]>
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 | * | |
881db7fb CL |
5 | * The allocator synchronizes using per slab locks or atomic operatios |
6 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 7 | * |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 9 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
10 | */ |
11 | ||
12 | #include <linux/mm.h> | |
1eb5ac64 | 13 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
14 | #include <linux/module.h> |
15 | #include <linux/bit_spinlock.h> | |
16 | #include <linux/interrupt.h> | |
17 | #include <linux/bitops.h> | |
18 | #include <linux/slab.h> | |
97d06609 | 19 | #include "slab.h" |
7b3c3a50 | 20 | #include <linux/proc_fs.h> |
3ac38faa | 21 | #include <linux/notifier.h> |
81819f0f | 22 | #include <linux/seq_file.h> |
5a896d9e | 23 | #include <linux/kmemcheck.h> |
81819f0f CL |
24 | #include <linux/cpu.h> |
25 | #include <linux/cpuset.h> | |
26 | #include <linux/mempolicy.h> | |
27 | #include <linux/ctype.h> | |
3ac7fe5a | 28 | #include <linux/debugobjects.h> |
81819f0f | 29 | #include <linux/kallsyms.h> |
b9049e23 | 30 | #include <linux/memory.h> |
f8bd2258 | 31 | #include <linux/math64.h> |
773ff60e | 32 | #include <linux/fault-inject.h> |
bfa71457 | 33 | #include <linux/stacktrace.h> |
4de900b4 | 34 | #include <linux/prefetch.h> |
2633d7a0 | 35 | #include <linux/memcontrol.h> |
81819f0f | 36 | |
4a92379b RK |
37 | #include <trace/events/kmem.h> |
38 | ||
072bb0aa MG |
39 | #include "internal.h" |
40 | ||
81819f0f CL |
41 | /* |
42 | * Lock order: | |
18004c5d | 43 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
44 | * 2. node->list_lock |
45 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 46 | * |
18004c5d | 47 | * slab_mutex |
881db7fb | 48 | * |
18004c5d | 49 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
50 | * and to synchronize major metadata changes to slab cache structures. |
51 | * | |
52 | * The slab_lock is only used for debugging and on arches that do not | |
53 | * have the ability to do a cmpxchg_double. It only protects the second | |
54 | * double word in the page struct. Meaning | |
55 | * A. page->freelist -> List of object free in a page | |
56 | * B. page->counters -> Counters of objects | |
57 | * C. page->frozen -> frozen state | |
58 | * | |
59 | * If a slab is frozen then it is exempt from list management. It is not | |
60 | * on any list. The processor that froze the slab is the one who can | |
61 | * perform list operations on the page. Other processors may put objects | |
62 | * onto the freelist but the processor that froze the slab is the only | |
63 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
64 | * |
65 | * The list_lock protects the partial and full list on each node and | |
66 | * the partial slab counter. If taken then no new slabs may be added or | |
67 | * removed from the lists nor make the number of partial slabs be modified. | |
68 | * (Note that the total number of slabs is an atomic value that may be | |
69 | * modified without taking the list lock). | |
70 | * | |
71 | * The list_lock is a centralized lock and thus we avoid taking it as | |
72 | * much as possible. As long as SLUB does not have to handle partial | |
73 | * slabs, operations can continue without any centralized lock. F.e. | |
74 | * allocating a long series of objects that fill up slabs does not require | |
75 | * the list lock. | |
81819f0f CL |
76 | * Interrupts are disabled during allocation and deallocation in order to |
77 | * make the slab allocator safe to use in the context of an irq. In addition | |
78 | * interrupts are disabled to ensure that the processor does not change | |
79 | * while handling per_cpu slabs, due to kernel preemption. | |
80 | * | |
81 | * SLUB assigns one slab for allocation to each processor. | |
82 | * Allocations only occur from these slabs called cpu slabs. | |
83 | * | |
672bba3a CL |
84 | * Slabs with free elements are kept on a partial list and during regular |
85 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 86 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
87 | * We track full slabs for debugging purposes though because otherwise we |
88 | * cannot scan all objects. | |
81819f0f CL |
89 | * |
90 | * Slabs are freed when they become empty. Teardown and setup is | |
91 | * minimal so we rely on the page allocators per cpu caches for | |
92 | * fast frees and allocs. | |
93 | * | |
94 | * Overloading of page flags that are otherwise used for LRU management. | |
95 | * | |
4b6f0750 CL |
96 | * PageActive The slab is frozen and exempt from list processing. |
97 | * This means that the slab is dedicated to a purpose | |
98 | * such as satisfying allocations for a specific | |
99 | * processor. Objects may be freed in the slab while | |
100 | * it is frozen but slab_free will then skip the usual | |
101 | * list operations. It is up to the processor holding | |
102 | * the slab to integrate the slab into the slab lists | |
103 | * when the slab is no longer needed. | |
104 | * | |
105 | * One use of this flag is to mark slabs that are | |
106 | * used for allocations. Then such a slab becomes a cpu | |
107 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 108 | * freelist that allows lockless access to |
894b8788 CL |
109 | * free objects in addition to the regular freelist |
110 | * that requires the slab lock. | |
81819f0f CL |
111 | * |
112 | * PageError Slab requires special handling due to debug | |
113 | * options set. This moves slab handling out of | |
894b8788 | 114 | * the fast path and disables lockless freelists. |
81819f0f CL |
115 | */ |
116 | ||
af537b0a CL |
117 | static inline int kmem_cache_debug(struct kmem_cache *s) |
118 | { | |
5577bd8a | 119 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 120 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 121 | #else |
af537b0a | 122 | return 0; |
5577bd8a | 123 | #endif |
af537b0a | 124 | } |
5577bd8a | 125 | |
345c905d JK |
126 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
127 | { | |
128 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
129 | return !kmem_cache_debug(s); | |
130 | #else | |
131 | return false; | |
132 | #endif | |
133 | } | |
134 | ||
81819f0f CL |
135 | /* |
136 | * Issues still to be resolved: | |
137 | * | |
81819f0f CL |
138 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
139 | * | |
81819f0f CL |
140 | * - Variable sizing of the per node arrays |
141 | */ | |
142 | ||
143 | /* Enable to test recovery from slab corruption on boot */ | |
144 | #undef SLUB_RESILIENCY_TEST | |
145 | ||
b789ef51 CL |
146 | /* Enable to log cmpxchg failures */ |
147 | #undef SLUB_DEBUG_CMPXCHG | |
148 | ||
2086d26a CL |
149 | /* |
150 | * Mininum number of partial slabs. These will be left on the partial | |
151 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
152 | */ | |
76be8950 | 153 | #define MIN_PARTIAL 5 |
e95eed57 | 154 | |
2086d26a CL |
155 | /* |
156 | * Maximum number of desirable partial slabs. | |
157 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 158 | * sort the partial list by the number of objects in use. |
2086d26a CL |
159 | */ |
160 | #define MAX_PARTIAL 10 | |
161 | ||
81819f0f CL |
162 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
163 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 164 | |
fa5ec8a1 | 165 | /* |
3de47213 DR |
166 | * Debugging flags that require metadata to be stored in the slab. These get |
167 | * disabled when slub_debug=O is used and a cache's min order increases with | |
168 | * metadata. | |
fa5ec8a1 | 169 | */ |
3de47213 | 170 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 171 | |
81819f0f CL |
172 | /* |
173 | * Set of flags that will prevent slab merging | |
174 | */ | |
175 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
176 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
177 | SLAB_FAILSLAB) | |
81819f0f CL |
178 | |
179 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 180 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 181 | |
210b5c06 CG |
182 | #define OO_SHIFT 16 |
183 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 184 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 185 | |
81819f0f | 186 | /* Internal SLUB flags */ |
f90ec390 | 187 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 188 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 189 | |
81819f0f CL |
190 | #ifdef CONFIG_SMP |
191 | static struct notifier_block slab_notifier; | |
192 | #endif | |
193 | ||
02cbc874 CL |
194 | /* |
195 | * Tracking user of a slab. | |
196 | */ | |
d6543e39 | 197 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 198 | struct track { |
ce71e27c | 199 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
200 | #ifdef CONFIG_STACKTRACE |
201 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
202 | #endif | |
02cbc874 CL |
203 | int cpu; /* Was running on cpu */ |
204 | int pid; /* Pid context */ | |
205 | unsigned long when; /* When did the operation occur */ | |
206 | }; | |
207 | ||
208 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
209 | ||
ab4d5ed5 | 210 | #ifdef CONFIG_SYSFS |
81819f0f CL |
211 | static int sysfs_slab_add(struct kmem_cache *); |
212 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
213 | static void sysfs_slab_remove(struct kmem_cache *); | |
107dab5c | 214 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
81819f0f | 215 | #else |
0c710013 CL |
216 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
217 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
218 | { return 0; } | |
db265eca | 219 | static inline void sysfs_slab_remove(struct kmem_cache *s) { } |
8ff12cfc | 220 | |
107dab5c | 221 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
222 | #endif |
223 | ||
4fdccdfb | 224 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
225 | { |
226 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 227 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
228 | #endif |
229 | } | |
230 | ||
81819f0f CL |
231 | /******************************************************************** |
232 | * Core slab cache functions | |
233 | *******************************************************************/ | |
234 | ||
81819f0f CL |
235 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) |
236 | { | |
81819f0f | 237 | return s->node[node]; |
81819f0f CL |
238 | } |
239 | ||
6446faa2 | 240 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
241 | static inline int check_valid_pointer(struct kmem_cache *s, |
242 | struct page *page, const void *object) | |
243 | { | |
244 | void *base; | |
245 | ||
a973e9dd | 246 | if (!object) |
02cbc874 CL |
247 | return 1; |
248 | ||
a973e9dd | 249 | base = page_address(page); |
39b26464 | 250 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
251 | (object - base) % s->size) { |
252 | return 0; | |
253 | } | |
254 | ||
255 | return 1; | |
256 | } | |
257 | ||
7656c72b CL |
258 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
259 | { | |
260 | return *(void **)(object + s->offset); | |
261 | } | |
262 | ||
0ad9500e ED |
263 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
264 | { | |
265 | prefetch(object + s->offset); | |
266 | } | |
267 | ||
1393d9a1 CL |
268 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
269 | { | |
270 | void *p; | |
271 | ||
272 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
273 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
274 | #else | |
275 | p = get_freepointer(s, object); | |
276 | #endif | |
277 | return p; | |
278 | } | |
279 | ||
7656c72b CL |
280 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
281 | { | |
282 | *(void **)(object + s->offset) = fp; | |
283 | } | |
284 | ||
285 | /* Loop over all objects in a slab */ | |
224a88be CL |
286 | #define for_each_object(__p, __s, __addr, __objects) \ |
287 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
288 | __p += (__s)->size) |
289 | ||
7656c72b CL |
290 | /* Determine object index from a given position */ |
291 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
292 | { | |
293 | return (p - addr) / s->size; | |
294 | } | |
295 | ||
d71f606f MK |
296 | static inline size_t slab_ksize(const struct kmem_cache *s) |
297 | { | |
298 | #ifdef CONFIG_SLUB_DEBUG | |
299 | /* | |
300 | * Debugging requires use of the padding between object | |
301 | * and whatever may come after it. | |
302 | */ | |
303 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
3b0efdfa | 304 | return s->object_size; |
d71f606f MK |
305 | |
306 | #endif | |
307 | /* | |
308 | * If we have the need to store the freelist pointer | |
309 | * back there or track user information then we can | |
310 | * only use the space before that information. | |
311 | */ | |
312 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
313 | return s->inuse; | |
314 | /* | |
315 | * Else we can use all the padding etc for the allocation | |
316 | */ | |
317 | return s->size; | |
318 | } | |
319 | ||
ab9a0f19 LJ |
320 | static inline int order_objects(int order, unsigned long size, int reserved) |
321 | { | |
322 | return ((PAGE_SIZE << order) - reserved) / size; | |
323 | } | |
324 | ||
834f3d11 | 325 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 326 | unsigned long size, int reserved) |
834f3d11 CL |
327 | { |
328 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 329 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
330 | }; |
331 | ||
332 | return x; | |
333 | } | |
334 | ||
335 | static inline int oo_order(struct kmem_cache_order_objects x) | |
336 | { | |
210b5c06 | 337 | return x.x >> OO_SHIFT; |
834f3d11 CL |
338 | } |
339 | ||
340 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
341 | { | |
210b5c06 | 342 | return x.x & OO_MASK; |
834f3d11 CL |
343 | } |
344 | ||
881db7fb CL |
345 | /* |
346 | * Per slab locking using the pagelock | |
347 | */ | |
348 | static __always_inline void slab_lock(struct page *page) | |
349 | { | |
350 | bit_spin_lock(PG_locked, &page->flags); | |
351 | } | |
352 | ||
353 | static __always_inline void slab_unlock(struct page *page) | |
354 | { | |
355 | __bit_spin_unlock(PG_locked, &page->flags); | |
356 | } | |
357 | ||
a0320865 DH |
358 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
359 | { | |
360 | struct page tmp; | |
361 | tmp.counters = counters_new; | |
362 | /* | |
363 | * page->counters can cover frozen/inuse/objects as well | |
364 | * as page->_count. If we assign to ->counters directly | |
365 | * we run the risk of losing updates to page->_count, so | |
366 | * be careful and only assign to the fields we need. | |
367 | */ | |
368 | page->frozen = tmp.frozen; | |
369 | page->inuse = tmp.inuse; | |
370 | page->objects = tmp.objects; | |
371 | } | |
372 | ||
1d07171c CL |
373 | /* Interrupts must be disabled (for the fallback code to work right) */ |
374 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
375 | void *freelist_old, unsigned long counters_old, | |
376 | void *freelist_new, unsigned long counters_new, | |
377 | const char *n) | |
378 | { | |
379 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
380 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
381 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 382 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 383 | if (cmpxchg_double(&page->freelist, &page->counters, |
1d07171c CL |
384 | freelist_old, counters_old, |
385 | freelist_new, counters_new)) | |
386 | return 1; | |
387 | } else | |
388 | #endif | |
389 | { | |
390 | slab_lock(page); | |
d0e0ac97 CG |
391 | if (page->freelist == freelist_old && |
392 | page->counters == counters_old) { | |
1d07171c | 393 | page->freelist = freelist_new; |
a0320865 | 394 | set_page_slub_counters(page, counters_new); |
1d07171c CL |
395 | slab_unlock(page); |
396 | return 1; | |
397 | } | |
398 | slab_unlock(page); | |
399 | } | |
400 | ||
401 | cpu_relax(); | |
402 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
403 | ||
404 | #ifdef SLUB_DEBUG_CMPXCHG | |
405 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
406 | #endif | |
407 | ||
408 | return 0; | |
409 | } | |
410 | ||
b789ef51 CL |
411 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
412 | void *freelist_old, unsigned long counters_old, | |
413 | void *freelist_new, unsigned long counters_new, | |
414 | const char *n) | |
415 | { | |
2565409f HC |
416 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
417 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 418 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 419 | if (cmpxchg_double(&page->freelist, &page->counters, |
b789ef51 CL |
420 | freelist_old, counters_old, |
421 | freelist_new, counters_new)) | |
422 | return 1; | |
423 | } else | |
424 | #endif | |
425 | { | |
1d07171c CL |
426 | unsigned long flags; |
427 | ||
428 | local_irq_save(flags); | |
881db7fb | 429 | slab_lock(page); |
d0e0ac97 CG |
430 | if (page->freelist == freelist_old && |
431 | page->counters == counters_old) { | |
b789ef51 | 432 | page->freelist = freelist_new; |
a0320865 | 433 | set_page_slub_counters(page, counters_new); |
881db7fb | 434 | slab_unlock(page); |
1d07171c | 435 | local_irq_restore(flags); |
b789ef51 CL |
436 | return 1; |
437 | } | |
881db7fb | 438 | slab_unlock(page); |
1d07171c | 439 | local_irq_restore(flags); |
b789ef51 CL |
440 | } |
441 | ||
442 | cpu_relax(); | |
443 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
444 | ||
445 | #ifdef SLUB_DEBUG_CMPXCHG | |
446 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
447 | #endif | |
448 | ||
449 | return 0; | |
450 | } | |
451 | ||
41ecc55b | 452 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
453 | /* |
454 | * Determine a map of object in use on a page. | |
455 | * | |
881db7fb | 456 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
457 | * not vanish from under us. |
458 | */ | |
459 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
460 | { | |
461 | void *p; | |
462 | void *addr = page_address(page); | |
463 | ||
464 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
465 | set_bit(slab_index(p, s, addr), map); | |
466 | } | |
467 | ||
41ecc55b CL |
468 | /* |
469 | * Debug settings: | |
470 | */ | |
f0630fff CL |
471 | #ifdef CONFIG_SLUB_DEBUG_ON |
472 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
473 | #else | |
41ecc55b | 474 | static int slub_debug; |
f0630fff | 475 | #endif |
41ecc55b CL |
476 | |
477 | static char *slub_debug_slabs; | |
fa5ec8a1 | 478 | static int disable_higher_order_debug; |
41ecc55b | 479 | |
81819f0f CL |
480 | /* |
481 | * Object debugging | |
482 | */ | |
483 | static void print_section(char *text, u8 *addr, unsigned int length) | |
484 | { | |
ffc79d28 SAS |
485 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
486 | length, 1); | |
81819f0f CL |
487 | } |
488 | ||
81819f0f CL |
489 | static struct track *get_track(struct kmem_cache *s, void *object, |
490 | enum track_item alloc) | |
491 | { | |
492 | struct track *p; | |
493 | ||
494 | if (s->offset) | |
495 | p = object + s->offset + sizeof(void *); | |
496 | else | |
497 | p = object + s->inuse; | |
498 | ||
499 | return p + alloc; | |
500 | } | |
501 | ||
502 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 503 | enum track_item alloc, unsigned long addr) |
81819f0f | 504 | { |
1a00df4a | 505 | struct track *p = get_track(s, object, alloc); |
81819f0f | 506 | |
81819f0f | 507 | if (addr) { |
d6543e39 BG |
508 | #ifdef CONFIG_STACKTRACE |
509 | struct stack_trace trace; | |
510 | int i; | |
511 | ||
512 | trace.nr_entries = 0; | |
513 | trace.max_entries = TRACK_ADDRS_COUNT; | |
514 | trace.entries = p->addrs; | |
515 | trace.skip = 3; | |
516 | save_stack_trace(&trace); | |
517 | ||
518 | /* See rant in lockdep.c */ | |
519 | if (trace.nr_entries != 0 && | |
520 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
521 | trace.nr_entries--; | |
522 | ||
523 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
524 | p->addrs[i] = 0; | |
525 | #endif | |
81819f0f CL |
526 | p->addr = addr; |
527 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 528 | p->pid = current->pid; |
81819f0f CL |
529 | p->when = jiffies; |
530 | } else | |
531 | memset(p, 0, sizeof(struct track)); | |
532 | } | |
533 | ||
81819f0f CL |
534 | static void init_tracking(struct kmem_cache *s, void *object) |
535 | { | |
24922684 CL |
536 | if (!(s->flags & SLAB_STORE_USER)) |
537 | return; | |
538 | ||
ce71e27c EGM |
539 | set_track(s, object, TRACK_FREE, 0UL); |
540 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
541 | } |
542 | ||
543 | static void print_track(const char *s, struct track *t) | |
544 | { | |
545 | if (!t->addr) | |
546 | return; | |
547 | ||
7daf705f | 548 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 549 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
d6543e39 BG |
550 | #ifdef CONFIG_STACKTRACE |
551 | { | |
552 | int i; | |
553 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
554 | if (t->addrs[i]) | |
555 | printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]); | |
556 | else | |
557 | break; | |
558 | } | |
559 | #endif | |
24922684 CL |
560 | } |
561 | ||
562 | static void print_tracking(struct kmem_cache *s, void *object) | |
563 | { | |
564 | if (!(s->flags & SLAB_STORE_USER)) | |
565 | return; | |
566 | ||
567 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
568 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
569 | } | |
570 | ||
571 | static void print_page_info(struct page *page) | |
572 | { | |
d0e0ac97 CG |
573 | printk(KERN_ERR |
574 | "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", | |
575 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
576 | |
577 | } | |
578 | ||
579 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
580 | { | |
581 | va_list args; | |
582 | char buf[100]; | |
583 | ||
584 | va_start(args, fmt); | |
585 | vsnprintf(buf, sizeof(buf), fmt, args); | |
586 | va_end(args); | |
587 | printk(KERN_ERR "========================================" | |
588 | "=====================================\n"); | |
265d47e7 | 589 | printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf); |
24922684 CL |
590 | printk(KERN_ERR "----------------------------------------" |
591 | "-------------------------------------\n\n"); | |
645df230 | 592 | |
373d4d09 | 593 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
594 | } |
595 | ||
24922684 CL |
596 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
597 | { | |
598 | va_list args; | |
599 | char buf[100]; | |
600 | ||
601 | va_start(args, fmt); | |
602 | vsnprintf(buf, sizeof(buf), fmt, args); | |
603 | va_end(args); | |
604 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
605 | } | |
606 | ||
607 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
608 | { |
609 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 610 | u8 *addr = page_address(page); |
24922684 CL |
611 | |
612 | print_tracking(s, p); | |
613 | ||
614 | print_page_info(page); | |
615 | ||
616 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
617 | p, p - addr, get_freepointer(s, p)); | |
618 | ||
619 | if (p > addr + 16) | |
ffc79d28 | 620 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 621 | |
3b0efdfa | 622 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 623 | PAGE_SIZE)); |
81819f0f | 624 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
625 | print_section("Redzone ", p + s->object_size, |
626 | s->inuse - s->object_size); | |
81819f0f | 627 | |
81819f0f CL |
628 | if (s->offset) |
629 | off = s->offset + sizeof(void *); | |
630 | else | |
631 | off = s->inuse; | |
632 | ||
24922684 | 633 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 634 | off += 2 * sizeof(struct track); |
81819f0f CL |
635 | |
636 | if (off != s->size) | |
637 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 638 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
639 | |
640 | dump_stack(); | |
81819f0f CL |
641 | } |
642 | ||
643 | static void object_err(struct kmem_cache *s, struct page *page, | |
644 | u8 *object, char *reason) | |
645 | { | |
3dc50637 | 646 | slab_bug(s, "%s", reason); |
24922684 | 647 | print_trailer(s, page, object); |
81819f0f CL |
648 | } |
649 | ||
d0e0ac97 CG |
650 | static void slab_err(struct kmem_cache *s, struct page *page, |
651 | const char *fmt, ...) | |
81819f0f CL |
652 | { |
653 | va_list args; | |
654 | char buf[100]; | |
655 | ||
24922684 CL |
656 | va_start(args, fmt); |
657 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 658 | va_end(args); |
3dc50637 | 659 | slab_bug(s, "%s", buf); |
24922684 | 660 | print_page_info(page); |
81819f0f CL |
661 | dump_stack(); |
662 | } | |
663 | ||
f7cb1933 | 664 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
665 | { |
666 | u8 *p = object; | |
667 | ||
668 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
669 | memset(p, POISON_FREE, s->object_size - 1); |
670 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
671 | } |
672 | ||
673 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 674 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
675 | } |
676 | ||
24922684 CL |
677 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
678 | void *from, void *to) | |
679 | { | |
680 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
681 | memset(from, data, to - from); | |
682 | } | |
683 | ||
684 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
685 | u8 *object, char *what, | |
06428780 | 686 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
687 | { |
688 | u8 *fault; | |
689 | u8 *end; | |
690 | ||
79824820 | 691 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
692 | if (!fault) |
693 | return 1; | |
694 | ||
695 | end = start + bytes; | |
696 | while (end > fault && end[-1] == value) | |
697 | end--; | |
698 | ||
699 | slab_bug(s, "%s overwritten", what); | |
700 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
701 | fault, end - 1, fault[0], value); | |
702 | print_trailer(s, page, object); | |
703 | ||
704 | restore_bytes(s, what, value, fault, end); | |
705 | return 0; | |
81819f0f CL |
706 | } |
707 | ||
81819f0f CL |
708 | /* |
709 | * Object layout: | |
710 | * | |
711 | * object address | |
712 | * Bytes of the object to be managed. | |
713 | * If the freepointer may overlay the object then the free | |
714 | * pointer is the first word of the object. | |
672bba3a | 715 | * |
81819f0f CL |
716 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
717 | * 0xa5 (POISON_END) | |
718 | * | |
3b0efdfa | 719 | * object + s->object_size |
81819f0f | 720 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 721 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 722 | * object_size == inuse. |
672bba3a | 723 | * |
81819f0f CL |
724 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
725 | * 0xcc (RED_ACTIVE) for objects in use. | |
726 | * | |
727 | * object + s->inuse | |
672bba3a CL |
728 | * Meta data starts here. |
729 | * | |
81819f0f CL |
730 | * A. Free pointer (if we cannot overwrite object on free) |
731 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 732 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 733 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
734 | * before the word boundary. |
735 | * | |
736 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
737 | * |
738 | * object + s->size | |
672bba3a | 739 | * Nothing is used beyond s->size. |
81819f0f | 740 | * |
3b0efdfa | 741 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 742 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
743 | * may be used with merged slabcaches. |
744 | */ | |
745 | ||
81819f0f CL |
746 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
747 | { | |
748 | unsigned long off = s->inuse; /* The end of info */ | |
749 | ||
750 | if (s->offset) | |
751 | /* Freepointer is placed after the object. */ | |
752 | off += sizeof(void *); | |
753 | ||
754 | if (s->flags & SLAB_STORE_USER) | |
755 | /* We also have user information there */ | |
756 | off += 2 * sizeof(struct track); | |
757 | ||
758 | if (s->size == off) | |
759 | return 1; | |
760 | ||
24922684 CL |
761 | return check_bytes_and_report(s, page, p, "Object padding", |
762 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
763 | } |
764 | ||
39b26464 | 765 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
766 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
767 | { | |
24922684 CL |
768 | u8 *start; |
769 | u8 *fault; | |
770 | u8 *end; | |
771 | int length; | |
772 | int remainder; | |
81819f0f CL |
773 | |
774 | if (!(s->flags & SLAB_POISON)) | |
775 | return 1; | |
776 | ||
a973e9dd | 777 | start = page_address(page); |
ab9a0f19 | 778 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
779 | end = start + length; |
780 | remainder = length % s->size; | |
81819f0f CL |
781 | if (!remainder) |
782 | return 1; | |
783 | ||
79824820 | 784 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
785 | if (!fault) |
786 | return 1; | |
787 | while (end > fault && end[-1] == POISON_INUSE) | |
788 | end--; | |
789 | ||
790 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 791 | print_section("Padding ", end - remainder, remainder); |
24922684 | 792 | |
8a3d271d | 793 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 794 | return 0; |
81819f0f CL |
795 | } |
796 | ||
797 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 798 | void *object, u8 val) |
81819f0f CL |
799 | { |
800 | u8 *p = object; | |
3b0efdfa | 801 | u8 *endobject = object + s->object_size; |
81819f0f CL |
802 | |
803 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 804 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 805 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 806 | return 0; |
81819f0f | 807 | } else { |
3b0efdfa | 808 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 809 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
810 | endobject, POISON_INUSE, |
811 | s->inuse - s->object_size); | |
3adbefee | 812 | } |
81819f0f CL |
813 | } |
814 | ||
815 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 816 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 817 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 818 | POISON_FREE, s->object_size - 1) || |
24922684 | 819 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 820 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 821 | return 0; |
81819f0f CL |
822 | /* |
823 | * check_pad_bytes cleans up on its own. | |
824 | */ | |
825 | check_pad_bytes(s, page, p); | |
826 | } | |
827 | ||
f7cb1933 | 828 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
829 | /* |
830 | * Object and freepointer overlap. Cannot check | |
831 | * freepointer while object is allocated. | |
832 | */ | |
833 | return 1; | |
834 | ||
835 | /* Check free pointer validity */ | |
836 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
837 | object_err(s, page, p, "Freepointer corrupt"); | |
838 | /* | |
9f6c708e | 839 | * No choice but to zap it and thus lose the remainder |
81819f0f | 840 | * of the free objects in this slab. May cause |
672bba3a | 841 | * another error because the object count is now wrong. |
81819f0f | 842 | */ |
a973e9dd | 843 | set_freepointer(s, p, NULL); |
81819f0f CL |
844 | return 0; |
845 | } | |
846 | return 1; | |
847 | } | |
848 | ||
849 | static int check_slab(struct kmem_cache *s, struct page *page) | |
850 | { | |
39b26464 CL |
851 | int maxobj; |
852 | ||
81819f0f CL |
853 | VM_BUG_ON(!irqs_disabled()); |
854 | ||
855 | if (!PageSlab(page)) { | |
24922684 | 856 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
857 | return 0; |
858 | } | |
39b26464 | 859 | |
ab9a0f19 | 860 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
861 | if (page->objects > maxobj) { |
862 | slab_err(s, page, "objects %u > max %u", | |
863 | s->name, page->objects, maxobj); | |
864 | return 0; | |
865 | } | |
866 | if (page->inuse > page->objects) { | |
24922684 | 867 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 868 | s->name, page->inuse, page->objects); |
81819f0f CL |
869 | return 0; |
870 | } | |
871 | /* Slab_pad_check fixes things up after itself */ | |
872 | slab_pad_check(s, page); | |
873 | return 1; | |
874 | } | |
875 | ||
876 | /* | |
672bba3a CL |
877 | * Determine if a certain object on a page is on the freelist. Must hold the |
878 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
879 | */ |
880 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
881 | { | |
882 | int nr = 0; | |
881db7fb | 883 | void *fp; |
81819f0f | 884 | void *object = NULL; |
224a88be | 885 | unsigned long max_objects; |
81819f0f | 886 | |
881db7fb | 887 | fp = page->freelist; |
39b26464 | 888 | while (fp && nr <= page->objects) { |
81819f0f CL |
889 | if (fp == search) |
890 | return 1; | |
891 | if (!check_valid_pointer(s, page, fp)) { | |
892 | if (object) { | |
893 | object_err(s, page, object, | |
894 | "Freechain corrupt"); | |
a973e9dd | 895 | set_freepointer(s, object, NULL); |
81819f0f | 896 | } else { |
24922684 | 897 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 898 | page->freelist = NULL; |
39b26464 | 899 | page->inuse = page->objects; |
24922684 | 900 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
901 | return 0; |
902 | } | |
903 | break; | |
904 | } | |
905 | object = fp; | |
906 | fp = get_freepointer(s, object); | |
907 | nr++; | |
908 | } | |
909 | ||
ab9a0f19 | 910 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
911 | if (max_objects > MAX_OBJS_PER_PAGE) |
912 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
913 | |
914 | if (page->objects != max_objects) { | |
915 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
916 | "should be %d", page->objects, max_objects); | |
917 | page->objects = max_objects; | |
918 | slab_fix(s, "Number of objects adjusted."); | |
919 | } | |
39b26464 | 920 | if (page->inuse != page->objects - nr) { |
70d71228 | 921 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
922 | "counted were %d", page->inuse, page->objects - nr); |
923 | page->inuse = page->objects - nr; | |
24922684 | 924 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
925 | } |
926 | return search == NULL; | |
927 | } | |
928 | ||
0121c619 CL |
929 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
930 | int alloc) | |
3ec09742 CL |
931 | { |
932 | if (s->flags & SLAB_TRACE) { | |
933 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
934 | s->name, | |
935 | alloc ? "alloc" : "free", | |
936 | object, page->inuse, | |
937 | page->freelist); | |
938 | ||
939 | if (!alloc) | |
d0e0ac97 CG |
940 | print_section("Object ", (void *)object, |
941 | s->object_size); | |
3ec09742 CL |
942 | |
943 | dump_stack(); | |
944 | } | |
945 | } | |
946 | ||
c016b0bd CL |
947 | /* |
948 | * Hooks for other subsystems that check memory allocations. In a typical | |
949 | * production configuration these hooks all should produce no code at all. | |
950 | */ | |
d56791b3 RB |
951 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
952 | { | |
953 | kmemleak_alloc(ptr, size, 1, flags); | |
954 | } | |
955 | ||
956 | static inline void kfree_hook(const void *x) | |
957 | { | |
958 | kmemleak_free(x); | |
959 | } | |
960 | ||
c016b0bd CL |
961 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) |
962 | { | |
c1d50836 | 963 | flags &= gfp_allowed_mask; |
c016b0bd CL |
964 | lockdep_trace_alloc(flags); |
965 | might_sleep_if(flags & __GFP_WAIT); | |
966 | ||
3b0efdfa | 967 | return should_failslab(s->object_size, flags, s->flags); |
c016b0bd CL |
968 | } |
969 | ||
d0e0ac97 CG |
970 | static inline void slab_post_alloc_hook(struct kmem_cache *s, |
971 | gfp_t flags, void *object) | |
c016b0bd | 972 | { |
c1d50836 | 973 | flags &= gfp_allowed_mask; |
b3d41885 | 974 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
3b0efdfa | 975 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); |
c016b0bd CL |
976 | } |
977 | ||
978 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
979 | { | |
980 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 981 | |
d3f661d6 | 982 | /* |
d1756174 | 983 | * Trouble is that we may no longer disable interrupts in the fast path |
d3f661d6 CL |
984 | * So in order to make the debug calls that expect irqs to be |
985 | * disabled we need to disable interrupts temporarily. | |
986 | */ | |
987 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
988 | { | |
989 | unsigned long flags; | |
990 | ||
991 | local_irq_save(flags); | |
3b0efdfa CL |
992 | kmemcheck_slab_free(s, x, s->object_size); |
993 | debug_check_no_locks_freed(x, s->object_size); | |
d3f661d6 CL |
994 | local_irq_restore(flags); |
995 | } | |
996 | #endif | |
f9b615de | 997 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
3b0efdfa | 998 | debug_check_no_obj_freed(x, s->object_size); |
c016b0bd CL |
999 | } |
1000 | ||
643b1138 | 1001 | /* |
672bba3a | 1002 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1003 | */ |
5cc6eee8 CL |
1004 | static void add_full(struct kmem_cache *s, |
1005 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1006 | { |
5cc6eee8 CL |
1007 | if (!(s->flags & SLAB_STORE_USER)) |
1008 | return; | |
1009 | ||
255d0884 | 1010 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1011 | list_add(&page->lru, &n->full); |
643b1138 CL |
1012 | } |
1013 | ||
c65c1877 | 1014 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1015 | { |
643b1138 CL |
1016 | if (!(s->flags & SLAB_STORE_USER)) |
1017 | return; | |
1018 | ||
255d0884 | 1019 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1020 | list_del(&page->lru); |
643b1138 CL |
1021 | } |
1022 | ||
0f389ec6 CL |
1023 | /* Tracking of the number of slabs for debugging purposes */ |
1024 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1025 | { | |
1026 | struct kmem_cache_node *n = get_node(s, node); | |
1027 | ||
1028 | return atomic_long_read(&n->nr_slabs); | |
1029 | } | |
1030 | ||
26c02cf0 AB |
1031 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1032 | { | |
1033 | return atomic_long_read(&n->nr_slabs); | |
1034 | } | |
1035 | ||
205ab99d | 1036 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1037 | { |
1038 | struct kmem_cache_node *n = get_node(s, node); | |
1039 | ||
1040 | /* | |
1041 | * May be called early in order to allocate a slab for the | |
1042 | * kmem_cache_node structure. Solve the chicken-egg | |
1043 | * dilemma by deferring the increment of the count during | |
1044 | * bootstrap (see early_kmem_cache_node_alloc). | |
1045 | */ | |
338b2642 | 1046 | if (likely(n)) { |
0f389ec6 | 1047 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1048 | atomic_long_add(objects, &n->total_objects); |
1049 | } | |
0f389ec6 | 1050 | } |
205ab99d | 1051 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1052 | { |
1053 | struct kmem_cache_node *n = get_node(s, node); | |
1054 | ||
1055 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1056 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1057 | } |
1058 | ||
1059 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1060 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1061 | void *object) | |
1062 | { | |
1063 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1064 | return; | |
1065 | ||
f7cb1933 | 1066 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1067 | init_tracking(s, object); |
1068 | } | |
1069 | ||
d0e0ac97 CG |
1070 | static noinline int alloc_debug_processing(struct kmem_cache *s, |
1071 | struct page *page, | |
ce71e27c | 1072 | void *object, unsigned long addr) |
81819f0f CL |
1073 | { |
1074 | if (!check_slab(s, page)) | |
1075 | goto bad; | |
1076 | ||
81819f0f CL |
1077 | if (!check_valid_pointer(s, page, object)) { |
1078 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1079 | goto bad; |
81819f0f CL |
1080 | } |
1081 | ||
f7cb1933 | 1082 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1083 | goto bad; |
81819f0f | 1084 | |
3ec09742 CL |
1085 | /* Success perform special debug activities for allocs */ |
1086 | if (s->flags & SLAB_STORE_USER) | |
1087 | set_track(s, object, TRACK_ALLOC, addr); | |
1088 | trace(s, page, object, 1); | |
f7cb1933 | 1089 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1090 | return 1; |
3ec09742 | 1091 | |
81819f0f CL |
1092 | bad: |
1093 | if (PageSlab(page)) { | |
1094 | /* | |
1095 | * If this is a slab page then lets do the best we can | |
1096 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1097 | * as used avoids touching the remaining objects. |
81819f0f | 1098 | */ |
24922684 | 1099 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1100 | page->inuse = page->objects; |
a973e9dd | 1101 | page->freelist = NULL; |
81819f0f CL |
1102 | } |
1103 | return 0; | |
1104 | } | |
1105 | ||
19c7ff9e CL |
1106 | static noinline struct kmem_cache_node *free_debug_processing( |
1107 | struct kmem_cache *s, struct page *page, void *object, | |
1108 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1109 | { |
19c7ff9e | 1110 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1111 | |
19c7ff9e | 1112 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1113 | slab_lock(page); |
1114 | ||
81819f0f CL |
1115 | if (!check_slab(s, page)) |
1116 | goto fail; | |
1117 | ||
1118 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1119 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1120 | goto fail; |
1121 | } | |
1122 | ||
1123 | if (on_freelist(s, page, object)) { | |
24922684 | 1124 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1125 | goto fail; |
1126 | } | |
1127 | ||
f7cb1933 | 1128 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1129 | goto out; |
81819f0f | 1130 | |
1b4f59e3 | 1131 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1132 | if (!PageSlab(page)) { |
70d71228 CL |
1133 | slab_err(s, page, "Attempt to free object(0x%p) " |
1134 | "outside of slab", object); | |
1b4f59e3 | 1135 | } else if (!page->slab_cache) { |
81819f0f | 1136 | printk(KERN_ERR |
70d71228 | 1137 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1138 | object); |
70d71228 | 1139 | dump_stack(); |
06428780 | 1140 | } else |
24922684 CL |
1141 | object_err(s, page, object, |
1142 | "page slab pointer corrupt."); | |
81819f0f CL |
1143 | goto fail; |
1144 | } | |
3ec09742 | 1145 | |
3ec09742 CL |
1146 | if (s->flags & SLAB_STORE_USER) |
1147 | set_track(s, object, TRACK_FREE, addr); | |
1148 | trace(s, page, object, 0); | |
f7cb1933 | 1149 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1150 | out: |
881db7fb | 1151 | slab_unlock(page); |
19c7ff9e CL |
1152 | /* |
1153 | * Keep node_lock to preserve integrity | |
1154 | * until the object is actually freed | |
1155 | */ | |
1156 | return n; | |
3ec09742 | 1157 | |
81819f0f | 1158 | fail: |
19c7ff9e CL |
1159 | slab_unlock(page); |
1160 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1161 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1162 | return NULL; |
81819f0f CL |
1163 | } |
1164 | ||
41ecc55b CL |
1165 | static int __init setup_slub_debug(char *str) |
1166 | { | |
f0630fff CL |
1167 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1168 | if (*str++ != '=' || !*str) | |
1169 | /* | |
1170 | * No options specified. Switch on full debugging. | |
1171 | */ | |
1172 | goto out; | |
1173 | ||
1174 | if (*str == ',') | |
1175 | /* | |
1176 | * No options but restriction on slabs. This means full | |
1177 | * debugging for slabs matching a pattern. | |
1178 | */ | |
1179 | goto check_slabs; | |
1180 | ||
fa5ec8a1 DR |
1181 | if (tolower(*str) == 'o') { |
1182 | /* | |
1183 | * Avoid enabling debugging on caches if its minimum order | |
1184 | * would increase as a result. | |
1185 | */ | |
1186 | disable_higher_order_debug = 1; | |
1187 | goto out; | |
1188 | } | |
1189 | ||
f0630fff CL |
1190 | slub_debug = 0; |
1191 | if (*str == '-') | |
1192 | /* | |
1193 | * Switch off all debugging measures. | |
1194 | */ | |
1195 | goto out; | |
1196 | ||
1197 | /* | |
1198 | * Determine which debug features should be switched on | |
1199 | */ | |
06428780 | 1200 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1201 | switch (tolower(*str)) { |
1202 | case 'f': | |
1203 | slub_debug |= SLAB_DEBUG_FREE; | |
1204 | break; | |
1205 | case 'z': | |
1206 | slub_debug |= SLAB_RED_ZONE; | |
1207 | break; | |
1208 | case 'p': | |
1209 | slub_debug |= SLAB_POISON; | |
1210 | break; | |
1211 | case 'u': | |
1212 | slub_debug |= SLAB_STORE_USER; | |
1213 | break; | |
1214 | case 't': | |
1215 | slub_debug |= SLAB_TRACE; | |
1216 | break; | |
4c13dd3b DM |
1217 | case 'a': |
1218 | slub_debug |= SLAB_FAILSLAB; | |
1219 | break; | |
f0630fff CL |
1220 | default: |
1221 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1222 | "unknown. skipped\n", *str); |
f0630fff | 1223 | } |
41ecc55b CL |
1224 | } |
1225 | ||
f0630fff | 1226 | check_slabs: |
41ecc55b CL |
1227 | if (*str == ',') |
1228 | slub_debug_slabs = str + 1; | |
f0630fff | 1229 | out: |
41ecc55b CL |
1230 | return 1; |
1231 | } | |
1232 | ||
1233 | __setup("slub_debug", setup_slub_debug); | |
1234 | ||
3b0efdfa | 1235 | static unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1236 | unsigned long flags, const char *name, |
51cc5068 | 1237 | void (*ctor)(void *)) |
41ecc55b CL |
1238 | { |
1239 | /* | |
e153362a | 1240 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1241 | */ |
c6f58d9b CL |
1242 | if (slub_debug && (!slub_debug_slabs || (name && |
1243 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1244 | flags |= slub_debug; |
ba0268a8 CL |
1245 | |
1246 | return flags; | |
41ecc55b CL |
1247 | } |
1248 | #else | |
3ec09742 CL |
1249 | static inline void setup_object_debug(struct kmem_cache *s, |
1250 | struct page *page, void *object) {} | |
41ecc55b | 1251 | |
3ec09742 | 1252 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1253 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1254 | |
19c7ff9e CL |
1255 | static inline struct kmem_cache_node *free_debug_processing( |
1256 | struct kmem_cache *s, struct page *page, void *object, | |
1257 | unsigned long addr, unsigned long *flags) { return NULL; } | |
41ecc55b | 1258 | |
41ecc55b CL |
1259 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1260 | { return 1; } | |
1261 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1262 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1263 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1264 | struct page *page) {} | |
c65c1877 PZ |
1265 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1266 | struct page *page) {} | |
3b0efdfa | 1267 | static inline unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1268 | unsigned long flags, const char *name, |
51cc5068 | 1269 | void (*ctor)(void *)) |
ba0268a8 CL |
1270 | { |
1271 | return flags; | |
1272 | } | |
41ecc55b | 1273 | #define slub_debug 0 |
0f389ec6 | 1274 | |
fdaa45e9 IM |
1275 | #define disable_higher_order_debug 0 |
1276 | ||
0f389ec6 CL |
1277 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1278 | { return 0; } | |
26c02cf0 AB |
1279 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1280 | { return 0; } | |
205ab99d CL |
1281 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1282 | int objects) {} | |
1283 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1284 | int objects) {} | |
7d550c56 | 1285 | |
d56791b3 RB |
1286 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1287 | { | |
1288 | kmemleak_alloc(ptr, size, 1, flags); | |
1289 | } | |
1290 | ||
1291 | static inline void kfree_hook(const void *x) | |
1292 | { | |
1293 | kmemleak_free(x); | |
1294 | } | |
1295 | ||
7d550c56 CL |
1296 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) |
1297 | { return 0; } | |
1298 | ||
1299 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
d56791b3 RB |
1300 | void *object) |
1301 | { | |
1302 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, | |
1303 | flags & gfp_allowed_mask); | |
1304 | } | |
7d550c56 | 1305 | |
d56791b3 RB |
1306 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1307 | { | |
1308 | kmemleak_free_recursive(x, s->flags); | |
1309 | } | |
7d550c56 | 1310 | |
ab4d5ed5 | 1311 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1312 | |
81819f0f CL |
1313 | /* |
1314 | * Slab allocation and freeing | |
1315 | */ | |
65c3376a CL |
1316 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1317 | struct kmem_cache_order_objects oo) | |
1318 | { | |
1319 | int order = oo_order(oo); | |
1320 | ||
b1eeab67 VN |
1321 | flags |= __GFP_NOTRACK; |
1322 | ||
2154a336 | 1323 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1324 | return alloc_pages(flags, order); |
1325 | else | |
6b65aaf3 | 1326 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1327 | } |
1328 | ||
81819f0f CL |
1329 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1330 | { | |
06428780 | 1331 | struct page *page; |
834f3d11 | 1332 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1333 | gfp_t alloc_gfp; |
81819f0f | 1334 | |
7e0528da CL |
1335 | flags &= gfp_allowed_mask; |
1336 | ||
1337 | if (flags & __GFP_WAIT) | |
1338 | local_irq_enable(); | |
1339 | ||
b7a49f0d | 1340 | flags |= s->allocflags; |
e12ba74d | 1341 | |
ba52270d PE |
1342 | /* |
1343 | * Let the initial higher-order allocation fail under memory pressure | |
1344 | * so we fall-back to the minimum order allocation. | |
1345 | */ | |
1346 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1347 | ||
1348 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1349 | if (unlikely(!page)) { |
1350 | oo = s->min; | |
1351 | /* | |
1352 | * Allocation may have failed due to fragmentation. | |
1353 | * Try a lower order alloc if possible | |
1354 | */ | |
1355 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1356 | |
7e0528da CL |
1357 | if (page) |
1358 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1359 | } |
5a896d9e | 1360 | |
737b719e | 1361 | if (kmemcheck_enabled && page |
5086c389 | 1362 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1363 | int pages = 1 << oo_order(oo); |
1364 | ||
1365 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1366 | ||
1367 | /* | |
1368 | * Objects from caches that have a constructor don't get | |
1369 | * cleared when they're allocated, so we need to do it here. | |
1370 | */ | |
1371 | if (s->ctor) | |
1372 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1373 | else | |
1374 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1375 | } |
1376 | ||
737b719e DR |
1377 | if (flags & __GFP_WAIT) |
1378 | local_irq_disable(); | |
1379 | if (!page) | |
1380 | return NULL; | |
1381 | ||
834f3d11 | 1382 | page->objects = oo_objects(oo); |
81819f0f CL |
1383 | mod_zone_page_state(page_zone(page), |
1384 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1385 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1386 | 1 << oo_order(oo)); |
81819f0f CL |
1387 | |
1388 | return page; | |
1389 | } | |
1390 | ||
1391 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1392 | void *object) | |
1393 | { | |
3ec09742 | 1394 | setup_object_debug(s, page, object); |
4f104934 | 1395 | if (unlikely(s->ctor)) |
51cc5068 | 1396 | s->ctor(object); |
81819f0f CL |
1397 | } |
1398 | ||
1399 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1400 | { | |
1401 | struct page *page; | |
81819f0f | 1402 | void *start; |
81819f0f CL |
1403 | void *last; |
1404 | void *p; | |
1f458cbf | 1405 | int order; |
81819f0f | 1406 | |
6cb06229 | 1407 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1408 | |
6cb06229 CL |
1409 | page = allocate_slab(s, |
1410 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1411 | if (!page) |
1412 | goto out; | |
1413 | ||
1f458cbf | 1414 | order = compound_order(page); |
205ab99d | 1415 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1f458cbf | 1416 | memcg_bind_pages(s, order); |
1b4f59e3 | 1417 | page->slab_cache = s; |
c03f94cc | 1418 | __SetPageSlab(page); |
072bb0aa MG |
1419 | if (page->pfmemalloc) |
1420 | SetPageSlabPfmemalloc(page); | |
81819f0f CL |
1421 | |
1422 | start = page_address(page); | |
81819f0f CL |
1423 | |
1424 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1425 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f CL |
1426 | |
1427 | last = start; | |
224a88be | 1428 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1429 | setup_object(s, page, last); |
1430 | set_freepointer(s, last, p); | |
1431 | last = p; | |
1432 | } | |
1433 | setup_object(s, page, last); | |
a973e9dd | 1434 | set_freepointer(s, last, NULL); |
81819f0f CL |
1435 | |
1436 | page->freelist = start; | |
e6e82ea1 | 1437 | page->inuse = page->objects; |
8cb0a506 | 1438 | page->frozen = 1; |
81819f0f | 1439 | out: |
81819f0f CL |
1440 | return page; |
1441 | } | |
1442 | ||
1443 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1444 | { | |
834f3d11 CL |
1445 | int order = compound_order(page); |
1446 | int pages = 1 << order; | |
81819f0f | 1447 | |
af537b0a | 1448 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1449 | void *p; |
1450 | ||
1451 | slab_pad_check(s, page); | |
224a88be CL |
1452 | for_each_object(p, s, page_address(page), |
1453 | page->objects) | |
f7cb1933 | 1454 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1455 | } |
1456 | ||
b1eeab67 | 1457 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1458 | |
81819f0f CL |
1459 | mod_zone_page_state(page_zone(page), |
1460 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1461 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1462 | -pages); |
81819f0f | 1463 | |
072bb0aa | 1464 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1465 | __ClearPageSlab(page); |
1f458cbf GC |
1466 | |
1467 | memcg_release_pages(s, order); | |
22b751c3 | 1468 | page_mapcount_reset(page); |
1eb5ac64 NP |
1469 | if (current->reclaim_state) |
1470 | current->reclaim_state->reclaimed_slab += pages; | |
d79923fa | 1471 | __free_memcg_kmem_pages(page, order); |
81819f0f CL |
1472 | } |
1473 | ||
da9a638c LJ |
1474 | #define need_reserve_slab_rcu \ |
1475 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1476 | ||
81819f0f CL |
1477 | static void rcu_free_slab(struct rcu_head *h) |
1478 | { | |
1479 | struct page *page; | |
1480 | ||
da9a638c LJ |
1481 | if (need_reserve_slab_rcu) |
1482 | page = virt_to_head_page(h); | |
1483 | else | |
1484 | page = container_of((struct list_head *)h, struct page, lru); | |
1485 | ||
1b4f59e3 | 1486 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1487 | } |
1488 | ||
1489 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1490 | { | |
1491 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1492 | struct rcu_head *head; |
1493 | ||
1494 | if (need_reserve_slab_rcu) { | |
1495 | int order = compound_order(page); | |
1496 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1497 | ||
1498 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1499 | head = page_address(page) + offset; | |
1500 | } else { | |
1501 | /* | |
1502 | * RCU free overloads the RCU head over the LRU | |
1503 | */ | |
1504 | head = (void *)&page->lru; | |
1505 | } | |
81819f0f CL |
1506 | |
1507 | call_rcu(head, rcu_free_slab); | |
1508 | } else | |
1509 | __free_slab(s, page); | |
1510 | } | |
1511 | ||
1512 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1513 | { | |
205ab99d | 1514 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1515 | free_slab(s, page); |
1516 | } | |
1517 | ||
1518 | /* | |
5cc6eee8 | 1519 | * Management of partially allocated slabs. |
81819f0f | 1520 | */ |
5cc6eee8 | 1521 | static inline void add_partial(struct kmem_cache_node *n, |
7c2e132c | 1522 | struct page *page, int tail) |
81819f0f | 1523 | { |
c65c1877 PZ |
1524 | lockdep_assert_held(&n->list_lock); |
1525 | ||
e95eed57 | 1526 | n->nr_partial++; |
136333d1 | 1527 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1528 | list_add_tail(&page->lru, &n->partial); |
1529 | else | |
1530 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1531 | } |
1532 | ||
5cc6eee8 | 1533 | static inline void remove_partial(struct kmem_cache_node *n, |
62e346a8 CL |
1534 | struct page *page) |
1535 | { | |
c65c1877 PZ |
1536 | lockdep_assert_held(&n->list_lock); |
1537 | ||
62e346a8 CL |
1538 | list_del(&page->lru); |
1539 | n->nr_partial--; | |
1540 | } | |
1541 | ||
81819f0f | 1542 | /* |
7ced3719 CL |
1543 | * Remove slab from the partial list, freeze it and |
1544 | * return the pointer to the freelist. | |
81819f0f | 1545 | * |
497b66f2 | 1546 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1547 | */ |
497b66f2 | 1548 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1549 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1550 | int mode, int *objects) |
81819f0f | 1551 | { |
2cfb7455 CL |
1552 | void *freelist; |
1553 | unsigned long counters; | |
1554 | struct page new; | |
1555 | ||
c65c1877 PZ |
1556 | lockdep_assert_held(&n->list_lock); |
1557 | ||
2cfb7455 CL |
1558 | /* |
1559 | * Zap the freelist and set the frozen bit. | |
1560 | * The old freelist is the list of objects for the | |
1561 | * per cpu allocation list. | |
1562 | */ | |
7ced3719 CL |
1563 | freelist = page->freelist; |
1564 | counters = page->counters; | |
1565 | new.counters = counters; | |
633b0764 | 1566 | *objects = new.objects - new.inuse; |
23910c50 | 1567 | if (mode) { |
7ced3719 | 1568 | new.inuse = page->objects; |
23910c50 PE |
1569 | new.freelist = NULL; |
1570 | } else { | |
1571 | new.freelist = freelist; | |
1572 | } | |
2cfb7455 | 1573 | |
a0132ac0 | 1574 | VM_BUG_ON(new.frozen); |
7ced3719 | 1575 | new.frozen = 1; |
2cfb7455 | 1576 | |
7ced3719 | 1577 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1578 | freelist, counters, |
02d7633f | 1579 | new.freelist, new.counters, |
7ced3719 | 1580 | "acquire_slab")) |
7ced3719 | 1581 | return NULL; |
2cfb7455 CL |
1582 | |
1583 | remove_partial(n, page); | |
7ced3719 | 1584 | WARN_ON(!freelist); |
49e22585 | 1585 | return freelist; |
81819f0f CL |
1586 | } |
1587 | ||
633b0764 | 1588 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1589 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1590 | |
81819f0f | 1591 | /* |
672bba3a | 1592 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1593 | */ |
8ba00bb6 JK |
1594 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1595 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1596 | { |
49e22585 CL |
1597 | struct page *page, *page2; |
1598 | void *object = NULL; | |
633b0764 JK |
1599 | int available = 0; |
1600 | int objects; | |
81819f0f CL |
1601 | |
1602 | /* | |
1603 | * Racy check. If we mistakenly see no partial slabs then we | |
1604 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1605 | * partial slab and there is none available then get_partials() |
1606 | * will return NULL. | |
81819f0f CL |
1607 | */ |
1608 | if (!n || !n->nr_partial) | |
1609 | return NULL; | |
1610 | ||
1611 | spin_lock(&n->list_lock); | |
49e22585 | 1612 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1613 | void *t; |
49e22585 | 1614 | |
8ba00bb6 JK |
1615 | if (!pfmemalloc_match(page, flags)) |
1616 | continue; | |
1617 | ||
633b0764 | 1618 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1619 | if (!t) |
1620 | break; | |
1621 | ||
633b0764 | 1622 | available += objects; |
12d79634 | 1623 | if (!object) { |
49e22585 | 1624 | c->page = page; |
49e22585 | 1625 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1626 | object = t; |
49e22585 | 1627 | } else { |
633b0764 | 1628 | put_cpu_partial(s, page, 0); |
8028dcea | 1629 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1630 | } |
345c905d JK |
1631 | if (!kmem_cache_has_cpu_partial(s) |
1632 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1633 | break; |
1634 | ||
497b66f2 | 1635 | } |
81819f0f | 1636 | spin_unlock(&n->list_lock); |
497b66f2 | 1637 | return object; |
81819f0f CL |
1638 | } |
1639 | ||
1640 | /* | |
672bba3a | 1641 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1642 | */ |
de3ec035 | 1643 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1644 | struct kmem_cache_cpu *c) |
81819f0f CL |
1645 | { |
1646 | #ifdef CONFIG_NUMA | |
1647 | struct zonelist *zonelist; | |
dd1a239f | 1648 | struct zoneref *z; |
54a6eb5c MG |
1649 | struct zone *zone; |
1650 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1651 | void *object; |
cc9a6c87 | 1652 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1653 | |
1654 | /* | |
672bba3a CL |
1655 | * The defrag ratio allows a configuration of the tradeoffs between |
1656 | * inter node defragmentation and node local allocations. A lower | |
1657 | * defrag_ratio increases the tendency to do local allocations | |
1658 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1659 | * |
672bba3a CL |
1660 | * If the defrag_ratio is set to 0 then kmalloc() always |
1661 | * returns node local objects. If the ratio is higher then kmalloc() | |
1662 | * may return off node objects because partial slabs are obtained | |
1663 | * from other nodes and filled up. | |
81819f0f | 1664 | * |
6446faa2 | 1665 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1666 | * defrag_ratio = 1000) then every (well almost) allocation will |
1667 | * first attempt to defrag slab caches on other nodes. This means | |
1668 | * scanning over all nodes to look for partial slabs which may be | |
1669 | * expensive if we do it every time we are trying to find a slab | |
1670 | * with available objects. | |
81819f0f | 1671 | */ |
9824601e CL |
1672 | if (!s->remote_node_defrag_ratio || |
1673 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1674 | return NULL; |
1675 | ||
cc9a6c87 MG |
1676 | do { |
1677 | cpuset_mems_cookie = get_mems_allowed(); | |
e7b691b0 | 1678 | zonelist = node_zonelist(slab_node(), flags); |
cc9a6c87 MG |
1679 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1680 | struct kmem_cache_node *n; | |
1681 | ||
1682 | n = get_node(s, zone_to_nid(zone)); | |
1683 | ||
1684 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && | |
1685 | n->nr_partial > s->min_partial) { | |
8ba00bb6 | 1686 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1687 | if (object) { |
1688 | /* | |
1689 | * Return the object even if | |
1690 | * put_mems_allowed indicated that | |
1691 | * the cpuset mems_allowed was | |
1692 | * updated in parallel. It's a | |
1693 | * harmless race between the alloc | |
1694 | * and the cpuset update. | |
1695 | */ | |
1696 | put_mems_allowed(cpuset_mems_cookie); | |
1697 | return object; | |
1698 | } | |
c0ff7453 | 1699 | } |
81819f0f | 1700 | } |
cc9a6c87 | 1701 | } while (!put_mems_allowed(cpuset_mems_cookie)); |
81819f0f CL |
1702 | #endif |
1703 | return NULL; | |
1704 | } | |
1705 | ||
1706 | /* | |
1707 | * Get a partial page, lock it and return it. | |
1708 | */ | |
497b66f2 | 1709 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1710 | struct kmem_cache_cpu *c) |
81819f0f | 1711 | { |
497b66f2 | 1712 | void *object; |
2154a336 | 1713 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f | 1714 | |
8ba00bb6 | 1715 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1716 | if (object || node != NUMA_NO_NODE) |
1717 | return object; | |
81819f0f | 1718 | |
acd19fd1 | 1719 | return get_any_partial(s, flags, c); |
81819f0f CL |
1720 | } |
1721 | ||
8a5ec0ba CL |
1722 | #ifdef CONFIG_PREEMPT |
1723 | /* | |
1724 | * Calculate the next globally unique transaction for disambiguiation | |
1725 | * during cmpxchg. The transactions start with the cpu number and are then | |
1726 | * incremented by CONFIG_NR_CPUS. | |
1727 | */ | |
1728 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1729 | #else | |
1730 | /* | |
1731 | * No preemption supported therefore also no need to check for | |
1732 | * different cpus. | |
1733 | */ | |
1734 | #define TID_STEP 1 | |
1735 | #endif | |
1736 | ||
1737 | static inline unsigned long next_tid(unsigned long tid) | |
1738 | { | |
1739 | return tid + TID_STEP; | |
1740 | } | |
1741 | ||
1742 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1743 | { | |
1744 | return tid % TID_STEP; | |
1745 | } | |
1746 | ||
1747 | static inline unsigned long tid_to_event(unsigned long tid) | |
1748 | { | |
1749 | return tid / TID_STEP; | |
1750 | } | |
1751 | ||
1752 | static inline unsigned int init_tid(int cpu) | |
1753 | { | |
1754 | return cpu; | |
1755 | } | |
1756 | ||
1757 | static inline void note_cmpxchg_failure(const char *n, | |
1758 | const struct kmem_cache *s, unsigned long tid) | |
1759 | { | |
1760 | #ifdef SLUB_DEBUG_CMPXCHG | |
1761 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1762 | ||
1763 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1764 | ||
1765 | #ifdef CONFIG_PREEMPT | |
1766 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1767 | printk("due to cpu change %d -> %d\n", | |
1768 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1769 | else | |
1770 | #endif | |
1771 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1772 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1773 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1774 | else | |
1775 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1776 | actual_tid, tid, next_tid(tid)); | |
1777 | #endif | |
4fdccdfb | 1778 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1779 | } |
1780 | ||
788e1aad | 1781 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1782 | { |
8a5ec0ba CL |
1783 | int cpu; |
1784 | ||
1785 | for_each_possible_cpu(cpu) | |
1786 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1787 | } |
2cfb7455 | 1788 | |
81819f0f CL |
1789 | /* |
1790 | * Remove the cpu slab | |
1791 | */ | |
d0e0ac97 CG |
1792 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1793 | void *freelist) | |
81819f0f | 1794 | { |
2cfb7455 | 1795 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1796 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1797 | int lock = 0; | |
1798 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1799 | void *nextfree; |
136333d1 | 1800 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1801 | struct page new; |
1802 | struct page old; | |
1803 | ||
1804 | if (page->freelist) { | |
84e554e6 | 1805 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1806 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1807 | } |
1808 | ||
894b8788 | 1809 | /* |
2cfb7455 CL |
1810 | * Stage one: Free all available per cpu objects back |
1811 | * to the page freelist while it is still frozen. Leave the | |
1812 | * last one. | |
1813 | * | |
1814 | * There is no need to take the list->lock because the page | |
1815 | * is still frozen. | |
1816 | */ | |
1817 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1818 | void *prior; | |
1819 | unsigned long counters; | |
1820 | ||
1821 | do { | |
1822 | prior = page->freelist; | |
1823 | counters = page->counters; | |
1824 | set_freepointer(s, freelist, prior); | |
1825 | new.counters = counters; | |
1826 | new.inuse--; | |
a0132ac0 | 1827 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1828 | |
1d07171c | 1829 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1830 | prior, counters, |
1831 | freelist, new.counters, | |
1832 | "drain percpu freelist")); | |
1833 | ||
1834 | freelist = nextfree; | |
1835 | } | |
1836 | ||
894b8788 | 1837 | /* |
2cfb7455 CL |
1838 | * Stage two: Ensure that the page is unfrozen while the |
1839 | * list presence reflects the actual number of objects | |
1840 | * during unfreeze. | |
1841 | * | |
1842 | * We setup the list membership and then perform a cmpxchg | |
1843 | * with the count. If there is a mismatch then the page | |
1844 | * is not unfrozen but the page is on the wrong list. | |
1845 | * | |
1846 | * Then we restart the process which may have to remove | |
1847 | * the page from the list that we just put it on again | |
1848 | * because the number of objects in the slab may have | |
1849 | * changed. | |
894b8788 | 1850 | */ |
2cfb7455 | 1851 | redo: |
894b8788 | 1852 | |
2cfb7455 CL |
1853 | old.freelist = page->freelist; |
1854 | old.counters = page->counters; | |
a0132ac0 | 1855 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1856 | |
2cfb7455 CL |
1857 | /* Determine target state of the slab */ |
1858 | new.counters = old.counters; | |
1859 | if (freelist) { | |
1860 | new.inuse--; | |
1861 | set_freepointer(s, freelist, old.freelist); | |
1862 | new.freelist = freelist; | |
1863 | } else | |
1864 | new.freelist = old.freelist; | |
1865 | ||
1866 | new.frozen = 0; | |
1867 | ||
81107188 | 1868 | if (!new.inuse && n->nr_partial > s->min_partial) |
2cfb7455 CL |
1869 | m = M_FREE; |
1870 | else if (new.freelist) { | |
1871 | m = M_PARTIAL; | |
1872 | if (!lock) { | |
1873 | lock = 1; | |
1874 | /* | |
1875 | * Taking the spinlock removes the possiblity | |
1876 | * that acquire_slab() will see a slab page that | |
1877 | * is frozen | |
1878 | */ | |
1879 | spin_lock(&n->list_lock); | |
1880 | } | |
1881 | } else { | |
1882 | m = M_FULL; | |
1883 | if (kmem_cache_debug(s) && !lock) { | |
1884 | lock = 1; | |
1885 | /* | |
1886 | * This also ensures that the scanning of full | |
1887 | * slabs from diagnostic functions will not see | |
1888 | * any frozen slabs. | |
1889 | */ | |
1890 | spin_lock(&n->list_lock); | |
1891 | } | |
1892 | } | |
1893 | ||
1894 | if (l != m) { | |
1895 | ||
1896 | if (l == M_PARTIAL) | |
1897 | ||
1898 | remove_partial(n, page); | |
1899 | ||
1900 | else if (l == M_FULL) | |
894b8788 | 1901 | |
c65c1877 | 1902 | remove_full(s, n, page); |
2cfb7455 CL |
1903 | |
1904 | if (m == M_PARTIAL) { | |
1905 | ||
1906 | add_partial(n, page, tail); | |
136333d1 | 1907 | stat(s, tail); |
2cfb7455 CL |
1908 | |
1909 | } else if (m == M_FULL) { | |
894b8788 | 1910 | |
2cfb7455 CL |
1911 | stat(s, DEACTIVATE_FULL); |
1912 | add_full(s, n, page); | |
1913 | ||
1914 | } | |
1915 | } | |
1916 | ||
1917 | l = m; | |
1d07171c | 1918 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1919 | old.freelist, old.counters, |
1920 | new.freelist, new.counters, | |
1921 | "unfreezing slab")) | |
1922 | goto redo; | |
1923 | ||
2cfb7455 CL |
1924 | if (lock) |
1925 | spin_unlock(&n->list_lock); | |
1926 | ||
1927 | if (m == M_FREE) { | |
1928 | stat(s, DEACTIVATE_EMPTY); | |
1929 | discard_slab(s, page); | |
1930 | stat(s, FREE_SLAB); | |
894b8788 | 1931 | } |
81819f0f CL |
1932 | } |
1933 | ||
d24ac77f JK |
1934 | /* |
1935 | * Unfreeze all the cpu partial slabs. | |
1936 | * | |
59a09917 CL |
1937 | * This function must be called with interrupts disabled |
1938 | * for the cpu using c (or some other guarantee must be there | |
1939 | * to guarantee no concurrent accesses). | |
d24ac77f | 1940 | */ |
59a09917 CL |
1941 | static void unfreeze_partials(struct kmem_cache *s, |
1942 | struct kmem_cache_cpu *c) | |
49e22585 | 1943 | { |
345c905d | 1944 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 1945 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1946 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1947 | |
1948 | while ((page = c->partial)) { | |
49e22585 CL |
1949 | struct page new; |
1950 | struct page old; | |
1951 | ||
1952 | c->partial = page->next; | |
43d77867 JK |
1953 | |
1954 | n2 = get_node(s, page_to_nid(page)); | |
1955 | if (n != n2) { | |
1956 | if (n) | |
1957 | spin_unlock(&n->list_lock); | |
1958 | ||
1959 | n = n2; | |
1960 | spin_lock(&n->list_lock); | |
1961 | } | |
49e22585 CL |
1962 | |
1963 | do { | |
1964 | ||
1965 | old.freelist = page->freelist; | |
1966 | old.counters = page->counters; | |
a0132ac0 | 1967 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
1968 | |
1969 | new.counters = old.counters; | |
1970 | new.freelist = old.freelist; | |
1971 | ||
1972 | new.frozen = 0; | |
1973 | ||
d24ac77f | 1974 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1975 | old.freelist, old.counters, |
1976 | new.freelist, new.counters, | |
1977 | "unfreezing slab")); | |
1978 | ||
43d77867 | 1979 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) { |
9ada1934 SL |
1980 | page->next = discard_page; |
1981 | discard_page = page; | |
43d77867 JK |
1982 | } else { |
1983 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1984 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1985 | } |
1986 | } | |
1987 | ||
1988 | if (n) | |
1989 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
1990 | |
1991 | while (discard_page) { | |
1992 | page = discard_page; | |
1993 | discard_page = discard_page->next; | |
1994 | ||
1995 | stat(s, DEACTIVATE_EMPTY); | |
1996 | discard_slab(s, page); | |
1997 | stat(s, FREE_SLAB); | |
1998 | } | |
345c905d | 1999 | #endif |
49e22585 CL |
2000 | } |
2001 | ||
2002 | /* | |
2003 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2004 | * slot if available. This is done without interrupts disabled and without | |
2005 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2006 | * onto a random cpus partial slot. | |
2007 | * | |
2008 | * If we did not find a slot then simply move all the partials to the | |
2009 | * per node partial list. | |
2010 | */ | |
633b0764 | 2011 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2012 | { |
345c905d | 2013 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2014 | struct page *oldpage; |
2015 | int pages; | |
2016 | int pobjects; | |
2017 | ||
2018 | do { | |
2019 | pages = 0; | |
2020 | pobjects = 0; | |
2021 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2022 | ||
2023 | if (oldpage) { | |
2024 | pobjects = oldpage->pobjects; | |
2025 | pages = oldpage->pages; | |
2026 | if (drain && pobjects > s->cpu_partial) { | |
2027 | unsigned long flags; | |
2028 | /* | |
2029 | * partial array is full. Move the existing | |
2030 | * set to the per node partial list. | |
2031 | */ | |
2032 | local_irq_save(flags); | |
59a09917 | 2033 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2034 | local_irq_restore(flags); |
e24fc410 | 2035 | oldpage = NULL; |
49e22585 CL |
2036 | pobjects = 0; |
2037 | pages = 0; | |
8028dcea | 2038 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2039 | } |
2040 | } | |
2041 | ||
2042 | pages++; | |
2043 | pobjects += page->objects - page->inuse; | |
2044 | ||
2045 | page->pages = pages; | |
2046 | page->pobjects = pobjects; | |
2047 | page->next = oldpage; | |
2048 | ||
d0e0ac97 CG |
2049 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2050 | != oldpage); | |
345c905d | 2051 | #endif |
49e22585 CL |
2052 | } |
2053 | ||
dfb4f096 | 2054 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2055 | { |
84e554e6 | 2056 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2057 | deactivate_slab(s, c->page, c->freelist); |
2058 | ||
2059 | c->tid = next_tid(c->tid); | |
2060 | c->page = NULL; | |
2061 | c->freelist = NULL; | |
81819f0f CL |
2062 | } |
2063 | ||
2064 | /* | |
2065 | * Flush cpu slab. | |
6446faa2 | 2066 | * |
81819f0f CL |
2067 | * Called from IPI handler with interrupts disabled. |
2068 | */ | |
0c710013 | 2069 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2070 | { |
9dfc6e68 | 2071 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2072 | |
49e22585 CL |
2073 | if (likely(c)) { |
2074 | if (c->page) | |
2075 | flush_slab(s, c); | |
2076 | ||
59a09917 | 2077 | unfreeze_partials(s, c); |
49e22585 | 2078 | } |
81819f0f CL |
2079 | } |
2080 | ||
2081 | static void flush_cpu_slab(void *d) | |
2082 | { | |
2083 | struct kmem_cache *s = d; | |
81819f0f | 2084 | |
dfb4f096 | 2085 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2086 | } |
2087 | ||
a8364d55 GBY |
2088 | static bool has_cpu_slab(int cpu, void *info) |
2089 | { | |
2090 | struct kmem_cache *s = info; | |
2091 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2092 | ||
02e1a9cd | 2093 | return c->page || c->partial; |
a8364d55 GBY |
2094 | } |
2095 | ||
81819f0f CL |
2096 | static void flush_all(struct kmem_cache *s) |
2097 | { | |
a8364d55 | 2098 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2099 | } |
2100 | ||
dfb4f096 CL |
2101 | /* |
2102 | * Check if the objects in a per cpu structure fit numa | |
2103 | * locality expectations. | |
2104 | */ | |
57d437d2 | 2105 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2106 | { |
2107 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2108 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2109 | return 0; |
2110 | #endif | |
2111 | return 1; | |
2112 | } | |
2113 | ||
781b2ba6 PE |
2114 | static int count_free(struct page *page) |
2115 | { | |
2116 | return page->objects - page->inuse; | |
2117 | } | |
2118 | ||
2119 | static unsigned long count_partial(struct kmem_cache_node *n, | |
2120 | int (*get_count)(struct page *)) | |
2121 | { | |
2122 | unsigned long flags; | |
2123 | unsigned long x = 0; | |
2124 | struct page *page; | |
2125 | ||
2126 | spin_lock_irqsave(&n->list_lock, flags); | |
2127 | list_for_each_entry(page, &n->partial, lru) | |
2128 | x += get_count(page); | |
2129 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2130 | return x; | |
2131 | } | |
2132 | ||
26c02cf0 AB |
2133 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2134 | { | |
2135 | #ifdef CONFIG_SLUB_DEBUG | |
2136 | return atomic_long_read(&n->total_objects); | |
2137 | #else | |
2138 | return 0; | |
2139 | #endif | |
2140 | } | |
2141 | ||
781b2ba6 PE |
2142 | static noinline void |
2143 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2144 | { | |
2145 | int node; | |
2146 | ||
2147 | printk(KERN_WARNING | |
2148 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
2149 | nid, gfpflags); | |
2150 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
3b0efdfa | 2151 | "default order: %d, min order: %d\n", s->name, s->object_size, |
781b2ba6 PE |
2152 | s->size, oo_order(s->oo), oo_order(s->min)); |
2153 | ||
3b0efdfa | 2154 | if (oo_order(s->min) > get_order(s->object_size)) |
fa5ec8a1 DR |
2155 | printk(KERN_WARNING " %s debugging increased min order, use " |
2156 | "slub_debug=O to disable.\n", s->name); | |
2157 | ||
781b2ba6 PE |
2158 | for_each_online_node(node) { |
2159 | struct kmem_cache_node *n = get_node(s, node); | |
2160 | unsigned long nr_slabs; | |
2161 | unsigned long nr_objs; | |
2162 | unsigned long nr_free; | |
2163 | ||
2164 | if (!n) | |
2165 | continue; | |
2166 | ||
26c02cf0 AB |
2167 | nr_free = count_partial(n, count_free); |
2168 | nr_slabs = node_nr_slabs(n); | |
2169 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
2170 | |
2171 | printk(KERN_WARNING | |
2172 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2173 | node, nr_slabs, nr_objs, nr_free); | |
2174 | } | |
2175 | } | |
2176 | ||
497b66f2 CL |
2177 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2178 | int node, struct kmem_cache_cpu **pc) | |
2179 | { | |
6faa6833 | 2180 | void *freelist; |
188fd063 CL |
2181 | struct kmem_cache_cpu *c = *pc; |
2182 | struct page *page; | |
497b66f2 | 2183 | |
188fd063 | 2184 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2185 | |
188fd063 CL |
2186 | if (freelist) |
2187 | return freelist; | |
2188 | ||
2189 | page = new_slab(s, flags, node); | |
497b66f2 CL |
2190 | if (page) { |
2191 | c = __this_cpu_ptr(s->cpu_slab); | |
2192 | if (c->page) | |
2193 | flush_slab(s, c); | |
2194 | ||
2195 | /* | |
2196 | * No other reference to the page yet so we can | |
2197 | * muck around with it freely without cmpxchg | |
2198 | */ | |
6faa6833 | 2199 | freelist = page->freelist; |
497b66f2 CL |
2200 | page->freelist = NULL; |
2201 | ||
2202 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2203 | c->page = page; |
2204 | *pc = c; | |
2205 | } else | |
6faa6833 | 2206 | freelist = NULL; |
497b66f2 | 2207 | |
6faa6833 | 2208 | return freelist; |
497b66f2 CL |
2209 | } |
2210 | ||
072bb0aa MG |
2211 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2212 | { | |
2213 | if (unlikely(PageSlabPfmemalloc(page))) | |
2214 | return gfp_pfmemalloc_allowed(gfpflags); | |
2215 | ||
2216 | return true; | |
2217 | } | |
2218 | ||
213eeb9f | 2219 | /* |
d0e0ac97 CG |
2220 | * Check the page->freelist of a page and either transfer the freelist to the |
2221 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2222 | * |
2223 | * The page is still frozen if the return value is not NULL. | |
2224 | * | |
2225 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2226 | * |
2227 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2228 | */ |
2229 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2230 | { | |
2231 | struct page new; | |
2232 | unsigned long counters; | |
2233 | void *freelist; | |
2234 | ||
2235 | do { | |
2236 | freelist = page->freelist; | |
2237 | counters = page->counters; | |
6faa6833 | 2238 | |
213eeb9f | 2239 | new.counters = counters; |
a0132ac0 | 2240 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2241 | |
2242 | new.inuse = page->objects; | |
2243 | new.frozen = freelist != NULL; | |
2244 | ||
d24ac77f | 2245 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2246 | freelist, counters, |
2247 | NULL, new.counters, | |
2248 | "get_freelist")); | |
2249 | ||
2250 | return freelist; | |
2251 | } | |
2252 | ||
81819f0f | 2253 | /* |
894b8788 CL |
2254 | * Slow path. The lockless freelist is empty or we need to perform |
2255 | * debugging duties. | |
2256 | * | |
894b8788 CL |
2257 | * Processing is still very fast if new objects have been freed to the |
2258 | * regular freelist. In that case we simply take over the regular freelist | |
2259 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2260 | * |
894b8788 CL |
2261 | * If that is not working then we fall back to the partial lists. We take the |
2262 | * first element of the freelist as the object to allocate now and move the | |
2263 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2264 | * |
894b8788 | 2265 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2266 | * we need to allocate a new slab. This is the slowest path since it involves |
2267 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2268 | */ |
ce71e27c EGM |
2269 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2270 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2271 | { |
6faa6833 | 2272 | void *freelist; |
f6e7def7 | 2273 | struct page *page; |
8a5ec0ba CL |
2274 | unsigned long flags; |
2275 | ||
2276 | local_irq_save(flags); | |
2277 | #ifdef CONFIG_PREEMPT | |
2278 | /* | |
2279 | * We may have been preempted and rescheduled on a different | |
2280 | * cpu before disabling interrupts. Need to reload cpu area | |
2281 | * pointer. | |
2282 | */ | |
2283 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2284 | #endif |
81819f0f | 2285 | |
f6e7def7 CL |
2286 | page = c->page; |
2287 | if (!page) | |
81819f0f | 2288 | goto new_slab; |
49e22585 | 2289 | redo: |
6faa6833 | 2290 | |
57d437d2 | 2291 | if (unlikely(!node_match(page, node))) { |
e36a2652 | 2292 | stat(s, ALLOC_NODE_MISMATCH); |
f6e7def7 | 2293 | deactivate_slab(s, page, c->freelist); |
c17dda40 CL |
2294 | c->page = NULL; |
2295 | c->freelist = NULL; | |
fc59c053 CL |
2296 | goto new_slab; |
2297 | } | |
6446faa2 | 2298 | |
072bb0aa MG |
2299 | /* |
2300 | * By rights, we should be searching for a slab page that was | |
2301 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2302 | * information when the page leaves the per-cpu allocator | |
2303 | */ | |
2304 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2305 | deactivate_slab(s, page, c->freelist); | |
2306 | c->page = NULL; | |
2307 | c->freelist = NULL; | |
2308 | goto new_slab; | |
2309 | } | |
2310 | ||
73736e03 | 2311 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2312 | freelist = c->freelist; |
2313 | if (freelist) | |
73736e03 | 2314 | goto load_freelist; |
03e404af | 2315 | |
2cfb7455 | 2316 | stat(s, ALLOC_SLOWPATH); |
03e404af | 2317 | |
f6e7def7 | 2318 | freelist = get_freelist(s, page); |
6446faa2 | 2319 | |
6faa6833 | 2320 | if (!freelist) { |
03e404af CL |
2321 | c->page = NULL; |
2322 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2323 | goto new_slab; |
03e404af | 2324 | } |
6446faa2 | 2325 | |
84e554e6 | 2326 | stat(s, ALLOC_REFILL); |
6446faa2 | 2327 | |
894b8788 | 2328 | load_freelist: |
507effea CL |
2329 | /* |
2330 | * freelist is pointing to the list of objects to be used. | |
2331 | * page is pointing to the page from which the objects are obtained. | |
2332 | * That page must be frozen for per cpu allocations to work. | |
2333 | */ | |
a0132ac0 | 2334 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2335 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2336 | c->tid = next_tid(c->tid); |
2337 | local_irq_restore(flags); | |
6faa6833 | 2338 | return freelist; |
81819f0f | 2339 | |
81819f0f | 2340 | new_slab: |
2cfb7455 | 2341 | |
49e22585 | 2342 | if (c->partial) { |
f6e7def7 CL |
2343 | page = c->page = c->partial; |
2344 | c->partial = page->next; | |
49e22585 CL |
2345 | stat(s, CPU_PARTIAL_ALLOC); |
2346 | c->freelist = NULL; | |
2347 | goto redo; | |
81819f0f CL |
2348 | } |
2349 | ||
188fd063 | 2350 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2351 | |
f4697436 CL |
2352 | if (unlikely(!freelist)) { |
2353 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) | |
2354 | slab_out_of_memory(s, gfpflags, node); | |
2cfb7455 | 2355 | |
f4697436 CL |
2356 | local_irq_restore(flags); |
2357 | return NULL; | |
81819f0f | 2358 | } |
2cfb7455 | 2359 | |
f6e7def7 | 2360 | page = c->page; |
5091b74a | 2361 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2362 | goto load_freelist; |
2cfb7455 | 2363 | |
497b66f2 | 2364 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2365 | if (kmem_cache_debug(s) && |
2366 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2367 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2368 | |
f6e7def7 | 2369 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2370 | c->page = NULL; |
2371 | c->freelist = NULL; | |
a71ae47a | 2372 | local_irq_restore(flags); |
6faa6833 | 2373 | return freelist; |
894b8788 CL |
2374 | } |
2375 | ||
2376 | /* | |
2377 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2378 | * have the fastpath folded into their functions. So no function call | |
2379 | * overhead for requests that can be satisfied on the fastpath. | |
2380 | * | |
2381 | * The fastpath works by first checking if the lockless freelist can be used. | |
2382 | * If not then __slab_alloc is called for slow processing. | |
2383 | * | |
2384 | * Otherwise we can simply pick the next object from the lockless free list. | |
2385 | */ | |
2b847c3c | 2386 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2387 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2388 | { |
894b8788 | 2389 | void **object; |
dfb4f096 | 2390 | struct kmem_cache_cpu *c; |
57d437d2 | 2391 | struct page *page; |
8a5ec0ba | 2392 | unsigned long tid; |
1f84260c | 2393 | |
c016b0bd | 2394 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2395 | return NULL; |
1f84260c | 2396 | |
d79923fa | 2397 | s = memcg_kmem_get_cache(s, gfpflags); |
8a5ec0ba | 2398 | redo: |
8a5ec0ba CL |
2399 | /* |
2400 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2401 | * enabled. We may switch back and forth between cpus while | |
2402 | * reading from one cpu area. That does not matter as long | |
2403 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b CL |
2404 | * |
2405 | * Preemption is disabled for the retrieval of the tid because that | |
2406 | * must occur from the current processor. We cannot allow rescheduling | |
2407 | * on a different processor between the determination of the pointer | |
2408 | * and the retrieval of the tid. | |
8a5ec0ba | 2409 | */ |
7cccd80b | 2410 | preempt_disable(); |
9dfc6e68 | 2411 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2412 | |
8a5ec0ba CL |
2413 | /* |
2414 | * The transaction ids are globally unique per cpu and per operation on | |
2415 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2416 | * occurs on the right processor and that there was no operation on the | |
2417 | * linked list in between. | |
2418 | */ | |
2419 | tid = c->tid; | |
7cccd80b | 2420 | preempt_enable(); |
8a5ec0ba | 2421 | |
9dfc6e68 | 2422 | object = c->freelist; |
57d437d2 | 2423 | page = c->page; |
ac6434e6 | 2424 | if (unlikely(!object || !node_match(page, node))) |
dfb4f096 | 2425 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2426 | |
2427 | else { | |
0ad9500e ED |
2428 | void *next_object = get_freepointer_safe(s, object); |
2429 | ||
8a5ec0ba | 2430 | /* |
25985edc | 2431 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2432 | * operation and if we are on the right processor. |
2433 | * | |
d0e0ac97 CG |
2434 | * The cmpxchg does the following atomically (without lock |
2435 | * semantics!) | |
8a5ec0ba CL |
2436 | * 1. Relocate first pointer to the current per cpu area. |
2437 | * 2. Verify that tid and freelist have not been changed | |
2438 | * 3. If they were not changed replace tid and freelist | |
2439 | * | |
d0e0ac97 CG |
2440 | * Since this is without lock semantics the protection is only |
2441 | * against code executing on this cpu *not* from access by | |
2442 | * other cpus. | |
8a5ec0ba | 2443 | */ |
933393f5 | 2444 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2445 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2446 | object, tid, | |
0ad9500e | 2447 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2448 | |
2449 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2450 | goto redo; | |
2451 | } | |
0ad9500e | 2452 | prefetch_freepointer(s, next_object); |
84e554e6 | 2453 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2454 | } |
8a5ec0ba | 2455 | |
74e2134f | 2456 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2457 | memset(object, 0, s->object_size); |
d07dbea4 | 2458 | |
c016b0bd | 2459 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2460 | |
894b8788 | 2461 | return object; |
81819f0f CL |
2462 | } |
2463 | ||
2b847c3c EG |
2464 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2465 | gfp_t gfpflags, unsigned long addr) | |
2466 | { | |
2467 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2468 | } | |
2469 | ||
81819f0f CL |
2470 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2471 | { | |
2b847c3c | 2472 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2473 | |
d0e0ac97 CG |
2474 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2475 | s->size, gfpflags); | |
5b882be4 EGM |
2476 | |
2477 | return ret; | |
81819f0f CL |
2478 | } |
2479 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2480 | ||
0f24f128 | 2481 | #ifdef CONFIG_TRACING |
4a92379b RK |
2482 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2483 | { | |
2b847c3c | 2484 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b RK |
2485 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
2486 | return ret; | |
2487 | } | |
2488 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2489 | #endif |
2490 | ||
81819f0f CL |
2491 | #ifdef CONFIG_NUMA |
2492 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2493 | { | |
2b847c3c | 2494 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2495 | |
ca2b84cb | 2496 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2497 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2498 | |
2499 | return ret; | |
81819f0f CL |
2500 | } |
2501 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2502 | |
0f24f128 | 2503 | #ifdef CONFIG_TRACING |
4a92379b | 2504 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2505 | gfp_t gfpflags, |
4a92379b | 2506 | int node, size_t size) |
5b882be4 | 2507 | { |
2b847c3c | 2508 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2509 | |
2510 | trace_kmalloc_node(_RET_IP_, ret, | |
2511 | size, s->size, gfpflags, node); | |
2512 | return ret; | |
5b882be4 | 2513 | } |
4a92379b | 2514 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2515 | #endif |
5d1f57e4 | 2516 | #endif |
5b882be4 | 2517 | |
81819f0f | 2518 | /* |
894b8788 CL |
2519 | * Slow patch handling. This may still be called frequently since objects |
2520 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2521 | * |
894b8788 CL |
2522 | * So we still attempt to reduce cache line usage. Just take the slab |
2523 | * lock and free the item. If there is no additional partial page | |
2524 | * handling required then we can return immediately. | |
81819f0f | 2525 | */ |
894b8788 | 2526 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2527 | void *x, unsigned long addr) |
81819f0f CL |
2528 | { |
2529 | void *prior; | |
2530 | void **object = (void *)x; | |
2cfb7455 | 2531 | int was_frozen; |
2cfb7455 CL |
2532 | struct page new; |
2533 | unsigned long counters; | |
2534 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2535 | unsigned long uninitialized_var(flags); |
81819f0f | 2536 | |
8a5ec0ba | 2537 | stat(s, FREE_SLOWPATH); |
81819f0f | 2538 | |
19c7ff9e CL |
2539 | if (kmem_cache_debug(s) && |
2540 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2541 | return; |
6446faa2 | 2542 | |
2cfb7455 | 2543 | do { |
837d678d JK |
2544 | if (unlikely(n)) { |
2545 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2546 | n = NULL; | |
2547 | } | |
2cfb7455 CL |
2548 | prior = page->freelist; |
2549 | counters = page->counters; | |
2550 | set_freepointer(s, object, prior); | |
2551 | new.counters = counters; | |
2552 | was_frozen = new.frozen; | |
2553 | new.inuse--; | |
837d678d | 2554 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2555 | |
c65c1877 | 2556 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2557 | |
2558 | /* | |
d0e0ac97 CG |
2559 | * Slab was on no list before and will be |
2560 | * partially empty | |
2561 | * We can defer the list move and instead | |
2562 | * freeze it. | |
49e22585 CL |
2563 | */ |
2564 | new.frozen = 1; | |
2565 | ||
c65c1877 | 2566 | } else { /* Needs to be taken off a list */ |
49e22585 CL |
2567 | |
2568 | n = get_node(s, page_to_nid(page)); | |
2569 | /* | |
2570 | * Speculatively acquire the list_lock. | |
2571 | * If the cmpxchg does not succeed then we may | |
2572 | * drop the list_lock without any processing. | |
2573 | * | |
2574 | * Otherwise the list_lock will synchronize with | |
2575 | * other processors updating the list of slabs. | |
2576 | */ | |
2577 | spin_lock_irqsave(&n->list_lock, flags); | |
2578 | ||
2579 | } | |
2cfb7455 | 2580 | } |
81819f0f | 2581 | |
2cfb7455 CL |
2582 | } while (!cmpxchg_double_slab(s, page, |
2583 | prior, counters, | |
2584 | object, new.counters, | |
2585 | "__slab_free")); | |
81819f0f | 2586 | |
2cfb7455 | 2587 | if (likely(!n)) { |
49e22585 CL |
2588 | |
2589 | /* | |
2590 | * If we just froze the page then put it onto the | |
2591 | * per cpu partial list. | |
2592 | */ | |
8028dcea | 2593 | if (new.frozen && !was_frozen) { |
49e22585 | 2594 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2595 | stat(s, CPU_PARTIAL_FREE); |
2596 | } | |
49e22585 | 2597 | /* |
2cfb7455 CL |
2598 | * The list lock was not taken therefore no list |
2599 | * activity can be necessary. | |
2600 | */ | |
2601 | if (was_frozen) | |
2602 | stat(s, FREE_FROZEN); | |
80f08c19 | 2603 | return; |
2cfb7455 | 2604 | } |
81819f0f | 2605 | |
837d678d JK |
2606 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) |
2607 | goto slab_empty; | |
2608 | ||
81819f0f | 2609 | /* |
837d678d JK |
2610 | * Objects left in the slab. If it was not on the partial list before |
2611 | * then add it. | |
81819f0f | 2612 | */ |
345c905d JK |
2613 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2614 | if (kmem_cache_debug(s)) | |
c65c1877 | 2615 | remove_full(s, n, page); |
837d678d JK |
2616 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2617 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2618 | } |
80f08c19 | 2619 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2620 | return; |
2621 | ||
2622 | slab_empty: | |
a973e9dd | 2623 | if (prior) { |
81819f0f | 2624 | /* |
6fbabb20 | 2625 | * Slab on the partial list. |
81819f0f | 2626 | */ |
5cc6eee8 | 2627 | remove_partial(n, page); |
84e554e6 | 2628 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2629 | } else { |
6fbabb20 | 2630 | /* Slab must be on the full list */ |
c65c1877 PZ |
2631 | remove_full(s, n, page); |
2632 | } | |
2cfb7455 | 2633 | |
80f08c19 | 2634 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2635 | stat(s, FREE_SLAB); |
81819f0f | 2636 | discard_slab(s, page); |
81819f0f CL |
2637 | } |
2638 | ||
894b8788 CL |
2639 | /* |
2640 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2641 | * can perform fastpath freeing without additional function calls. | |
2642 | * | |
2643 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2644 | * of this processor. This typically the case if we have just allocated | |
2645 | * the item before. | |
2646 | * | |
2647 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2648 | * with all sorts of special processing. | |
2649 | */ | |
06428780 | 2650 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2651 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2652 | { |
2653 | void **object = (void *)x; | |
dfb4f096 | 2654 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2655 | unsigned long tid; |
1f84260c | 2656 | |
c016b0bd CL |
2657 | slab_free_hook(s, x); |
2658 | ||
8a5ec0ba CL |
2659 | redo: |
2660 | /* | |
2661 | * Determine the currently cpus per cpu slab. | |
2662 | * The cpu may change afterward. However that does not matter since | |
2663 | * data is retrieved via this pointer. If we are on the same cpu | |
2664 | * during the cmpxchg then the free will succedd. | |
2665 | */ | |
7cccd80b | 2666 | preempt_disable(); |
9dfc6e68 | 2667 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2668 | |
8a5ec0ba | 2669 | tid = c->tid; |
7cccd80b | 2670 | preempt_enable(); |
c016b0bd | 2671 | |
442b06bc | 2672 | if (likely(page == c->page)) { |
ff12059e | 2673 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2674 | |
933393f5 | 2675 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2676 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2677 | c->freelist, tid, | |
2678 | object, next_tid(tid)))) { | |
2679 | ||
2680 | note_cmpxchg_failure("slab_free", s, tid); | |
2681 | goto redo; | |
2682 | } | |
84e554e6 | 2683 | stat(s, FREE_FASTPATH); |
894b8788 | 2684 | } else |
ff12059e | 2685 | __slab_free(s, page, x, addr); |
894b8788 | 2686 | |
894b8788 CL |
2687 | } |
2688 | ||
81819f0f CL |
2689 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2690 | { | |
b9ce5ef4 GC |
2691 | s = cache_from_obj(s, x); |
2692 | if (!s) | |
79576102 | 2693 | return; |
b9ce5ef4 | 2694 | slab_free(s, virt_to_head_page(x), x, _RET_IP_); |
ca2b84cb | 2695 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2696 | } |
2697 | EXPORT_SYMBOL(kmem_cache_free); | |
2698 | ||
81819f0f | 2699 | /* |
672bba3a CL |
2700 | * Object placement in a slab is made very easy because we always start at |
2701 | * offset 0. If we tune the size of the object to the alignment then we can | |
2702 | * get the required alignment by putting one properly sized object after | |
2703 | * another. | |
81819f0f CL |
2704 | * |
2705 | * Notice that the allocation order determines the sizes of the per cpu | |
2706 | * caches. Each processor has always one slab available for allocations. | |
2707 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2708 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2709 | * locking overhead. |
81819f0f CL |
2710 | */ |
2711 | ||
2712 | /* | |
2713 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2714 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2715 | * and increases the number of allocations possible without having to | |
2716 | * take the list_lock. | |
2717 | */ | |
2718 | static int slub_min_order; | |
114e9e89 | 2719 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2720 | static int slub_min_objects; |
81819f0f CL |
2721 | |
2722 | /* | |
2723 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2724 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2725 | */ |
2726 | static int slub_nomerge; | |
2727 | ||
81819f0f CL |
2728 | /* |
2729 | * Calculate the order of allocation given an slab object size. | |
2730 | * | |
672bba3a CL |
2731 | * The order of allocation has significant impact on performance and other |
2732 | * system components. Generally order 0 allocations should be preferred since | |
2733 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2734 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2735 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2736 | * would be wasted. |
2737 | * | |
2738 | * In order to reach satisfactory performance we must ensure that a minimum | |
2739 | * number of objects is in one slab. Otherwise we may generate too much | |
2740 | * activity on the partial lists which requires taking the list_lock. This is | |
2741 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2742 | * |
672bba3a CL |
2743 | * slub_max_order specifies the order where we begin to stop considering the |
2744 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2745 | * we try to keep the page order as low as possible. So we accept more waste | |
2746 | * of space in favor of a small page order. | |
81819f0f | 2747 | * |
672bba3a CL |
2748 | * Higher order allocations also allow the placement of more objects in a |
2749 | * slab and thereby reduce object handling overhead. If the user has | |
2750 | * requested a higher mininum order then we start with that one instead of | |
2751 | * the smallest order which will fit the object. | |
81819f0f | 2752 | */ |
5e6d444e | 2753 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2754 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2755 | { |
2756 | int order; | |
2757 | int rem; | |
6300ea75 | 2758 | int min_order = slub_min_order; |
81819f0f | 2759 | |
ab9a0f19 | 2760 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2761 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2762 | |
6300ea75 | 2763 | for (order = max(min_order, |
5e6d444e CL |
2764 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2765 | order <= max_order; order++) { | |
81819f0f | 2766 | |
5e6d444e | 2767 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2768 | |
ab9a0f19 | 2769 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2770 | continue; |
2771 | ||
ab9a0f19 | 2772 | rem = (slab_size - reserved) % size; |
81819f0f | 2773 | |
5e6d444e | 2774 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2775 | break; |
2776 | ||
2777 | } | |
672bba3a | 2778 | |
81819f0f CL |
2779 | return order; |
2780 | } | |
2781 | ||
ab9a0f19 | 2782 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2783 | { |
2784 | int order; | |
2785 | int min_objects; | |
2786 | int fraction; | |
e8120ff1 | 2787 | int max_objects; |
5e6d444e CL |
2788 | |
2789 | /* | |
2790 | * Attempt to find best configuration for a slab. This | |
2791 | * works by first attempting to generate a layout with | |
2792 | * the best configuration and backing off gradually. | |
2793 | * | |
2794 | * First we reduce the acceptable waste in a slab. Then | |
2795 | * we reduce the minimum objects required in a slab. | |
2796 | */ | |
2797 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2798 | if (!min_objects) |
2799 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2800 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2801 | min_objects = min(min_objects, max_objects); |
2802 | ||
5e6d444e | 2803 | while (min_objects > 1) { |
c124f5b5 | 2804 | fraction = 16; |
5e6d444e CL |
2805 | while (fraction >= 4) { |
2806 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2807 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2808 | if (order <= slub_max_order) |
2809 | return order; | |
2810 | fraction /= 2; | |
2811 | } | |
5086c389 | 2812 | min_objects--; |
5e6d444e CL |
2813 | } |
2814 | ||
2815 | /* | |
2816 | * We were unable to place multiple objects in a slab. Now | |
2817 | * lets see if we can place a single object there. | |
2818 | */ | |
ab9a0f19 | 2819 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2820 | if (order <= slub_max_order) |
2821 | return order; | |
2822 | ||
2823 | /* | |
2824 | * Doh this slab cannot be placed using slub_max_order. | |
2825 | */ | |
ab9a0f19 | 2826 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2827 | if (order < MAX_ORDER) |
5e6d444e CL |
2828 | return order; |
2829 | return -ENOSYS; | |
2830 | } | |
2831 | ||
5595cffc | 2832 | static void |
4053497d | 2833 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2834 | { |
2835 | n->nr_partial = 0; | |
81819f0f CL |
2836 | spin_lock_init(&n->list_lock); |
2837 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2838 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2839 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2840 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2841 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2842 | #endif |
81819f0f CL |
2843 | } |
2844 | ||
55136592 | 2845 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2846 | { |
6c182dc0 | 2847 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 2848 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 2849 | |
8a5ec0ba | 2850 | /* |
d4d84fef CM |
2851 | * Must align to double word boundary for the double cmpxchg |
2852 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2853 | */ |
d4d84fef CM |
2854 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2855 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2856 | |
2857 | if (!s->cpu_slab) | |
2858 | return 0; | |
2859 | ||
2860 | init_kmem_cache_cpus(s); | |
4c93c355 | 2861 | |
8a5ec0ba | 2862 | return 1; |
4c93c355 | 2863 | } |
4c93c355 | 2864 | |
51df1142 CL |
2865 | static struct kmem_cache *kmem_cache_node; |
2866 | ||
81819f0f CL |
2867 | /* |
2868 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2869 | * slab on the node for this slabcache. There are no concurrent accesses | |
2870 | * possible. | |
2871 | * | |
721ae22a ZYW |
2872 | * Note that this function only works on the kmem_cache_node |
2873 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 2874 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 2875 | */ |
55136592 | 2876 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2877 | { |
2878 | struct page *page; | |
2879 | struct kmem_cache_node *n; | |
2880 | ||
51df1142 | 2881 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2882 | |
51df1142 | 2883 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2884 | |
2885 | BUG_ON(!page); | |
a2f92ee7 CL |
2886 | if (page_to_nid(page) != node) { |
2887 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2888 | "node %d\n", node); | |
2889 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2890 | "in order to be able to continue\n"); | |
2891 | } | |
2892 | ||
81819f0f CL |
2893 | n = page->freelist; |
2894 | BUG_ON(!n); | |
51df1142 | 2895 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2896 | page->inuse = 1; |
8cb0a506 | 2897 | page->frozen = 0; |
51df1142 | 2898 | kmem_cache_node->node[node] = n; |
8ab1372f | 2899 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2900 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2901 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2902 | #endif |
4053497d | 2903 | init_kmem_cache_node(n); |
51df1142 | 2904 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2905 | |
67b6c900 DH |
2906 | /* |
2907 | * the lock is for lockdep's sake, not for any actual | |
2908 | * race protection | |
2909 | */ | |
2910 | spin_lock(&n->list_lock); | |
136333d1 | 2911 | add_partial(n, page, DEACTIVATE_TO_HEAD); |
67b6c900 | 2912 | spin_unlock(&n->list_lock); |
81819f0f CL |
2913 | } |
2914 | ||
2915 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2916 | { | |
2917 | int node; | |
2918 | ||
f64dc58c | 2919 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2920 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2921 | |
73367bd8 | 2922 | if (n) |
51df1142 CL |
2923 | kmem_cache_free(kmem_cache_node, n); |
2924 | ||
81819f0f CL |
2925 | s->node[node] = NULL; |
2926 | } | |
2927 | } | |
2928 | ||
55136592 | 2929 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2930 | { |
2931 | int node; | |
81819f0f | 2932 | |
f64dc58c | 2933 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2934 | struct kmem_cache_node *n; |
2935 | ||
73367bd8 | 2936 | if (slab_state == DOWN) { |
55136592 | 2937 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2938 | continue; |
2939 | } | |
51df1142 | 2940 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2941 | GFP_KERNEL, node); |
81819f0f | 2942 | |
73367bd8 AD |
2943 | if (!n) { |
2944 | free_kmem_cache_nodes(s); | |
2945 | return 0; | |
81819f0f | 2946 | } |
73367bd8 | 2947 | |
81819f0f | 2948 | s->node[node] = n; |
4053497d | 2949 | init_kmem_cache_node(n); |
81819f0f CL |
2950 | } |
2951 | return 1; | |
2952 | } | |
81819f0f | 2953 | |
c0bdb232 | 2954 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2955 | { |
2956 | if (min < MIN_PARTIAL) | |
2957 | min = MIN_PARTIAL; | |
2958 | else if (min > MAX_PARTIAL) | |
2959 | min = MAX_PARTIAL; | |
2960 | s->min_partial = min; | |
2961 | } | |
2962 | ||
81819f0f CL |
2963 | /* |
2964 | * calculate_sizes() determines the order and the distribution of data within | |
2965 | * a slab object. | |
2966 | */ | |
06b285dc | 2967 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2968 | { |
2969 | unsigned long flags = s->flags; | |
3b0efdfa | 2970 | unsigned long size = s->object_size; |
834f3d11 | 2971 | int order; |
81819f0f | 2972 | |
d8b42bf5 CL |
2973 | /* |
2974 | * Round up object size to the next word boundary. We can only | |
2975 | * place the free pointer at word boundaries and this determines | |
2976 | * the possible location of the free pointer. | |
2977 | */ | |
2978 | size = ALIGN(size, sizeof(void *)); | |
2979 | ||
2980 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2981 | /* |
2982 | * Determine if we can poison the object itself. If the user of | |
2983 | * the slab may touch the object after free or before allocation | |
2984 | * then we should never poison the object itself. | |
2985 | */ | |
2986 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2987 | !s->ctor) |
81819f0f CL |
2988 | s->flags |= __OBJECT_POISON; |
2989 | else | |
2990 | s->flags &= ~__OBJECT_POISON; | |
2991 | ||
81819f0f CL |
2992 | |
2993 | /* | |
672bba3a | 2994 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2995 | * end of the object and the free pointer. If not then add an |
672bba3a | 2996 | * additional word to have some bytes to store Redzone information. |
81819f0f | 2997 | */ |
3b0efdfa | 2998 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 2999 | size += sizeof(void *); |
41ecc55b | 3000 | #endif |
81819f0f CL |
3001 | |
3002 | /* | |
672bba3a CL |
3003 | * With that we have determined the number of bytes in actual use |
3004 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3005 | */ |
3006 | s->inuse = size; | |
3007 | ||
3008 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3009 | s->ctor)) { |
81819f0f CL |
3010 | /* |
3011 | * Relocate free pointer after the object if it is not | |
3012 | * permitted to overwrite the first word of the object on | |
3013 | * kmem_cache_free. | |
3014 | * | |
3015 | * This is the case if we do RCU, have a constructor or | |
3016 | * destructor or are poisoning the objects. | |
3017 | */ | |
3018 | s->offset = size; | |
3019 | size += sizeof(void *); | |
3020 | } | |
3021 | ||
c12b3c62 | 3022 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3023 | if (flags & SLAB_STORE_USER) |
3024 | /* | |
3025 | * Need to store information about allocs and frees after | |
3026 | * the object. | |
3027 | */ | |
3028 | size += 2 * sizeof(struct track); | |
3029 | ||
be7b3fbc | 3030 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3031 | /* |
3032 | * Add some empty padding so that we can catch | |
3033 | * overwrites from earlier objects rather than let | |
3034 | * tracking information or the free pointer be | |
0211a9c8 | 3035 | * corrupted if a user writes before the start |
81819f0f CL |
3036 | * of the object. |
3037 | */ | |
3038 | size += sizeof(void *); | |
41ecc55b | 3039 | #endif |
672bba3a | 3040 | |
81819f0f CL |
3041 | /* |
3042 | * SLUB stores one object immediately after another beginning from | |
3043 | * offset 0. In order to align the objects we have to simply size | |
3044 | * each object to conform to the alignment. | |
3045 | */ | |
45906855 | 3046 | size = ALIGN(size, s->align); |
81819f0f | 3047 | s->size = size; |
06b285dc CL |
3048 | if (forced_order >= 0) |
3049 | order = forced_order; | |
3050 | else | |
ab9a0f19 | 3051 | order = calculate_order(size, s->reserved); |
81819f0f | 3052 | |
834f3d11 | 3053 | if (order < 0) |
81819f0f CL |
3054 | return 0; |
3055 | ||
b7a49f0d | 3056 | s->allocflags = 0; |
834f3d11 | 3057 | if (order) |
b7a49f0d CL |
3058 | s->allocflags |= __GFP_COMP; |
3059 | ||
3060 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3061 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3062 | |
3063 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3064 | s->allocflags |= __GFP_RECLAIMABLE; | |
3065 | ||
81819f0f CL |
3066 | /* |
3067 | * Determine the number of objects per slab | |
3068 | */ | |
ab9a0f19 LJ |
3069 | s->oo = oo_make(order, size, s->reserved); |
3070 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3071 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3072 | s->max = s->oo; | |
81819f0f | 3073 | |
834f3d11 | 3074 | return !!oo_objects(s->oo); |
81819f0f CL |
3075 | } |
3076 | ||
8a13a4cc | 3077 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3078 | { |
8a13a4cc | 3079 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3080 | s->reserved = 0; |
81819f0f | 3081 | |
da9a638c LJ |
3082 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3083 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3084 | |
06b285dc | 3085 | if (!calculate_sizes(s, -1)) |
81819f0f | 3086 | goto error; |
3de47213 DR |
3087 | if (disable_higher_order_debug) { |
3088 | /* | |
3089 | * Disable debugging flags that store metadata if the min slab | |
3090 | * order increased. | |
3091 | */ | |
3b0efdfa | 3092 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3093 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3094 | s->offset = 0; | |
3095 | if (!calculate_sizes(s, -1)) | |
3096 | goto error; | |
3097 | } | |
3098 | } | |
81819f0f | 3099 | |
2565409f HC |
3100 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3101 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3102 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3103 | /* Enable fast mode */ | |
3104 | s->flags |= __CMPXCHG_DOUBLE; | |
3105 | #endif | |
3106 | ||
3b89d7d8 DR |
3107 | /* |
3108 | * The larger the object size is, the more pages we want on the partial | |
3109 | * list to avoid pounding the page allocator excessively. | |
3110 | */ | |
49e22585 CL |
3111 | set_min_partial(s, ilog2(s->size) / 2); |
3112 | ||
3113 | /* | |
3114 | * cpu_partial determined the maximum number of objects kept in the | |
3115 | * per cpu partial lists of a processor. | |
3116 | * | |
3117 | * Per cpu partial lists mainly contain slabs that just have one | |
3118 | * object freed. If they are used for allocation then they can be | |
3119 | * filled up again with minimal effort. The slab will never hit the | |
3120 | * per node partial lists and therefore no locking will be required. | |
3121 | * | |
3122 | * This setting also determines | |
3123 | * | |
3124 | * A) The number of objects from per cpu partial slabs dumped to the | |
3125 | * per node list when we reach the limit. | |
9f264904 | 3126 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3127 | * per node list when we run out of per cpu objects. We only fetch |
3128 | * 50% to keep some capacity around for frees. | |
49e22585 | 3129 | */ |
345c905d | 3130 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3131 | s->cpu_partial = 0; |
3132 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3133 | s->cpu_partial = 2; |
3134 | else if (s->size >= 1024) | |
3135 | s->cpu_partial = 6; | |
3136 | else if (s->size >= 256) | |
3137 | s->cpu_partial = 13; | |
3138 | else | |
3139 | s->cpu_partial = 30; | |
3140 | ||
81819f0f | 3141 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3142 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3143 | #endif |
55136592 | 3144 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3145 | goto error; |
81819f0f | 3146 | |
55136592 | 3147 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3148 | return 0; |
ff12059e | 3149 | |
4c93c355 | 3150 | free_kmem_cache_nodes(s); |
81819f0f CL |
3151 | error: |
3152 | if (flags & SLAB_PANIC) | |
3153 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3154 | "order=%u offset=%u flags=%lx\n", | |
d0e0ac97 CG |
3155 | s->name, (unsigned long)s->size, s->size, |
3156 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3157 | return -EINVAL; |
81819f0f | 3158 | } |
81819f0f | 3159 | |
33b12c38 CL |
3160 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3161 | const char *text) | |
3162 | { | |
3163 | #ifdef CONFIG_SLUB_DEBUG | |
3164 | void *addr = page_address(page); | |
3165 | void *p; | |
a5dd5c11 NK |
3166 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3167 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3168 | if (!map) |
3169 | return; | |
945cf2b6 | 3170 | slab_err(s, page, text, s->name); |
33b12c38 | 3171 | slab_lock(page); |
33b12c38 | 3172 | |
5f80b13a | 3173 | get_map(s, page, map); |
33b12c38 CL |
3174 | for_each_object(p, s, addr, page->objects) { |
3175 | ||
3176 | if (!test_bit(slab_index(p, s, addr), map)) { | |
3177 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
3178 | p, p - addr); | |
3179 | print_tracking(s, p); | |
3180 | } | |
3181 | } | |
3182 | slab_unlock(page); | |
bbd7d57b | 3183 | kfree(map); |
33b12c38 CL |
3184 | #endif |
3185 | } | |
3186 | ||
81819f0f | 3187 | /* |
599870b1 | 3188 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3189 | * This is called from kmem_cache_close(). We must be the last thread |
3190 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3191 | */ |
599870b1 | 3192 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3193 | { |
81819f0f CL |
3194 | struct page *page, *h; |
3195 | ||
33b12c38 | 3196 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3197 | if (!page->inuse) { |
5cc6eee8 | 3198 | remove_partial(n, page); |
81819f0f | 3199 | discard_slab(s, page); |
33b12c38 CL |
3200 | } else { |
3201 | list_slab_objects(s, page, | |
945cf2b6 | 3202 | "Objects remaining in %s on kmem_cache_close()"); |
599870b1 | 3203 | } |
33b12c38 | 3204 | } |
81819f0f CL |
3205 | } |
3206 | ||
3207 | /* | |
672bba3a | 3208 | * Release all resources used by a slab cache. |
81819f0f | 3209 | */ |
0c710013 | 3210 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3211 | { |
3212 | int node; | |
3213 | ||
3214 | flush_all(s); | |
81819f0f | 3215 | /* Attempt to free all objects */ |
f64dc58c | 3216 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3217 | struct kmem_cache_node *n = get_node(s, node); |
3218 | ||
599870b1 CL |
3219 | free_partial(s, n); |
3220 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3221 | return 1; |
3222 | } | |
945cf2b6 | 3223 | free_percpu(s->cpu_slab); |
81819f0f CL |
3224 | free_kmem_cache_nodes(s); |
3225 | return 0; | |
3226 | } | |
3227 | ||
945cf2b6 | 3228 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f | 3229 | { |
12c3667f | 3230 | int rc = kmem_cache_close(s); |
945cf2b6 | 3231 | |
5413dfba GC |
3232 | if (!rc) { |
3233 | /* | |
3234 | * We do the same lock strategy around sysfs_slab_add, see | |
3235 | * __kmem_cache_create. Because this is pretty much the last | |
3236 | * operation we do and the lock will be released shortly after | |
3237 | * that in slab_common.c, we could just move sysfs_slab_remove | |
3238 | * to a later point in common code. We should do that when we | |
3239 | * have a common sysfs framework for all allocators. | |
3240 | */ | |
3241 | mutex_unlock(&slab_mutex); | |
81819f0f | 3242 | sysfs_slab_remove(s); |
5413dfba GC |
3243 | mutex_lock(&slab_mutex); |
3244 | } | |
12c3667f CL |
3245 | |
3246 | return rc; | |
81819f0f | 3247 | } |
81819f0f CL |
3248 | |
3249 | /******************************************************************** | |
3250 | * Kmalloc subsystem | |
3251 | *******************************************************************/ | |
3252 | ||
81819f0f CL |
3253 | static int __init setup_slub_min_order(char *str) |
3254 | { | |
06428780 | 3255 | get_option(&str, &slub_min_order); |
81819f0f CL |
3256 | |
3257 | return 1; | |
3258 | } | |
3259 | ||
3260 | __setup("slub_min_order=", setup_slub_min_order); | |
3261 | ||
3262 | static int __init setup_slub_max_order(char *str) | |
3263 | { | |
06428780 | 3264 | get_option(&str, &slub_max_order); |
818cf590 | 3265 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3266 | |
3267 | return 1; | |
3268 | } | |
3269 | ||
3270 | __setup("slub_max_order=", setup_slub_max_order); | |
3271 | ||
3272 | static int __init setup_slub_min_objects(char *str) | |
3273 | { | |
06428780 | 3274 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3275 | |
3276 | return 1; | |
3277 | } | |
3278 | ||
3279 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3280 | ||
3281 | static int __init setup_slub_nomerge(char *str) | |
3282 | { | |
3283 | slub_nomerge = 1; | |
3284 | return 1; | |
3285 | } | |
3286 | ||
3287 | __setup("slub_nomerge", setup_slub_nomerge); | |
3288 | ||
81819f0f CL |
3289 | void *__kmalloc(size_t size, gfp_t flags) |
3290 | { | |
aadb4bc4 | 3291 | struct kmem_cache *s; |
5b882be4 | 3292 | void *ret; |
81819f0f | 3293 | |
95a05b42 | 3294 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3295 | return kmalloc_large(size, flags); |
aadb4bc4 | 3296 | |
2c59dd65 | 3297 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3298 | |
3299 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3300 | return s; |
3301 | ||
2b847c3c | 3302 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3303 | |
ca2b84cb | 3304 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3305 | |
3306 | return ret; | |
81819f0f CL |
3307 | } |
3308 | EXPORT_SYMBOL(__kmalloc); | |
3309 | ||
5d1f57e4 | 3310 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3311 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3312 | { | |
b1eeab67 | 3313 | struct page *page; |
e4f7c0b4 | 3314 | void *ptr = NULL; |
f619cfe1 | 3315 | |
d79923fa | 3316 | flags |= __GFP_COMP | __GFP_NOTRACK | __GFP_KMEMCG; |
b1eeab67 | 3317 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3318 | if (page) |
e4f7c0b4 CM |
3319 | ptr = page_address(page); |
3320 | ||
d56791b3 | 3321 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3322 | return ptr; |
f619cfe1 CL |
3323 | } |
3324 | ||
81819f0f CL |
3325 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3326 | { | |
aadb4bc4 | 3327 | struct kmem_cache *s; |
5b882be4 | 3328 | void *ret; |
81819f0f | 3329 | |
95a05b42 | 3330 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3331 | ret = kmalloc_large_node(size, flags, node); |
3332 | ||
ca2b84cb EGM |
3333 | trace_kmalloc_node(_RET_IP_, ret, |
3334 | size, PAGE_SIZE << get_order(size), | |
3335 | flags, node); | |
5b882be4 EGM |
3336 | |
3337 | return ret; | |
3338 | } | |
aadb4bc4 | 3339 | |
2c59dd65 | 3340 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3341 | |
3342 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3343 | return s; |
3344 | ||
2b847c3c | 3345 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3346 | |
ca2b84cb | 3347 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3348 | |
3349 | return ret; | |
81819f0f CL |
3350 | } |
3351 | EXPORT_SYMBOL(__kmalloc_node); | |
3352 | #endif | |
3353 | ||
3354 | size_t ksize(const void *object) | |
3355 | { | |
272c1d21 | 3356 | struct page *page; |
81819f0f | 3357 | |
ef8b4520 | 3358 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3359 | return 0; |
3360 | ||
294a80a8 | 3361 | page = virt_to_head_page(object); |
294a80a8 | 3362 | |
76994412 PE |
3363 | if (unlikely(!PageSlab(page))) { |
3364 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3365 | return PAGE_SIZE << compound_order(page); |
76994412 | 3366 | } |
81819f0f | 3367 | |
1b4f59e3 | 3368 | return slab_ksize(page->slab_cache); |
81819f0f | 3369 | } |
b1aabecd | 3370 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3371 | |
3372 | void kfree(const void *x) | |
3373 | { | |
81819f0f | 3374 | struct page *page; |
5bb983b0 | 3375 | void *object = (void *)x; |
81819f0f | 3376 | |
2121db74 PE |
3377 | trace_kfree(_RET_IP_, x); |
3378 | ||
2408c550 | 3379 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3380 | return; |
3381 | ||
b49af68f | 3382 | page = virt_to_head_page(x); |
aadb4bc4 | 3383 | if (unlikely(!PageSlab(page))) { |
0937502a | 3384 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3385 | kfree_hook(x); |
d79923fa | 3386 | __free_memcg_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3387 | return; |
3388 | } | |
1b4f59e3 | 3389 | slab_free(page->slab_cache, page, object, _RET_IP_); |
81819f0f CL |
3390 | } |
3391 | EXPORT_SYMBOL(kfree); | |
3392 | ||
2086d26a | 3393 | /* |
672bba3a CL |
3394 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3395 | * the remaining slabs by the number of items in use. The slabs with the | |
3396 | * most items in use come first. New allocations will then fill those up | |
3397 | * and thus they can be removed from the partial lists. | |
3398 | * | |
3399 | * The slabs with the least items are placed last. This results in them | |
3400 | * being allocated from last increasing the chance that the last objects | |
3401 | * are freed in them. | |
2086d26a CL |
3402 | */ |
3403 | int kmem_cache_shrink(struct kmem_cache *s) | |
3404 | { | |
3405 | int node; | |
3406 | int i; | |
3407 | struct kmem_cache_node *n; | |
3408 | struct page *page; | |
3409 | struct page *t; | |
205ab99d | 3410 | int objects = oo_objects(s->max); |
2086d26a | 3411 | struct list_head *slabs_by_inuse = |
834f3d11 | 3412 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3413 | unsigned long flags; |
3414 | ||
3415 | if (!slabs_by_inuse) | |
3416 | return -ENOMEM; | |
3417 | ||
3418 | flush_all(s); | |
f64dc58c | 3419 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3420 | n = get_node(s, node); |
3421 | ||
3422 | if (!n->nr_partial) | |
3423 | continue; | |
3424 | ||
834f3d11 | 3425 | for (i = 0; i < objects; i++) |
2086d26a CL |
3426 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3427 | ||
3428 | spin_lock_irqsave(&n->list_lock, flags); | |
3429 | ||
3430 | /* | |
672bba3a | 3431 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3432 | * |
672bba3a CL |
3433 | * Note that concurrent frees may occur while we hold the |
3434 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3435 | */ |
3436 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3437 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3438 | if (!page->inuse) | |
3439 | n->nr_partial--; | |
2086d26a CL |
3440 | } |
3441 | ||
2086d26a | 3442 | /* |
672bba3a CL |
3443 | * Rebuild the partial list with the slabs filled up most |
3444 | * first and the least used slabs at the end. | |
2086d26a | 3445 | */ |
69cb8e6b | 3446 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3447 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3448 | ||
2086d26a | 3449 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3450 | |
3451 | /* Release empty slabs */ | |
3452 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3453 | discard_slab(s, page); | |
2086d26a CL |
3454 | } |
3455 | ||
3456 | kfree(slabs_by_inuse); | |
3457 | return 0; | |
3458 | } | |
3459 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3460 | ||
b9049e23 YG |
3461 | static int slab_mem_going_offline_callback(void *arg) |
3462 | { | |
3463 | struct kmem_cache *s; | |
3464 | ||
18004c5d | 3465 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3466 | list_for_each_entry(s, &slab_caches, list) |
3467 | kmem_cache_shrink(s); | |
18004c5d | 3468 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3469 | |
3470 | return 0; | |
3471 | } | |
3472 | ||
3473 | static void slab_mem_offline_callback(void *arg) | |
3474 | { | |
3475 | struct kmem_cache_node *n; | |
3476 | struct kmem_cache *s; | |
3477 | struct memory_notify *marg = arg; | |
3478 | int offline_node; | |
3479 | ||
b9d5ab25 | 3480 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3481 | |
3482 | /* | |
3483 | * If the node still has available memory. we need kmem_cache_node | |
3484 | * for it yet. | |
3485 | */ | |
3486 | if (offline_node < 0) | |
3487 | return; | |
3488 | ||
18004c5d | 3489 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3490 | list_for_each_entry(s, &slab_caches, list) { |
3491 | n = get_node(s, offline_node); | |
3492 | if (n) { | |
3493 | /* | |
3494 | * if n->nr_slabs > 0, slabs still exist on the node | |
3495 | * that is going down. We were unable to free them, | |
c9404c9c | 3496 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3497 | * callback. So, we must fail. |
3498 | */ | |
0f389ec6 | 3499 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3500 | |
3501 | s->node[offline_node] = NULL; | |
8de66a0c | 3502 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3503 | } |
3504 | } | |
18004c5d | 3505 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3506 | } |
3507 | ||
3508 | static int slab_mem_going_online_callback(void *arg) | |
3509 | { | |
3510 | struct kmem_cache_node *n; | |
3511 | struct kmem_cache *s; | |
3512 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3513 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3514 | int ret = 0; |
3515 | ||
3516 | /* | |
3517 | * If the node's memory is already available, then kmem_cache_node is | |
3518 | * already created. Nothing to do. | |
3519 | */ | |
3520 | if (nid < 0) | |
3521 | return 0; | |
3522 | ||
3523 | /* | |
0121c619 | 3524 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3525 | * allocate a kmem_cache_node structure in order to bring the node |
3526 | * online. | |
3527 | */ | |
18004c5d | 3528 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3529 | list_for_each_entry(s, &slab_caches, list) { |
3530 | /* | |
3531 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3532 | * since memory is not yet available from the node that | |
3533 | * is brought up. | |
3534 | */ | |
8de66a0c | 3535 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3536 | if (!n) { |
3537 | ret = -ENOMEM; | |
3538 | goto out; | |
3539 | } | |
4053497d | 3540 | init_kmem_cache_node(n); |
b9049e23 YG |
3541 | s->node[nid] = n; |
3542 | } | |
3543 | out: | |
18004c5d | 3544 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3545 | return ret; |
3546 | } | |
3547 | ||
3548 | static int slab_memory_callback(struct notifier_block *self, | |
3549 | unsigned long action, void *arg) | |
3550 | { | |
3551 | int ret = 0; | |
3552 | ||
3553 | switch (action) { | |
3554 | case MEM_GOING_ONLINE: | |
3555 | ret = slab_mem_going_online_callback(arg); | |
3556 | break; | |
3557 | case MEM_GOING_OFFLINE: | |
3558 | ret = slab_mem_going_offline_callback(arg); | |
3559 | break; | |
3560 | case MEM_OFFLINE: | |
3561 | case MEM_CANCEL_ONLINE: | |
3562 | slab_mem_offline_callback(arg); | |
3563 | break; | |
3564 | case MEM_ONLINE: | |
3565 | case MEM_CANCEL_OFFLINE: | |
3566 | break; | |
3567 | } | |
dc19f9db KH |
3568 | if (ret) |
3569 | ret = notifier_from_errno(ret); | |
3570 | else | |
3571 | ret = NOTIFY_OK; | |
b9049e23 YG |
3572 | return ret; |
3573 | } | |
3574 | ||
3ac38faa AM |
3575 | static struct notifier_block slab_memory_callback_nb = { |
3576 | .notifier_call = slab_memory_callback, | |
3577 | .priority = SLAB_CALLBACK_PRI, | |
3578 | }; | |
b9049e23 | 3579 | |
81819f0f CL |
3580 | /******************************************************************** |
3581 | * Basic setup of slabs | |
3582 | *******************************************************************/ | |
3583 | ||
51df1142 CL |
3584 | /* |
3585 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3586 | * the page allocator. Allocate them properly then fix up the pointers |
3587 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3588 | */ |
3589 | ||
dffb4d60 | 3590 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3591 | { |
3592 | int node; | |
dffb4d60 | 3593 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
51df1142 | 3594 | |
dffb4d60 | 3595 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3596 | |
7d557b3c GC |
3597 | /* |
3598 | * This runs very early, and only the boot processor is supposed to be | |
3599 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3600 | * IPIs around. | |
3601 | */ | |
3602 | __flush_cpu_slab(s, smp_processor_id()); | |
51df1142 CL |
3603 | for_each_node_state(node, N_NORMAL_MEMORY) { |
3604 | struct kmem_cache_node *n = get_node(s, node); | |
3605 | struct page *p; | |
3606 | ||
3607 | if (n) { | |
3608 | list_for_each_entry(p, &n->partial, lru) | |
1b4f59e3 | 3609 | p->slab_cache = s; |
51df1142 | 3610 | |
607bf324 | 3611 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 | 3612 | list_for_each_entry(p, &n->full, lru) |
1b4f59e3 | 3613 | p->slab_cache = s; |
51df1142 CL |
3614 | #endif |
3615 | } | |
3616 | } | |
dffb4d60 CL |
3617 | list_add(&s->list, &slab_caches); |
3618 | return s; | |
51df1142 CL |
3619 | } |
3620 | ||
81819f0f CL |
3621 | void __init kmem_cache_init(void) |
3622 | { | |
dffb4d60 CL |
3623 | static __initdata struct kmem_cache boot_kmem_cache, |
3624 | boot_kmem_cache_node; | |
51df1142 | 3625 | |
fc8d8620 SG |
3626 | if (debug_guardpage_minorder()) |
3627 | slub_max_order = 0; | |
3628 | ||
dffb4d60 CL |
3629 | kmem_cache_node = &boot_kmem_cache_node; |
3630 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3631 | |
dffb4d60 CL |
3632 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3633 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3634 | |
3ac38faa | 3635 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3636 | |
3637 | /* Able to allocate the per node structures */ | |
3638 | slab_state = PARTIAL; | |
3639 | ||
dffb4d60 CL |
3640 | create_boot_cache(kmem_cache, "kmem_cache", |
3641 | offsetof(struct kmem_cache, node) + | |
3642 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3643 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3644 | |
dffb4d60 | 3645 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3646 | |
51df1142 CL |
3647 | /* |
3648 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3649 | * kmem_cache_node is separately allocated so no need to | |
3650 | * update any list pointers. | |
3651 | */ | |
dffb4d60 | 3652 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3653 | |
3654 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f97d5f63 | 3655 | create_kmalloc_caches(0); |
81819f0f CL |
3656 | |
3657 | #ifdef CONFIG_SMP | |
3658 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3659 | #endif |
81819f0f | 3660 | |
3adbefee | 3661 | printk(KERN_INFO |
f97d5f63 | 3662 | "SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d," |
4b356be0 | 3663 | " CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3664 | cache_line_size(), |
81819f0f CL |
3665 | slub_min_order, slub_max_order, slub_min_objects, |
3666 | nr_cpu_ids, nr_node_ids); | |
3667 | } | |
3668 | ||
7e85ee0c PE |
3669 | void __init kmem_cache_init_late(void) |
3670 | { | |
7e85ee0c PE |
3671 | } |
3672 | ||
81819f0f CL |
3673 | /* |
3674 | * Find a mergeable slab cache | |
3675 | */ | |
3676 | static int slab_unmergeable(struct kmem_cache *s) | |
3677 | { | |
3678 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3679 | return 1; | |
3680 | ||
c59def9f | 3681 | if (s->ctor) |
81819f0f CL |
3682 | return 1; |
3683 | ||
8ffa6875 CL |
3684 | /* |
3685 | * We may have set a slab to be unmergeable during bootstrap. | |
3686 | */ | |
3687 | if (s->refcount < 0) | |
3688 | return 1; | |
3689 | ||
81819f0f CL |
3690 | return 0; |
3691 | } | |
3692 | ||
2633d7a0 | 3693 | static struct kmem_cache *find_mergeable(struct mem_cgroup *memcg, size_t size, |
ba0268a8 | 3694 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3695 | void (*ctor)(void *)) |
81819f0f | 3696 | { |
5b95a4ac | 3697 | struct kmem_cache *s; |
81819f0f CL |
3698 | |
3699 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3700 | return NULL; | |
3701 | ||
c59def9f | 3702 | if (ctor) |
81819f0f CL |
3703 | return NULL; |
3704 | ||
3705 | size = ALIGN(size, sizeof(void *)); | |
3706 | align = calculate_alignment(flags, align, size); | |
3707 | size = ALIGN(size, align); | |
ba0268a8 | 3708 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3709 | |
5b95a4ac | 3710 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3711 | if (slab_unmergeable(s)) |
3712 | continue; | |
3713 | ||
3714 | if (size > s->size) | |
3715 | continue; | |
3716 | ||
ba0268a8 | 3717 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3718 | continue; |
3719 | /* | |
3720 | * Check if alignment is compatible. | |
3721 | * Courtesy of Adrian Drzewiecki | |
3722 | */ | |
06428780 | 3723 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3724 | continue; |
3725 | ||
3726 | if (s->size - size >= sizeof(void *)) | |
3727 | continue; | |
3728 | ||
2633d7a0 GC |
3729 | if (!cache_match_memcg(s, memcg)) |
3730 | continue; | |
3731 | ||
81819f0f CL |
3732 | return s; |
3733 | } | |
3734 | return NULL; | |
3735 | } | |
3736 | ||
2633d7a0 GC |
3737 | struct kmem_cache * |
3738 | __kmem_cache_alias(struct mem_cgroup *memcg, const char *name, size_t size, | |
3739 | size_t align, unsigned long flags, void (*ctor)(void *)) | |
81819f0f CL |
3740 | { |
3741 | struct kmem_cache *s; | |
3742 | ||
2633d7a0 | 3743 | s = find_mergeable(memcg, size, align, flags, name, ctor); |
81819f0f CL |
3744 | if (s) { |
3745 | s->refcount++; | |
3746 | /* | |
3747 | * Adjust the object sizes so that we clear | |
3748 | * the complete object on kzalloc. | |
3749 | */ | |
3b0efdfa | 3750 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3751 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3752 | |
7b8f3b66 | 3753 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3754 | s->refcount--; |
cbb79694 | 3755 | s = NULL; |
7b8f3b66 | 3756 | } |
a0e1d1be | 3757 | } |
6446faa2 | 3758 | |
cbb79694 CL |
3759 | return s; |
3760 | } | |
84c1cf62 | 3761 | |
8a13a4cc | 3762 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3763 | { |
aac3a166 PE |
3764 | int err; |
3765 | ||
3766 | err = kmem_cache_open(s, flags); | |
3767 | if (err) | |
3768 | return err; | |
20cea968 | 3769 | |
45530c44 CL |
3770 | /* Mutex is not taken during early boot */ |
3771 | if (slab_state <= UP) | |
3772 | return 0; | |
3773 | ||
107dab5c | 3774 | memcg_propagate_slab_attrs(s); |
aac3a166 PE |
3775 | mutex_unlock(&slab_mutex); |
3776 | err = sysfs_slab_add(s); | |
3777 | mutex_lock(&slab_mutex); | |
20cea968 | 3778 | |
aac3a166 PE |
3779 | if (err) |
3780 | kmem_cache_close(s); | |
20cea968 | 3781 | |
aac3a166 | 3782 | return err; |
81819f0f | 3783 | } |
81819f0f | 3784 | |
81819f0f | 3785 | #ifdef CONFIG_SMP |
81819f0f | 3786 | /* |
672bba3a CL |
3787 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3788 | * necessary. | |
81819f0f | 3789 | */ |
0db0628d | 3790 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3791 | unsigned long action, void *hcpu) |
3792 | { | |
3793 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3794 | struct kmem_cache *s; |
3795 | unsigned long flags; | |
81819f0f CL |
3796 | |
3797 | switch (action) { | |
3798 | case CPU_UP_CANCELED: | |
8bb78442 | 3799 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3800 | case CPU_DEAD: |
8bb78442 | 3801 | case CPU_DEAD_FROZEN: |
18004c5d | 3802 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3803 | list_for_each_entry(s, &slab_caches, list) { |
3804 | local_irq_save(flags); | |
3805 | __flush_cpu_slab(s, cpu); | |
3806 | local_irq_restore(flags); | |
3807 | } | |
18004c5d | 3808 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3809 | break; |
3810 | default: | |
3811 | break; | |
3812 | } | |
3813 | return NOTIFY_OK; | |
3814 | } | |
3815 | ||
0db0628d | 3816 | static struct notifier_block slab_notifier = { |
3adbefee | 3817 | .notifier_call = slab_cpuup_callback |
06428780 | 3818 | }; |
81819f0f CL |
3819 | |
3820 | #endif | |
3821 | ||
ce71e27c | 3822 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3823 | { |
aadb4bc4 | 3824 | struct kmem_cache *s; |
94b528d0 | 3825 | void *ret; |
aadb4bc4 | 3826 | |
95a05b42 | 3827 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
3828 | return kmalloc_large(size, gfpflags); |
3829 | ||
2c59dd65 | 3830 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3831 | |
2408c550 | 3832 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3833 | return s; |
81819f0f | 3834 | |
2b847c3c | 3835 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 3836 | |
25985edc | 3837 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3838 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3839 | |
3840 | return ret; | |
81819f0f CL |
3841 | } |
3842 | ||
5d1f57e4 | 3843 | #ifdef CONFIG_NUMA |
81819f0f | 3844 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3845 | int node, unsigned long caller) |
81819f0f | 3846 | { |
aadb4bc4 | 3847 | struct kmem_cache *s; |
94b528d0 | 3848 | void *ret; |
aadb4bc4 | 3849 | |
95a05b42 | 3850 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
3851 | ret = kmalloc_large_node(size, gfpflags, node); |
3852 | ||
3853 | trace_kmalloc_node(caller, ret, | |
3854 | size, PAGE_SIZE << get_order(size), | |
3855 | gfpflags, node); | |
3856 | ||
3857 | return ret; | |
3858 | } | |
eada35ef | 3859 | |
2c59dd65 | 3860 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3861 | |
2408c550 | 3862 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3863 | return s; |
81819f0f | 3864 | |
2b847c3c | 3865 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 3866 | |
25985edc | 3867 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3868 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3869 | |
3870 | return ret; | |
81819f0f | 3871 | } |
5d1f57e4 | 3872 | #endif |
81819f0f | 3873 | |
ab4d5ed5 | 3874 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3875 | static int count_inuse(struct page *page) |
3876 | { | |
3877 | return page->inuse; | |
3878 | } | |
3879 | ||
3880 | static int count_total(struct page *page) | |
3881 | { | |
3882 | return page->objects; | |
3883 | } | |
ab4d5ed5 | 3884 | #endif |
205ab99d | 3885 | |
ab4d5ed5 | 3886 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3887 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3888 | unsigned long *map) | |
53e15af0 CL |
3889 | { |
3890 | void *p; | |
a973e9dd | 3891 | void *addr = page_address(page); |
53e15af0 CL |
3892 | |
3893 | if (!check_slab(s, page) || | |
3894 | !on_freelist(s, page, NULL)) | |
3895 | return 0; | |
3896 | ||
3897 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3898 | bitmap_zero(map, page->objects); |
53e15af0 | 3899 | |
5f80b13a CL |
3900 | get_map(s, page, map); |
3901 | for_each_object(p, s, addr, page->objects) { | |
3902 | if (test_bit(slab_index(p, s, addr), map)) | |
3903 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3904 | return 0; | |
53e15af0 CL |
3905 | } |
3906 | ||
224a88be | 3907 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3908 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3909 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3910 | return 0; |
3911 | return 1; | |
3912 | } | |
3913 | ||
434e245d CL |
3914 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3915 | unsigned long *map) | |
53e15af0 | 3916 | { |
881db7fb CL |
3917 | slab_lock(page); |
3918 | validate_slab(s, page, map); | |
3919 | slab_unlock(page); | |
53e15af0 CL |
3920 | } |
3921 | ||
434e245d CL |
3922 | static int validate_slab_node(struct kmem_cache *s, |
3923 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3924 | { |
3925 | unsigned long count = 0; | |
3926 | struct page *page; | |
3927 | unsigned long flags; | |
3928 | ||
3929 | spin_lock_irqsave(&n->list_lock, flags); | |
3930 | ||
3931 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3932 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3933 | count++; |
3934 | } | |
3935 | if (count != n->nr_partial) | |
3936 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3937 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3938 | ||
3939 | if (!(s->flags & SLAB_STORE_USER)) | |
3940 | goto out; | |
3941 | ||
3942 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3943 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3944 | count++; |
3945 | } | |
3946 | if (count != atomic_long_read(&n->nr_slabs)) | |
3947 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3948 | "counter=%ld\n", s->name, count, | |
3949 | atomic_long_read(&n->nr_slabs)); | |
3950 | ||
3951 | out: | |
3952 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3953 | return count; | |
3954 | } | |
3955 | ||
434e245d | 3956 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3957 | { |
3958 | int node; | |
3959 | unsigned long count = 0; | |
205ab99d | 3960 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3961 | sizeof(unsigned long), GFP_KERNEL); |
3962 | ||
3963 | if (!map) | |
3964 | return -ENOMEM; | |
53e15af0 CL |
3965 | |
3966 | flush_all(s); | |
f64dc58c | 3967 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3968 | struct kmem_cache_node *n = get_node(s, node); |
3969 | ||
434e245d | 3970 | count += validate_slab_node(s, n, map); |
53e15af0 | 3971 | } |
434e245d | 3972 | kfree(map); |
53e15af0 CL |
3973 | return count; |
3974 | } | |
88a420e4 | 3975 | /* |
672bba3a | 3976 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3977 | * and freed. |
3978 | */ | |
3979 | ||
3980 | struct location { | |
3981 | unsigned long count; | |
ce71e27c | 3982 | unsigned long addr; |
45edfa58 CL |
3983 | long long sum_time; |
3984 | long min_time; | |
3985 | long max_time; | |
3986 | long min_pid; | |
3987 | long max_pid; | |
174596a0 | 3988 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3989 | nodemask_t nodes; |
88a420e4 CL |
3990 | }; |
3991 | ||
3992 | struct loc_track { | |
3993 | unsigned long max; | |
3994 | unsigned long count; | |
3995 | struct location *loc; | |
3996 | }; | |
3997 | ||
3998 | static void free_loc_track(struct loc_track *t) | |
3999 | { | |
4000 | if (t->max) | |
4001 | free_pages((unsigned long)t->loc, | |
4002 | get_order(sizeof(struct location) * t->max)); | |
4003 | } | |
4004 | ||
68dff6a9 | 4005 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4006 | { |
4007 | struct location *l; | |
4008 | int order; | |
4009 | ||
88a420e4 CL |
4010 | order = get_order(sizeof(struct location) * max); |
4011 | ||
68dff6a9 | 4012 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4013 | if (!l) |
4014 | return 0; | |
4015 | ||
4016 | if (t->count) { | |
4017 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4018 | free_loc_track(t); | |
4019 | } | |
4020 | t->max = max; | |
4021 | t->loc = l; | |
4022 | return 1; | |
4023 | } | |
4024 | ||
4025 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4026 | const struct track *track) |
88a420e4 CL |
4027 | { |
4028 | long start, end, pos; | |
4029 | struct location *l; | |
ce71e27c | 4030 | unsigned long caddr; |
45edfa58 | 4031 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4032 | |
4033 | start = -1; | |
4034 | end = t->count; | |
4035 | ||
4036 | for ( ; ; ) { | |
4037 | pos = start + (end - start + 1) / 2; | |
4038 | ||
4039 | /* | |
4040 | * There is nothing at "end". If we end up there | |
4041 | * we need to add something to before end. | |
4042 | */ | |
4043 | if (pos == end) | |
4044 | break; | |
4045 | ||
4046 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4047 | if (track->addr == caddr) { |
4048 | ||
4049 | l = &t->loc[pos]; | |
4050 | l->count++; | |
4051 | if (track->when) { | |
4052 | l->sum_time += age; | |
4053 | if (age < l->min_time) | |
4054 | l->min_time = age; | |
4055 | if (age > l->max_time) | |
4056 | l->max_time = age; | |
4057 | ||
4058 | if (track->pid < l->min_pid) | |
4059 | l->min_pid = track->pid; | |
4060 | if (track->pid > l->max_pid) | |
4061 | l->max_pid = track->pid; | |
4062 | ||
174596a0 RR |
4063 | cpumask_set_cpu(track->cpu, |
4064 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4065 | } |
4066 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4067 | return 1; |
4068 | } | |
4069 | ||
45edfa58 | 4070 | if (track->addr < caddr) |
88a420e4 CL |
4071 | end = pos; |
4072 | else | |
4073 | start = pos; | |
4074 | } | |
4075 | ||
4076 | /* | |
672bba3a | 4077 | * Not found. Insert new tracking element. |
88a420e4 | 4078 | */ |
68dff6a9 | 4079 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4080 | return 0; |
4081 | ||
4082 | l = t->loc + pos; | |
4083 | if (pos < t->count) | |
4084 | memmove(l + 1, l, | |
4085 | (t->count - pos) * sizeof(struct location)); | |
4086 | t->count++; | |
4087 | l->count = 1; | |
45edfa58 CL |
4088 | l->addr = track->addr; |
4089 | l->sum_time = age; | |
4090 | l->min_time = age; | |
4091 | l->max_time = age; | |
4092 | l->min_pid = track->pid; | |
4093 | l->max_pid = track->pid; | |
174596a0 RR |
4094 | cpumask_clear(to_cpumask(l->cpus)); |
4095 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4096 | nodes_clear(l->nodes); |
4097 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4098 | return 1; |
4099 | } | |
4100 | ||
4101 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4102 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4103 | unsigned long *map) |
88a420e4 | 4104 | { |
a973e9dd | 4105 | void *addr = page_address(page); |
88a420e4 CL |
4106 | void *p; |
4107 | ||
39b26464 | 4108 | bitmap_zero(map, page->objects); |
5f80b13a | 4109 | get_map(s, page, map); |
88a420e4 | 4110 | |
224a88be | 4111 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4112 | if (!test_bit(slab_index(p, s, addr), map)) |
4113 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4114 | } |
4115 | ||
4116 | static int list_locations(struct kmem_cache *s, char *buf, | |
4117 | enum track_item alloc) | |
4118 | { | |
e374d483 | 4119 | int len = 0; |
88a420e4 | 4120 | unsigned long i; |
68dff6a9 | 4121 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4122 | int node; |
bbd7d57b ED |
4123 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4124 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4125 | |
bbd7d57b ED |
4126 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4127 | GFP_TEMPORARY)) { | |
4128 | kfree(map); | |
68dff6a9 | 4129 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4130 | } |
88a420e4 CL |
4131 | /* Push back cpu slabs */ |
4132 | flush_all(s); | |
4133 | ||
f64dc58c | 4134 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4135 | struct kmem_cache_node *n = get_node(s, node); |
4136 | unsigned long flags; | |
4137 | struct page *page; | |
4138 | ||
9e86943b | 4139 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4140 | continue; |
4141 | ||
4142 | spin_lock_irqsave(&n->list_lock, flags); | |
4143 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4144 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4145 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4146 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4147 | spin_unlock_irqrestore(&n->list_lock, flags); |
4148 | } | |
4149 | ||
4150 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4151 | struct location *l = &t.loc[i]; |
88a420e4 | 4152 | |
9c246247 | 4153 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4154 | break; |
e374d483 | 4155 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4156 | |
4157 | if (l->addr) | |
62c70bce | 4158 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4159 | else |
e374d483 | 4160 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4161 | |
4162 | if (l->sum_time != l->min_time) { | |
e374d483 | 4163 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4164 | l->min_time, |
4165 | (long)div_u64(l->sum_time, l->count), | |
4166 | l->max_time); | |
45edfa58 | 4167 | } else |
e374d483 | 4168 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4169 | l->min_time); |
4170 | ||
4171 | if (l->min_pid != l->max_pid) | |
e374d483 | 4172 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4173 | l->min_pid, l->max_pid); |
4174 | else | |
e374d483 | 4175 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4176 | l->min_pid); |
4177 | ||
174596a0 RR |
4178 | if (num_online_cpus() > 1 && |
4179 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4180 | len < PAGE_SIZE - 60) { |
4181 | len += sprintf(buf + len, " cpus="); | |
d0e0ac97 CG |
4182 | len += cpulist_scnprintf(buf + len, |
4183 | PAGE_SIZE - len - 50, | |
174596a0 | 4184 | to_cpumask(l->cpus)); |
45edfa58 CL |
4185 | } |
4186 | ||
62bc62a8 | 4187 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4188 | len < PAGE_SIZE - 60) { |
4189 | len += sprintf(buf + len, " nodes="); | |
d0e0ac97 CG |
4190 | len += nodelist_scnprintf(buf + len, |
4191 | PAGE_SIZE - len - 50, | |
4192 | l->nodes); | |
45edfa58 CL |
4193 | } |
4194 | ||
e374d483 | 4195 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4196 | } |
4197 | ||
4198 | free_loc_track(&t); | |
bbd7d57b | 4199 | kfree(map); |
88a420e4 | 4200 | if (!t.count) |
e374d483 HH |
4201 | len += sprintf(buf, "No data\n"); |
4202 | return len; | |
88a420e4 | 4203 | } |
ab4d5ed5 | 4204 | #endif |
88a420e4 | 4205 | |
a5a84755 CL |
4206 | #ifdef SLUB_RESILIENCY_TEST |
4207 | static void resiliency_test(void) | |
4208 | { | |
4209 | u8 *p; | |
4210 | ||
95a05b42 | 4211 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 CL |
4212 | |
4213 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4214 | printk(KERN_ERR "-----------------------\n"); | |
4215 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4216 | ||
4217 | p = kzalloc(16, GFP_KERNEL); | |
4218 | p[16] = 0x12; | |
4219 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4220 | " 0x12->0x%p\n\n", p + 16); | |
4221 | ||
4222 | validate_slab_cache(kmalloc_caches[4]); | |
4223 | ||
4224 | /* Hmmm... The next two are dangerous */ | |
4225 | p = kzalloc(32, GFP_KERNEL); | |
4226 | p[32 + sizeof(void *)] = 0x34; | |
4227 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4228 | " 0x34 -> -0x%p\n", p); | |
4229 | printk(KERN_ERR | |
4230 | "If allocated object is overwritten then not detectable\n\n"); | |
4231 | ||
4232 | validate_slab_cache(kmalloc_caches[5]); | |
4233 | p = kzalloc(64, GFP_KERNEL); | |
4234 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4235 | *p = 0x56; | |
4236 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4237 | p); | |
4238 | printk(KERN_ERR | |
4239 | "If allocated object is overwritten then not detectable\n\n"); | |
4240 | validate_slab_cache(kmalloc_caches[6]); | |
4241 | ||
4242 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4243 | p = kzalloc(128, GFP_KERNEL); | |
4244 | kfree(p); | |
4245 | *p = 0x78; | |
4246 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4247 | validate_slab_cache(kmalloc_caches[7]); | |
4248 | ||
4249 | p = kzalloc(256, GFP_KERNEL); | |
4250 | kfree(p); | |
4251 | p[50] = 0x9a; | |
4252 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4253 | p); | |
4254 | validate_slab_cache(kmalloc_caches[8]); | |
4255 | ||
4256 | p = kzalloc(512, GFP_KERNEL); | |
4257 | kfree(p); | |
4258 | p[512] = 0xab; | |
4259 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4260 | validate_slab_cache(kmalloc_caches[9]); | |
4261 | } | |
4262 | #else | |
4263 | #ifdef CONFIG_SYSFS | |
4264 | static void resiliency_test(void) {}; | |
4265 | #endif | |
4266 | #endif | |
4267 | ||
ab4d5ed5 | 4268 | #ifdef CONFIG_SYSFS |
81819f0f | 4269 | enum slab_stat_type { |
205ab99d CL |
4270 | SL_ALL, /* All slabs */ |
4271 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4272 | SL_CPU, /* Only slabs used for cpu caches */ | |
4273 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4274 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4275 | }; |
4276 | ||
205ab99d | 4277 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4278 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4279 | #define SO_CPU (1 << SL_CPU) | |
4280 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4281 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4282 | |
62e5c4b4 CG |
4283 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4284 | char *buf, unsigned long flags) | |
81819f0f CL |
4285 | { |
4286 | unsigned long total = 0; | |
81819f0f CL |
4287 | int node; |
4288 | int x; | |
4289 | unsigned long *nodes; | |
81819f0f | 4290 | |
e35e1a97 | 4291 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4292 | if (!nodes) |
4293 | return -ENOMEM; | |
81819f0f | 4294 | |
205ab99d CL |
4295 | if (flags & SO_CPU) { |
4296 | int cpu; | |
81819f0f | 4297 | |
205ab99d | 4298 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4299 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4300 | cpu); | |
ec3ab083 | 4301 | int node; |
49e22585 | 4302 | struct page *page; |
dfb4f096 | 4303 | |
bc6697d8 | 4304 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4305 | if (!page) |
4306 | continue; | |
205ab99d | 4307 | |
ec3ab083 CL |
4308 | node = page_to_nid(page); |
4309 | if (flags & SO_TOTAL) | |
4310 | x = page->objects; | |
4311 | else if (flags & SO_OBJECTS) | |
4312 | x = page->inuse; | |
4313 | else | |
4314 | x = 1; | |
49e22585 | 4315 | |
ec3ab083 CL |
4316 | total += x; |
4317 | nodes[node] += x; | |
4318 | ||
4319 | page = ACCESS_ONCE(c->partial); | |
49e22585 | 4320 | if (page) { |
8afb1474 LZ |
4321 | node = page_to_nid(page); |
4322 | if (flags & SO_TOTAL) | |
4323 | WARN_ON_ONCE(1); | |
4324 | else if (flags & SO_OBJECTS) | |
4325 | WARN_ON_ONCE(1); | |
4326 | else | |
4327 | x = page->pages; | |
bc6697d8 ED |
4328 | total += x; |
4329 | nodes[node] += x; | |
49e22585 | 4330 | } |
81819f0f CL |
4331 | } |
4332 | } | |
4333 | ||
04d94879 | 4334 | lock_memory_hotplug(); |
ab4d5ed5 | 4335 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4336 | if (flags & SO_ALL) { |
4337 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4338 | struct kmem_cache_node *n = get_node(s, node); | |
4339 | ||
d0e0ac97 CG |
4340 | if (flags & SO_TOTAL) |
4341 | x = atomic_long_read(&n->total_objects); | |
4342 | else if (flags & SO_OBJECTS) | |
4343 | x = atomic_long_read(&n->total_objects) - | |
4344 | count_partial(n, count_free); | |
81819f0f | 4345 | else |
205ab99d | 4346 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4347 | total += x; |
4348 | nodes[node] += x; | |
4349 | } | |
4350 | ||
ab4d5ed5 CL |
4351 | } else |
4352 | #endif | |
4353 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4354 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4355 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4356 | |
205ab99d CL |
4357 | if (flags & SO_TOTAL) |
4358 | x = count_partial(n, count_total); | |
4359 | else if (flags & SO_OBJECTS) | |
4360 | x = count_partial(n, count_inuse); | |
81819f0f | 4361 | else |
205ab99d | 4362 | x = n->nr_partial; |
81819f0f CL |
4363 | total += x; |
4364 | nodes[node] += x; | |
4365 | } | |
4366 | } | |
81819f0f CL |
4367 | x = sprintf(buf, "%lu", total); |
4368 | #ifdef CONFIG_NUMA | |
f64dc58c | 4369 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4370 | if (nodes[node]) |
4371 | x += sprintf(buf + x, " N%d=%lu", | |
4372 | node, nodes[node]); | |
4373 | #endif | |
04d94879 | 4374 | unlock_memory_hotplug(); |
81819f0f CL |
4375 | kfree(nodes); |
4376 | return x + sprintf(buf + x, "\n"); | |
4377 | } | |
4378 | ||
ab4d5ed5 | 4379 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4380 | static int any_slab_objects(struct kmem_cache *s) |
4381 | { | |
4382 | int node; | |
81819f0f | 4383 | |
dfb4f096 | 4384 | for_each_online_node(node) { |
81819f0f CL |
4385 | struct kmem_cache_node *n = get_node(s, node); |
4386 | ||
dfb4f096 CL |
4387 | if (!n) |
4388 | continue; | |
4389 | ||
4ea33e2d | 4390 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4391 | return 1; |
4392 | } | |
4393 | return 0; | |
4394 | } | |
ab4d5ed5 | 4395 | #endif |
81819f0f CL |
4396 | |
4397 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4398 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4399 | |
4400 | struct slab_attribute { | |
4401 | struct attribute attr; | |
4402 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4403 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4404 | }; | |
4405 | ||
4406 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4407 | static struct slab_attribute _name##_attr = \ |
4408 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4409 | |
4410 | #define SLAB_ATTR(_name) \ | |
4411 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4412 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4413 | |
81819f0f CL |
4414 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4415 | { | |
4416 | return sprintf(buf, "%d\n", s->size); | |
4417 | } | |
4418 | SLAB_ATTR_RO(slab_size); | |
4419 | ||
4420 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4421 | { | |
4422 | return sprintf(buf, "%d\n", s->align); | |
4423 | } | |
4424 | SLAB_ATTR_RO(align); | |
4425 | ||
4426 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4427 | { | |
3b0efdfa | 4428 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4429 | } |
4430 | SLAB_ATTR_RO(object_size); | |
4431 | ||
4432 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4433 | { | |
834f3d11 | 4434 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4435 | } |
4436 | SLAB_ATTR_RO(objs_per_slab); | |
4437 | ||
06b285dc CL |
4438 | static ssize_t order_store(struct kmem_cache *s, |
4439 | const char *buf, size_t length) | |
4440 | { | |
0121c619 CL |
4441 | unsigned long order; |
4442 | int err; | |
4443 | ||
3dbb95f7 | 4444 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4445 | if (err) |
4446 | return err; | |
06b285dc CL |
4447 | |
4448 | if (order > slub_max_order || order < slub_min_order) | |
4449 | return -EINVAL; | |
4450 | ||
4451 | calculate_sizes(s, order); | |
4452 | return length; | |
4453 | } | |
4454 | ||
81819f0f CL |
4455 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4456 | { | |
834f3d11 | 4457 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4458 | } |
06b285dc | 4459 | SLAB_ATTR(order); |
81819f0f | 4460 | |
73d342b1 DR |
4461 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4462 | { | |
4463 | return sprintf(buf, "%lu\n", s->min_partial); | |
4464 | } | |
4465 | ||
4466 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4467 | size_t length) | |
4468 | { | |
4469 | unsigned long min; | |
4470 | int err; | |
4471 | ||
3dbb95f7 | 4472 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4473 | if (err) |
4474 | return err; | |
4475 | ||
c0bdb232 | 4476 | set_min_partial(s, min); |
73d342b1 DR |
4477 | return length; |
4478 | } | |
4479 | SLAB_ATTR(min_partial); | |
4480 | ||
49e22585 CL |
4481 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4482 | { | |
4483 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4484 | } | |
4485 | ||
4486 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4487 | size_t length) | |
4488 | { | |
4489 | unsigned long objects; | |
4490 | int err; | |
4491 | ||
3dbb95f7 | 4492 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4493 | if (err) |
4494 | return err; | |
345c905d | 4495 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4496 | return -EINVAL; |
49e22585 CL |
4497 | |
4498 | s->cpu_partial = objects; | |
4499 | flush_all(s); | |
4500 | return length; | |
4501 | } | |
4502 | SLAB_ATTR(cpu_partial); | |
4503 | ||
81819f0f CL |
4504 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4505 | { | |
62c70bce JP |
4506 | if (!s->ctor) |
4507 | return 0; | |
4508 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4509 | } |
4510 | SLAB_ATTR_RO(ctor); | |
4511 | ||
81819f0f CL |
4512 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4513 | { | |
4514 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4515 | } | |
4516 | SLAB_ATTR_RO(aliases); | |
4517 | ||
81819f0f CL |
4518 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4519 | { | |
d9acf4b7 | 4520 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4521 | } |
4522 | SLAB_ATTR_RO(partial); | |
4523 | ||
4524 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4525 | { | |
d9acf4b7 | 4526 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4527 | } |
4528 | SLAB_ATTR_RO(cpu_slabs); | |
4529 | ||
4530 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4531 | { | |
205ab99d | 4532 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4533 | } |
4534 | SLAB_ATTR_RO(objects); | |
4535 | ||
205ab99d CL |
4536 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4537 | { | |
4538 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4539 | } | |
4540 | SLAB_ATTR_RO(objects_partial); | |
4541 | ||
49e22585 CL |
4542 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4543 | { | |
4544 | int objects = 0; | |
4545 | int pages = 0; | |
4546 | int cpu; | |
4547 | int len; | |
4548 | ||
4549 | for_each_online_cpu(cpu) { | |
4550 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4551 | ||
4552 | if (page) { | |
4553 | pages += page->pages; | |
4554 | objects += page->pobjects; | |
4555 | } | |
4556 | } | |
4557 | ||
4558 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4559 | ||
4560 | #ifdef CONFIG_SMP | |
4561 | for_each_online_cpu(cpu) { | |
4562 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4563 | ||
4564 | if (page && len < PAGE_SIZE - 20) | |
4565 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4566 | page->pobjects, page->pages); | |
4567 | } | |
4568 | #endif | |
4569 | return len + sprintf(buf + len, "\n"); | |
4570 | } | |
4571 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4572 | ||
a5a84755 CL |
4573 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4574 | { | |
4575 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4576 | } | |
4577 | ||
4578 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4579 | const char *buf, size_t length) | |
4580 | { | |
4581 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4582 | if (buf[0] == '1') | |
4583 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4584 | return length; | |
4585 | } | |
4586 | SLAB_ATTR(reclaim_account); | |
4587 | ||
4588 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4589 | { | |
4590 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4591 | } | |
4592 | SLAB_ATTR_RO(hwcache_align); | |
4593 | ||
4594 | #ifdef CONFIG_ZONE_DMA | |
4595 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4596 | { | |
4597 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4598 | } | |
4599 | SLAB_ATTR_RO(cache_dma); | |
4600 | #endif | |
4601 | ||
4602 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4603 | { | |
4604 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4605 | } | |
4606 | SLAB_ATTR_RO(destroy_by_rcu); | |
4607 | ||
ab9a0f19 LJ |
4608 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4609 | { | |
4610 | return sprintf(buf, "%d\n", s->reserved); | |
4611 | } | |
4612 | SLAB_ATTR_RO(reserved); | |
4613 | ||
ab4d5ed5 | 4614 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4615 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4616 | { | |
4617 | return show_slab_objects(s, buf, SO_ALL); | |
4618 | } | |
4619 | SLAB_ATTR_RO(slabs); | |
4620 | ||
205ab99d CL |
4621 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4622 | { | |
4623 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4624 | } | |
4625 | SLAB_ATTR_RO(total_objects); | |
4626 | ||
81819f0f CL |
4627 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4628 | { | |
4629 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4630 | } | |
4631 | ||
4632 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4633 | const char *buf, size_t length) | |
4634 | { | |
4635 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4636 | if (buf[0] == '1') { |
4637 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4638 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4639 | } |
81819f0f CL |
4640 | return length; |
4641 | } | |
4642 | SLAB_ATTR(sanity_checks); | |
4643 | ||
4644 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4645 | { | |
4646 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4647 | } | |
4648 | ||
4649 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4650 | size_t length) | |
4651 | { | |
4652 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4653 | if (buf[0] == '1') { |
4654 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4655 | s->flags |= SLAB_TRACE; |
b789ef51 | 4656 | } |
81819f0f CL |
4657 | return length; |
4658 | } | |
4659 | SLAB_ATTR(trace); | |
4660 | ||
81819f0f CL |
4661 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4662 | { | |
4663 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4664 | } | |
4665 | ||
4666 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4667 | const char *buf, size_t length) | |
4668 | { | |
4669 | if (any_slab_objects(s)) | |
4670 | return -EBUSY; | |
4671 | ||
4672 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4673 | if (buf[0] == '1') { |
4674 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4675 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4676 | } |
06b285dc | 4677 | calculate_sizes(s, -1); |
81819f0f CL |
4678 | return length; |
4679 | } | |
4680 | SLAB_ATTR(red_zone); | |
4681 | ||
4682 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4683 | { | |
4684 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4685 | } | |
4686 | ||
4687 | static ssize_t poison_store(struct kmem_cache *s, | |
4688 | const char *buf, size_t length) | |
4689 | { | |
4690 | if (any_slab_objects(s)) | |
4691 | return -EBUSY; | |
4692 | ||
4693 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4694 | if (buf[0] == '1') { |
4695 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4696 | s->flags |= SLAB_POISON; |
b789ef51 | 4697 | } |
06b285dc | 4698 | calculate_sizes(s, -1); |
81819f0f CL |
4699 | return length; |
4700 | } | |
4701 | SLAB_ATTR(poison); | |
4702 | ||
4703 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4704 | { | |
4705 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4706 | } | |
4707 | ||
4708 | static ssize_t store_user_store(struct kmem_cache *s, | |
4709 | const char *buf, size_t length) | |
4710 | { | |
4711 | if (any_slab_objects(s)) | |
4712 | return -EBUSY; | |
4713 | ||
4714 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4715 | if (buf[0] == '1') { |
4716 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4717 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4718 | } |
06b285dc | 4719 | calculate_sizes(s, -1); |
81819f0f CL |
4720 | return length; |
4721 | } | |
4722 | SLAB_ATTR(store_user); | |
4723 | ||
53e15af0 CL |
4724 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4725 | { | |
4726 | return 0; | |
4727 | } | |
4728 | ||
4729 | static ssize_t validate_store(struct kmem_cache *s, | |
4730 | const char *buf, size_t length) | |
4731 | { | |
434e245d CL |
4732 | int ret = -EINVAL; |
4733 | ||
4734 | if (buf[0] == '1') { | |
4735 | ret = validate_slab_cache(s); | |
4736 | if (ret >= 0) | |
4737 | ret = length; | |
4738 | } | |
4739 | return ret; | |
53e15af0 CL |
4740 | } |
4741 | SLAB_ATTR(validate); | |
a5a84755 CL |
4742 | |
4743 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4744 | { | |
4745 | if (!(s->flags & SLAB_STORE_USER)) | |
4746 | return -ENOSYS; | |
4747 | return list_locations(s, buf, TRACK_ALLOC); | |
4748 | } | |
4749 | SLAB_ATTR_RO(alloc_calls); | |
4750 | ||
4751 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4752 | { | |
4753 | if (!(s->flags & SLAB_STORE_USER)) | |
4754 | return -ENOSYS; | |
4755 | return list_locations(s, buf, TRACK_FREE); | |
4756 | } | |
4757 | SLAB_ATTR_RO(free_calls); | |
4758 | #endif /* CONFIG_SLUB_DEBUG */ | |
4759 | ||
4760 | #ifdef CONFIG_FAILSLAB | |
4761 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4762 | { | |
4763 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4764 | } | |
4765 | ||
4766 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4767 | size_t length) | |
4768 | { | |
4769 | s->flags &= ~SLAB_FAILSLAB; | |
4770 | if (buf[0] == '1') | |
4771 | s->flags |= SLAB_FAILSLAB; | |
4772 | return length; | |
4773 | } | |
4774 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4775 | #endif |
53e15af0 | 4776 | |
2086d26a CL |
4777 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4778 | { | |
4779 | return 0; | |
4780 | } | |
4781 | ||
4782 | static ssize_t shrink_store(struct kmem_cache *s, | |
4783 | const char *buf, size_t length) | |
4784 | { | |
4785 | if (buf[0] == '1') { | |
4786 | int rc = kmem_cache_shrink(s); | |
4787 | ||
4788 | if (rc) | |
4789 | return rc; | |
4790 | } else | |
4791 | return -EINVAL; | |
4792 | return length; | |
4793 | } | |
4794 | SLAB_ATTR(shrink); | |
4795 | ||
81819f0f | 4796 | #ifdef CONFIG_NUMA |
9824601e | 4797 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4798 | { |
9824601e | 4799 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4800 | } |
4801 | ||
9824601e | 4802 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4803 | const char *buf, size_t length) |
4804 | { | |
0121c619 CL |
4805 | unsigned long ratio; |
4806 | int err; | |
4807 | ||
3dbb95f7 | 4808 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
4809 | if (err) |
4810 | return err; | |
4811 | ||
e2cb96b7 | 4812 | if (ratio <= 100) |
0121c619 | 4813 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4814 | |
81819f0f CL |
4815 | return length; |
4816 | } | |
9824601e | 4817 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4818 | #endif |
4819 | ||
8ff12cfc | 4820 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4821 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4822 | { | |
4823 | unsigned long sum = 0; | |
4824 | int cpu; | |
4825 | int len; | |
4826 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4827 | ||
4828 | if (!data) | |
4829 | return -ENOMEM; | |
4830 | ||
4831 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4832 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4833 | |
4834 | data[cpu] = x; | |
4835 | sum += x; | |
4836 | } | |
4837 | ||
4838 | len = sprintf(buf, "%lu", sum); | |
4839 | ||
50ef37b9 | 4840 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4841 | for_each_online_cpu(cpu) { |
4842 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4843 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4844 | } |
50ef37b9 | 4845 | #endif |
8ff12cfc CL |
4846 | kfree(data); |
4847 | return len + sprintf(buf + len, "\n"); | |
4848 | } | |
4849 | ||
78eb00cc DR |
4850 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4851 | { | |
4852 | int cpu; | |
4853 | ||
4854 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4855 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4856 | } |
4857 | ||
8ff12cfc CL |
4858 | #define STAT_ATTR(si, text) \ |
4859 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4860 | { \ | |
4861 | return show_stat(s, buf, si); \ | |
4862 | } \ | |
78eb00cc DR |
4863 | static ssize_t text##_store(struct kmem_cache *s, \ |
4864 | const char *buf, size_t length) \ | |
4865 | { \ | |
4866 | if (buf[0] != '0') \ | |
4867 | return -EINVAL; \ | |
4868 | clear_stat(s, si); \ | |
4869 | return length; \ | |
4870 | } \ | |
4871 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4872 | |
4873 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4874 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4875 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4876 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4877 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4878 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4879 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4880 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4881 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4882 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4883 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4884 | STAT_ATTR(FREE_SLAB, free_slab); |
4885 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4886 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4887 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4888 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4889 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4890 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4891 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4892 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4893 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4894 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
4895 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
4896 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
4897 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
4898 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
4899 | #endif |
4900 | ||
06428780 | 4901 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4902 | &slab_size_attr.attr, |
4903 | &object_size_attr.attr, | |
4904 | &objs_per_slab_attr.attr, | |
4905 | &order_attr.attr, | |
73d342b1 | 4906 | &min_partial_attr.attr, |
49e22585 | 4907 | &cpu_partial_attr.attr, |
81819f0f | 4908 | &objects_attr.attr, |
205ab99d | 4909 | &objects_partial_attr.attr, |
81819f0f CL |
4910 | &partial_attr.attr, |
4911 | &cpu_slabs_attr.attr, | |
4912 | &ctor_attr.attr, | |
81819f0f CL |
4913 | &aliases_attr.attr, |
4914 | &align_attr.attr, | |
81819f0f CL |
4915 | &hwcache_align_attr.attr, |
4916 | &reclaim_account_attr.attr, | |
4917 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4918 | &shrink_attr.attr, |
ab9a0f19 | 4919 | &reserved_attr.attr, |
49e22585 | 4920 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 4921 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4922 | &total_objects_attr.attr, |
4923 | &slabs_attr.attr, | |
4924 | &sanity_checks_attr.attr, | |
4925 | &trace_attr.attr, | |
81819f0f CL |
4926 | &red_zone_attr.attr, |
4927 | &poison_attr.attr, | |
4928 | &store_user_attr.attr, | |
53e15af0 | 4929 | &validate_attr.attr, |
88a420e4 CL |
4930 | &alloc_calls_attr.attr, |
4931 | &free_calls_attr.attr, | |
ab4d5ed5 | 4932 | #endif |
81819f0f CL |
4933 | #ifdef CONFIG_ZONE_DMA |
4934 | &cache_dma_attr.attr, | |
4935 | #endif | |
4936 | #ifdef CONFIG_NUMA | |
9824601e | 4937 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4938 | #endif |
4939 | #ifdef CONFIG_SLUB_STATS | |
4940 | &alloc_fastpath_attr.attr, | |
4941 | &alloc_slowpath_attr.attr, | |
4942 | &free_fastpath_attr.attr, | |
4943 | &free_slowpath_attr.attr, | |
4944 | &free_frozen_attr.attr, | |
4945 | &free_add_partial_attr.attr, | |
4946 | &free_remove_partial_attr.attr, | |
4947 | &alloc_from_partial_attr.attr, | |
4948 | &alloc_slab_attr.attr, | |
4949 | &alloc_refill_attr.attr, | |
e36a2652 | 4950 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
4951 | &free_slab_attr.attr, |
4952 | &cpuslab_flush_attr.attr, | |
4953 | &deactivate_full_attr.attr, | |
4954 | &deactivate_empty_attr.attr, | |
4955 | &deactivate_to_head_attr.attr, | |
4956 | &deactivate_to_tail_attr.attr, | |
4957 | &deactivate_remote_frees_attr.attr, | |
03e404af | 4958 | &deactivate_bypass_attr.attr, |
65c3376a | 4959 | &order_fallback_attr.attr, |
b789ef51 CL |
4960 | &cmpxchg_double_fail_attr.attr, |
4961 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
4962 | &cpu_partial_alloc_attr.attr, |
4963 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
4964 | &cpu_partial_node_attr.attr, |
4965 | &cpu_partial_drain_attr.attr, | |
81819f0f | 4966 | #endif |
4c13dd3b DM |
4967 | #ifdef CONFIG_FAILSLAB |
4968 | &failslab_attr.attr, | |
4969 | #endif | |
4970 | ||
81819f0f CL |
4971 | NULL |
4972 | }; | |
4973 | ||
4974 | static struct attribute_group slab_attr_group = { | |
4975 | .attrs = slab_attrs, | |
4976 | }; | |
4977 | ||
4978 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4979 | struct attribute *attr, | |
4980 | char *buf) | |
4981 | { | |
4982 | struct slab_attribute *attribute; | |
4983 | struct kmem_cache *s; | |
4984 | int err; | |
4985 | ||
4986 | attribute = to_slab_attr(attr); | |
4987 | s = to_slab(kobj); | |
4988 | ||
4989 | if (!attribute->show) | |
4990 | return -EIO; | |
4991 | ||
4992 | err = attribute->show(s, buf); | |
4993 | ||
4994 | return err; | |
4995 | } | |
4996 | ||
4997 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4998 | struct attribute *attr, | |
4999 | const char *buf, size_t len) | |
5000 | { | |
5001 | struct slab_attribute *attribute; | |
5002 | struct kmem_cache *s; | |
5003 | int err; | |
5004 | ||
5005 | attribute = to_slab_attr(attr); | |
5006 | s = to_slab(kobj); | |
5007 | ||
5008 | if (!attribute->store) | |
5009 | return -EIO; | |
5010 | ||
5011 | err = attribute->store(s, buf, len); | |
107dab5c GC |
5012 | #ifdef CONFIG_MEMCG_KMEM |
5013 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { | |
5014 | int i; | |
81819f0f | 5015 | |
107dab5c GC |
5016 | mutex_lock(&slab_mutex); |
5017 | if (s->max_attr_size < len) | |
5018 | s->max_attr_size = len; | |
5019 | ||
ebe945c2 GC |
5020 | /* |
5021 | * This is a best effort propagation, so this function's return | |
5022 | * value will be determined by the parent cache only. This is | |
5023 | * basically because not all attributes will have a well | |
5024 | * defined semantics for rollbacks - most of the actions will | |
5025 | * have permanent effects. | |
5026 | * | |
5027 | * Returning the error value of any of the children that fail | |
5028 | * is not 100 % defined, in the sense that users seeing the | |
5029 | * error code won't be able to know anything about the state of | |
5030 | * the cache. | |
5031 | * | |
5032 | * Only returning the error code for the parent cache at least | |
5033 | * has well defined semantics. The cache being written to | |
5034 | * directly either failed or succeeded, in which case we loop | |
5035 | * through the descendants with best-effort propagation. | |
5036 | */ | |
107dab5c | 5037 | for_each_memcg_cache_index(i) { |
2ade4de8 | 5038 | struct kmem_cache *c = cache_from_memcg_idx(s, i); |
107dab5c GC |
5039 | if (c) |
5040 | attribute->store(c, buf, len); | |
5041 | } | |
5042 | mutex_unlock(&slab_mutex); | |
5043 | } | |
5044 | #endif | |
81819f0f CL |
5045 | return err; |
5046 | } | |
5047 | ||
107dab5c GC |
5048 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5049 | { | |
5050 | #ifdef CONFIG_MEMCG_KMEM | |
5051 | int i; | |
5052 | char *buffer = NULL; | |
5053 | ||
5054 | if (!is_root_cache(s)) | |
5055 | return; | |
5056 | ||
5057 | /* | |
5058 | * This mean this cache had no attribute written. Therefore, no point | |
5059 | * in copying default values around | |
5060 | */ | |
5061 | if (!s->max_attr_size) | |
5062 | return; | |
5063 | ||
5064 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5065 | char mbuf[64]; | |
5066 | char *buf; | |
5067 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5068 | ||
5069 | if (!attr || !attr->store || !attr->show) | |
5070 | continue; | |
5071 | ||
5072 | /* | |
5073 | * It is really bad that we have to allocate here, so we will | |
5074 | * do it only as a fallback. If we actually allocate, though, | |
5075 | * we can just use the allocated buffer until the end. | |
5076 | * | |
5077 | * Most of the slub attributes will tend to be very small in | |
5078 | * size, but sysfs allows buffers up to a page, so they can | |
5079 | * theoretically happen. | |
5080 | */ | |
5081 | if (buffer) | |
5082 | buf = buffer; | |
5083 | else if (s->max_attr_size < ARRAY_SIZE(mbuf)) | |
5084 | buf = mbuf; | |
5085 | else { | |
5086 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5087 | if (WARN_ON(!buffer)) | |
5088 | continue; | |
5089 | buf = buffer; | |
5090 | } | |
5091 | ||
5092 | attr->show(s->memcg_params->root_cache, buf); | |
5093 | attr->store(s, buf, strlen(buf)); | |
5094 | } | |
5095 | ||
5096 | if (buffer) | |
5097 | free_page((unsigned long)buffer); | |
5098 | #endif | |
5099 | } | |
5100 | ||
52cf25d0 | 5101 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5102 | .show = slab_attr_show, |
5103 | .store = slab_attr_store, | |
5104 | }; | |
5105 | ||
5106 | static struct kobj_type slab_ktype = { | |
5107 | .sysfs_ops = &slab_sysfs_ops, | |
5108 | }; | |
5109 | ||
5110 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5111 | { | |
5112 | struct kobj_type *ktype = get_ktype(kobj); | |
5113 | ||
5114 | if (ktype == &slab_ktype) | |
5115 | return 1; | |
5116 | return 0; | |
5117 | } | |
5118 | ||
9cd43611 | 5119 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5120 | .filter = uevent_filter, |
5121 | }; | |
5122 | ||
27c3a314 | 5123 | static struct kset *slab_kset; |
81819f0f CL |
5124 | |
5125 | #define ID_STR_LENGTH 64 | |
5126 | ||
5127 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5128 | * |
5129 | * Format :[flags-]size | |
81819f0f CL |
5130 | */ |
5131 | static char *create_unique_id(struct kmem_cache *s) | |
5132 | { | |
5133 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5134 | char *p = name; | |
5135 | ||
5136 | BUG_ON(!name); | |
5137 | ||
5138 | *p++ = ':'; | |
5139 | /* | |
5140 | * First flags affecting slabcache operations. We will only | |
5141 | * get here for aliasable slabs so we do not need to support | |
5142 | * too many flags. The flags here must cover all flags that | |
5143 | * are matched during merging to guarantee that the id is | |
5144 | * unique. | |
5145 | */ | |
5146 | if (s->flags & SLAB_CACHE_DMA) | |
5147 | *p++ = 'd'; | |
5148 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5149 | *p++ = 'a'; | |
5150 | if (s->flags & SLAB_DEBUG_FREE) | |
5151 | *p++ = 'F'; | |
5a896d9e VN |
5152 | if (!(s->flags & SLAB_NOTRACK)) |
5153 | *p++ = 't'; | |
81819f0f CL |
5154 | if (p != name + 1) |
5155 | *p++ = '-'; | |
5156 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 GC |
5157 | |
5158 | #ifdef CONFIG_MEMCG_KMEM | |
5159 | if (!is_root_cache(s)) | |
d0e0ac97 CG |
5160 | p += sprintf(p, "-%08d", |
5161 | memcg_cache_id(s->memcg_params->memcg)); | |
2633d7a0 GC |
5162 | #endif |
5163 | ||
81819f0f CL |
5164 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5165 | return name; | |
5166 | } | |
5167 | ||
5168 | static int sysfs_slab_add(struct kmem_cache *s) | |
5169 | { | |
5170 | int err; | |
5171 | const char *name; | |
45530c44 | 5172 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5173 | |
81819f0f CL |
5174 | if (unmergeable) { |
5175 | /* | |
5176 | * Slabcache can never be merged so we can use the name proper. | |
5177 | * This is typically the case for debug situations. In that | |
5178 | * case we can catch duplicate names easily. | |
5179 | */ | |
27c3a314 | 5180 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5181 | name = s->name; |
5182 | } else { | |
5183 | /* | |
5184 | * Create a unique name for the slab as a target | |
5185 | * for the symlinks. | |
5186 | */ | |
5187 | name = create_unique_id(s); | |
5188 | } | |
5189 | ||
27c3a314 | 5190 | s->kobj.kset = slab_kset; |
26e4f205 | 5191 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
1eada11c GKH |
5192 | if (err) { |
5193 | kobject_put(&s->kobj); | |
81819f0f | 5194 | return err; |
1eada11c | 5195 | } |
81819f0f CL |
5196 | |
5197 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
5198 | if (err) { |
5199 | kobject_del(&s->kobj); | |
5200 | kobject_put(&s->kobj); | |
81819f0f | 5201 | return err; |
5788d8ad | 5202 | } |
81819f0f CL |
5203 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5204 | if (!unmergeable) { | |
5205 | /* Setup first alias */ | |
5206 | sysfs_slab_alias(s, s->name); | |
5207 | kfree(name); | |
5208 | } | |
5209 | return 0; | |
5210 | } | |
5211 | ||
5212 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5213 | { | |
97d06609 | 5214 | if (slab_state < FULL) |
2bce6485 CL |
5215 | /* |
5216 | * Sysfs has not been setup yet so no need to remove the | |
5217 | * cache from sysfs. | |
5218 | */ | |
5219 | return; | |
5220 | ||
81819f0f CL |
5221 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5222 | kobject_del(&s->kobj); | |
151c602f | 5223 | kobject_put(&s->kobj); |
81819f0f CL |
5224 | } |
5225 | ||
5226 | /* | |
5227 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5228 | * available lest we lose that information. |
81819f0f CL |
5229 | */ |
5230 | struct saved_alias { | |
5231 | struct kmem_cache *s; | |
5232 | const char *name; | |
5233 | struct saved_alias *next; | |
5234 | }; | |
5235 | ||
5af328a5 | 5236 | static struct saved_alias *alias_list; |
81819f0f CL |
5237 | |
5238 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5239 | { | |
5240 | struct saved_alias *al; | |
5241 | ||
97d06609 | 5242 | if (slab_state == FULL) { |
81819f0f CL |
5243 | /* |
5244 | * If we have a leftover link then remove it. | |
5245 | */ | |
27c3a314 GKH |
5246 | sysfs_remove_link(&slab_kset->kobj, name); |
5247 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5248 | } |
5249 | ||
5250 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5251 | if (!al) | |
5252 | return -ENOMEM; | |
5253 | ||
5254 | al->s = s; | |
5255 | al->name = name; | |
5256 | al->next = alias_list; | |
5257 | alias_list = al; | |
5258 | return 0; | |
5259 | } | |
5260 | ||
5261 | static int __init slab_sysfs_init(void) | |
5262 | { | |
5b95a4ac | 5263 | struct kmem_cache *s; |
81819f0f CL |
5264 | int err; |
5265 | ||
18004c5d | 5266 | mutex_lock(&slab_mutex); |
2bce6485 | 5267 | |
0ff21e46 | 5268 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5269 | if (!slab_kset) { |
18004c5d | 5270 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5271 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
5272 | return -ENOSYS; | |
5273 | } | |
5274 | ||
97d06609 | 5275 | slab_state = FULL; |
26a7bd03 | 5276 | |
5b95a4ac | 5277 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5278 | err = sysfs_slab_add(s); |
5d540fb7 CL |
5279 | if (err) |
5280 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
5281 | " to sysfs\n", s->name); | |
26a7bd03 | 5282 | } |
81819f0f CL |
5283 | |
5284 | while (alias_list) { | |
5285 | struct saved_alias *al = alias_list; | |
5286 | ||
5287 | alias_list = alias_list->next; | |
5288 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5289 | if (err) |
5290 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
068ce415 | 5291 | " %s to sysfs\n", al->name); |
81819f0f CL |
5292 | kfree(al); |
5293 | } | |
5294 | ||
18004c5d | 5295 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5296 | resiliency_test(); |
5297 | return 0; | |
5298 | } | |
5299 | ||
5300 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5301 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5302 | |
5303 | /* | |
5304 | * The /proc/slabinfo ABI | |
5305 | */ | |
158a9624 | 5306 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5307 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5308 | { |
57ed3eda | 5309 | unsigned long nr_slabs = 0; |
205ab99d CL |
5310 | unsigned long nr_objs = 0; |
5311 | unsigned long nr_free = 0; | |
57ed3eda PE |
5312 | int node; |
5313 | ||
57ed3eda PE |
5314 | for_each_online_node(node) { |
5315 | struct kmem_cache_node *n = get_node(s, node); | |
5316 | ||
5317 | if (!n) | |
5318 | continue; | |
5319 | ||
c17fd13e WL |
5320 | nr_slabs += node_nr_slabs(n); |
5321 | nr_objs += node_nr_objs(n); | |
205ab99d | 5322 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5323 | } |
5324 | ||
0d7561c6 GC |
5325 | sinfo->active_objs = nr_objs - nr_free; |
5326 | sinfo->num_objs = nr_objs; | |
5327 | sinfo->active_slabs = nr_slabs; | |
5328 | sinfo->num_slabs = nr_slabs; | |
5329 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5330 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5331 | } |
5332 | ||
0d7561c6 | 5333 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5334 | { |
7b3c3a50 AD |
5335 | } |
5336 | ||
b7454ad3 GC |
5337 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5338 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5339 | { |
b7454ad3 | 5340 | return -EIO; |
7b3c3a50 | 5341 | } |
158a9624 | 5342 | #endif /* CONFIG_SLABINFO */ |