412 | ||
1413 | spin_lock(&n->list_lock); | |
62e346a8 | 1414 | __remove_partial(n, page); |
81819f0f CL |
1415 | spin_unlock(&n->list_lock); |
1416 | } | |
1417 | ||
1418 | /* | |
672bba3a | 1419 | * Lock slab and remove from the partial list. |
81819f0f | 1420 | * |
672bba3a | 1421 | * Must hold list_lock. |
81819f0f | 1422 | */ |
0121c619 CL |
1423 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1424 | struct page *page) | |
81819f0f CL |
1425 | { |
1426 | if (slab_trylock(page)) { | |
62e346a8 | 1427 | __remove_partial(n, page); |
81819f0f CL |
1428 | return 1; |
1429 | } | |
1430 | return 0; | |
1431 | } | |
1432 | ||
1433 | /* | |
672bba3a | 1434 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1435 | */ |
1436 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1437 | { | |
1438 | struct page *page; | |
1439 | ||
1440 | /* | |
1441 | * Racy check. If we mistakenly see no partial slabs then we | |
1442 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1443 | * partial slab and there is none available then get_partials() |
1444 | * will return NULL. | |
81819f0f CL |
1445 | */ |
1446 | if (!n || !n->nr_partial) | |
1447 | return NULL; | |
1448 | ||
1449 | spin_lock(&n->list_lock); | |
1450 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1451 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1452 | goto out; |
1453 | page = NULL; | |
1454 | out: | |
1455 | spin_unlock(&n->list_lock); | |
1456 | return page; | |
1457 | } | |
1458 | ||
1459 | /* | |
672bba3a | 1460 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1461 | */ |
1462 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1463 | { | |
1464 | #ifdef CONFIG_NUMA | |
1465 | struct zonelist *zonelist; | |
dd1a239f | 1466 | struct zoneref *z; |
54a6eb5c MG |
1467 | struct zone *zone; |
1468 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1469 | struct page *page; |
1470 | ||
1471 | /* | |
672bba3a CL |
1472 | * The defrag ratio allows a configuration of the tradeoffs between |
1473 | * inter node defragmentation and node local allocations. A lower | |
1474 | * defrag_ratio increases the tendency to do local allocations | |
1475 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1476 | * |
672bba3a CL |
1477 | * If the defrag_ratio is set to 0 then kmalloc() always |
1478 | * returns node local objects. If the ratio is higher then kmalloc() | |
1479 | * may return off node objects because partial slabs are obtained | |
1480 | * from other nodes and filled up. | |
81819f0f | 1481 | * |
6446faa2 | 1482 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1483 | * defrag_ratio = 1000) then every (well almost) allocation will |
1484 | * first attempt to defrag slab caches on other nodes. This means | |
1485 | * scanning over all nodes to look for partial slabs which may be | |
1486 | * expensive if we do it every time we are trying to find a slab | |
1487 | * with available objects. | |
81819f0f | 1488 | */ |
9824601e CL |
1489 | if (!s->remote_node_defrag_ratio || |
1490 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1491 | return NULL; |
1492 | ||
c0ff7453 | 1493 | get_mems_allowed(); |
0e88460d | 1494 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1495 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1496 | struct kmem_cache_node *n; |
1497 | ||
54a6eb5c | 1498 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1499 | |
54a6eb5c | 1500 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1501 | n->nr_partial > s->min_partial) { |
81819f0f | 1502 | page = get_partial_node(n); |
c0ff7453 MX |
1503 | if (page) { |
1504 | put_mems_allowed(); | |
81819f0f | 1505 | return page; |
c0ff7453 | 1506 | } |
81819f0f CL |
1507 | } |
1508 | } | |
c0ff7453 | 1509 | put_mems_allowed(); |
81819f0f CL |
1510 | #endif |
1511 | return NULL; | |
1512 | } | |
1513 | ||
1514 | /* | |
1515 | * Get a partial page, lock it and return it. | |
1516 | */ | |
1517 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1518 | { | |
1519 | struct page *page; | |
2154a336 | 1520 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f CL |
1521 | |
1522 | page = get_partial_node(get_node(s, searchnode)); | |
33de04ec | 1523 | if (page || node != NUMA_NO_NODE) |
81819f0f CL |
1524 | return page; |
1525 | ||
1526 | return get_any_partial(s, flags); | |
1527 | } | |
1528 | ||
1529 | /* | |
1530 | * Move a page back to the lists. | |
1531 | * | |
1532 | * Must be called with the slab lock held. | |
1533 | * | |
1534 | * On exit the slab lock will have been dropped. | |
1535 | */ | |
7c2e132c | 1536 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
3478973d | 1537 | __releases(bitlock) |
81819f0f | 1538 | { |
e95eed57 CL |
1539 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1540 | ||
81819f0f | 1541 | if (page->inuse) { |
e95eed57 | 1542 | |
a973e9dd | 1543 | if (page->freelist) { |
7c2e132c | 1544 | add_partial(n, page, tail); |
84e554e6 | 1545 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
8ff12cfc | 1546 | } else { |
84e554e6 | 1547 | stat(s, DEACTIVATE_FULL); |
af537b0a | 1548 | if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER)) |
8ff12cfc CL |
1549 | add_full(n, page); |
1550 | } | |
81819f0f CL |
1551 | slab_unlock(page); |
1552 | } else { | |
84e554e6 | 1553 | stat(s, DEACTIVATE_EMPTY); |
3b89d7d8 | 1554 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1555 | /* |
672bba3a CL |
1556 | * Adding an empty slab to the partial slabs in order |
1557 | * to avoid page allocator overhead. This slab needs | |
1558 | * to come after the other slabs with objects in | |
6446faa2 CL |
1559 | * so that the others get filled first. That way the |
1560 | * size of the partial list stays small. | |
1561 | * | |
0121c619 CL |
1562 | * kmem_cache_shrink can reclaim any empty slabs from |
1563 | * the partial list. | |
e95eed57 | 1564 | */ |
7c2e132c | 1565 | add_partial(n, page, 1); |
e95eed57 CL |
1566 | slab_unlock(page); |
1567 | } else { | |
1568 | slab_unlock(page); | |
84e554e6 | 1569 | stat(s, FREE_SLAB); |
e95eed57 CL |
1570 | discard_slab(s, page); |
1571 | } | |
81819f0f CL |
1572 | } |
1573 | } | |
1574 | ||
8a5ec0ba CL |
1575 | #ifdef CONFIG_PREEMPT |
1576 | /* | |
1577 | * Calculate the next globally unique transaction for disambiguiation | |
1578 | * during cmpxchg. The transactions start with the cpu number and are then | |
1579 | * incremented by CONFIG_NR_CPUS. | |
1580 | */ | |
1581 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1582 | #else | |
1583 | /* | |
1584 | * No preemption supported therefore also no need to check for | |
1585 | * different cpus. | |
1586 | */ | |
1587 | #define TID_STEP 1 | |
1588 | #endif | |
1589 | ||
1590 | static inline unsigned long next_tid(unsigned long tid) | |
1591 | { | |
1592 | return tid + TID_STEP; | |
1593 | } | |
1594 | ||
1595 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1596 | { | |
1597 | return tid % TID_STEP; | |
1598 | } | |
1599 | ||
1600 | static inline unsigned long tid_to_event(unsigned long tid) | |
1601 | { | |
1602 | return tid / TID_STEP; | |
1603 | } | |
1604 | ||
1605 | static inline unsigned int init_tid(int cpu) | |
1606 | { | |
1607 | return cpu; | |
1608 | } | |
1609 | ||
1610 | static inline void note_cmpxchg_failure(const char *n, | |
1611 | const struct kmem_cache *s, unsigned long tid) | |
1612 | { | |
1613 | #ifdef SLUB_DEBUG_CMPXCHG | |
1614 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1615 | ||
1616 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1617 | ||
1618 | #ifdef CONFIG_PREEMPT | |
1619 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1620 | printk("due to cpu change %d -> %d\n", | |
1621 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1622 | else | |
1623 | #endif | |
1624 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1625 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1626 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1627 | else | |
1628 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1629 | actual_tid, tid, next_tid(tid)); | |
1630 | #endif | |
4fdccdfb | 1631 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1632 | } |
1633 | ||
8a5ec0ba CL |
1634 | void init_kmem_cache_cpus(struct kmem_cache *s) |
1635 | { | |
8a5ec0ba CL |
1636 | int cpu; |
1637 | ||
1638 | for_each_possible_cpu(cpu) | |
1639 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1640 | } |
81819f0f CL |
1641 | /* |
1642 | * Remove the cpu slab | |
1643 | */ | |
dfb4f096 | 1644 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
3478973d | 1645 | __releases(bitlock) |
81819f0f | 1646 | { |
dfb4f096 | 1647 | struct page *page = c->page; |
7c2e132c | 1648 | int tail = 1; |
8ff12cfc | 1649 | |
b773ad73 | 1650 | if (page->freelist) |
84e554e6 | 1651 | stat(s, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1652 | /* |
6446faa2 | 1653 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1654 | * because both freelists are empty. So this is unlikely |
1655 | * to occur. | |
1656 | */ | |
a973e9dd | 1657 | while (unlikely(c->freelist)) { |
894b8788 CL |
1658 | void **object; |
1659 | ||
7c2e132c CL |
1660 | tail = 0; /* Hot objects. Put the slab first */ |
1661 | ||
894b8788 | 1662 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1663 | object = c->freelist; |
ff12059e | 1664 | c->freelist = get_freepointer(s, c->freelist); |
894b8788 CL |
1665 | |
1666 | /* And put onto the regular freelist */ | |
ff12059e | 1667 | set_freepointer(s, object, page->freelist); |
894b8788 CL |
1668 | page->freelist = object; |
1669 | page->inuse--; | |
1670 | } | |
dfb4f096 | 1671 | c->page = NULL; |
8a5ec0ba | 1672 | c->tid = next_tid(c->tid); |
8cb0a506 | 1673 | page->frozen = 0; |
7c2e132c | 1674 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1675 | } |
1676 | ||
dfb4f096 | 1677 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1678 | { |
84e554e6 | 1679 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 CL |
1680 | slab_lock(c->page); |
1681 | deactivate_slab(s, c); | |
81819f0f CL |
1682 | } |
1683 | ||
1684 | /* | |
1685 | * Flush cpu slab. | |
6446faa2 | 1686 | * |
81819f0f CL |
1687 | * Called from IPI handler with interrupts disabled. |
1688 | */ | |
0c710013 | 1689 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1690 | { |
9dfc6e68 | 1691 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1692 | |
dfb4f096 CL |
1693 | if (likely(c && c->page)) |
1694 | flush_slab(s, c); | |
81819f0f CL |
1695 | } |
1696 | ||
1697 | static void flush_cpu_slab(void *d) | |
1698 | { | |
1699 | struct kmem_cache *s = d; | |
81819f0f | 1700 | |
dfb4f096 | 1701 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1702 | } |
1703 | ||
1704 | static void flush_all(struct kmem_cache *s) | |
1705 | { | |
15c8b6c1 | 1706 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1707 | } |
1708 | ||
dfb4f096 CL |
1709 | /* |
1710 | * Check if the objects in a per cpu structure fit numa | |
1711 | * locality expectations. | |
1712 | */ | |
1713 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1714 | { | |
1715 | #ifdef CONFIG_NUMA | |
2154a336 | 1716 | if (node != NUMA_NO_NODE && c->node != node) |
dfb4f096 CL |
1717 | return 0; |
1718 | #endif | |
1719 | return 1; | |
1720 | } | |
1721 | ||
781b2ba6 PE |
1722 | static int count_free(struct page *page) |
1723 | { | |
1724 | return page->objects - page->inuse; | |
1725 | } | |
1726 | ||
1727 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1728 | int (*get_count)(struct page *)) | |
1729 | { | |
1730 | unsigned long flags; | |
1731 | unsigned long x = 0; | |
1732 | struct page *page; | |
1733 | ||
1734 | spin_lock_irqsave(&n->list_lock, flags); | |
1735 | list_for_each_entry(page, &n->partial, lru) | |
1736 | x += get_count(page); | |
1737 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1738 | return x; | |
1739 | } | |
1740 | ||
26c02cf0 AB |
1741 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1742 | { | |
1743 | #ifdef CONFIG_SLUB_DEBUG | |
1744 | return atomic_long_read(&n->total_objects); | |
1745 | #else | |
1746 | return 0; | |
1747 | #endif | |
1748 | } | |
1749 | ||
781b2ba6 PE |
1750 | static noinline void |
1751 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1752 | { | |
1753 | int node; | |
1754 | ||
1755 | printk(KERN_WARNING | |
1756 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1757 | nid, gfpflags); | |
1758 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1759 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1760 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1761 | ||
fa5ec8a1 DR |
1762 | if (oo_order(s->min) > get_order(s->objsize)) |
1763 | printk(KERN_WARNING " %s debugging increased min order, use " | |
1764 | "slub_debug=O to disable.\n", s->name); | |
1765 | ||
781b2ba6 PE |
1766 | for_each_online_node(node) { |
1767 | struct kmem_cache_node *n = get_node(s, node); | |
1768 | unsigned long nr_slabs; | |
1769 | unsigned long nr_objs; | |
1770 | unsigned long nr_free; | |
1771 | ||
1772 | if (!n) | |
1773 | continue; | |
1774 | ||
26c02cf0 AB |
1775 | nr_free = count_partial(n, count_free); |
1776 | nr_slabs = node_nr_slabs(n); | |
1777 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1778 | |
1779 | printk(KERN_WARNING | |
1780 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1781 | node, nr_slabs, nr_objs, nr_free); | |
1782 | } | |
1783 | } | |
1784 | ||
81819f0f | 1785 | /* |
894b8788 CL |
1786 | * Slow path. The lockless freelist is empty or we need to perform |
1787 | * debugging duties. | |
1788 | * | |
1789 | * Interrupts are disabled. | |
81819f0f | 1790 | * |
894b8788 CL |
1791 | * Processing is still very fast if new objects have been freed to the |
1792 | * regular freelist. In that case we simply take over the regular freelist | |
1793 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1794 | * |
894b8788 CL |
1795 | * If that is not working then we fall back to the partial lists. We take the |
1796 | * first element of the freelist as the object to allocate now and move the | |
1797 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1798 | * |
894b8788 | 1799 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1800 | * we need to allocate a new slab. This is the slowest path since it involves |
1801 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1802 | */ |
ce71e27c EGM |
1803 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1804 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1805 | { |
81819f0f | 1806 | void **object; |
01ad8a7b | 1807 | struct page *page; |
8a5ec0ba CL |
1808 | unsigned long flags; |
1809 | ||
1810 | local_irq_save(flags); | |
1811 | #ifdef CONFIG_PREEMPT | |
1812 | /* | |
1813 | * We may have been preempted and rescheduled on a different | |
1814 | * cpu before disabling interrupts. Need to reload cpu area | |
1815 | * pointer. | |
1816 | */ | |
1817 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 1818 | #endif |
81819f0f | 1819 | |
e72e9c23 LT |
1820 | /* We handle __GFP_ZERO in the caller */ |
1821 | gfpflags &= ~__GFP_ZERO; | |
1822 | ||
01ad8a7b CL |
1823 | page = c->page; |
1824 | if (!page) | |
81819f0f CL |
1825 | goto new_slab; |
1826 | ||
01ad8a7b | 1827 | slab_lock(page); |
dfb4f096 | 1828 | if (unlikely(!node_match(c, node))) |
81819f0f | 1829 | goto another_slab; |
6446faa2 | 1830 | |
84e554e6 | 1831 | stat(s, ALLOC_REFILL); |
6446faa2 | 1832 | |
894b8788 | 1833 | load_freelist: |
8cb0a506 CL |
1834 | VM_BUG_ON(!page->frozen); |
1835 | ||
01ad8a7b | 1836 | object = page->freelist; |
a973e9dd | 1837 | if (unlikely(!object)) |
81819f0f | 1838 | goto another_slab; |
af537b0a | 1839 | if (kmem_cache_debug(s)) |
81819f0f CL |
1840 | goto debug; |
1841 | ||
ff12059e | 1842 | c->freelist = get_freepointer(s, object); |
01ad8a7b CL |
1843 | page->inuse = page->objects; |
1844 | page->freelist = NULL; | |
01ad8a7b | 1845 | |
01ad8a7b | 1846 | slab_unlock(page); |
8a5ec0ba CL |
1847 | c->tid = next_tid(c->tid); |
1848 | local_irq_restore(flags); | |
84e554e6 | 1849 | stat(s, ALLOC_SLOWPATH); |
81819f0f CL |
1850 | return object; |
1851 | ||
1852 | another_slab: | |
dfb4f096 | 1853 | deactivate_slab(s, c); |
81819f0f CL |
1854 | |
1855 | new_slab: | |
01ad8a7b CL |
1856 | page = get_partial(s, gfpflags, node); |
1857 | if (page) { | |
84e554e6 | 1858 | stat(s, ALLOC_FROM_PARTIAL); |
8cb0a506 | 1859 | page->frozen = 1; |
dc1fb7f4 CL |
1860 | c->node = page_to_nid(page); |
1861 | c->page = page; | |
894b8788 | 1862 | goto load_freelist; |
81819f0f CL |
1863 | } |
1864 | ||
01ad8a7b | 1865 | page = new_slab(s, gfpflags, node); |
b811c202 | 1866 | |
01ad8a7b | 1867 | if (page) { |
9dfc6e68 | 1868 | c = __this_cpu_ptr(s->cpu_slab); |
84e554e6 | 1869 | stat(s, ALLOC_SLAB); |
05aa3450 | 1870 | if (c->page) |
dfb4f096 | 1871 | flush_slab(s, c); |
01ad8a7b CL |
1872 | |
1873 | slab_lock(page); | |
50d5c41c | 1874 | page->frozen = 1; |
bd07d87f DR |
1875 | c->node = page_to_nid(page); |
1876 | c->page = page; | |
4b6f0750 | 1877 | goto load_freelist; |
81819f0f | 1878 | } |
95f85989 PE |
1879 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
1880 | slab_out_of_memory(s, gfpflags, node); | |
2fd66c51 | 1881 | local_irq_restore(flags); |
71c7a06f | 1882 | return NULL; |
81819f0f | 1883 | debug: |
01ad8a7b | 1884 | if (!alloc_debug_processing(s, page, object, addr)) |
81819f0f | 1885 | goto another_slab; |
894b8788 | 1886 | |
01ad8a7b CL |
1887 | page->inuse++; |
1888 | page->freelist = get_freepointer(s, object); | |
442b06bc CL |
1889 | deactivate_slab(s, c); |
1890 | c->page = NULL; | |
15b7c514 | 1891 | c->node = NUMA_NO_NODE; |
a71ae47a CL |
1892 | local_irq_restore(flags); |
1893 | return object; | |
894b8788 CL |
1894 | } |
1895 | ||
1896 | /* | |
1897 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1898 | * have the fastpath folded into their functions. So no function call | |
1899 | * overhead for requests that can be satisfied on the fastpath. | |
1900 | * | |
1901 | * The fastpath works by first checking if the lockless freelist can be used. | |
1902 | * If not then __slab_alloc is called for slow processing. | |
1903 | * | |
1904 | * Otherwise we can simply pick the next object from the lockless free list. | |
1905 | */ | |
06428780 | 1906 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1907 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1908 | { |
894b8788 | 1909 | void **object; |
dfb4f096 | 1910 | struct kmem_cache_cpu *c; |
8a5ec0ba | 1911 | unsigned long tid; |
1f84260c | 1912 | |
c016b0bd | 1913 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 1914 | return NULL; |
1f84260c | 1915 | |
8a5ec0ba | 1916 | redo: |
8a5ec0ba CL |
1917 | |
1918 | /* | |
1919 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
1920 | * enabled. We may switch back and forth between cpus while | |
1921 | * reading from one cpu area. That does not matter as long | |
1922 | * as we end up on the original cpu again when doing the cmpxchg. | |
1923 | */ | |
9dfc6e68 | 1924 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 1925 | |
8a5ec0ba CL |
1926 | /* |
1927 | * The transaction ids are globally unique per cpu and per operation on | |
1928 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
1929 | * occurs on the right processor and that there was no operation on the | |
1930 | * linked list in between. | |
1931 | */ | |
1932 | tid = c->tid; | |
1933 | barrier(); | |
8a5ec0ba | 1934 | |
9dfc6e68 | 1935 | object = c->freelist; |
9dfc6e68 | 1936 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 1937 | |
dfb4f096 | 1938 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1939 | |
1940 | else { | |
8a5ec0ba | 1941 | /* |
25985edc | 1942 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
1943 | * operation and if we are on the right processor. |
1944 | * | |
1945 | * The cmpxchg does the following atomically (without lock semantics!) | |
1946 | * 1. Relocate first pointer to the current per cpu area. | |
1947 | * 2. Verify that tid and freelist have not been changed | |
1948 | * 3. If they were not changed replace tid and freelist | |
1949 | * | |
1950 | * Since this is without lock semantics the protection is only against | |
1951 | * code executing on this cpu *not* from access by other cpus. | |
1952 | */ | |
30106b8c | 1953 | if (unlikely(!irqsafe_cpu_cmpxchg_double( |
8a5ec0ba CL |
1954 | s->cpu_slab->freelist, s->cpu_slab->tid, |
1955 | object, tid, | |
1393d9a1 | 1956 | get_freepointer_safe(s, object), next_tid(tid)))) { |
8a5ec0ba CL |
1957 | |
1958 | note_cmpxchg_failure("slab_alloc", s, tid); | |
1959 | goto redo; | |
1960 | } | |
84e554e6 | 1961 | stat(s, ALLOC_FASTPATH); |
894b8788 | 1962 | } |
8a5ec0ba | 1963 | |
74e2134f | 1964 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 1965 | memset(object, 0, s->objsize); |
d07dbea4 | 1966 | |
c016b0bd | 1967 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 1968 | |
894b8788 | 1969 | return object; |
81819f0f CL |
1970 | } |
1971 | ||
1972 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1973 | { | |
2154a336 | 1974 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 1975 | |
ca2b84cb | 1976 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1977 | |
1978 | return ret; | |
81819f0f CL |
1979 | } |
1980 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1981 | ||
0f24f128 | 1982 | #ifdef CONFIG_TRACING |
4a92379b RK |
1983 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
1984 | { | |
1985 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
1986 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
1987 | return ret; | |
1988 | } | |
1989 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
1990 | ||
1991 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 1992 | { |
4a92379b RK |
1993 | void *ret = kmalloc_order(size, flags, order); |
1994 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1995 | return ret; | |
5b882be4 | 1996 | } |
4a92379b | 1997 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
1998 | #endif |
1999 | ||
81819f0f CL |
2000 | #ifdef CONFIG_NUMA |
2001 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2002 | { | |
5b882be4 EGM |
2003 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2004 | ||
ca2b84cb EGM |
2005 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
2006 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
2007 | |
2008 | return ret; | |
81819f0f CL |
2009 | } |
2010 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2011 | |
0f24f128 | 2012 | #ifdef CONFIG_TRACING |
4a92379b | 2013 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2014 | gfp_t gfpflags, |
4a92379b | 2015 | int node, size_t size) |
5b882be4 | 2016 | { |
4a92379b RK |
2017 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2018 | ||
2019 | trace_kmalloc_node(_RET_IP_, ret, | |
2020 | size, s->size, gfpflags, node); | |
2021 | return ret; | |
5b882be4 | 2022 | } |
4a92379b | 2023 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2024 | #endif |
5d1f57e4 | 2025 | #endif |
5b882be4 | 2026 | |
81819f0f | 2027 | /* |
894b8788 CL |
2028 | * Slow patch handling. This may still be called frequently since objects |
2029 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2030 | * |
894b8788 CL |
2031 | * So we still attempt to reduce cache line usage. Just take the slab |
2032 | * lock and free the item. If there is no additional partial page | |
2033 | * handling required then we can return immediately. | |
81819f0f | 2034 | */ |
894b8788 | 2035 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2036 | void *x, unsigned long addr) |
81819f0f CL |
2037 | { |
2038 | void *prior; | |
2039 | void **object = (void *)x; | |
8a5ec0ba | 2040 | unsigned long flags; |
81819f0f | 2041 | |
8a5ec0ba | 2042 | local_irq_save(flags); |
81819f0f | 2043 | slab_lock(page); |
8a5ec0ba | 2044 | stat(s, FREE_SLOWPATH); |
81819f0f | 2045 | |
8dc16c6c CL |
2046 | if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr)) |
2047 | goto out_unlock; | |
6446faa2 | 2048 | |
ff12059e CL |
2049 | prior = page->freelist; |
2050 | set_freepointer(s, object, prior); | |
81819f0f CL |
2051 | page->freelist = object; |
2052 | page->inuse--; | |
2053 | ||
50d5c41c | 2054 | if (unlikely(page->frozen)) { |
84e554e6 | 2055 | stat(s, FREE_FROZEN); |
81819f0f | 2056 | goto out_unlock; |
8ff12cfc | 2057 | } |
81819f0f CL |
2058 | |
2059 | if (unlikely(!page->inuse)) | |
2060 | goto slab_empty; | |
2061 | ||
2062 | /* | |
6446faa2 | 2063 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
2064 | * then add it. |
2065 | */ | |
a973e9dd | 2066 | if (unlikely(!prior)) { |
7c2e132c | 2067 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
84e554e6 | 2068 | stat(s, FREE_ADD_PARTIAL); |
8ff12cfc | 2069 | } |
81819f0f CL |
2070 | |
2071 | out_unlock: | |
2072 | slab_unlock(page); | |
8a5ec0ba | 2073 | local_irq_restore(flags); |
81819f0f CL |
2074 | return; |
2075 | ||
2076 | slab_empty: | |
a973e9dd | 2077 | if (prior) { |
81819f0f | 2078 | /* |
672bba3a | 2079 | * Slab still on the partial list. |
81819f0f CL |
2080 | */ |
2081 | remove_partial(s, page); | |
84e554e6 | 2082 | stat(s, FREE_REMOVE_PARTIAL); |
8ff12cfc | 2083 | } |
81819f0f | 2084 | slab_unlock(page); |
8a5ec0ba | 2085 | local_irq_restore(flags); |
84e554e6 | 2086 | stat(s, FREE_SLAB); |
81819f0f | 2087 | discard_slab(s, page); |
81819f0f CL |
2088 | } |
2089 | ||
894b8788 CL |
2090 | /* |
2091 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2092 | * can perform fastpath freeing without additional function calls. | |
2093 | * | |
2094 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2095 | * of this processor. This typically the case if we have just allocated | |
2096 | * the item before. | |
2097 | * | |
2098 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2099 | * with all sorts of special processing. | |
2100 | */ | |
06428780 | 2101 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2102 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2103 | { |
2104 | void **object = (void *)x; | |
dfb4f096 | 2105 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2106 | unsigned long tid; |
1f84260c | 2107 | |
c016b0bd CL |
2108 | slab_free_hook(s, x); |
2109 | ||
8a5ec0ba | 2110 | redo: |
a24c5a0e | 2111 | |
8a5ec0ba CL |
2112 | /* |
2113 | * Determine the currently cpus per cpu slab. | |
2114 | * The cpu may change afterward. However that does not matter since | |
2115 | * data is retrieved via this pointer. If we are on the same cpu | |
2116 | * during the cmpxchg then the free will succedd. | |
2117 | */ | |
9dfc6e68 | 2118 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2119 | |
8a5ec0ba CL |
2120 | tid = c->tid; |
2121 | barrier(); | |
c016b0bd | 2122 | |
442b06bc | 2123 | if (likely(page == c->page)) { |
ff12059e | 2124 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2125 | |
30106b8c | 2126 | if (unlikely(!irqsafe_cpu_cmpxchg_double( |
8a5ec0ba CL |
2127 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2128 | c->freelist, tid, | |
2129 | object, next_tid(tid)))) { | |
2130 | ||
2131 | note_cmpxchg_failure("slab_free", s, tid); | |
2132 | goto redo; | |
2133 | } | |
84e554e6 | 2134 | stat(s, FREE_FASTPATH); |
894b8788 | 2135 | } else |
ff12059e | 2136 | __slab_free(s, page, x, addr); |
894b8788 | 2137 | |
894b8788 CL |
2138 | } |
2139 | ||
81819f0f CL |
2140 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2141 | { | |
77c5e2d0 | 2142 | struct page *page; |
81819f0f | 2143 | |
b49af68f | 2144 | page = virt_to_head_page(x); |
81819f0f | 2145 | |
ce71e27c | 2146 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 2147 | |
ca2b84cb | 2148 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2149 | } |
2150 | EXPORT_SYMBOL(kmem_cache_free); | |
2151 | ||
81819f0f | 2152 | /* |
672bba3a CL |
2153 | * Object placement in a slab is made very easy because we always start at |
2154 | * offset 0. If we tune the size of the object to the alignment then we can | |
2155 | * get the required alignment by putting one properly sized object after | |
2156 | * another. | |
81819f0f CL |
2157 | * |
2158 | * Notice that the allocation order determines the sizes of the per cpu | |
2159 | * caches. Each processor has always one slab available for allocations. | |
2160 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2161 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2162 | * locking overhead. |
81819f0f CL |
2163 | */ |
2164 | ||
2165 | /* | |
2166 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2167 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2168 | * and increases the number of allocations possible without having to | |
2169 | * take the list_lock. | |
2170 | */ | |
2171 | static int slub_min_order; | |
114e9e89 | 2172 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2173 | static int slub_min_objects; |
81819f0f CL |
2174 | |
2175 | /* | |
2176 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2177 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2178 | */ |
2179 | static int slub_nomerge; | |
2180 | ||
81819f0f CL |
2181 | /* |
2182 | * Calculate the order of allocation given an slab object size. | |
2183 | * | |
672bba3a CL |
2184 | * The order of allocation has significant impact on performance and other |
2185 | * system components. Generally order 0 allocations should be preferred since | |
2186 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2187 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2188 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2189 | * would be wasted. |
2190 | * | |
2191 | * In order to reach satisfactory performance we must ensure that a minimum | |
2192 | * number of objects is in one slab. Otherwise we may generate too much | |
2193 | * activity on the partial lists which requires taking the list_lock. This is | |
2194 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2195 | * |
672bba3a CL |
2196 | * slub_max_order specifies the order where we begin to stop considering the |
2197 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2198 | * we try to keep the page order as low as possible. So we accept more waste | |
2199 | * of space in favor of a small page order. | |
81819f0f | 2200 | * |
672bba3a CL |
2201 | * Higher order allocations also allow the placement of more objects in a |
2202 | * slab and thereby reduce object handling overhead. If the user has | |
2203 | * requested a higher mininum order then we start with that one instead of | |
2204 | * the smallest order which will fit the object. | |
81819f0f | 2205 | */ |
5e6d444e | 2206 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2207 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2208 | { |
2209 | int order; | |
2210 | int rem; | |
6300ea75 | 2211 | int min_order = slub_min_order; |
81819f0f | 2212 | |
ab9a0f19 | 2213 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2214 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2215 | |
6300ea75 | 2216 | for (order = max(min_order, |
5e6d444e CL |
2217 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2218 | order <= max_order; order++) { | |
81819f0f | 2219 | |
5e6d444e | 2220 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2221 | |
ab9a0f19 | 2222 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2223 | continue; |
2224 | ||
ab9a0f19 | 2225 | rem = (slab_size - reserved) % size; |
81819f0f | 2226 | |
5e6d444e | 2227 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2228 | break; |
2229 | ||
2230 | } | |
672bba3a | 2231 | |
81819f0f CL |
2232 | return order; |
2233 | } | |
2234 | ||
ab9a0f19 | 2235 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2236 | { |
2237 | int order; | |
2238 | int min_objects; | |
2239 | int fraction; | |
e8120ff1 | 2240 | int max_objects; |
5e6d444e CL |
2241 | |
2242 | /* | |
2243 | * Attempt to find best configuration for a slab. This | |
2244 | * works by first attempting to generate a layout with | |
2245 | * the best configuration and backing off gradually. | |
2246 | * | |
2247 | * First we reduce the acceptable waste in a slab. Then | |
2248 | * we reduce the minimum objects required in a slab. | |
2249 | */ | |
2250 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2251 | if (!min_objects) |
2252 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2253 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2254 | min_objects = min(min_objects, max_objects); |
2255 | ||
5e6d444e | 2256 | while (min_objects > 1) { |
c124f5b5 | 2257 | fraction = 16; |
5e6d444e CL |
2258 | while (fraction >= 4) { |
2259 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2260 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2261 | if (order <= slub_max_order) |
2262 | return order; | |
2263 | fraction /= 2; | |
2264 | } | |
5086c389 | 2265 | min_objects--; |
5e6d444e CL |
2266 | } |
2267 | ||
2268 | /* | |
2269 | * We were unable to place multiple objects in a slab. Now | |
2270 | * lets see if we can place a single object there. | |
2271 | */ | |
ab9a0f19 | 2272 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2273 | if (order <= slub_max_order) |
2274 | return order; | |
2275 | ||
2276 | /* | |
2277 | * Doh this slab cannot be placed using slub_max_order. | |
2278 | */ | |
ab9a0f19 | 2279 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2280 | if (order < MAX_ORDER) |
5e6d444e CL |
2281 | return order; |
2282 | return -ENOSYS; | |
2283 | } | |
2284 | ||
81819f0f | 2285 | /* |
672bba3a | 2286 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2287 | */ |
2288 | static unsigned long calculate_alignment(unsigned long flags, | |
2289 | unsigned long align, unsigned long size) | |
2290 | { | |
2291 | /* | |
6446faa2 CL |
2292 | * If the user wants hardware cache aligned objects then follow that |
2293 | * suggestion if the object is sufficiently large. | |
81819f0f | 2294 | * |
6446faa2 CL |
2295 | * The hardware cache alignment cannot override the specified |
2296 | * alignment though. If that is greater then use it. | |
81819f0f | 2297 | */ |
b6210386 NP |
2298 | if (flags & SLAB_HWCACHE_ALIGN) { |
2299 | unsigned long ralign = cache_line_size(); | |
2300 | while (size <= ralign / 2) | |
2301 | ralign /= 2; | |
2302 | align = max(align, ralign); | |
2303 | } | |
81819f0f CL |
2304 | |
2305 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2306 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2307 | |
2308 | return ALIGN(align, sizeof(void *)); | |
2309 | } | |
2310 | ||
5595cffc PE |
2311 | static void |
2312 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2313 | { |
2314 | n->nr_partial = 0; | |
81819f0f CL |
2315 | spin_lock_init(&n->list_lock); |
2316 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2317 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2318 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2319 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2320 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2321 | #endif |
81819f0f CL |
2322 | } |
2323 | ||
55136592 | 2324 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2325 | { |
6c182dc0 CL |
2326 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2327 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2328 | |
8a5ec0ba | 2329 | /* |
d4d84fef CM |
2330 | * Must align to double word boundary for the double cmpxchg |
2331 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2332 | */ |
d4d84fef CM |
2333 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2334 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2335 | |
2336 | if (!s->cpu_slab) | |
2337 | return 0; | |
2338 | ||
2339 | init_kmem_cache_cpus(s); | |
4c93c355 | 2340 | |
8a5ec0ba | 2341 | return 1; |
4c93c355 | 2342 | } |
4c93c355 | 2343 | |
51df1142 CL |
2344 | static struct kmem_cache *kmem_cache_node; |
2345 | ||
81819f0f CL |
2346 | /* |
2347 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2348 | * slab on the node for this slabcache. There are no concurrent accesses | |
2349 | * possible. | |
2350 | * | |
2351 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2352 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2353 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2354 | */ |
55136592 | 2355 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2356 | { |
2357 | struct page *page; | |
2358 | struct kmem_cache_node *n; | |
ba84c73c | 2359 | unsigned long flags; |
81819f0f | 2360 | |
51df1142 | 2361 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2362 | |
51df1142 | 2363 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2364 | |
2365 | BUG_ON(!page); | |
a2f92ee7 CL |
2366 | if (page_to_nid(page) != node) { |
2367 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2368 | "node %d\n", node); | |
2369 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2370 | "in order to be able to continue\n"); | |
2371 | } | |
2372 | ||
81819f0f CL |
2373 | n = page->freelist; |
2374 | BUG_ON(!n); | |
51df1142 | 2375 | page->freelist = get_freepointer(kmem_cache_node, n); |
81819f0f | 2376 | page->inuse++; |
8cb0a506 | 2377 | page->frozen = 0; |
51df1142 | 2378 | kmem_cache_node->node[node] = n; |
8ab1372f | 2379 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2380 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2381 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2382 | #endif |
51df1142 CL |
2383 | init_kmem_cache_node(n, kmem_cache_node); |
2384 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
6446faa2 | 2385 | |
ba84c73c | 2386 | /* |
2387 | * lockdep requires consistent irq usage for each lock | |
2388 | * so even though there cannot be a race this early in | |
2389 | * the boot sequence, we still disable irqs. | |
2390 | */ | |
2391 | local_irq_save(flags); | |
7c2e132c | 2392 | add_partial(n, page, 0); |
ba84c73c | 2393 | local_irq_restore(flags); |
81819f0f CL |
2394 | } |
2395 | ||
2396 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2397 | { | |
2398 | int node; | |
2399 | ||
f64dc58c | 2400 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2401 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2402 | |
73367bd8 | 2403 | if (n) |
51df1142 CL |
2404 | kmem_cache_free(kmem_cache_node, n); |
2405 | ||
81819f0f CL |
2406 | s->node[node] = NULL; |
2407 | } | |
2408 | } | |
2409 | ||
55136592 | 2410 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2411 | { |
2412 | int node; | |
81819f0f | 2413 | |
f64dc58c | 2414 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2415 | struct kmem_cache_node *n; |
2416 | ||
73367bd8 | 2417 | if (slab_state == DOWN) { |
55136592 | 2418 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2419 | continue; |
2420 | } | |
51df1142 | 2421 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2422 | GFP_KERNEL, node); |
81819f0f | 2423 | |
73367bd8 AD |
2424 | if (!n) { |
2425 | free_kmem_cache_nodes(s); | |
2426 | return 0; | |
81819f0f | 2427 | } |
73367bd8 | 2428 | |
81819f0f | 2429 | s->node[node] = n; |
5595cffc | 2430 | init_kmem_cache_node(n, s); |
81819f0f CL |
2431 | } |
2432 | return 1; | |
2433 | } | |
81819f0f | 2434 | |
c0bdb232 | 2435 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2436 | { |
2437 | if (min < MIN_PARTIAL) | |
2438 | min = MIN_PARTIAL; | |
2439 | else if (min > MAX_PARTIAL) | |
2440 | min = MAX_PARTIAL; | |
2441 | s->min_partial = min; | |
2442 | } | |
2443 | ||
81819f0f CL |
2444 | /* |
2445 | * calculate_sizes() determines the order and the distribution of data within | |
2446 | * a slab object. | |
2447 | */ | |
06b285dc | 2448 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2449 | { |
2450 | unsigned long flags = s->flags; | |
2451 | unsigned long size = s->objsize; | |
2452 | unsigned long align = s->align; | |
834f3d11 | 2453 | int order; |
81819f0f | 2454 | |
d8b42bf5 CL |
2455 | /* |
2456 | * Round up object size to the next word boundary. We can only | |
2457 | * place the free pointer at word boundaries and this determines | |
2458 | * the possible location of the free pointer. | |
2459 | */ | |
2460 | size = ALIGN(size, sizeof(void *)); | |
2461 | ||
2462 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2463 | /* |
2464 | * Determine if we can poison the object itself. If the user of | |
2465 | * the slab may touch the object after free or before allocation | |
2466 | * then we should never poison the object itself. | |
2467 | */ | |
2468 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2469 | !s->ctor) |
81819f0f CL |
2470 | s->flags |= __OBJECT_POISON; |
2471 | else | |
2472 | s->flags &= ~__OBJECT_POISON; | |
2473 | ||
81819f0f CL |
2474 | |
2475 | /* | |
672bba3a | 2476 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2477 | * end of the object and the free pointer. If not then add an |
672bba3a | 2478 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2479 | */ |
2480 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2481 | size += sizeof(void *); | |
41ecc55b | 2482 | #endif |
81819f0f CL |
2483 | |
2484 | /* | |
672bba3a CL |
2485 | * With that we have determined the number of bytes in actual use |
2486 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2487 | */ |
2488 | s->inuse = size; | |
2489 | ||
2490 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2491 | s->ctor)) { |
81819f0f CL |
2492 | /* |
2493 | * Relocate free pointer after the object if it is not | |
2494 | * permitted to overwrite the first word of the object on | |
2495 | * kmem_cache_free. | |
2496 | * | |
2497 | * This is the case if we do RCU, have a constructor or | |
2498 | * destructor or are poisoning the objects. | |
2499 | */ | |
2500 | s->offset = size; | |
2501 | size += sizeof(void *); | |
2502 | } | |
2503 | ||
c12b3c62 | 2504 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2505 | if (flags & SLAB_STORE_USER) |
2506 | /* | |
2507 | * Need to store information about allocs and frees after | |
2508 | * the object. | |
2509 | */ | |
2510 | size += 2 * sizeof(struct track); | |
2511 | ||
be7b3fbc | 2512 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2513 | /* |
2514 | * Add some empty padding so that we can catch | |
2515 | * overwrites from earlier objects rather than let | |
2516 | * tracking information or the free pointer be | |
0211a9c8 | 2517 | * corrupted if a user writes before the start |
81819f0f CL |
2518 | * of the object. |
2519 | */ | |
2520 | size += sizeof(void *); | |
41ecc55b | 2521 | #endif |
672bba3a | 2522 | |
81819f0f CL |
2523 | /* |
2524 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2525 | * user specified and the dynamic determination of cache line size |
2526 | * on bootup. | |
81819f0f CL |
2527 | */ |
2528 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2529 | s->align = align; |
81819f0f CL |
2530 | |
2531 | /* | |
2532 | * SLUB stores one object immediately after another beginning from | |
2533 | * offset 0. In order to align the objects we have to simply size | |
2534 | * each object to conform to the alignment. | |
2535 | */ | |
2536 | size = ALIGN(size, align); | |
2537 | s->size = size; | |
06b285dc CL |
2538 | if (forced_order >= 0) |
2539 | order = forced_order; | |
2540 | else | |
ab9a0f19 | 2541 | order = calculate_order(size, s->reserved); |
81819f0f | 2542 | |
834f3d11 | 2543 | if (order < 0) |
81819f0f CL |
2544 | return 0; |
2545 | ||
b7a49f0d | 2546 | s->allocflags = 0; |
834f3d11 | 2547 | if (order) |
b7a49f0d CL |
2548 | s->allocflags |= __GFP_COMP; |
2549 | ||
2550 | if (s->flags & SLAB_CACHE_DMA) | |
2551 | s->allocflags |= SLUB_DMA; | |
2552 | ||
2553 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2554 | s->allocflags |= __GFP_RECLAIMABLE; | |
2555 | ||
81819f0f CL |
2556 | /* |
2557 | * Determine the number of objects per slab | |
2558 | */ | |
ab9a0f19 LJ |
2559 | s->oo = oo_make(order, size, s->reserved); |
2560 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
2561 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2562 | s->max = s->oo; | |
81819f0f | 2563 | |
834f3d11 | 2564 | return !!oo_objects(s->oo); |
81819f0f CL |
2565 | |
2566 | } | |
2567 | ||
55136592 | 2568 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
2569 | const char *name, size_t size, |
2570 | size_t align, unsigned long flags, | |
51cc5068 | 2571 | void (*ctor)(void *)) |
81819f0f CL |
2572 | { |
2573 | memset(s, 0, kmem_size); | |
2574 | s->name = name; | |
2575 | s->ctor = ctor; | |
81819f0f | 2576 | s->objsize = size; |
81819f0f | 2577 | s->align = align; |
ba0268a8 | 2578 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
ab9a0f19 | 2579 | s->reserved = 0; |
81819f0f | 2580 | |
da9a638c LJ |
2581 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
2582 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 2583 | |
06b285dc | 2584 | if (!calculate_sizes(s, -1)) |
81819f0f | 2585 | goto error; |
3de47213 DR |
2586 | if (disable_higher_order_debug) { |
2587 | /* | |
2588 | * Disable debugging flags that store metadata if the min slab | |
2589 | * order increased. | |
2590 | */ | |
2591 | if (get_order(s->size) > get_order(s->objsize)) { | |
2592 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2593 | s->offset = 0; | |
2594 | if (!calculate_sizes(s, -1)) | |
2595 | goto error; | |
2596 | } | |
2597 | } | |
81819f0f | 2598 | |
3b89d7d8 DR |
2599 | /* |
2600 | * The larger the object size is, the more pages we want on the partial | |
2601 | * list to avoid pounding the page allocator excessively. | |
2602 | */ | |
c0bdb232 | 2603 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2604 | s->refcount = 1; |
2605 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2606 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2607 | #endif |
55136592 | 2608 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 2609 | goto error; |
81819f0f | 2610 | |
55136592 | 2611 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 2612 | return 1; |
ff12059e | 2613 | |
4c93c355 | 2614 | free_kmem_cache_nodes(s); |
81819f0f CL |
2615 | error: |
2616 | if (flags & SLAB_PANIC) | |
2617 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2618 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2619 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2620 | s->offset, flags); |
2621 | return 0; | |
2622 | } | |
81819f0f | 2623 | |
81819f0f CL |
2624 | /* |
2625 | * Determine the size of a slab object | |
2626 | */ | |
2627 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2628 | { | |
2629 | return s->objsize; | |
2630 | } | |
2631 | EXPORT_SYMBOL(kmem_cache_size); | |
2632 | ||
33b12c38 CL |
2633 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2634 | const char *text) | |
2635 | { | |
2636 | #ifdef CONFIG_SLUB_DEBUG | |
2637 | void *addr = page_address(page); | |
2638 | void *p; | |
a5dd5c11 NK |
2639 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
2640 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
2641 | if (!map) |
2642 | return; | |
33b12c38 CL |
2643 | slab_err(s, page, "%s", text); |
2644 | slab_lock(page); | |
33b12c38 | 2645 | |
5f80b13a | 2646 | get_map(s, page, map); |
33b12c38 CL |
2647 | for_each_object(p, s, addr, page->objects) { |
2648 | ||
2649 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2650 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2651 | p, p - addr); | |
2652 | print_tracking(s, p); | |
2653 | } | |
2654 | } | |
2655 | slab_unlock(page); | |
bbd7d57b | 2656 | kfree(map); |
33b12c38 CL |
2657 | #endif |
2658 | } | |
2659 | ||
81819f0f | 2660 | /* |
599870b1 | 2661 | * Attempt to free all partial slabs on a node. |
81819f0f | 2662 | */ |
599870b1 | 2663 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2664 | { |
81819f0f CL |
2665 | unsigned long flags; |
2666 | struct page *page, *h; | |
2667 | ||
2668 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2669 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 2670 | if (!page->inuse) { |
62e346a8 | 2671 | __remove_partial(n, page); |
81819f0f | 2672 | discard_slab(s, page); |
33b12c38 CL |
2673 | } else { |
2674 | list_slab_objects(s, page, | |
2675 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2676 | } |
33b12c38 | 2677 | } |
81819f0f | 2678 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2679 | } |
2680 | ||
2681 | /* | |
672bba3a | 2682 | * Release all resources used by a slab cache. |
81819f0f | 2683 | */ |
0c710013 | 2684 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2685 | { |
2686 | int node; | |
2687 | ||
2688 | flush_all(s); | |
9dfc6e68 | 2689 | free_percpu(s->cpu_slab); |
81819f0f | 2690 | /* Attempt to free all objects */ |
f64dc58c | 2691 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2692 | struct kmem_cache_node *n = get_node(s, node); |
2693 | ||
599870b1 CL |
2694 | free_partial(s, n); |
2695 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2696 | return 1; |
2697 | } | |
2698 | free_kmem_cache_nodes(s); | |
2699 | return 0; | |
2700 | } | |
2701 | ||
2702 | /* | |
2703 | * Close a cache and release the kmem_cache structure | |
2704 | * (must be used for caches created using kmem_cache_create) | |
2705 | */ | |
2706 | void kmem_cache_destroy(struct kmem_cache *s) | |
2707 | { | |
2708 | down_write(&slub_lock); | |
2709 | s->refcount--; | |
2710 | if (!s->refcount) { | |
2711 | list_del(&s->list); | |
d629d819 PE |
2712 | if (kmem_cache_close(s)) { |
2713 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2714 | "still has objects.\n", s->name, __func__); | |
2715 | dump_stack(); | |
2716 | } | |
d76b1590 ED |
2717 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2718 | rcu_barrier(); | |
81819f0f | 2719 | sysfs_slab_remove(s); |
2bce6485 CL |
2720 | } |
2721 | up_write(&slub_lock); | |
81819f0f CL |
2722 | } |
2723 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2724 | ||
2725 | /******************************************************************** | |
2726 | * Kmalloc subsystem | |
2727 | *******************************************************************/ | |
2728 | ||
51df1142 | 2729 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
2730 | EXPORT_SYMBOL(kmalloc_caches); |
2731 | ||
51df1142 CL |
2732 | static struct kmem_cache *kmem_cache; |
2733 | ||
55136592 | 2734 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 2735 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
2736 | #endif |
2737 | ||
81819f0f CL |
2738 | static int __init setup_slub_min_order(char *str) |
2739 | { | |
06428780 | 2740 | get_option(&str, &slub_min_order); |
81819f0f CL |
2741 | |
2742 | return 1; | |
2743 | } | |
2744 | ||
2745 | __setup("slub_min_order=", setup_slub_min_order); | |
2746 | ||
2747 | static int __init setup_slub_max_order(char *str) | |
2748 | { | |
06428780 | 2749 | get_option(&str, &slub_max_order); |
818cf590 | 2750 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2751 | |
2752 | return 1; | |
2753 | } | |
2754 | ||
2755 | __setup("slub_max_order=", setup_slub_max_order); | |
2756 | ||
2757 | static int __init setup_slub_min_objects(char *str) | |
2758 | { | |
06428780 | 2759 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2760 | |
2761 | return 1; | |
2762 | } | |
2763 | ||
2764 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2765 | ||
2766 | static int __init setup_slub_nomerge(char *str) | |
2767 | { | |
2768 | slub_nomerge = 1; | |
2769 | return 1; | |
2770 | } | |
2771 | ||
2772 | __setup("slub_nomerge", setup_slub_nomerge); | |
2773 | ||
51df1142 CL |
2774 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
2775 | int size, unsigned int flags) | |
81819f0f | 2776 | { |
51df1142 CL |
2777 | struct kmem_cache *s; |
2778 | ||
2779 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
2780 | ||
83b519e8 PE |
2781 | /* |
2782 | * This function is called with IRQs disabled during early-boot on | |
2783 | * single CPU so there's no need to take slub_lock here. | |
2784 | */ | |
55136592 | 2785 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2786 | flags, NULL)) |
81819f0f CL |
2787 | goto panic; |
2788 | ||
2789 | list_add(&s->list, &slab_caches); | |
51df1142 | 2790 | return s; |
81819f0f CL |
2791 | |
2792 | panic: | |
2793 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 2794 | return NULL; |
81819f0f CL |
2795 | } |
2796 | ||
f1b26339 CL |
2797 | /* |
2798 | * Conversion table for small slabs sizes / 8 to the index in the | |
2799 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2800 | * of two cache sizes there. The size of larger slabs can be determined using | |
2801 | * fls. | |
2802 | */ | |
2803 | static s8 size_index[24] = { | |
2804 | 3, /* 8 */ | |
2805 | 4, /* 16 */ | |
2806 | 5, /* 24 */ | |
2807 | 5, /* 32 */ | |
2808 | 6, /* 40 */ | |
2809 | 6, /* 48 */ | |
2810 | 6, /* 56 */ | |
2811 | 6, /* 64 */ | |
2812 | 1, /* 72 */ | |
2813 | 1, /* 80 */ | |
2814 | 1, /* 88 */ | |
2815 | 1, /* 96 */ | |
2816 | 7, /* 104 */ | |
2817 | 7, /* 112 */ | |
2818 | 7, /* 120 */ | |
2819 | 7, /* 128 */ | |
2820 | 2, /* 136 */ | |
2821 | 2, /* 144 */ | |
2822 | 2, /* 152 */ | |
2823 | 2, /* 160 */ | |
2824 | 2, /* 168 */ | |
2825 | 2, /* 176 */ | |
2826 | 2, /* 184 */ | |
2827 | 2 /* 192 */ | |
2828 | }; | |
2829 | ||
acdfcd04 AK |
2830 | static inline int size_index_elem(size_t bytes) |
2831 | { | |
2832 | return (bytes - 1) / 8; | |
2833 | } | |
2834 | ||
81819f0f CL |
2835 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2836 | { | |
f1b26339 | 2837 | int index; |
81819f0f | 2838 | |
f1b26339 CL |
2839 | if (size <= 192) { |
2840 | if (!size) | |
2841 | return ZERO_SIZE_PTR; | |
81819f0f | 2842 | |
acdfcd04 | 2843 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 2844 | } else |
f1b26339 | 2845 | index = fls(size - 1); |
81819f0f CL |
2846 | |
2847 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2848 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 2849 | return kmalloc_dma_caches[index]; |
f1b26339 | 2850 | |
81819f0f | 2851 | #endif |
51df1142 | 2852 | return kmalloc_caches[index]; |
81819f0f CL |
2853 | } |
2854 | ||
2855 | void *__kmalloc(size_t size, gfp_t flags) | |
2856 | { | |
aadb4bc4 | 2857 | struct kmem_cache *s; |
5b882be4 | 2858 | void *ret; |
81819f0f | 2859 | |
ffadd4d0 | 2860 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2861 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2862 | |
2863 | s = get_slab(size, flags); | |
2864 | ||
2865 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2866 | return s; |
2867 | ||
2154a336 | 2868 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2869 | |
ca2b84cb | 2870 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2871 | |
2872 | return ret; | |
81819f0f CL |
2873 | } |
2874 | EXPORT_SYMBOL(__kmalloc); | |
2875 | ||
5d1f57e4 | 2876 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
2877 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2878 | { | |
b1eeab67 | 2879 | struct page *page; |
e4f7c0b4 | 2880 | void *ptr = NULL; |
f619cfe1 | 2881 | |
b1eeab67 VN |
2882 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2883 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 2884 | if (page) |
e4f7c0b4 CM |
2885 | ptr = page_address(page); |
2886 | ||
2887 | kmemleak_alloc(ptr, size, 1, flags); | |
2888 | return ptr; | |
f619cfe1 CL |
2889 | } |
2890 | ||
81819f0f CL |
2891 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
2892 | { | |
aadb4bc4 | 2893 | struct kmem_cache *s; |
5b882be4 | 2894 | void *ret; |
81819f0f | 2895 | |
057685cf | 2896 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2897 | ret = kmalloc_large_node(size, flags, node); |
2898 | ||
ca2b84cb EGM |
2899 | trace_kmalloc_node(_RET_IP_, ret, |
2900 | size, PAGE_SIZE << get_order(size), | |
2901 | flags, node); | |
5b882be4 EGM |
2902 | |
2903 | return ret; | |
2904 | } | |
aadb4bc4 CL |
2905 | |
2906 | s = get_slab(size, flags); | |
2907 | ||
2908 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2909 | return s; |
2910 | ||
5b882be4 EGM |
2911 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2912 | ||
ca2b84cb | 2913 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2914 | |
2915 | return ret; | |
81819f0f CL |
2916 | } |
2917 | EXPORT_SYMBOL(__kmalloc_node); | |
2918 | #endif | |
2919 | ||
2920 | size_t ksize(const void *object) | |
2921 | { | |
272c1d21 | 2922 | struct page *page; |
81819f0f | 2923 | |
ef8b4520 | 2924 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2925 | return 0; |
2926 | ||
294a80a8 | 2927 | page = virt_to_head_page(object); |
294a80a8 | 2928 | |
76994412 PE |
2929 | if (unlikely(!PageSlab(page))) { |
2930 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2931 | return PAGE_SIZE << compound_order(page); |
76994412 | 2932 | } |
81819f0f | 2933 | |
b3d41885 | 2934 | return slab_ksize(page->slab); |
81819f0f | 2935 | } |
b1aabecd | 2936 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2937 | |
2938 | void kfree(const void *x) | |
2939 | { | |
81819f0f | 2940 | struct page *page; |
5bb983b0 | 2941 | void *object = (void *)x; |
81819f0f | 2942 | |
2121db74 PE |
2943 | trace_kfree(_RET_IP_, x); |
2944 | ||
2408c550 | 2945 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2946 | return; |
2947 | ||
b49af68f | 2948 | page = virt_to_head_page(x); |
aadb4bc4 | 2949 | if (unlikely(!PageSlab(page))) { |
0937502a | 2950 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 2951 | kmemleak_free(x); |
aadb4bc4 CL |
2952 | put_page(page); |
2953 | return; | |
2954 | } | |
ce71e27c | 2955 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2956 | } |
2957 | EXPORT_SYMBOL(kfree); | |
2958 | ||
2086d26a | 2959 | /* |
672bba3a CL |
2960 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2961 | * the remaining slabs by the number of items in use. The slabs with the | |
2962 | * most items in use come first. New allocations will then fill those up | |
2963 | * and thus they can be removed from the partial lists. | |
2964 | * | |
2965 | * The slabs with the least items are placed last. This results in them | |
2966 | * being allocated from last increasing the chance that the last objects | |
2967 | * are freed in them. | |
2086d26a CL |
2968 | */ |
2969 | int kmem_cache_shrink(struct kmem_cache *s) | |
2970 | { | |
2971 | int node; | |
2972 | int i; | |
2973 | struct kmem_cache_node *n; | |
2974 | struct page *page; | |
2975 | struct page *t; | |
205ab99d | 2976 | int objects = oo_objects(s->max); |
2086d26a | 2977 | struct list_head *slabs_by_inuse = |
834f3d11 | 2978 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2979 | unsigned long flags; |
2980 | ||
2981 | if (!slabs_by_inuse) | |
2982 | return -ENOMEM; | |
2983 | ||
2984 | flush_all(s); | |
f64dc58c | 2985 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2986 | n = get_node(s, node); |
2987 | ||
2988 | if (!n->nr_partial) | |
2989 | continue; | |
2990 | ||
834f3d11 | 2991 | for (i = 0; i < objects; i++) |
2086d26a CL |
2992 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2993 | ||
2994 | spin_lock_irqsave(&n->list_lock, flags); | |
2995 | ||
2996 | /* | |
672bba3a | 2997 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2998 | * |
672bba3a CL |
2999 | * Note that concurrent frees may occur while we hold the |
3000 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3001 | */ |
3002 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
3003 | if (!page->inuse && slab_trylock(page)) { | |
3004 | /* | |
3005 | * Must hold slab lock here because slab_free | |
3006 | * may have freed the last object and be | |
3007 | * waiting to release the slab. | |
3008 | */ | |
62e346a8 | 3009 | __remove_partial(n, page); |
2086d26a CL |
3010 | slab_unlock(page); |
3011 | discard_slab(s, page); | |
3012 | } else { | |
fcda3d89 CL |
3013 | list_move(&page->lru, |
3014 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
3015 | } |
3016 | } | |
3017 | ||
2086d26a | 3018 | /* |
672bba3a CL |
3019 | * Rebuild the partial list with the slabs filled up most |
3020 | * first and the least used slabs at the end. | |
2086d26a | 3021 | */ |
834f3d11 | 3022 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
3023 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3024 | ||
2086d26a CL |
3025 | spin_unlock_irqrestore(&n->list_lock, flags); |
3026 | } | |
3027 | ||
3028 | kfree(slabs_by_inuse); | |
3029 | return 0; | |
3030 | } | |
3031 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3032 | ||
92a5bbc1 | 3033 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
3034 | static int slab_mem_going_offline_callback(void *arg) |
3035 | { | |
3036 | struct kmem_cache *s; | |
3037 | ||
3038 | down_read(&slub_lock); | |
3039 | list_for_each_entry(s, &slab_caches, list) | |
3040 | kmem_cache_shrink(s); | |
3041 | up_read(&slub_lock); | |
3042 | ||
3043 | return 0; | |
3044 | } | |
3045 | ||
3046 | static void slab_mem_offline_callback(void *arg) | |
3047 | { | |
3048 | struct kmem_cache_node *n; | |
3049 | struct kmem_cache *s; | |
3050 | struct memory_notify *marg = arg; | |
3051 | int offline_node; | |
3052 | ||
3053 | offline_node = marg->status_change_nid; | |
3054 | ||
3055 | /* | |
3056 | * If the node still has available memory. we need kmem_cache_node | |
3057 | * for it yet. | |
3058 | */ | |
3059 | if (offline_node < 0) | |
3060 | return; | |
3061 | ||
3062 | down_read(&slub_lock); | |
3063 | list_for_each_entry(s, &slab_caches, list) { | |
3064 | n = get_node(s, offline_node); | |
3065 | if (n) { | |
3066 | /* | |
3067 | * if n->nr_slabs > 0, slabs still exist on the node | |
3068 | * that is going down. We were unable to free them, | |
c9404c9c | 3069 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3070 | * callback. So, we must fail. |
3071 | */ | |
0f389ec6 | 3072 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3073 | |
3074 | s->node[offline_node] = NULL; | |
8de66a0c | 3075 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3076 | } |
3077 | } | |
3078 | up_read(&slub_lock); | |
3079 | } | |
3080 | ||
3081 | static int slab_mem_going_online_callback(void *arg) | |
3082 | { | |
3083 | struct kmem_cache_node *n; | |
3084 | struct kmem_cache *s; | |
3085 | struct memory_notify *marg = arg; | |
3086 | int nid = marg->status_change_nid; | |
3087 | int ret = 0; | |
3088 | ||
3089 | /* | |
3090 | * If the node's memory is already available, then kmem_cache_node is | |
3091 | * already created. Nothing to do. | |
3092 | */ | |
3093 | if (nid < 0) | |
3094 | return 0; | |
3095 | ||
3096 | /* | |
0121c619 | 3097 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3098 | * allocate a kmem_cache_node structure in order to bring the node |
3099 | * online. | |
3100 | */ | |
3101 | down_read(&slub_lock); | |
3102 | list_for_each_entry(s, &slab_caches, list) { | |
3103 | /* | |
3104 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3105 | * since memory is not yet available from the node that | |
3106 | * is brought up. | |
3107 | */ | |
8de66a0c | 3108 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3109 | if (!n) { |
3110 | ret = -ENOMEM; | |
3111 | goto out; | |
3112 | } | |
5595cffc | 3113 | init_kmem_cache_node(n, s); |
b9049e23 YG |
3114 | s->node[nid] = n; |
3115 | } | |
3116 | out: | |
3117 | up_read(&slub_lock); | |
3118 | return ret; | |
3119 | } | |
3120 | ||
3121 | static int slab_memory_callback(struct notifier_block *self, | |
3122 | unsigned long action, void *arg) | |
3123 | { | |
3124 | int ret = 0; | |
3125 | ||
3126 | switch (action) { | |
3127 | case MEM_GOING_ONLINE: | |
3128 | ret = slab_mem_going_online_callback(arg); | |
3129 | break; | |
3130 | case MEM_GOING_OFFLINE: | |
3131 | ret = slab_mem_going_offline_callback(arg); | |
3132 | break; | |
3133 | case MEM_OFFLINE: | |
3134 | case MEM_CANCEL_ONLINE: | |
3135 | slab_mem_offline_callback(arg); | |
3136 | break; | |
3137 | case MEM_ONLINE: | |
3138 | case MEM_CANCEL_OFFLINE: | |
3139 | break; | |
3140 | } | |
dc19f9db KH |
3141 | if (ret) |
3142 | ret = notifier_from_errno(ret); | |
3143 | else | |
3144 | ret = NOTIFY_OK; | |
b9049e23 YG |
3145 | return ret; |
3146 | } | |
3147 | ||
3148 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3149 | ||
81819f0f CL |
3150 | /******************************************************************** |
3151 | * Basic setup of slabs | |
3152 | *******************************************************************/ | |
3153 | ||
51df1142 CL |
3154 | /* |
3155 | * Used for early kmem_cache structures that were allocated using | |
3156 | * the page allocator | |
3157 | */ | |
3158 | ||
3159 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
3160 | { | |
3161 | int node; | |
3162 | ||
3163 | list_add(&s->list, &slab_caches); | |
3164 | s->refcount = -1; | |
3165 | ||
3166 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3167 | struct kmem_cache_node *n = get_node(s, node); | |
3168 | struct page *p; | |
3169 | ||
3170 | if (n) { | |
3171 | list_for_each_entry(p, &n->partial, lru) | |
3172 | p->slab = s; | |
3173 | ||
607bf324 | 3174 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
3175 | list_for_each_entry(p, &n->full, lru) |
3176 | p->slab = s; | |
3177 | #endif | |
3178 | } | |
3179 | } | |
3180 | } | |
3181 | ||
81819f0f CL |
3182 | void __init kmem_cache_init(void) |
3183 | { | |
3184 | int i; | |
4b356be0 | 3185 | int caches = 0; |
51df1142 CL |
3186 | struct kmem_cache *temp_kmem_cache; |
3187 | int order; | |
51df1142 CL |
3188 | struct kmem_cache *temp_kmem_cache_node; |
3189 | unsigned long kmalloc_size; | |
3190 | ||
3191 | kmem_size = offsetof(struct kmem_cache, node) + | |
3192 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3193 | ||
3194 | /* Allocate two kmem_caches from the page allocator */ | |
3195 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3196 | order = get_order(2 * kmalloc_size); | |
3197 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3198 | ||
81819f0f CL |
3199 | /* |
3200 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3201 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3202 | * kmem_cache_open for slab_state == DOWN. |
3203 | */ | |
51df1142 CL |
3204 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3205 | ||
3206 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3207 | sizeof(struct kmem_cache_node), | |
3208 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3209 | |
0c40ba4f | 3210 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3211 | |
3212 | /* Able to allocate the per node structures */ | |
3213 | slab_state = PARTIAL; | |
3214 | ||
51df1142 CL |
3215 | temp_kmem_cache = kmem_cache; |
3216 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3217 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3218 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3219 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3220 | |
51df1142 CL |
3221 | /* |
3222 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3223 | * kmem_cache_node is separately allocated so no need to | |
3224 | * update any list pointers. | |
3225 | */ | |
3226 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3227 | |
51df1142 CL |
3228 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3229 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3230 | ||
3231 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3232 | ||
3233 | caches++; | |
51df1142 CL |
3234 | kmem_cache_bootstrap_fixup(kmem_cache); |
3235 | caches++; | |
3236 | /* Free temporary boot structure */ | |
3237 | free_pages((unsigned long)temp_kmem_cache, order); | |
3238 | ||
3239 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3240 | |
3241 | /* | |
3242 | * Patch up the size_index table if we have strange large alignment | |
3243 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3244 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3245 | * |
3246 | * Largest permitted alignment is 256 bytes due to the way we | |
3247 | * handle the index determination for the smaller caches. | |
3248 | * | |
3249 | * Make sure that nothing crazy happens if someone starts tinkering | |
3250 | * around with ARCH_KMALLOC_MINALIGN | |
3251 | */ | |
3252 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3253 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3254 | ||
acdfcd04 AK |
3255 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3256 | int elem = size_index_elem(i); | |
3257 | if (elem >= ARRAY_SIZE(size_index)) | |
3258 | break; | |
3259 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3260 | } | |
f1b26339 | 3261 | |
acdfcd04 AK |
3262 | if (KMALLOC_MIN_SIZE == 64) { |
3263 | /* | |
3264 | * The 96 byte size cache is not used if the alignment | |
3265 | * is 64 byte. | |
3266 | */ | |
3267 | for (i = 64 + 8; i <= 96; i += 8) | |
3268 | size_index[size_index_elem(i)] = 7; | |
3269 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3270 | /* |
3271 | * The 192 byte sized cache is not used if the alignment | |
3272 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3273 | * instead. | |
3274 | */ | |
3275 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3276 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3277 | } |
3278 | ||
51df1142 CL |
3279 | /* Caches that are not of the two-to-the-power-of size */ |
3280 | if (KMALLOC_MIN_SIZE <= 32) { | |
3281 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3282 | caches++; | |
3283 | } | |
3284 | ||
3285 | if (KMALLOC_MIN_SIZE <= 64) { | |
3286 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3287 | caches++; | |
3288 | } | |
3289 | ||
3290 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3291 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3292 | caches++; | |
3293 | } | |
3294 | ||
81819f0f CL |
3295 | slab_state = UP; |
3296 | ||
3297 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3298 | if (KMALLOC_MIN_SIZE <= 32) { |
3299 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3300 | BUG_ON(!kmalloc_caches[1]->name); | |
3301 | } | |
3302 | ||
3303 | if (KMALLOC_MIN_SIZE <= 64) { | |
3304 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3305 | BUG_ON(!kmalloc_caches[2]->name); | |
3306 | } | |
3307 | ||
d7278bd7 CL |
3308 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3309 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3310 | ||
3311 | BUG_ON(!s); | |
51df1142 | 3312 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3313 | } |
81819f0f CL |
3314 | |
3315 | #ifdef CONFIG_SMP | |
3316 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3317 | #endif |
81819f0f | 3318 | |
55136592 | 3319 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3320 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3321 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3322 | |
51df1142 | 3323 | if (s && s->size) { |
55136592 CL |
3324 | char *name = kasprintf(GFP_NOWAIT, |
3325 | "dma-kmalloc-%d", s->objsize); | |
3326 | ||
3327 | BUG_ON(!name); | |
51df1142 CL |
3328 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3329 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3330 | } |
3331 | } | |
3332 | #endif | |
3adbefee IM |
3333 | printk(KERN_INFO |
3334 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3335 | " CPUs=%d, Nodes=%d\n", |
3336 | caches, cache_line_size(), | |
81819f0f CL |
3337 | slub_min_order, slub_max_order, slub_min_objects, |
3338 | nr_cpu_ids, nr_node_ids); | |
3339 | } | |
3340 | ||
7e85ee0c PE |
3341 | void __init kmem_cache_init_late(void) |
3342 | { | |
7e85ee0c PE |
3343 | } |
3344 | ||
81819f0f CL |
3345 | /* |
3346 | * Find a mergeable slab cache | |
3347 | */ | |
3348 | static int slab_unmergeable(struct kmem_cache *s) | |
3349 | { | |
3350 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3351 | return 1; | |
3352 | ||
c59def9f | 3353 | if (s->ctor) |
81819f0f CL |
3354 | return 1; |
3355 | ||
8ffa6875 CL |
3356 | /* |
3357 | * We may have set a slab to be unmergeable during bootstrap. | |
3358 | */ | |
3359 | if (s->refcount < 0) | |
3360 | return 1; | |
3361 | ||
81819f0f CL |
3362 | return 0; |
3363 | } | |
3364 | ||
3365 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3366 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3367 | void (*ctor)(void *)) |
81819f0f | 3368 | { |
5b95a4ac | 3369 | struct kmem_cache *s; |
81819f0f CL |
3370 | |
3371 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3372 | return NULL; | |
3373 | ||
c59def9f | 3374 | if (ctor) |
81819f0f CL |
3375 | return NULL; |
3376 | ||
3377 | size = ALIGN(size, sizeof(void *)); | |
3378 | align = calculate_alignment(flags, align, size); | |
3379 | size = ALIGN(size, align); | |
ba0268a8 | 3380 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3381 | |
5b95a4ac | 3382 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3383 | if (slab_unmergeable(s)) |
3384 | continue; | |
3385 | ||
3386 | if (size > s->size) | |
3387 | continue; | |
3388 | ||
ba0268a8 | 3389 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3390 | continue; |
3391 | /* | |
3392 | * Check if alignment is compatible. | |
3393 | * Courtesy of Adrian Drzewiecki | |
3394 | */ | |
06428780 | 3395 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3396 | continue; |
3397 | ||
3398 | if (s->size - size >= sizeof(void *)) | |
3399 | continue; | |
3400 | ||
3401 | return s; | |
3402 | } | |
3403 | return NULL; | |
3404 | } | |
3405 | ||
3406 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3407 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3408 | { |
3409 | struct kmem_cache *s; | |
84c1cf62 | 3410 | char *n; |
81819f0f | 3411 | |
fe1ff49d BH |
3412 | if (WARN_ON(!name)) |
3413 | return NULL; | |
3414 | ||
81819f0f | 3415 | down_write(&slub_lock); |
ba0268a8 | 3416 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3417 | if (s) { |
3418 | s->refcount++; | |
3419 | /* | |
3420 | * Adjust the object sizes so that we clear | |
3421 | * the complete object on kzalloc. | |
3422 | */ | |
3423 | s->objsize = max(s->objsize, (int)size); | |
3424 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3425 | |
7b8f3b66 | 3426 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3427 | s->refcount--; |
81819f0f | 3428 | goto err; |
7b8f3b66 | 3429 | } |
2bce6485 | 3430 | up_write(&slub_lock); |
a0e1d1be CL |
3431 | return s; |
3432 | } | |
6446faa2 | 3433 | |
84c1cf62 PE |
3434 | n = kstrdup(name, GFP_KERNEL); |
3435 | if (!n) | |
3436 | goto err; | |
3437 | ||
a0e1d1be CL |
3438 | s = kmalloc(kmem_size, GFP_KERNEL); |
3439 | if (s) { | |
84c1cf62 | 3440 | if (kmem_cache_open(s, n, |
c59def9f | 3441 | size, align, flags, ctor)) { |
81819f0f | 3442 | list_add(&s->list, &slab_caches); |
7b8f3b66 | 3443 | if (sysfs_slab_add(s)) { |
7b8f3b66 | 3444 | list_del(&s->list); |
84c1cf62 | 3445 | kfree(n); |
7b8f3b66 | 3446 | kfree(s); |
a0e1d1be | 3447 | goto err; |
7b8f3b66 | 3448 | } |
2bce6485 | 3449 | up_write(&slub_lock); |
a0e1d1be CL |
3450 | return s; |
3451 | } | |
84c1cf62 | 3452 | kfree(n); |
a0e1d1be | 3453 | kfree(s); |
81819f0f | 3454 | } |
68cee4f1 | 3455 | err: |
81819f0f | 3456 | up_write(&slub_lock); |
81819f0f | 3457 | |
81819f0f CL |
3458 | if (flags & SLAB_PANIC) |
3459 | panic("Cannot create slabcache %s\n", name); | |
3460 | else | |
3461 | s = NULL; | |
3462 | return s; | |
3463 | } | |
3464 | EXPORT_SYMBOL(kmem_cache_create); | |
3465 | ||
81819f0f | 3466 | #ifdef CONFIG_SMP |
81819f0f | 3467 | /* |
672bba3a CL |
3468 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3469 | * necessary. | |
81819f0f CL |
3470 | */ |
3471 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3472 | unsigned long action, void *hcpu) | |
3473 | { | |
3474 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3475 | struct kmem_cache *s; |
3476 | unsigned long flags; | |
81819f0f CL |
3477 | |
3478 | switch (action) { | |
3479 | case CPU_UP_CANCELED: | |
8bb78442 | 3480 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3481 | case CPU_DEAD: |
8bb78442 | 3482 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3483 | down_read(&slub_lock); |
3484 | list_for_each_entry(s, &slab_caches, list) { | |
3485 | local_irq_save(flags); | |
3486 | __flush_cpu_slab(s, cpu); | |
3487 | local_irq_restore(flags); | |
3488 | } | |
3489 | up_read(&slub_lock); | |
81819f0f CL |
3490 | break; |
3491 | default: | |
3492 | break; | |
3493 | } | |
3494 | return NOTIFY_OK; | |
3495 | } | |
3496 | ||
06428780 | 3497 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3498 | .notifier_call = slab_cpuup_callback |
06428780 | 3499 | }; |
81819f0f CL |
3500 | |
3501 | #endif | |
3502 | ||
ce71e27c | 3503 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3504 | { |
aadb4bc4 | 3505 | struct kmem_cache *s; |
94b528d0 | 3506 | void *ret; |
aadb4bc4 | 3507 | |
ffadd4d0 | 3508 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3509 | return kmalloc_large(size, gfpflags); |
3510 | ||
aadb4bc4 | 3511 | s = get_slab(size, gfpflags); |
81819f0f | 3512 | |
2408c550 | 3513 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3514 | return s; |
81819f0f | 3515 | |
2154a336 | 3516 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 | 3517 | |
25985edc | 3518 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3519 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3520 | |
3521 | return ret; | |
81819f0f CL |
3522 | } |
3523 | ||
5d1f57e4 | 3524 | #ifdef CONFIG_NUMA |
81819f0f | 3525 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3526 | int node, unsigned long caller) |
81819f0f | 3527 | { |
aadb4bc4 | 3528 | struct kmem_cache *s; |
94b528d0 | 3529 | void *ret; |
aadb4bc4 | 3530 | |
d3e14aa3 XF |
3531 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3532 | ret = kmalloc_large_node(size, gfpflags, node); | |
3533 | ||
3534 | trace_kmalloc_node(caller, ret, | |
3535 | size, PAGE_SIZE << get_order(size), | |
3536 | gfpflags, node); | |
3537 | ||
3538 | return ret; | |
3539 | } | |
eada35ef | 3540 | |
aadb4bc4 | 3541 | s = get_slab(size, gfpflags); |
81819f0f | 3542 | |
2408c550 | 3543 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3544 | return s; |
81819f0f | 3545 | |
94b528d0 EGM |
3546 | ret = slab_alloc(s, gfpflags, node, caller); |
3547 | ||
25985edc | 3548 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3549 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3550 | |
3551 | return ret; | |
81819f0f | 3552 | } |
5d1f57e4 | 3553 | #endif |
81819f0f | 3554 | |
ab4d5ed5 | 3555 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3556 | static int count_inuse(struct page *page) |
3557 | { | |
3558 | return page->inuse; | |
3559 | } | |
3560 | ||
3561 | static int count_total(struct page *page) | |
3562 | { | |
3563 | return page->objects; | |
3564 | } | |
ab4d5ed5 | 3565 | #endif |
205ab99d | 3566 | |
ab4d5ed5 | 3567 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3568 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3569 | unsigned long *map) | |
53e15af0 CL |
3570 | { |
3571 | void *p; | |
a973e9dd | 3572 | void *addr = page_address(page); |
53e15af0 CL |
3573 | |
3574 | if (!check_slab(s, page) || | |
3575 | !on_freelist(s, page, NULL)) | |
3576 | return 0; | |
3577 | ||
3578 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3579 | bitmap_zero(map, page->objects); |
53e15af0 | 3580 | |
5f80b13a CL |
3581 | get_map(s, page, map); |
3582 | for_each_object(p, s, addr, page->objects) { | |
3583 | if (test_bit(slab_index(p, s, addr), map)) | |
3584 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3585 | return 0; | |
53e15af0 CL |
3586 | } |
3587 | ||
224a88be | 3588 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3589 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3590 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3591 | return 0; |
3592 | return 1; | |
3593 | } | |
3594 | ||
434e245d CL |
3595 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3596 | unsigned long *map) | |
53e15af0 CL |
3597 | { |
3598 | if (slab_trylock(page)) { | |
434e245d | 3599 | validate_slab(s, page, map); |
53e15af0 CL |
3600 | slab_unlock(page); |
3601 | } else | |
3602 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3603 | s->name, page); | |
53e15af0 CL |
3604 | } |
3605 | ||
434e245d CL |
3606 | static int validate_slab_node(struct kmem_cache *s, |
3607 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3608 | { |
3609 | unsigned long count = 0; | |
3610 | struct page *page; | |
3611 | unsigned long flags; | |
3612 | ||
3613 | spin_lock_irqsave(&n->list_lock, flags); | |
3614 | ||
3615 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3616 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3617 | count++; |
3618 | } | |
3619 | if (count != n->nr_partial) | |
3620 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3621 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3622 | ||
3623 | if (!(s->flags & SLAB_STORE_USER)) | |
3624 | goto out; | |
3625 | ||
3626 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3627 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3628 | count++; |
3629 | } | |
3630 | if (count != atomic_long_read(&n->nr_slabs)) | |
3631 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3632 | "counter=%ld\n", s->name, count, | |
3633 | atomic_long_read(&n->nr_slabs)); | |
3634 | ||
3635 | out: | |
3636 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3637 | return count; | |
3638 | } | |
3639 | ||
434e245d | 3640 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3641 | { |
3642 | int node; | |
3643 | unsigned long count = 0; | |
205ab99d | 3644 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3645 | sizeof(unsigned long), GFP_KERNEL); |
3646 | ||
3647 | if (!map) | |
3648 | return -ENOMEM; | |
53e15af0 CL |
3649 | |
3650 | flush_all(s); | |
f64dc58c | 3651 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3652 | struct kmem_cache_node *n = get_node(s, node); |
3653 | ||
434e245d | 3654 | count += validate_slab_node(s, n, map); |
53e15af0 | 3655 | } |
434e245d | 3656 | kfree(map); |
53e15af0 CL |
3657 | return count; |
3658 | } | |
88a420e4 | 3659 | /* |
672bba3a | 3660 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3661 | * and freed. |
3662 | */ | |
3663 | ||
3664 | struct location { | |
3665 | unsigned long count; | |
ce71e27c | 3666 | unsigned long addr; |
45edfa58 CL |
3667 | long long sum_time; |
3668 | long min_time; | |
3669 | long max_time; | |
3670 | long min_pid; | |
3671 | long max_pid; | |
174596a0 | 3672 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3673 | nodemask_t nodes; |
88a420e4 CL |
3674 | }; |
3675 | ||
3676 | struct loc_track { | |
3677 | unsigned long max; | |
3678 | unsigned long count; | |
3679 | struct location *loc; | |
3680 | }; | |
3681 | ||
3682 | static void free_loc_track(struct loc_track *t) | |
3683 | { | |
3684 | if (t->max) | |
3685 | free_pages((unsigned long)t->loc, | |
3686 | get_order(sizeof(struct location) * t->max)); | |
3687 | } | |
3688 | ||
68dff6a9 | 3689 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3690 | { |
3691 | struct location *l; | |
3692 | int order; | |
3693 | ||
88a420e4 CL |
3694 | order = get_order(sizeof(struct location) * max); |
3695 | ||
68dff6a9 | 3696 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3697 | if (!l) |
3698 | return 0; | |
3699 | ||
3700 | if (t->count) { | |
3701 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3702 | free_loc_track(t); | |
3703 | } | |
3704 | t->max = max; | |
3705 | t->loc = l; | |
3706 | return 1; | |
3707 | } | |
3708 | ||
3709 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3710 | const struct track *track) |
88a420e4 CL |
3711 | { |
3712 | long start, end, pos; | |
3713 | struct location *l; | |
ce71e27c | 3714 | unsigned long caddr; |
45edfa58 | 3715 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3716 | |
3717 | start = -1; | |
3718 | end = t->count; | |
3719 | ||
3720 | for ( ; ; ) { | |
3721 | pos = start + (end - start + 1) / 2; | |
3722 | ||
3723 | /* | |
3724 | * There is nothing at "end". If we end up there | |
3725 | * we need to add something to before end. | |
3726 | */ | |
3727 | if (pos == end) | |
3728 | break; | |
3729 | ||
3730 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3731 | if (track->addr == caddr) { |
3732 | ||
3733 | l = &t->loc[pos]; | |
3734 | l->count++; | |
3735 | if (track->when) { | |
3736 | l->sum_time += age; | |
3737 | if (age < l->min_time) | |
3738 | l->min_time = age; | |
3739 | if (age > l->max_time) | |
3740 | l->max_time = age; | |
3741 | ||
3742 | if (track->pid < l->min_pid) | |
3743 | l->min_pid = track->pid; | |
3744 | if (track->pid > l->max_pid) | |
3745 | l->max_pid = track->pid; | |
3746 | ||
174596a0 RR |
3747 | cpumask_set_cpu(track->cpu, |
3748 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3749 | } |
3750 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3751 | return 1; |
3752 | } | |
3753 | ||
45edfa58 | 3754 | if (track->addr < caddr) |
88a420e4 CL |
3755 | end = pos; |
3756 | else | |
3757 | start = pos; | |
3758 | } | |
3759 | ||
3760 | /* | |
672bba3a | 3761 | * Not found. Insert new tracking element. |
88a420e4 | 3762 | */ |
68dff6a9 | 3763 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3764 | return 0; |
3765 | ||
3766 | l = t->loc + pos; | |
3767 | if (pos < t->count) | |
3768 | memmove(l + 1, l, | |
3769 | (t->count - pos) * sizeof(struct location)); | |
3770 | t->count++; | |
3771 | l->count = 1; | |
45edfa58 CL |
3772 | l->addr = track->addr; |
3773 | l->sum_time = age; | |
3774 | l->min_time = age; | |
3775 | l->max_time = age; | |
3776 | l->min_pid = track->pid; | |
3777 | l->max_pid = track->pid; | |
174596a0 RR |
3778 | cpumask_clear(to_cpumask(l->cpus)); |
3779 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3780 | nodes_clear(l->nodes); |
3781 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3782 | return 1; |
3783 | } | |
3784 | ||
3785 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 3786 | struct page *page, enum track_item alloc, |
a5dd5c11 | 3787 | unsigned long *map) |
88a420e4 | 3788 | { |
a973e9dd | 3789 | void *addr = page_address(page); |
88a420e4 CL |
3790 | void *p; |
3791 | ||
39b26464 | 3792 | bitmap_zero(map, page->objects); |
5f80b13a | 3793 | get_map(s, page, map); |
88a420e4 | 3794 | |
224a88be | 3795 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3796 | if (!test_bit(slab_index(p, s, addr), map)) |
3797 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3798 | } |
3799 | ||
3800 | static int list_locations(struct kmem_cache *s, char *buf, | |
3801 | enum track_item alloc) | |
3802 | { | |
e374d483 | 3803 | int len = 0; |
88a420e4 | 3804 | unsigned long i; |
68dff6a9 | 3805 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 3806 | int node; |
bbd7d57b ED |
3807 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
3808 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 3809 | |
bbd7d57b ED |
3810 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
3811 | GFP_TEMPORARY)) { | |
3812 | kfree(map); | |
68dff6a9 | 3813 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 3814 | } |
88a420e4 CL |
3815 | /* Push back cpu slabs */ |
3816 | flush_all(s); | |
3817 | ||
f64dc58c | 3818 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3819 | struct kmem_cache_node *n = get_node(s, node); |
3820 | unsigned long flags; | |
3821 | struct page *page; | |
3822 | ||
9e86943b | 3823 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3824 | continue; |
3825 | ||
3826 | spin_lock_irqsave(&n->list_lock, flags); | |
3827 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 3828 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 3829 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 3830 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
3831 | spin_unlock_irqrestore(&n->list_lock, flags); |
3832 | } | |
3833 | ||
3834 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3835 | struct location *l = &t.loc[i]; |
88a420e4 | 3836 | |
9c246247 | 3837 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3838 | break; |
e374d483 | 3839 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3840 | |
3841 | if (l->addr) | |
62c70bce | 3842 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 3843 | else |
e374d483 | 3844 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3845 | |
3846 | if (l->sum_time != l->min_time) { | |
e374d483 | 3847 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3848 | l->min_time, |
3849 | (long)div_u64(l->sum_time, l->count), | |
3850 | l->max_time); | |
45edfa58 | 3851 | } else |
e374d483 | 3852 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3853 | l->min_time); |
3854 | ||
3855 | if (l->min_pid != l->max_pid) | |
e374d483 | 3856 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3857 | l->min_pid, l->max_pid); |
3858 | else | |
e374d483 | 3859 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3860 | l->min_pid); |
3861 | ||
174596a0 RR |
3862 | if (num_online_cpus() > 1 && |
3863 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3864 | len < PAGE_SIZE - 60) { |
3865 | len += sprintf(buf + len, " cpus="); | |
3866 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3867 | to_cpumask(l->cpus)); |
45edfa58 CL |
3868 | } |
3869 | ||
62bc62a8 | 3870 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3871 | len < PAGE_SIZE - 60) { |
3872 | len += sprintf(buf + len, " nodes="); | |
3873 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3874 | l->nodes); |
3875 | } | |
3876 | ||
e374d483 | 3877 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3878 | } |
3879 | ||
3880 | free_loc_track(&t); | |
bbd7d57b | 3881 | kfree(map); |
88a420e4 | 3882 | if (!t.count) |
e374d483 HH |
3883 | len += sprintf(buf, "No data\n"); |
3884 | return len; | |
88a420e4 | 3885 | } |
ab4d5ed5 | 3886 | #endif |
88a420e4 | 3887 | |
a5a84755 CL |
3888 | #ifdef SLUB_RESILIENCY_TEST |
3889 | static void resiliency_test(void) | |
3890 | { | |
3891 | u8 *p; | |
3892 | ||
3893 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
3894 | ||
3895 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3896 | printk(KERN_ERR "-----------------------\n"); | |
3897 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3898 | ||
3899 | p = kzalloc(16, GFP_KERNEL); | |
3900 | p[16] = 0x12; | |
3901 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3902 | " 0x12->0x%p\n\n", p + 16); | |
3903 | ||
3904 | validate_slab_cache(kmalloc_caches[4]); | |
3905 | ||
3906 | /* Hmmm... The next two are dangerous */ | |
3907 | p = kzalloc(32, GFP_KERNEL); | |
3908 | p[32 + sizeof(void *)] = 0x34; | |
3909 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3910 | " 0x34 -> -0x%p\n", p); | |
3911 | printk(KERN_ERR | |
3912 | "If allocated object is overwritten then not detectable\n\n"); | |
3913 | ||
3914 | validate_slab_cache(kmalloc_caches[5]); | |
3915 | p = kzalloc(64, GFP_KERNEL); | |
3916 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3917 | *p = 0x56; | |
3918 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3919 | p); | |
3920 | printk(KERN_ERR | |
3921 | "If allocated object is overwritten then not detectable\n\n"); | |
3922 | validate_slab_cache(kmalloc_caches[6]); | |
3923 | ||
3924 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3925 | p = kzalloc(128, GFP_KERNEL); | |
3926 | kfree(p); | |
3927 | *p = 0x78; | |
3928 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3929 | validate_slab_cache(kmalloc_caches[7]); | |
3930 | ||
3931 | p = kzalloc(256, GFP_KERNEL); | |
3932 | kfree(p); | |
3933 | p[50] = 0x9a; | |
3934 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
3935 | p); | |
3936 | validate_slab_cache(kmalloc_caches[8]); | |
3937 | ||
3938 | p = kzalloc(512, GFP_KERNEL); | |
3939 | kfree(p); | |
3940 | p[512] = 0xab; | |
3941 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3942 | validate_slab_cache(kmalloc_caches[9]); | |
3943 | } | |
3944 | #else | |
3945 | #ifdef CONFIG_SYSFS | |
3946 | static void resiliency_test(void) {}; | |
3947 | #endif | |
3948 | #endif | |
3949 | ||
ab4d5ed5 | 3950 | #ifdef CONFIG_SYSFS |
81819f0f | 3951 | enum slab_stat_type { |
205ab99d CL |
3952 | SL_ALL, /* All slabs */ |
3953 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3954 | SL_CPU, /* Only slabs used for cpu caches */ | |
3955 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3956 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3957 | }; |
3958 | ||
205ab99d | 3959 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3960 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3961 | #define SO_CPU (1 << SL_CPU) | |
3962 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3963 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3964 | |
62e5c4b4 CG |
3965 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3966 | char *buf, unsigned long flags) | |
81819f0f CL |
3967 | { |
3968 | unsigned long total = 0; | |
81819f0f CL |
3969 | int node; |
3970 | int x; | |
3971 | unsigned long *nodes; | |
3972 | unsigned long *per_cpu; | |
3973 | ||
3974 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3975 | if (!nodes) |
3976 | return -ENOMEM; | |
81819f0f CL |
3977 | per_cpu = nodes + nr_node_ids; |
3978 | ||
205ab99d CL |
3979 | if (flags & SO_CPU) { |
3980 | int cpu; | |
81819f0f | 3981 | |
205ab99d | 3982 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 3983 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
dfb4f096 | 3984 | |
205ab99d CL |
3985 | if (!c || c->node < 0) |
3986 | continue; | |
3987 | ||
3988 | if (c->page) { | |
3989 | if (flags & SO_TOTAL) | |
3990 | x = c->page->objects; | |
3991 | else if (flags & SO_OBJECTS) | |
3992 | x = c->page->inuse; | |
81819f0f CL |
3993 | else |
3994 | x = 1; | |
205ab99d | 3995 | |
81819f0f | 3996 | total += x; |
205ab99d | 3997 | nodes[c->node] += x; |
81819f0f | 3998 | } |
205ab99d | 3999 | per_cpu[c->node]++; |
81819f0f CL |
4000 | } |
4001 | } | |
4002 | ||
04d94879 | 4003 | lock_memory_hotplug(); |
ab4d5ed5 | 4004 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4005 | if (flags & SO_ALL) { |
4006 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4007 | struct kmem_cache_node *n = get_node(s, node); | |
4008 | ||
4009 | if (flags & SO_TOTAL) | |
4010 | x = atomic_long_read(&n->total_objects); | |
4011 | else if (flags & SO_OBJECTS) | |
4012 | x = atomic_long_read(&n->total_objects) - | |
4013 | count_partial(n, count_free); | |
81819f0f | 4014 | |
81819f0f | 4015 | else |
205ab99d | 4016 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4017 | total += x; |
4018 | nodes[node] += x; | |
4019 | } | |
4020 | ||
ab4d5ed5 CL |
4021 | } else |
4022 | #endif | |
4023 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4024 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4025 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4026 | |
205ab99d CL |
4027 | if (flags & SO_TOTAL) |
4028 | x = count_partial(n, count_total); | |
4029 | else if (flags & SO_OBJECTS) | |
4030 | x = count_partial(n, count_inuse); | |
81819f0f | 4031 | else |
205ab99d | 4032 | x = n->nr_partial; |
81819f0f CL |
4033 | total += x; |
4034 | nodes[node] += x; | |
4035 | } | |
4036 | } | |
81819f0f CL |
4037 | x = sprintf(buf, "%lu", total); |
4038 | #ifdef CONFIG_NUMA | |
f64dc58c | 4039 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4040 | if (nodes[node]) |
4041 | x += sprintf(buf + x, " N%d=%lu", | |
4042 | node, nodes[node]); | |
4043 | #endif | |
04d94879 | 4044 | unlock_memory_hotplug(); |
81819f0f CL |
4045 | kfree(nodes); |
4046 | return x + sprintf(buf + x, "\n"); | |
4047 | } | |
4048 | ||
ab4d5ed5 | 4049 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4050 | static int any_slab_objects(struct kmem_cache *s) |
4051 | { | |
4052 | int node; | |
81819f0f | 4053 | |
dfb4f096 | 4054 | for_each_online_node(node) { |
81819f0f CL |
4055 | struct kmem_cache_node *n = get_node(s, node); |
4056 | ||
dfb4f096 CL |
4057 | if (!n) |
4058 | continue; | |
4059 | ||
4ea33e2d | 4060 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4061 | return 1; |
4062 | } | |
4063 | return 0; | |
4064 | } | |
ab4d5ed5 | 4065 | #endif |
81819f0f CL |
4066 | |
4067 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
4068 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
4069 | ||
4070 | struct slab_attribute { | |
4071 | struct attribute attr; | |
4072 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4073 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4074 | }; | |
4075 | ||
4076 | #define SLAB_ATTR_RO(_name) \ | |
4077 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
4078 | ||
4079 | #define SLAB_ATTR(_name) \ | |
4080 | static struct slab_attribute _name##_attr = \ | |
4081 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
4082 | ||
81819f0f CL |
4083 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4084 | { | |
4085 | return sprintf(buf, "%d\n", s->size); | |
4086 | } | |
4087 | SLAB_ATTR_RO(slab_size); | |
4088 | ||
4089 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4090 | { | |
4091 | return sprintf(buf, "%d\n", s->align); | |
4092 | } | |
4093 | SLAB_ATTR_RO(align); | |
4094 | ||
4095 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4096 | { | |
4097 | return sprintf(buf, "%d\n", s->objsize); | |
4098 | } | |
4099 | SLAB_ATTR_RO(object_size); | |
4100 | ||
4101 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4102 | { | |
834f3d11 | 4103 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4104 | } |
4105 | SLAB_ATTR_RO(objs_per_slab); | |
4106 | ||
06b285dc CL |
4107 | static ssize_t order_store(struct kmem_cache *s, |
4108 | const char *buf, size_t length) | |
4109 | { | |
0121c619 CL |
4110 | unsigned long order; |
4111 | int err; | |
4112 | ||
4113 | err = strict_strtoul(buf, 10, &order); | |
4114 | if (err) | |
4115 | return err; | |
06b285dc CL |
4116 | |
4117 | if (order > slub_max_order || order < slub_min_order) | |
4118 | return -EINVAL; | |
4119 | ||
4120 | calculate_sizes(s, order); | |
4121 | return length; | |
4122 | } | |
4123 | ||
81819f0f CL |
4124 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4125 | { | |
834f3d11 | 4126 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4127 | } |
06b285dc | 4128 | SLAB_ATTR(order); |
81819f0f | 4129 | |
73d342b1 DR |
4130 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4131 | { | |
4132 | return sprintf(buf, "%lu\n", s->min_partial); | |
4133 | } | |
4134 | ||
4135 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4136 | size_t length) | |
4137 | { | |
4138 | unsigned long min; | |
4139 | int err; | |
4140 | ||
4141 | err = strict_strtoul(buf, 10, &min); | |
4142 | if (err) | |
4143 | return err; | |
4144 | ||
c0bdb232 | 4145 | set_min_partial(s, min); |
73d342b1 DR |
4146 | return length; |
4147 | } | |
4148 | SLAB_ATTR(min_partial); | |
4149 | ||
81819f0f CL |
4150 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4151 | { | |
62c70bce JP |
4152 | if (!s->ctor) |
4153 | return 0; | |
4154 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4155 | } |
4156 | SLAB_ATTR_RO(ctor); | |
4157 | ||
81819f0f CL |
4158 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4159 | { | |
4160 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4161 | } | |
4162 | SLAB_ATTR_RO(aliases); | |
4163 | ||
81819f0f CL |
4164 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4165 | { | |
d9acf4b7 | 4166 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4167 | } |
4168 | SLAB_ATTR_RO(partial); | |
4169 | ||
4170 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4171 | { | |
d9acf4b7 | 4172 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4173 | } |
4174 | SLAB_ATTR_RO(cpu_slabs); | |
4175 | ||
4176 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4177 | { | |
205ab99d | 4178 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4179 | } |
4180 | SLAB_ATTR_RO(objects); | |
4181 | ||
205ab99d CL |
4182 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4183 | { | |
4184 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4185 | } | |
4186 | SLAB_ATTR_RO(objects_partial); | |
4187 | ||
a5a84755 CL |
4188 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4189 | { | |
4190 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4191 | } | |
4192 | ||
4193 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4194 | const char *buf, size_t length) | |
4195 | { | |
4196 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4197 | if (buf[0] == '1') | |
4198 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4199 | return length; | |
4200 | } | |
4201 | SLAB_ATTR(reclaim_account); | |
4202 | ||
4203 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4204 | { | |
4205 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4206 | } | |
4207 | SLAB_ATTR_RO(hwcache_align); | |
4208 | ||
4209 | #ifdef CONFIG_ZONE_DMA | |
4210 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4211 | { | |
4212 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4213 | } | |
4214 | SLAB_ATTR_RO(cache_dma); | |
4215 | #endif | |
4216 | ||
4217 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4218 | { | |
4219 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4220 | } | |
4221 | SLAB_ATTR_RO(destroy_by_rcu); | |
4222 | ||
ab9a0f19 LJ |
4223 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4224 | { | |
4225 | return sprintf(buf, "%d\n", s->reserved); | |
4226 | } | |
4227 | SLAB_ATTR_RO(reserved); | |
4228 | ||
ab4d5ed5 | 4229 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4230 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4231 | { | |
4232 | return show_slab_objects(s, buf, SO_ALL); | |
4233 | } | |
4234 | SLAB_ATTR_RO(slabs); | |
4235 | ||
205ab99d CL |
4236 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4237 | { | |
4238 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4239 | } | |
4240 | SLAB_ATTR_RO(total_objects); | |
4241 | ||
81819f0f CL |
4242 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4243 | { | |
4244 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4245 | } | |
4246 | ||
4247 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4248 | const char *buf, size_t length) | |
4249 | { | |
4250 | s->flags &= ~SLAB_DEBUG_FREE; | |
4251 | if (buf[0] == '1') | |
4252 | s->flags |= SLAB_DEBUG_FREE; | |
4253 | return length; | |
4254 | } | |
4255 | SLAB_ATTR(sanity_checks); | |
4256 | ||
4257 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4258 | { | |
4259 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4260 | } | |
4261 | ||
4262 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4263 | size_t length) | |
4264 | { | |
4265 | s->flags &= ~SLAB_TRACE; | |
4266 | if (buf[0] == '1') | |
4267 | s->flags |= SLAB_TRACE; | |
4268 | return length; | |
4269 | } | |
4270 | SLAB_ATTR(trace); | |
4271 | ||
81819f0f CL |
4272 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4273 | { | |
4274 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4275 | } | |
4276 | ||
4277 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4278 | const char *buf, size_t length) | |
4279 | { | |
4280 | if (any_slab_objects(s)) | |
4281 | return -EBUSY; | |
4282 | ||
4283 | s->flags &= ~SLAB_RED_ZONE; | |
4284 | if (buf[0] == '1') | |
4285 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4286 | calculate_sizes(s, -1); |
81819f0f CL |
4287 | return length; |
4288 | } | |
4289 | SLAB_ATTR(red_zone); | |
4290 | ||
4291 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4292 | { | |
4293 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4294 | } | |
4295 | ||
4296 | static ssize_t poison_store(struct kmem_cache *s, | |
4297 | const char *buf, size_t length) | |
4298 | { | |
4299 | if (any_slab_objects(s)) | |
4300 | return -EBUSY; | |
4301 | ||
4302 | s->flags &= ~SLAB_POISON; | |
4303 | if (buf[0] == '1') | |
4304 | s->flags |= SLAB_POISON; | |
06b285dc | 4305 | calculate_sizes(s, -1); |
81819f0f CL |
4306 | return length; |
4307 | } | |
4308 | SLAB_ATTR(poison); | |
4309 | ||
4310 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4311 | { | |
4312 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4313 | } | |
4314 | ||
4315 | static ssize_t store_user_store(struct kmem_cache *s, | |
4316 | const char *buf, size_t length) | |
4317 | { | |
4318 | if (any_slab_objects(s)) | |
4319 | return -EBUSY; | |
4320 | ||
4321 | s->flags &= ~SLAB_STORE_USER; | |
4322 | if (buf[0] == '1') | |
4323 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4324 | calculate_sizes(s, -1); |
81819f0f CL |
4325 | return length; |
4326 | } | |
4327 | SLAB_ATTR(store_user); | |
4328 | ||
53e15af0 CL |
4329 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4330 | { | |
4331 | return 0; | |
4332 | } | |
4333 | ||
4334 | static ssize_t validate_store(struct kmem_cache *s, | |
4335 | const char *buf, size_t length) | |
4336 | { | |
434e245d CL |
4337 | int ret = -EINVAL; |
4338 | ||
4339 | if (buf[0] == '1') { | |
4340 | ret = validate_slab_cache(s); | |
4341 | if (ret >= 0) | |
4342 | ret = length; | |
4343 | } | |
4344 | return ret; | |
53e15af0 CL |
4345 | } |
4346 | SLAB_ATTR(validate); | |
a5a84755 CL |
4347 | |
4348 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4349 | { | |
4350 | if (!(s->flags & SLAB_STORE_USER)) | |
4351 | return -ENOSYS; | |
4352 | return list_locations(s, buf, TRACK_ALLOC); | |
4353 | } | |
4354 | SLAB_ATTR_RO(alloc_calls); | |
4355 | ||
4356 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4357 | { | |
4358 | if (!(s->flags & SLAB_STORE_USER)) | |
4359 | return -ENOSYS; | |
4360 | return list_locations(s, buf, TRACK_FREE); | |
4361 | } | |
4362 | SLAB_ATTR_RO(free_calls); | |
4363 | #endif /* CONFIG_SLUB_DEBUG */ | |
4364 | ||
4365 | #ifdef CONFIG_FAILSLAB | |
4366 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4367 | { | |
4368 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4369 | } | |
4370 | ||
4371 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4372 | size_t length) | |
4373 | { | |
4374 | s->flags &= ~SLAB_FAILSLAB; | |
4375 | if (buf[0] == '1') | |
4376 | s->flags |= SLAB_FAILSLAB; | |
4377 | return length; | |
4378 | } | |
4379 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4380 | #endif |
53e15af0 | 4381 | |
2086d26a CL |
4382 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4383 | { | |
4384 | return 0; | |
4385 | } | |
4386 | ||
4387 | static ssize_t shrink_store(struct kmem_cache *s, | |
4388 | const char *buf, size_t length) | |
4389 | { | |
4390 | if (buf[0] == '1') { | |
4391 | int rc = kmem_cache_shrink(s); | |
4392 | ||
4393 | if (rc) | |
4394 | return rc; | |
4395 | } else | |
4396 | return -EINVAL; | |
4397 | return length; | |
4398 | } | |
4399 | SLAB_ATTR(shrink); | |
4400 | ||
81819f0f | 4401 | #ifdef CONFIG_NUMA |
9824601e | 4402 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4403 | { |
9824601e | 4404 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4405 | } |
4406 | ||
9824601e | 4407 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4408 | const char *buf, size_t length) |
4409 | { | |
0121c619 CL |
4410 | unsigned long ratio; |
4411 | int err; | |
4412 | ||
4413 | err = strict_strtoul(buf, 10, &ratio); | |
4414 | if (err) | |
4415 | return err; | |
4416 | ||
e2cb96b7 | 4417 | if (ratio <= 100) |
0121c619 | 4418 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4419 | |
81819f0f CL |
4420 | return length; |
4421 | } | |
9824601e | 4422 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4423 | #endif |
4424 | ||
8ff12cfc | 4425 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4426 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4427 | { | |
4428 | unsigned long sum = 0; | |
4429 | int cpu; | |
4430 | int len; | |
4431 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4432 | ||
4433 | if (!data) | |
4434 | return -ENOMEM; | |
4435 | ||
4436 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4437 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4438 | |
4439 | data[cpu] = x; | |
4440 | sum += x; | |
4441 | } | |
4442 | ||
4443 | len = sprintf(buf, "%lu", sum); | |
4444 | ||
50ef37b9 | 4445 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4446 | for_each_online_cpu(cpu) { |
4447 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4448 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4449 | } |
50ef37b9 | 4450 | #endif |
8ff12cfc CL |
4451 | kfree(data); |
4452 | return len + sprintf(buf + len, "\n"); | |
4453 | } | |
4454 | ||
78eb00cc DR |
4455 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4456 | { | |
4457 | int cpu; | |
4458 | ||
4459 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4460 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4461 | } |
4462 | ||
8ff12cfc CL |
4463 | #define STAT_ATTR(si, text) \ |
4464 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4465 | { \ | |
4466 | return show_stat(s, buf, si); \ | |
4467 | } \ | |
78eb00cc DR |
4468 | static ssize_t text##_store(struct kmem_cache *s, \ |
4469 | const char *buf, size_t length) \ | |
4470 | { \ | |
4471 | if (buf[0] != '0') \ | |
4472 | return -EINVAL; \ | |
4473 | clear_stat(s, si); \ | |
4474 | return length; \ | |
4475 | } \ | |
4476 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4477 | |
4478 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4479 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4480 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4481 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4482 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4483 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4484 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4485 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4486 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4487 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4488 | STAT_ATTR(FREE_SLAB, free_slab); | |
4489 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4490 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4491 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4492 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4493 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4494 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4495 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4496 | #endif |
4497 | ||
06428780 | 4498 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4499 | &slab_size_attr.attr, |
4500 | &object_size_attr.attr, | |
4501 | &objs_per_slab_attr.attr, | |
4502 | &order_attr.attr, | |
73d342b1 | 4503 | &min_partial_attr.attr, |
81819f0f | 4504 | &objects_attr.attr, |
205ab99d | 4505 | &objects_partial_attr.attr, |
81819f0f CL |
4506 | &partial_attr.attr, |
4507 | &cpu_slabs_attr.attr, | |
4508 | &ctor_attr.attr, | |
81819f0f CL |
4509 | &aliases_attr.attr, |
4510 | &align_attr.attr, | |
81819f0f CL |
4511 | &hwcache_align_attr.attr, |
4512 | &reclaim_account_attr.attr, | |
4513 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4514 | &shrink_attr.attr, |
ab9a0f19 | 4515 | &reserved_attr.attr, |
ab4d5ed5 | 4516 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4517 | &total_objects_attr.attr, |
4518 | &slabs_attr.attr, | |
4519 | &sanity_checks_attr.attr, | |
4520 | &trace_attr.attr, | |
81819f0f CL |
4521 | &red_zone_attr.attr, |
4522 | &poison_attr.attr, | |
4523 | &store_user_attr.attr, | |
53e15af0 | 4524 | &validate_attr.attr, |
88a420e4 CL |
4525 | &alloc_calls_attr.attr, |
4526 | &free_calls_attr.attr, | |
ab4d5ed5 | 4527 | #endif |
81819f0f CL |
4528 | #ifdef CONFIG_ZONE_DMA |
4529 | &cache_dma_attr.attr, | |
4530 | #endif | |
4531 | #ifdef CONFIG_NUMA | |
9824601e | 4532 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4533 | #endif |
4534 | #ifdef CONFIG_SLUB_STATS | |
4535 | &alloc_fastpath_attr.attr, | |
4536 | &alloc_slowpath_attr.attr, | |
4537 | &free_fastpath_attr.attr, | |
4538 | &free_slowpath_attr.attr, | |
4539 | &free_frozen_attr.attr, | |
4540 | &free_add_partial_attr.attr, | |
4541 | &free_remove_partial_attr.attr, | |
4542 | &alloc_from_partial_attr.attr, | |
4543 | &alloc_slab_attr.attr, | |
4544 | &alloc_refill_attr.attr, | |
4545 | &free_slab_attr.attr, | |
4546 | &cpuslab_flush_attr.attr, | |
4547 | &deactivate_full_attr.attr, | |
4548 | &deactivate_empty_attr.attr, | |
4549 | &deactivate_to_head_attr.attr, | |
4550 | &deactivate_to_tail_attr.attr, | |
4551 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4552 | &order_fallback_attr.attr, |
81819f0f | 4553 | #endif |
4c13dd3b DM |
4554 | #ifdef CONFIG_FAILSLAB |
4555 | &failslab_attr.attr, | |
4556 | #endif | |
4557 | ||
81819f0f CL |
4558 | NULL |
4559 | }; | |
4560 | ||
4561 | static struct attribute_group slab_attr_group = { | |
4562 | .attrs = slab_attrs, | |
4563 | }; | |
4564 | ||
4565 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4566 | struct attribute *attr, | |
4567 | char *buf) | |
4568 | { | |
4569 | struct slab_attribute *attribute; | |
4570 | struct kmem_cache *s; | |
4571 | int err; | |
4572 | ||
4573 | attribute = to_slab_attr(attr); | |
4574 | s = to_slab(kobj); | |
4575 | ||
4576 | if (!attribute->show) | |
4577 | return -EIO; | |
4578 | ||
4579 | err = attribute->show(s, buf); | |
4580 | ||
4581 | return err; | |
4582 | } | |
4583 | ||
4584 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4585 | struct attribute *attr, | |
4586 | const char *buf, size_t len) | |
4587 | { | |
4588 | struct slab_attribute *attribute; | |
4589 | struct kmem_cache *s; | |
4590 | int err; | |
4591 | ||
4592 | attribute = to_slab_attr(attr); | |
4593 | s = to_slab(kobj); | |
4594 | ||
4595 | if (!attribute->store) | |
4596 | return -EIO; | |
4597 | ||
4598 | err = attribute->store(s, buf, len); | |
4599 | ||
4600 | return err; | |
4601 | } | |
4602 | ||
151c602f CL |
4603 | static void kmem_cache_release(struct kobject *kobj) |
4604 | { | |
4605 | struct kmem_cache *s = to_slab(kobj); | |
4606 | ||
84c1cf62 | 4607 | kfree(s->name); |
151c602f CL |
4608 | kfree(s); |
4609 | } | |
4610 | ||
52cf25d0 | 4611 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
4612 | .show = slab_attr_show, |
4613 | .store = slab_attr_store, | |
4614 | }; | |
4615 | ||
4616 | static struct kobj_type slab_ktype = { | |
4617 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4618 | .release = kmem_cache_release |
81819f0f CL |
4619 | }; |
4620 | ||
4621 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4622 | { | |
4623 | struct kobj_type *ktype = get_ktype(kobj); | |
4624 | ||
4625 | if (ktype == &slab_ktype) | |
4626 | return 1; | |
4627 | return 0; | |
4628 | } | |
4629 | ||
9cd43611 | 4630 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
4631 | .filter = uevent_filter, |
4632 | }; | |
4633 | ||
27c3a314 | 4634 | static struct kset *slab_kset; |
81819f0f CL |
4635 | |
4636 | #define ID_STR_LENGTH 64 | |
4637 | ||
4638 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4639 | * |
4640 | * Format :[flags-]size | |
81819f0f CL |
4641 | */ |
4642 | static char *create_unique_id(struct kmem_cache *s) | |
4643 | { | |
4644 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4645 | char *p = name; | |
4646 | ||
4647 | BUG_ON(!name); | |
4648 | ||
4649 | *p++ = ':'; | |
4650 | /* | |
4651 | * First flags affecting slabcache operations. We will only | |
4652 | * get here for aliasable slabs so we do not need to support | |
4653 | * too many flags. The flags here must cover all flags that | |
4654 | * are matched during merging to guarantee that the id is | |
4655 | * unique. | |
4656 | */ | |
4657 | if (s->flags & SLAB_CACHE_DMA) | |
4658 | *p++ = 'd'; | |
4659 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4660 | *p++ = 'a'; | |
4661 | if (s->flags & SLAB_DEBUG_FREE) | |
4662 | *p++ = 'F'; | |
5a896d9e VN |
4663 | if (!(s->flags & SLAB_NOTRACK)) |
4664 | *p++ = 't'; | |
81819f0f CL |
4665 | if (p != name + 1) |
4666 | *p++ = '-'; | |
4667 | p += sprintf(p, "%07d", s->size); | |
4668 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4669 | return name; | |
4670 | } | |
4671 | ||
4672 | static int sysfs_slab_add(struct kmem_cache *s) | |
4673 | { | |
4674 | int err; | |
4675 | const char *name; | |
4676 | int unmergeable; | |
4677 | ||
4678 | if (slab_state < SYSFS) | |
4679 | /* Defer until later */ | |
4680 | return 0; | |
4681 | ||
4682 | unmergeable = slab_unmergeable(s); | |
4683 | if (unmergeable) { | |
4684 | /* | |
4685 | * Slabcache can never be merged so we can use the name proper. | |
4686 | * This is typically the case for debug situations. In that | |
4687 | * case we can catch duplicate names easily. | |
4688 | */ | |
27c3a314 | 4689 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4690 | name = s->name; |
4691 | } else { | |
4692 | /* | |
4693 | * Create a unique name for the slab as a target | |
4694 | * for the symlinks. | |
4695 | */ | |
4696 | name = create_unique_id(s); | |
4697 | } | |
4698 | ||
27c3a314 | 4699 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4700 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4701 | if (err) { | |
4702 | kobject_put(&s->kobj); | |
81819f0f | 4703 | return err; |
1eada11c | 4704 | } |
81819f0f CL |
4705 | |
4706 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
4707 | if (err) { |
4708 | kobject_del(&s->kobj); | |
4709 | kobject_put(&s->kobj); | |
81819f0f | 4710 | return err; |
5788d8ad | 4711 | } |
81819f0f CL |
4712 | kobject_uevent(&s->kobj, KOBJ_ADD); |
4713 | if (!unmergeable) { | |
4714 | /* Setup first alias */ | |
4715 | sysfs_slab_alias(s, s->name); | |
4716 | kfree(name); | |
4717 | } | |
4718 | return 0; | |
4719 | } | |
4720 | ||
4721 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4722 | { | |
2bce6485 CL |
4723 | if (slab_state < SYSFS) |
4724 | /* | |
4725 | * Sysfs has not been setup yet so no need to remove the | |
4726 | * cache from sysfs. | |
4727 | */ | |
4728 | return; | |
4729 | ||
81819f0f CL |
4730 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
4731 | kobject_del(&s->kobj); | |
151c602f | 4732 | kobject_put(&s->kobj); |
81819f0f CL |
4733 | } |
4734 | ||
4735 | /* | |
4736 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4737 | * available lest we lose that information. |
81819f0f CL |
4738 | */ |
4739 | struct saved_alias { | |
4740 | struct kmem_cache *s; | |
4741 | const char *name; | |
4742 | struct saved_alias *next; | |
4743 | }; | |
4744 | ||
5af328a5 | 4745 | static struct saved_alias *alias_list; |
81819f0f CL |
4746 | |
4747 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4748 | { | |
4749 | struct saved_alias *al; | |
4750 | ||
4751 | if (slab_state == SYSFS) { | |
4752 | /* | |
4753 | * If we have a leftover link then remove it. | |
4754 | */ | |
27c3a314 GKH |
4755 | sysfs_remove_link(&slab_kset->kobj, name); |
4756 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4757 | } |
4758 | ||
4759 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4760 | if (!al) | |
4761 | return -ENOMEM; | |
4762 | ||
4763 | al->s = s; | |
4764 | al->name = name; | |
4765 | al->next = alias_list; | |
4766 | alias_list = al; | |
4767 | return 0; | |
4768 | } | |
4769 | ||
4770 | static int __init slab_sysfs_init(void) | |
4771 | { | |
5b95a4ac | 4772 | struct kmem_cache *s; |
81819f0f CL |
4773 | int err; |
4774 | ||
2bce6485 CL |
4775 | down_write(&slub_lock); |
4776 | ||
0ff21e46 | 4777 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4778 | if (!slab_kset) { |
2bce6485 | 4779 | up_write(&slub_lock); |
81819f0f CL |
4780 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4781 | return -ENOSYS; | |
4782 | } | |
4783 | ||
26a7bd03 CL |
4784 | slab_state = SYSFS; |
4785 | ||
5b95a4ac | 4786 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4787 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4788 | if (err) |
4789 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4790 | " to sysfs\n", s->name); | |
26a7bd03 | 4791 | } |
81819f0f CL |
4792 | |
4793 | while (alias_list) { | |
4794 | struct saved_alias *al = alias_list; | |
4795 | ||
4796 | alias_list = alias_list->next; | |
4797 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4798 | if (err) |
4799 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4800 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4801 | kfree(al); |
4802 | } | |
4803 | ||
2bce6485 | 4804 | up_write(&slub_lock); |
81819f0f CL |
4805 | resiliency_test(); |
4806 | return 0; | |
4807 | } | |
4808 | ||
4809 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 4810 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
4811 | |
4812 | /* | |
4813 | * The /proc/slabinfo ABI | |
4814 | */ | |
158a9624 | 4815 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4816 | static void print_slabinfo_header(struct seq_file *m) |
4817 | { | |
4818 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4819 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4820 | "<objperslab> <pagesperslab>"); | |
4821 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4822 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4823 | seq_putc(m, '\n'); | |
4824 | } | |
4825 | ||
4826 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4827 | { | |
4828 | loff_t n = *pos; | |
4829 | ||
4830 | down_read(&slub_lock); | |
4831 | if (!n) | |
4832 | print_slabinfo_header(m); | |
4833 | ||
4834 | return seq_list_start(&slab_caches, *pos); | |
4835 | } | |
4836 | ||
4837 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4838 | { | |
4839 | return seq_list_next(p, &slab_caches, pos); | |
4840 | } | |
4841 | ||
4842 | static void s_stop(struct seq_file *m, void *p) | |
4843 | { | |
4844 | up_read(&slub_lock); | |
4845 | } | |
4846 | ||
4847 | static int s_show(struct seq_file *m, void *p) | |
4848 | { | |
4849 | unsigned long nr_partials = 0; | |
4850 | unsigned long nr_slabs = 0; | |
4851 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4852 | unsigned long nr_objs = 0; |
4853 | unsigned long nr_free = 0; | |
57ed3eda PE |
4854 | struct kmem_cache *s; |
4855 | int node; | |
4856 | ||
4857 | s = list_entry(p, struct kmem_cache, list); | |
4858 | ||
4859 | for_each_online_node(node) { | |
4860 | struct kmem_cache_node *n = get_node(s, node); | |
4861 | ||
4862 | if (!n) | |
4863 | continue; | |
4864 | ||
4865 | nr_partials += n->nr_partial; | |
4866 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4867 | nr_objs += atomic_long_read(&n->total_objects); |
4868 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4869 | } |
4870 | ||
205ab99d | 4871 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4872 | |
4873 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4874 | nr_objs, s->size, oo_objects(s->oo), |
4875 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4876 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4877 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4878 | 0UL); | |
4879 | seq_putc(m, '\n'); | |
4880 | return 0; | |
4881 | } | |
4882 | ||
7b3c3a50 | 4883 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4884 | .start = s_start, |
4885 | .next = s_next, | |
4886 | .stop = s_stop, | |
4887 | .show = s_show, | |
4888 | }; | |
4889 | ||
7b3c3a50 AD |
4890 | static int slabinfo_open(struct inode *inode, struct file *file) |
4891 | { | |
4892 | return seq_open(file, &slabinfo_op); | |
4893 | } | |
4894 | ||
4895 | static const struct file_operations proc_slabinfo_operations = { | |
4896 | .open = slabinfo_open, | |
4897 | .read = seq_read, | |
4898 | .llseek = seq_lseek, | |
4899 | .release = seq_release, | |
4900 | }; | |
4901 | ||
4902 | static int __init slab_proc_init(void) | |
4903 | { | |
cf5d1131 | 4904 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
4905 | return 0; |
4906 | } | |
4907 | module_init(slab_proc_init); | |
158a9624 | 4908 | #endif /* CONFIG_SLABINFO */ |