Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
81819f0f CL |
2 | /* |
3 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
4 | * objects in per cpu and per node lists. | |
5 | * | |
881db7fb CL |
6 | * The allocator synchronizes using per slab locks or atomic operatios |
7 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 8 | * |
cde53535 | 9 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 10 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
11 | */ |
12 | ||
13 | #include <linux/mm.h> | |
1eb5ac64 | 14 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
15 | #include <linux/module.h> |
16 | #include <linux/bit_spinlock.h> | |
17 | #include <linux/interrupt.h> | |
18 | #include <linux/bitops.h> | |
19 | #include <linux/slab.h> | |
97d06609 | 20 | #include "slab.h" |
7b3c3a50 | 21 | #include <linux/proc_fs.h> |
81819f0f | 22 | #include <linux/seq_file.h> |
a79316c6 | 23 | #include <linux/kasan.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> |
2482ddec | 36 | #include <linux/random.h> |
81819f0f | 37 | |
4a92379b RK |
38 | #include <trace/events/kmem.h> |
39 | ||
072bb0aa MG |
40 | #include "internal.h" |
41 | ||
81819f0f CL |
42 | /* |
43 | * Lock order: | |
18004c5d | 44 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
45 | * 2. node->list_lock |
46 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 47 | * |
18004c5d | 48 | * slab_mutex |
881db7fb | 49 | * |
18004c5d | 50 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
51 | * and to synchronize major metadata changes to slab cache structures. |
52 | * | |
53 | * The slab_lock is only used for debugging and on arches that do not | |
b7ccc7f8 | 54 | * have the ability to do a cmpxchg_double. It only protects: |
881db7fb | 55 | * A. page->freelist -> List of object free in a page |
b7ccc7f8 MW |
56 | * B. page->inuse -> Number of objects in use |
57 | * C. page->objects -> Number of objects in page | |
58 | * D. page->frozen -> frozen state | |
881db7fb CL |
59 | * |
60 | * If a slab is frozen then it is exempt from list management. It is not | |
61 | * on any list. The processor that froze the slab is the one who can | |
62 | * perform list operations on the page. Other processors may put objects | |
63 | * onto the freelist but the processor that froze the slab is the only | |
64 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
65 | * |
66 | * The list_lock protects the partial and full list on each node and | |
67 | * the partial slab counter. If taken then no new slabs may be added or | |
68 | * removed from the lists nor make the number of partial slabs be modified. | |
69 | * (Note that the total number of slabs is an atomic value that may be | |
70 | * modified without taking the list lock). | |
71 | * | |
72 | * The list_lock is a centralized lock and thus we avoid taking it as | |
73 | * much as possible. As long as SLUB does not have to handle partial | |
74 | * slabs, operations can continue without any centralized lock. F.e. | |
75 | * allocating a long series of objects that fill up slabs does not require | |
76 | * the list lock. | |
81819f0f CL |
77 | * Interrupts are disabled during allocation and deallocation in order to |
78 | * make the slab allocator safe to use in the context of an irq. In addition | |
79 | * interrupts are disabled to ensure that the processor does not change | |
80 | * while handling per_cpu slabs, due to kernel preemption. | |
81 | * | |
82 | * SLUB assigns one slab for allocation to each processor. | |
83 | * Allocations only occur from these slabs called cpu slabs. | |
84 | * | |
672bba3a CL |
85 | * Slabs with free elements are kept on a partial list and during regular |
86 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 87 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
88 | * We track full slabs for debugging purposes though because otherwise we |
89 | * cannot scan all objects. | |
81819f0f CL |
90 | * |
91 | * Slabs are freed when they become empty. Teardown and setup is | |
92 | * minimal so we rely on the page allocators per cpu caches for | |
93 | * fast frees and allocs. | |
94 | * | |
95 | * Overloading of page flags that are otherwise used for LRU management. | |
96 | * | |
4b6f0750 CL |
97 | * PageActive The slab is frozen and exempt from list processing. |
98 | * This means that the slab is dedicated to a purpose | |
99 | * such as satisfying allocations for a specific | |
100 | * processor. Objects may be freed in the slab while | |
101 | * it is frozen but slab_free will then skip the usual | |
102 | * list operations. It is up to the processor holding | |
103 | * the slab to integrate the slab into the slab lists | |
104 | * when the slab is no longer needed. | |
105 | * | |
106 | * One use of this flag is to mark slabs that are | |
107 | * used for allocations. Then such a slab becomes a cpu | |
108 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 109 | * freelist that allows lockless access to |
894b8788 CL |
110 | * free objects in addition to the regular freelist |
111 | * that requires the slab lock. | |
81819f0f CL |
112 | * |
113 | * PageError Slab requires special handling due to debug | |
114 | * options set. This moves slab handling out of | |
894b8788 | 115 | * the fast path and disables lockless freelists. |
81819f0f CL |
116 | */ |
117 | ||
af537b0a CL |
118 | static inline int kmem_cache_debug(struct kmem_cache *s) |
119 | { | |
5577bd8a | 120 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 121 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 122 | #else |
af537b0a | 123 | return 0; |
5577bd8a | 124 | #endif |
af537b0a | 125 | } |
5577bd8a | 126 | |
117d54df | 127 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be JK |
128 | { |
129 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) | |
130 | p += s->red_left_pad; | |
131 | ||
132 | return p; | |
133 | } | |
134 | ||
345c905d JK |
135 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
136 | { | |
137 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
138 | return !kmem_cache_debug(s); | |
139 | #else | |
140 | return false; | |
141 | #endif | |
142 | } | |
143 | ||
81819f0f CL |
144 | /* |
145 | * Issues still to be resolved: | |
146 | * | |
81819f0f CL |
147 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
148 | * | |
81819f0f CL |
149 | * - Variable sizing of the per node arrays |
150 | */ | |
151 | ||
152 | /* Enable to test recovery from slab corruption on boot */ | |
153 | #undef SLUB_RESILIENCY_TEST | |
154 | ||
b789ef51 CL |
155 | /* Enable to log cmpxchg failures */ |
156 | #undef SLUB_DEBUG_CMPXCHG | |
157 | ||
2086d26a CL |
158 | /* |
159 | * Mininum number of partial slabs. These will be left on the partial | |
160 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
161 | */ | |
76be8950 | 162 | #define MIN_PARTIAL 5 |
e95eed57 | 163 | |
2086d26a CL |
164 | /* |
165 | * Maximum number of desirable partial slabs. | |
166 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 167 | * sort the partial list by the number of objects in use. |
2086d26a CL |
168 | */ |
169 | #define MAX_PARTIAL 10 | |
170 | ||
becfda68 | 171 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 172 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 173 | |
149daaf3 LA |
174 | /* |
175 | * These debug flags cannot use CMPXCHG because there might be consistency | |
176 | * issues when checking or reading debug information | |
177 | */ | |
178 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
179 | SLAB_TRACE) | |
180 | ||
181 | ||
fa5ec8a1 | 182 | /* |
3de47213 DR |
183 | * Debugging flags that require metadata to be stored in the slab. These get |
184 | * disabled when slub_debug=O is used and a cache's min order increases with | |
185 | * metadata. | |
fa5ec8a1 | 186 | */ |
3de47213 | 187 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 188 | |
210b5c06 CG |
189 | #define OO_SHIFT 16 |
190 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 191 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 192 | |
81819f0f | 193 | /* Internal SLUB flags */ |
d50112ed | 194 | /* Poison object */ |
4fd0b46e | 195 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) |
d50112ed | 196 | /* Use cmpxchg_double */ |
4fd0b46e | 197 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) |
81819f0f | 198 | |
02cbc874 CL |
199 | /* |
200 | * Tracking user of a slab. | |
201 | */ | |
d6543e39 | 202 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 203 | struct track { |
ce71e27c | 204 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
205 | #ifdef CONFIG_STACKTRACE |
206 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
207 | #endif | |
02cbc874 CL |
208 | int cpu; /* Was running on cpu */ |
209 | int pid; /* Pid context */ | |
210 | unsigned long when; /* When did the operation occur */ | |
211 | }; | |
212 | ||
213 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
214 | ||
ab4d5ed5 | 215 | #ifdef CONFIG_SYSFS |
81819f0f CL |
216 | static int sysfs_slab_add(struct kmem_cache *); |
217 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 218 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
bf5eb3de | 219 | static void sysfs_slab_remove(struct kmem_cache *s); |
81819f0f | 220 | #else |
0c710013 CL |
221 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
222 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
223 | { return 0; } | |
107dab5c | 224 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
bf5eb3de | 225 | static inline void sysfs_slab_remove(struct kmem_cache *s) { } |
81819f0f CL |
226 | #endif |
227 | ||
4fdccdfb | 228 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
229 | { |
230 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
231 | /* |
232 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
233 | * avoid this_cpu_add()'s irq-disable overhead. | |
234 | */ | |
235 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
236 | #endif |
237 | } | |
238 | ||
81819f0f CL |
239 | /******************************************************************** |
240 | * Core slab cache functions | |
241 | *******************************************************************/ | |
242 | ||
2482ddec KC |
243 | /* |
244 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
245 | * with an XOR of the address where the pointer is held and a per-cache | |
246 | * random number. | |
247 | */ | |
248 | static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr, | |
249 | unsigned long ptr_addr) | |
250 | { | |
251 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
d36a63a9 AK |
252 | /* |
253 | * When CONFIG_KASAN_SW_TAGS is enabled, ptr_addr might be tagged. | |
254 | * Normally, this doesn't cause any issues, as both set_freepointer() | |
255 | * and get_freepointer() are called with a pointer with the same tag. | |
256 | * However, there are some issues with CONFIG_SLUB_DEBUG code. For | |
257 | * example, when __free_slub() iterates over objects in a cache, it | |
258 | * passes untagged pointers to check_object(). check_object() in turns | |
259 | * calls get_freepointer() with an untagged pointer, which causes the | |
260 | * freepointer to be restored incorrectly. | |
261 | */ | |
262 | return (void *)((unsigned long)ptr ^ s->random ^ | |
263 | (unsigned long)kasan_reset_tag((void *)ptr_addr)); | |
2482ddec KC |
264 | #else |
265 | return ptr; | |
266 | #endif | |
267 | } | |
268 | ||
269 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
270 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
271 | void *ptr_addr) | |
272 | { | |
273 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
274 | (unsigned long)ptr_addr); | |
275 | } | |
276 | ||
7656c72b CL |
277 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
278 | { | |
2482ddec | 279 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
280 | } |
281 | ||
0ad9500e ED |
282 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
283 | { | |
0882ff91 | 284 | prefetch(object + s->offset); |
0ad9500e ED |
285 | } |
286 | ||
1393d9a1 CL |
287 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
288 | { | |
2482ddec | 289 | unsigned long freepointer_addr; |
1393d9a1 CL |
290 | void *p; |
291 | ||
922d566c JK |
292 | if (!debug_pagealloc_enabled()) |
293 | return get_freepointer(s, object); | |
294 | ||
2482ddec KC |
295 | freepointer_addr = (unsigned long)object + s->offset; |
296 | probe_kernel_read(&p, (void **)freepointer_addr, sizeof(p)); | |
297 | return freelist_ptr(s, p, freepointer_addr); | |
1393d9a1 CL |
298 | } |
299 | ||
7656c72b CL |
300 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
301 | { | |
2482ddec KC |
302 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
303 | ||
ce6fa91b AP |
304 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
305 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
306 | #endif | |
307 | ||
2482ddec | 308 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
309 | } |
310 | ||
311 | /* Loop over all objects in a slab */ | |
224a88be | 312 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
313 | for (__p = fixup_red_left(__s, __addr); \ |
314 | __p < (__addr) + (__objects) * (__s)->size; \ | |
315 | __p += (__s)->size) | |
7656c72b | 316 | |
7656c72b | 317 | /* Determine object index from a given position */ |
284b50dd | 318 | static inline unsigned int slab_index(void *p, struct kmem_cache *s, void *addr) |
7656c72b | 319 | { |
6373dca1 | 320 | return (kasan_reset_tag(p) - addr) / s->size; |
7656c72b CL |
321 | } |
322 | ||
9736d2a9 | 323 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 324 | { |
9736d2a9 | 325 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
326 | } |
327 | ||
19af27af | 328 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 329 | unsigned int size) |
834f3d11 CL |
330 | { |
331 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 332 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
333 | }; |
334 | ||
335 | return x; | |
336 | } | |
337 | ||
19af27af | 338 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 339 | { |
210b5c06 | 340 | return x.x >> OO_SHIFT; |
834f3d11 CL |
341 | } |
342 | ||
19af27af | 343 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 344 | { |
210b5c06 | 345 | return x.x & OO_MASK; |
834f3d11 CL |
346 | } |
347 | ||
881db7fb CL |
348 | /* |
349 | * Per slab locking using the pagelock | |
350 | */ | |
351 | static __always_inline void slab_lock(struct page *page) | |
352 | { | |
48c935ad | 353 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
354 | bit_spin_lock(PG_locked, &page->flags); |
355 | } | |
356 | ||
357 | static __always_inline void slab_unlock(struct page *page) | |
358 | { | |
48c935ad | 359 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
360 | __bit_spin_unlock(PG_locked, &page->flags); |
361 | } | |
362 | ||
1d07171c CL |
363 | /* Interrupts must be disabled (for the fallback code to work right) */ |
364 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
365 | void *freelist_old, unsigned long counters_old, | |
366 | void *freelist_new, unsigned long counters_new, | |
367 | const char *n) | |
368 | { | |
369 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
370 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
371 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 372 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 373 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
374 | freelist_old, counters_old, |
375 | freelist_new, counters_new)) | |
6f6528a1 | 376 | return true; |
1d07171c CL |
377 | } else |
378 | #endif | |
379 | { | |
380 | slab_lock(page); | |
d0e0ac97 CG |
381 | if (page->freelist == freelist_old && |
382 | page->counters == counters_old) { | |
1d07171c | 383 | page->freelist = freelist_new; |
7d27a04b | 384 | page->counters = counters_new; |
1d07171c | 385 | slab_unlock(page); |
6f6528a1 | 386 | return true; |
1d07171c CL |
387 | } |
388 | slab_unlock(page); | |
389 | } | |
390 | ||
391 | cpu_relax(); | |
392 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
393 | ||
394 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 395 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
396 | #endif |
397 | ||
6f6528a1 | 398 | return false; |
1d07171c CL |
399 | } |
400 | ||
b789ef51 CL |
401 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
402 | void *freelist_old, unsigned long counters_old, | |
403 | void *freelist_new, unsigned long counters_new, | |
404 | const char *n) | |
405 | { | |
2565409f HC |
406 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
407 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 408 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 409 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
410 | freelist_old, counters_old, |
411 | freelist_new, counters_new)) | |
6f6528a1 | 412 | return true; |
b789ef51 CL |
413 | } else |
414 | #endif | |
415 | { | |
1d07171c CL |
416 | unsigned long flags; |
417 | ||
418 | local_irq_save(flags); | |
881db7fb | 419 | slab_lock(page); |
d0e0ac97 CG |
420 | if (page->freelist == freelist_old && |
421 | page->counters == counters_old) { | |
b789ef51 | 422 | page->freelist = freelist_new; |
7d27a04b | 423 | page->counters = counters_new; |
881db7fb | 424 | slab_unlock(page); |
1d07171c | 425 | local_irq_restore(flags); |
6f6528a1 | 426 | return true; |
b789ef51 | 427 | } |
881db7fb | 428 | slab_unlock(page); |
1d07171c | 429 | local_irq_restore(flags); |
b789ef51 CL |
430 | } |
431 | ||
432 | cpu_relax(); | |
433 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
434 | ||
435 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 436 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
437 | #endif |
438 | ||
6f6528a1 | 439 | return false; |
b789ef51 CL |
440 | } |
441 | ||
41ecc55b | 442 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
443 | /* |
444 | * Determine a map of object in use on a page. | |
445 | * | |
881db7fb | 446 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
447 | * not vanish from under us. |
448 | */ | |
449 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
450 | { | |
451 | void *p; | |
452 | void *addr = page_address(page); | |
453 | ||
454 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
455 | set_bit(slab_index(p, s, addr), map); | |
456 | } | |
457 | ||
870b1fbb | 458 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
459 | { |
460 | if (s->flags & SLAB_RED_ZONE) | |
461 | return s->size - s->red_left_pad; | |
462 | ||
463 | return s->size; | |
464 | } | |
465 | ||
466 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
467 | { | |
468 | if (s->flags & SLAB_RED_ZONE) | |
469 | p -= s->red_left_pad; | |
470 | ||
471 | return p; | |
472 | } | |
473 | ||
41ecc55b CL |
474 | /* |
475 | * Debug settings: | |
476 | */ | |
89d3c87e | 477 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 478 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 479 | #else |
d50112ed | 480 | static slab_flags_t slub_debug; |
f0630fff | 481 | #endif |
41ecc55b CL |
482 | |
483 | static char *slub_debug_slabs; | |
fa5ec8a1 | 484 | static int disable_higher_order_debug; |
41ecc55b | 485 | |
a79316c6 AR |
486 | /* |
487 | * slub is about to manipulate internal object metadata. This memory lies | |
488 | * outside the range of the allocated object, so accessing it would normally | |
489 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
490 | * to tell kasan that these accesses are OK. | |
491 | */ | |
492 | static inline void metadata_access_enable(void) | |
493 | { | |
494 | kasan_disable_current(); | |
495 | } | |
496 | ||
497 | static inline void metadata_access_disable(void) | |
498 | { | |
499 | kasan_enable_current(); | |
500 | } | |
501 | ||
81819f0f CL |
502 | /* |
503 | * Object debugging | |
504 | */ | |
d86bd1be JK |
505 | |
506 | /* Verify that a pointer has an address that is valid within a slab page */ | |
507 | static inline int check_valid_pointer(struct kmem_cache *s, | |
508 | struct page *page, void *object) | |
509 | { | |
510 | void *base; | |
511 | ||
512 | if (!object) | |
513 | return 1; | |
514 | ||
515 | base = page_address(page); | |
338cfaad | 516 | object = kasan_reset_tag(object); |
d86bd1be JK |
517 | object = restore_red_left(s, object); |
518 | if (object < base || object >= base + page->objects * s->size || | |
519 | (object - base) % s->size) { | |
520 | return 0; | |
521 | } | |
522 | ||
523 | return 1; | |
524 | } | |
525 | ||
aa2efd5e DT |
526 | static void print_section(char *level, char *text, u8 *addr, |
527 | unsigned int length) | |
81819f0f | 528 | { |
a79316c6 | 529 | metadata_access_enable(); |
aa2efd5e | 530 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
ffc79d28 | 531 | length, 1); |
a79316c6 | 532 | metadata_access_disable(); |
81819f0f CL |
533 | } |
534 | ||
81819f0f CL |
535 | static struct track *get_track(struct kmem_cache *s, void *object, |
536 | enum track_item alloc) | |
537 | { | |
538 | struct track *p; | |
539 | ||
540 | if (s->offset) | |
541 | p = object + s->offset + sizeof(void *); | |
542 | else | |
543 | p = object + s->inuse; | |
544 | ||
545 | return p + alloc; | |
546 | } | |
547 | ||
548 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 549 | enum track_item alloc, unsigned long addr) |
81819f0f | 550 | { |
1a00df4a | 551 | struct track *p = get_track(s, object, alloc); |
81819f0f | 552 | |
81819f0f | 553 | if (addr) { |
d6543e39 | 554 | #ifdef CONFIG_STACKTRACE |
79716799 | 555 | unsigned int nr_entries; |
d6543e39 | 556 | |
a79316c6 | 557 | metadata_access_enable(); |
79716799 | 558 | nr_entries = stack_trace_save(p->addrs, TRACK_ADDRS_COUNT, 3); |
a79316c6 | 559 | metadata_access_disable(); |
d6543e39 | 560 | |
79716799 TG |
561 | if (nr_entries < TRACK_ADDRS_COUNT) |
562 | p->addrs[nr_entries] = 0; | |
d6543e39 | 563 | #endif |
81819f0f CL |
564 | p->addr = addr; |
565 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 566 | p->pid = current->pid; |
81819f0f | 567 | p->when = jiffies; |
b8ca7ff7 | 568 | } else { |
81819f0f | 569 | memset(p, 0, sizeof(struct track)); |
b8ca7ff7 | 570 | } |
81819f0f CL |
571 | } |
572 | ||
81819f0f CL |
573 | static void init_tracking(struct kmem_cache *s, void *object) |
574 | { | |
24922684 CL |
575 | if (!(s->flags & SLAB_STORE_USER)) |
576 | return; | |
577 | ||
ce71e27c EGM |
578 | set_track(s, object, TRACK_FREE, 0UL); |
579 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
580 | } |
581 | ||
86609d33 | 582 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f CL |
583 | { |
584 | if (!t->addr) | |
585 | return; | |
586 | ||
f9f58285 | 587 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 588 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
d6543e39 BG |
589 | #ifdef CONFIG_STACKTRACE |
590 | { | |
591 | int i; | |
592 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
593 | if (t->addrs[i]) | |
f9f58285 | 594 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
595 | else |
596 | break; | |
597 | } | |
598 | #endif | |
24922684 CL |
599 | } |
600 | ||
601 | static void print_tracking(struct kmem_cache *s, void *object) | |
602 | { | |
86609d33 | 603 | unsigned long pr_time = jiffies; |
24922684 CL |
604 | if (!(s->flags & SLAB_STORE_USER)) |
605 | return; | |
606 | ||
86609d33 CP |
607 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
608 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
609 | } |
610 | ||
611 | static void print_page_info(struct page *page) | |
612 | { | |
f9f58285 | 613 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 614 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
615 | |
616 | } | |
617 | ||
618 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
619 | { | |
ecc42fbe | 620 | struct va_format vaf; |
24922684 | 621 | va_list args; |
24922684 CL |
622 | |
623 | va_start(args, fmt); | |
ecc42fbe FF |
624 | vaf.fmt = fmt; |
625 | vaf.va = &args; | |
f9f58285 | 626 | pr_err("=============================================================================\n"); |
ecc42fbe | 627 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 628 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 629 | |
373d4d09 | 630 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 631 | va_end(args); |
81819f0f CL |
632 | } |
633 | ||
24922684 CL |
634 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
635 | { | |
ecc42fbe | 636 | struct va_format vaf; |
24922684 | 637 | va_list args; |
24922684 CL |
638 | |
639 | va_start(args, fmt); | |
ecc42fbe FF |
640 | vaf.fmt = fmt; |
641 | vaf.va = &args; | |
642 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 643 | va_end(args); |
24922684 CL |
644 | } |
645 | ||
646 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
647 | { |
648 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 649 | u8 *addr = page_address(page); |
24922684 CL |
650 | |
651 | print_tracking(s, p); | |
652 | ||
653 | print_page_info(page); | |
654 | ||
f9f58285 FF |
655 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
656 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 657 | |
d86bd1be | 658 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e DT |
659 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
660 | s->red_left_pad); | |
d86bd1be | 661 | else if (p > addr + 16) |
aa2efd5e | 662 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 663 | |
aa2efd5e | 664 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 665 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 666 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e | 667 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 668 | s->inuse - s->object_size); |
81819f0f | 669 | |
81819f0f CL |
670 | if (s->offset) |
671 | off = s->offset + sizeof(void *); | |
672 | else | |
673 | off = s->inuse; | |
674 | ||
24922684 | 675 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 676 | off += 2 * sizeof(struct track); |
81819f0f | 677 | |
80a9201a AP |
678 | off += kasan_metadata_size(s); |
679 | ||
d86bd1be | 680 | if (off != size_from_object(s)) |
81819f0f | 681 | /* Beginning of the filler is the free pointer */ |
aa2efd5e DT |
682 | print_section(KERN_ERR, "Padding ", p + off, |
683 | size_from_object(s) - off); | |
24922684 CL |
684 | |
685 | dump_stack(); | |
81819f0f CL |
686 | } |
687 | ||
75c66def | 688 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
689 | u8 *object, char *reason) |
690 | { | |
3dc50637 | 691 | slab_bug(s, "%s", reason); |
24922684 | 692 | print_trailer(s, page, object); |
81819f0f CL |
693 | } |
694 | ||
a38965bf | 695 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, |
d0e0ac97 | 696 | const char *fmt, ...) |
81819f0f CL |
697 | { |
698 | va_list args; | |
699 | char buf[100]; | |
700 | ||
24922684 CL |
701 | va_start(args, fmt); |
702 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 703 | va_end(args); |
3dc50637 | 704 | slab_bug(s, "%s", buf); |
24922684 | 705 | print_page_info(page); |
81819f0f CL |
706 | dump_stack(); |
707 | } | |
708 | ||
f7cb1933 | 709 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
710 | { |
711 | u8 *p = object; | |
712 | ||
d86bd1be JK |
713 | if (s->flags & SLAB_RED_ZONE) |
714 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
715 | ||
81819f0f | 716 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
717 | memset(p, POISON_FREE, s->object_size - 1); |
718 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
719 | } |
720 | ||
721 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 722 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
723 | } |
724 | ||
24922684 CL |
725 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
726 | void *from, void *to) | |
727 | { | |
728 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
729 | memset(from, data, to - from); | |
730 | } | |
731 | ||
732 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
733 | u8 *object, char *what, | |
06428780 | 734 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
735 | { |
736 | u8 *fault; | |
737 | u8 *end; | |
738 | ||
a79316c6 | 739 | metadata_access_enable(); |
79824820 | 740 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 741 | metadata_access_disable(); |
24922684 CL |
742 | if (!fault) |
743 | return 1; | |
744 | ||
745 | end = start + bytes; | |
746 | while (end > fault && end[-1] == value) | |
747 | end--; | |
748 | ||
749 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 750 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
751 | fault, end - 1, fault[0], value); |
752 | print_trailer(s, page, object); | |
753 | ||
754 | restore_bytes(s, what, value, fault, end); | |
755 | return 0; | |
81819f0f CL |
756 | } |
757 | ||
81819f0f CL |
758 | /* |
759 | * Object layout: | |
760 | * | |
761 | * object address | |
762 | * Bytes of the object to be managed. | |
763 | * If the freepointer may overlay the object then the free | |
764 | * pointer is the first word of the object. | |
672bba3a | 765 | * |
81819f0f CL |
766 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
767 | * 0xa5 (POISON_END) | |
768 | * | |
3b0efdfa | 769 | * object + s->object_size |
81819f0f | 770 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 771 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 772 | * object_size == inuse. |
672bba3a | 773 | * |
81819f0f CL |
774 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
775 | * 0xcc (RED_ACTIVE) for objects in use. | |
776 | * | |
777 | * object + s->inuse | |
672bba3a CL |
778 | * Meta data starts here. |
779 | * | |
81819f0f CL |
780 | * A. Free pointer (if we cannot overwrite object on free) |
781 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 782 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 783 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
784 | * before the word boundary. |
785 | * | |
786 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
787 | * |
788 | * object + s->size | |
672bba3a | 789 | * Nothing is used beyond s->size. |
81819f0f | 790 | * |
3b0efdfa | 791 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 792 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
793 | * may be used with merged slabcaches. |
794 | */ | |
795 | ||
81819f0f CL |
796 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
797 | { | |
798 | unsigned long off = s->inuse; /* The end of info */ | |
799 | ||
800 | if (s->offset) | |
801 | /* Freepointer is placed after the object. */ | |
802 | off += sizeof(void *); | |
803 | ||
804 | if (s->flags & SLAB_STORE_USER) | |
805 | /* We also have user information there */ | |
806 | off += 2 * sizeof(struct track); | |
807 | ||
80a9201a AP |
808 | off += kasan_metadata_size(s); |
809 | ||
d86bd1be | 810 | if (size_from_object(s) == off) |
81819f0f CL |
811 | return 1; |
812 | ||
24922684 | 813 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 814 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
815 | } |
816 | ||
39b26464 | 817 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
818 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
819 | { | |
24922684 CL |
820 | u8 *start; |
821 | u8 *fault; | |
822 | u8 *end; | |
5d682681 | 823 | u8 *pad; |
24922684 CL |
824 | int length; |
825 | int remainder; | |
81819f0f CL |
826 | |
827 | if (!(s->flags & SLAB_POISON)) | |
828 | return 1; | |
829 | ||
a973e9dd | 830 | start = page_address(page); |
9736d2a9 | 831 | length = PAGE_SIZE << compound_order(page); |
39b26464 CL |
832 | end = start + length; |
833 | remainder = length % s->size; | |
81819f0f CL |
834 | if (!remainder) |
835 | return 1; | |
836 | ||
5d682681 | 837 | pad = end - remainder; |
a79316c6 | 838 | metadata_access_enable(); |
5d682681 | 839 | fault = memchr_inv(pad, POISON_INUSE, remainder); |
a79316c6 | 840 | metadata_access_disable(); |
24922684 CL |
841 | if (!fault) |
842 | return 1; | |
843 | while (end > fault && end[-1] == POISON_INUSE) | |
844 | end--; | |
845 | ||
846 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
5d682681 | 847 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 848 | |
5d682681 | 849 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
24922684 | 850 | return 0; |
81819f0f CL |
851 | } |
852 | ||
853 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 854 | void *object, u8 val) |
81819f0f CL |
855 | { |
856 | u8 *p = object; | |
3b0efdfa | 857 | u8 *endobject = object + s->object_size; |
81819f0f CL |
858 | |
859 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
860 | if (!check_bytes_and_report(s, page, object, "Redzone", |
861 | object - s->red_left_pad, val, s->red_left_pad)) | |
862 | return 0; | |
863 | ||
24922684 | 864 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 865 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 866 | return 0; |
81819f0f | 867 | } else { |
3b0efdfa | 868 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 869 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
870 | endobject, POISON_INUSE, |
871 | s->inuse - s->object_size); | |
3adbefee | 872 | } |
81819f0f CL |
873 | } |
874 | ||
875 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 876 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 877 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 878 | POISON_FREE, s->object_size - 1) || |
24922684 | 879 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 880 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 881 | return 0; |
81819f0f CL |
882 | /* |
883 | * check_pad_bytes cleans up on its own. | |
884 | */ | |
885 | check_pad_bytes(s, page, p); | |
886 | } | |
887 | ||
f7cb1933 | 888 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
889 | /* |
890 | * Object and freepointer overlap. Cannot check | |
891 | * freepointer while object is allocated. | |
892 | */ | |
893 | return 1; | |
894 | ||
895 | /* Check free pointer validity */ | |
896 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
897 | object_err(s, page, p, "Freepointer corrupt"); | |
898 | /* | |
9f6c708e | 899 | * No choice but to zap it and thus lose the remainder |
81819f0f | 900 | * of the free objects in this slab. May cause |
672bba3a | 901 | * another error because the object count is now wrong. |
81819f0f | 902 | */ |
a973e9dd | 903 | set_freepointer(s, p, NULL); |
81819f0f CL |
904 | return 0; |
905 | } | |
906 | return 1; | |
907 | } | |
908 | ||
909 | static int check_slab(struct kmem_cache *s, struct page *page) | |
910 | { | |
39b26464 CL |
911 | int maxobj; |
912 | ||
81819f0f CL |
913 | VM_BUG_ON(!irqs_disabled()); |
914 | ||
915 | if (!PageSlab(page)) { | |
24922684 | 916 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
917 | return 0; |
918 | } | |
39b26464 | 919 | |
9736d2a9 | 920 | maxobj = order_objects(compound_order(page), s->size); |
39b26464 CL |
921 | if (page->objects > maxobj) { |
922 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 923 | page->objects, maxobj); |
39b26464 CL |
924 | return 0; |
925 | } | |
926 | if (page->inuse > page->objects) { | |
24922684 | 927 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 928 | page->inuse, page->objects); |
81819f0f CL |
929 | return 0; |
930 | } | |
931 | /* Slab_pad_check fixes things up after itself */ | |
932 | slab_pad_check(s, page); | |
933 | return 1; | |
934 | } | |
935 | ||
936 | /* | |
672bba3a CL |
937 | * Determine if a certain object on a page is on the freelist. Must hold the |
938 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
939 | */ |
940 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
941 | { | |
942 | int nr = 0; | |
881db7fb | 943 | void *fp; |
81819f0f | 944 | void *object = NULL; |
f6edde9c | 945 | int max_objects; |
81819f0f | 946 | |
881db7fb | 947 | fp = page->freelist; |
39b26464 | 948 | while (fp && nr <= page->objects) { |
81819f0f CL |
949 | if (fp == search) |
950 | return 1; | |
951 | if (!check_valid_pointer(s, page, fp)) { | |
952 | if (object) { | |
953 | object_err(s, page, object, | |
954 | "Freechain corrupt"); | |
a973e9dd | 955 | set_freepointer(s, object, NULL); |
81819f0f | 956 | } else { |
24922684 | 957 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 958 | page->freelist = NULL; |
39b26464 | 959 | page->inuse = page->objects; |
24922684 | 960 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
961 | return 0; |
962 | } | |
963 | break; | |
964 | } | |
965 | object = fp; | |
966 | fp = get_freepointer(s, object); | |
967 | nr++; | |
968 | } | |
969 | ||
9736d2a9 | 970 | max_objects = order_objects(compound_order(page), s->size); |
210b5c06 CG |
971 | if (max_objects > MAX_OBJS_PER_PAGE) |
972 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
973 | |
974 | if (page->objects != max_objects) { | |
756a025f JP |
975 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
976 | page->objects, max_objects); | |
224a88be CL |
977 | page->objects = max_objects; |
978 | slab_fix(s, "Number of objects adjusted."); | |
979 | } | |
39b26464 | 980 | if (page->inuse != page->objects - nr) { |
756a025f JP |
981 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
982 | page->inuse, page->objects - nr); | |
39b26464 | 983 | page->inuse = page->objects - nr; |
24922684 | 984 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
985 | } |
986 | return search == NULL; | |
987 | } | |
988 | ||
0121c619 CL |
989 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
990 | int alloc) | |
3ec09742 CL |
991 | { |
992 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 993 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
994 | s->name, |
995 | alloc ? "alloc" : "free", | |
996 | object, page->inuse, | |
997 | page->freelist); | |
998 | ||
999 | if (!alloc) | |
aa2efd5e | 1000 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1001 | s->object_size); |
3ec09742 CL |
1002 | |
1003 | dump_stack(); | |
1004 | } | |
1005 | } | |
1006 | ||
643b1138 | 1007 | /* |
672bba3a | 1008 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1009 | */ |
5cc6eee8 CL |
1010 | static void add_full(struct kmem_cache *s, |
1011 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1012 | { |
5cc6eee8 CL |
1013 | if (!(s->flags & SLAB_STORE_USER)) |
1014 | return; | |
1015 | ||
255d0884 | 1016 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1017 | list_add(&page->lru, &n->full); |
643b1138 CL |
1018 | } |
1019 | ||
c65c1877 | 1020 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1021 | { |
643b1138 CL |
1022 | if (!(s->flags & SLAB_STORE_USER)) |
1023 | return; | |
1024 | ||
255d0884 | 1025 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1026 | list_del(&page->lru); |
643b1138 CL |
1027 | } |
1028 | ||
0f389ec6 CL |
1029 | /* Tracking of the number of slabs for debugging purposes */ |
1030 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1031 | { | |
1032 | struct kmem_cache_node *n = get_node(s, node); | |
1033 | ||
1034 | return atomic_long_read(&n->nr_slabs); | |
1035 | } | |
1036 | ||
26c02cf0 AB |
1037 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1038 | { | |
1039 | return atomic_long_read(&n->nr_slabs); | |
1040 | } | |
1041 | ||
205ab99d | 1042 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1043 | { |
1044 | struct kmem_cache_node *n = get_node(s, node); | |
1045 | ||
1046 | /* | |
1047 | * May be called early in order to allocate a slab for the | |
1048 | * kmem_cache_node structure. Solve the chicken-egg | |
1049 | * dilemma by deferring the increment of the count during | |
1050 | * bootstrap (see early_kmem_cache_node_alloc). | |
1051 | */ | |
338b2642 | 1052 | if (likely(n)) { |
0f389ec6 | 1053 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1054 | atomic_long_add(objects, &n->total_objects); |
1055 | } | |
0f389ec6 | 1056 | } |
205ab99d | 1057 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1058 | { |
1059 | struct kmem_cache_node *n = get_node(s, node); | |
1060 | ||
1061 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1062 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1063 | } |
1064 | ||
1065 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1066 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1067 | void *object) | |
1068 | { | |
1069 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1070 | return; | |
1071 | ||
f7cb1933 | 1072 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1073 | init_tracking(s, object); |
1074 | } | |
1075 | ||
a7101224 AK |
1076 | static void setup_page_debug(struct kmem_cache *s, void *addr, int order) |
1077 | { | |
1078 | if (!(s->flags & SLAB_POISON)) | |
1079 | return; | |
1080 | ||
1081 | metadata_access_enable(); | |
1082 | memset(addr, POISON_INUSE, PAGE_SIZE << order); | |
1083 | metadata_access_disable(); | |
1084 | } | |
1085 | ||
becfda68 | 1086 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
278d7756 | 1087 | struct page *page, void *object) |
81819f0f CL |
1088 | { |
1089 | if (!check_slab(s, page)) | |
becfda68 | 1090 | return 0; |
81819f0f | 1091 | |
81819f0f CL |
1092 | if (!check_valid_pointer(s, page, object)) { |
1093 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1094 | return 0; |
81819f0f CL |
1095 | } |
1096 | ||
f7cb1933 | 1097 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1098 | return 0; |
1099 | ||
1100 | return 1; | |
1101 | } | |
1102 | ||
1103 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1104 | struct page *page, | |
1105 | void *object, unsigned long addr) | |
1106 | { | |
1107 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
278d7756 | 1108 | if (!alloc_consistency_checks(s, page, object)) |
becfda68 LA |
1109 | goto bad; |
1110 | } | |
81819f0f | 1111 | |
3ec09742 CL |
1112 | /* Success perform special debug activities for allocs */ |
1113 | if (s->flags & SLAB_STORE_USER) | |
1114 | set_track(s, object, TRACK_ALLOC, addr); | |
1115 | trace(s, page, object, 1); | |
f7cb1933 | 1116 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1117 | return 1; |
3ec09742 | 1118 | |
81819f0f CL |
1119 | bad: |
1120 | if (PageSlab(page)) { | |
1121 | /* | |
1122 | * If this is a slab page then lets do the best we can | |
1123 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1124 | * as used avoids touching the remaining objects. |
81819f0f | 1125 | */ |
24922684 | 1126 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1127 | page->inuse = page->objects; |
a973e9dd | 1128 | page->freelist = NULL; |
81819f0f CL |
1129 | } |
1130 | return 0; | |
1131 | } | |
1132 | ||
becfda68 LA |
1133 | static inline int free_consistency_checks(struct kmem_cache *s, |
1134 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1135 | { |
81819f0f | 1136 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1137 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1138 | return 0; |
81819f0f CL |
1139 | } |
1140 | ||
1141 | if (on_freelist(s, page, object)) { | |
24922684 | 1142 | object_err(s, page, object, "Object already free"); |
becfda68 | 1143 | return 0; |
81819f0f CL |
1144 | } |
1145 | ||
f7cb1933 | 1146 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1147 | return 0; |
81819f0f | 1148 | |
1b4f59e3 | 1149 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1150 | if (!PageSlab(page)) { |
756a025f JP |
1151 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1152 | object); | |
1b4f59e3 | 1153 | } else if (!page->slab_cache) { |
f9f58285 FF |
1154 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1155 | object); | |
70d71228 | 1156 | dump_stack(); |
06428780 | 1157 | } else |
24922684 CL |
1158 | object_err(s, page, object, |
1159 | "page slab pointer corrupt."); | |
becfda68 LA |
1160 | return 0; |
1161 | } | |
1162 | return 1; | |
1163 | } | |
1164 | ||
1165 | /* Supports checking bulk free of a constructed freelist */ | |
1166 | static noinline int free_debug_processing( | |
1167 | struct kmem_cache *s, struct page *page, | |
1168 | void *head, void *tail, int bulk_cnt, | |
1169 | unsigned long addr) | |
1170 | { | |
1171 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1172 | void *object = head; | |
1173 | int cnt = 0; | |
1174 | unsigned long uninitialized_var(flags); | |
1175 | int ret = 0; | |
1176 | ||
1177 | spin_lock_irqsave(&n->list_lock, flags); | |
1178 | slab_lock(page); | |
1179 | ||
1180 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1181 | if (!check_slab(s, page)) | |
1182 | goto out; | |
1183 | } | |
1184 | ||
1185 | next_object: | |
1186 | cnt++; | |
1187 | ||
1188 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1189 | if (!free_consistency_checks(s, page, object, addr)) | |
1190 | goto out; | |
81819f0f | 1191 | } |
3ec09742 | 1192 | |
3ec09742 CL |
1193 | if (s->flags & SLAB_STORE_USER) |
1194 | set_track(s, object, TRACK_FREE, addr); | |
1195 | trace(s, page, object, 0); | |
81084651 | 1196 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1197 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1198 | |
1199 | /* Reached end of constructed freelist yet? */ | |
1200 | if (object != tail) { | |
1201 | object = get_freepointer(s, object); | |
1202 | goto next_object; | |
1203 | } | |
804aa132 LA |
1204 | ret = 1; |
1205 | ||
5c2e4bbb | 1206 | out: |
81084651 JDB |
1207 | if (cnt != bulk_cnt) |
1208 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1209 | bulk_cnt, cnt); | |
1210 | ||
881db7fb | 1211 | slab_unlock(page); |
282acb43 | 1212 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1213 | if (!ret) |
1214 | slab_fix(s, "Object at 0x%p not freed", object); | |
1215 | return ret; | |
81819f0f CL |
1216 | } |
1217 | ||
41ecc55b CL |
1218 | static int __init setup_slub_debug(char *str) |
1219 | { | |
f0630fff CL |
1220 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1221 | if (*str++ != '=' || !*str) | |
1222 | /* | |
1223 | * No options specified. Switch on full debugging. | |
1224 | */ | |
1225 | goto out; | |
1226 | ||
1227 | if (*str == ',') | |
1228 | /* | |
1229 | * No options but restriction on slabs. This means full | |
1230 | * debugging for slabs matching a pattern. | |
1231 | */ | |
1232 | goto check_slabs; | |
1233 | ||
1234 | slub_debug = 0; | |
1235 | if (*str == '-') | |
1236 | /* | |
1237 | * Switch off all debugging measures. | |
1238 | */ | |
1239 | goto out; | |
1240 | ||
1241 | /* | |
1242 | * Determine which debug features should be switched on | |
1243 | */ | |
06428780 | 1244 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1245 | switch (tolower(*str)) { |
1246 | case 'f': | |
becfda68 | 1247 | slub_debug |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1248 | break; |
1249 | case 'z': | |
1250 | slub_debug |= SLAB_RED_ZONE; | |
1251 | break; | |
1252 | case 'p': | |
1253 | slub_debug |= SLAB_POISON; | |
1254 | break; | |
1255 | case 'u': | |
1256 | slub_debug |= SLAB_STORE_USER; | |
1257 | break; | |
1258 | case 't': | |
1259 | slub_debug |= SLAB_TRACE; | |
1260 | break; | |
4c13dd3b DM |
1261 | case 'a': |
1262 | slub_debug |= SLAB_FAILSLAB; | |
1263 | break; | |
08303a73 CA |
1264 | case 'o': |
1265 | /* | |
1266 | * Avoid enabling debugging on caches if its minimum | |
1267 | * order would increase as a result. | |
1268 | */ | |
1269 | disable_higher_order_debug = 1; | |
1270 | break; | |
f0630fff | 1271 | default: |
f9f58285 FF |
1272 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1273 | *str); | |
f0630fff | 1274 | } |
41ecc55b CL |
1275 | } |
1276 | ||
f0630fff | 1277 | check_slabs: |
41ecc55b CL |
1278 | if (*str == ',') |
1279 | slub_debug_slabs = str + 1; | |
f0630fff | 1280 | out: |
41ecc55b CL |
1281 | return 1; |
1282 | } | |
1283 | ||
1284 | __setup("slub_debug", setup_slub_debug); | |
1285 | ||
c5fd3ca0 AT |
1286 | /* |
1287 | * kmem_cache_flags - apply debugging options to the cache | |
1288 | * @object_size: the size of an object without meta data | |
1289 | * @flags: flags to set | |
1290 | * @name: name of the cache | |
1291 | * @ctor: constructor function | |
1292 | * | |
1293 | * Debug option(s) are applied to @flags. In addition to the debug | |
1294 | * option(s), if a slab name (or multiple) is specified i.e. | |
1295 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1296 | * then only the select slabs will receive the debug option(s). | |
1297 | */ | |
0293d1fd | 1298 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1299 | slab_flags_t flags, const char *name, |
51cc5068 | 1300 | void (*ctor)(void *)) |
41ecc55b | 1301 | { |
c5fd3ca0 AT |
1302 | char *iter; |
1303 | size_t len; | |
1304 | ||
1305 | /* If slub_debug = 0, it folds into the if conditional. */ | |
1306 | if (!slub_debug_slabs) | |
1307 | return flags | slub_debug; | |
1308 | ||
1309 | len = strlen(name); | |
1310 | iter = slub_debug_slabs; | |
1311 | while (*iter) { | |
1312 | char *end, *glob; | |
1313 | size_t cmplen; | |
1314 | ||
1315 | end = strchr(iter, ','); | |
1316 | if (!end) | |
1317 | end = iter + strlen(iter); | |
1318 | ||
1319 | glob = strnchr(iter, end - iter, '*'); | |
1320 | if (glob) | |
1321 | cmplen = glob - iter; | |
1322 | else | |
1323 | cmplen = max_t(size_t, len, (end - iter)); | |
1324 | ||
1325 | if (!strncmp(name, iter, cmplen)) { | |
1326 | flags |= slub_debug; | |
1327 | break; | |
1328 | } | |
1329 | ||
1330 | if (!*end) | |
1331 | break; | |
1332 | iter = end + 1; | |
1333 | } | |
ba0268a8 CL |
1334 | |
1335 | return flags; | |
41ecc55b | 1336 | } |
b4a64718 | 1337 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1338 | static inline void setup_object_debug(struct kmem_cache *s, |
1339 | struct page *page, void *object) {} | |
a7101224 AK |
1340 | static inline void setup_page_debug(struct kmem_cache *s, |
1341 | void *addr, int order) {} | |
41ecc55b | 1342 | |
3ec09742 | 1343 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1344 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1345 | |
282acb43 | 1346 | static inline int free_debug_processing( |
81084651 JDB |
1347 | struct kmem_cache *s, struct page *page, |
1348 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1349 | unsigned long addr) { return 0; } |
41ecc55b | 1350 | |
41ecc55b CL |
1351 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1352 | { return 1; } | |
1353 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1354 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1355 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1356 | struct page *page) {} | |
c65c1877 PZ |
1357 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1358 | struct page *page) {} | |
0293d1fd | 1359 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1360 | slab_flags_t flags, const char *name, |
51cc5068 | 1361 | void (*ctor)(void *)) |
ba0268a8 CL |
1362 | { |
1363 | return flags; | |
1364 | } | |
41ecc55b | 1365 | #define slub_debug 0 |
0f389ec6 | 1366 | |
fdaa45e9 IM |
1367 | #define disable_higher_order_debug 0 |
1368 | ||
0f389ec6 CL |
1369 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1370 | { return 0; } | |
26c02cf0 AB |
1371 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1372 | { return 0; } | |
205ab99d CL |
1373 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1374 | int objects) {} | |
1375 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1376 | int objects) {} | |
7d550c56 | 1377 | |
02e72cc6 AR |
1378 | #endif /* CONFIG_SLUB_DEBUG */ |
1379 | ||
1380 | /* | |
1381 | * Hooks for other subsystems that check memory allocations. In a typical | |
1382 | * production configuration these hooks all should produce no code at all. | |
1383 | */ | |
0116523c | 1384 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
d56791b3 | 1385 | { |
53128245 | 1386 | ptr = kasan_kmalloc_large(ptr, size, flags); |
a2f77575 | 1387 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ |
d56791b3 | 1388 | kmemleak_alloc(ptr, size, 1, flags); |
53128245 | 1389 | return ptr; |
d56791b3 RB |
1390 | } |
1391 | ||
ee3ce779 | 1392 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1393 | { |
1394 | kmemleak_free(x); | |
ee3ce779 | 1395 | kasan_kfree_large(x, _RET_IP_); |
d56791b3 RB |
1396 | } |
1397 | ||
c3895391 | 1398 | static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x) |
d56791b3 RB |
1399 | { |
1400 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1401 | |
02e72cc6 AR |
1402 | /* |
1403 | * Trouble is that we may no longer disable interrupts in the fast path | |
1404 | * So in order to make the debug calls that expect irqs to be | |
1405 | * disabled we need to disable interrupts temporarily. | |
1406 | */ | |
4675ff05 | 1407 | #ifdef CONFIG_LOCKDEP |
02e72cc6 AR |
1408 | { |
1409 | unsigned long flags; | |
1410 | ||
1411 | local_irq_save(flags); | |
02e72cc6 AR |
1412 | debug_check_no_locks_freed(x, s->object_size); |
1413 | local_irq_restore(flags); | |
1414 | } | |
1415 | #endif | |
1416 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1417 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1418 | |
c3895391 AK |
1419 | /* KASAN might put x into memory quarantine, delaying its reuse */ |
1420 | return kasan_slab_free(s, x, _RET_IP_); | |
02e72cc6 | 1421 | } |
205ab99d | 1422 | |
c3895391 AK |
1423 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
1424 | void **head, void **tail) | |
81084651 JDB |
1425 | { |
1426 | /* | |
1427 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1428 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1429 | */ | |
4675ff05 | 1430 | #if defined(CONFIG_LOCKDEP) || \ |
81084651 JDB |
1431 | defined(CONFIG_DEBUG_KMEMLEAK) || \ |
1432 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1433 | defined(CONFIG_KASAN) | |
1434 | ||
c3895391 AK |
1435 | void *object; |
1436 | void *next = *head; | |
1437 | void *old_tail = *tail ? *tail : *head; | |
1438 | ||
1439 | /* Head and tail of the reconstructed freelist */ | |
1440 | *head = NULL; | |
1441 | *tail = NULL; | |
81084651 JDB |
1442 | |
1443 | do { | |
c3895391 AK |
1444 | object = next; |
1445 | next = get_freepointer(s, object); | |
1446 | /* If object's reuse doesn't have to be delayed */ | |
1447 | if (!slab_free_hook(s, object)) { | |
1448 | /* Move object to the new freelist */ | |
1449 | set_freepointer(s, object, *head); | |
1450 | *head = object; | |
1451 | if (!*tail) | |
1452 | *tail = object; | |
1453 | } | |
1454 | } while (object != old_tail); | |
1455 | ||
1456 | if (*head == *tail) | |
1457 | *tail = NULL; | |
1458 | ||
1459 | return *head != NULL; | |
1460 | #else | |
1461 | return true; | |
81084651 JDB |
1462 | #endif |
1463 | } | |
1464 | ||
4d176711 | 1465 | static void *setup_object(struct kmem_cache *s, struct page *page, |
588f8ba9 TG |
1466 | void *object) |
1467 | { | |
1468 | setup_object_debug(s, page, object); | |
4d176711 | 1469 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1470 | if (unlikely(s->ctor)) { |
1471 | kasan_unpoison_object_data(s, object); | |
1472 | s->ctor(object); | |
1473 | kasan_poison_object_data(s, object); | |
1474 | } | |
4d176711 | 1475 | return object; |
588f8ba9 TG |
1476 | } |
1477 | ||
81819f0f CL |
1478 | /* |
1479 | * Slab allocation and freeing | |
1480 | */ | |
5dfb4175 VD |
1481 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1482 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1483 | { |
5dfb4175 | 1484 | struct page *page; |
19af27af | 1485 | unsigned int order = oo_order(oo); |
65c3376a | 1486 | |
2154a336 | 1487 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1488 | page = alloc_pages(flags, order); |
65c3376a | 1489 | else |
96db800f | 1490 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1491 | |
f3ccb2c4 VD |
1492 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1493 | __free_pages(page, order); | |
1494 | page = NULL; | |
1495 | } | |
5dfb4175 VD |
1496 | |
1497 | return page; | |
65c3376a CL |
1498 | } |
1499 | ||
210e7a43 TG |
1500 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1501 | /* Pre-initialize the random sequence cache */ | |
1502 | static int init_cache_random_seq(struct kmem_cache *s) | |
1503 | { | |
19af27af | 1504 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1505 | int err; |
210e7a43 | 1506 | |
a810007a SR |
1507 | /* Bailout if already initialised */ |
1508 | if (s->random_seq) | |
1509 | return 0; | |
1510 | ||
210e7a43 TG |
1511 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1512 | if (err) { | |
1513 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1514 | s->name); | |
1515 | return err; | |
1516 | } | |
1517 | ||
1518 | /* Transform to an offset on the set of pages */ | |
1519 | if (s->random_seq) { | |
19af27af AD |
1520 | unsigned int i; |
1521 | ||
210e7a43 TG |
1522 | for (i = 0; i < count; i++) |
1523 | s->random_seq[i] *= s->size; | |
1524 | } | |
1525 | return 0; | |
1526 | } | |
1527 | ||
1528 | /* Initialize each random sequence freelist per cache */ | |
1529 | static void __init init_freelist_randomization(void) | |
1530 | { | |
1531 | struct kmem_cache *s; | |
1532 | ||
1533 | mutex_lock(&slab_mutex); | |
1534 | ||
1535 | list_for_each_entry(s, &slab_caches, list) | |
1536 | init_cache_random_seq(s); | |
1537 | ||
1538 | mutex_unlock(&slab_mutex); | |
1539 | } | |
1540 | ||
1541 | /* Get the next entry on the pre-computed freelist randomized */ | |
1542 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1543 | unsigned long *pos, void *start, | |
1544 | unsigned long page_limit, | |
1545 | unsigned long freelist_count) | |
1546 | { | |
1547 | unsigned int idx; | |
1548 | ||
1549 | /* | |
1550 | * If the target page allocation failed, the number of objects on the | |
1551 | * page might be smaller than the usual size defined by the cache. | |
1552 | */ | |
1553 | do { | |
1554 | idx = s->random_seq[*pos]; | |
1555 | *pos += 1; | |
1556 | if (*pos >= freelist_count) | |
1557 | *pos = 0; | |
1558 | } while (unlikely(idx >= page_limit)); | |
1559 | ||
1560 | return (char *)start + idx; | |
1561 | } | |
1562 | ||
1563 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1564 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1565 | { | |
1566 | void *start; | |
1567 | void *cur; | |
1568 | void *next; | |
1569 | unsigned long idx, pos, page_limit, freelist_count; | |
1570 | ||
1571 | if (page->objects < 2 || !s->random_seq) | |
1572 | return false; | |
1573 | ||
1574 | freelist_count = oo_objects(s->oo); | |
1575 | pos = get_random_int() % freelist_count; | |
1576 | ||
1577 | page_limit = page->objects * s->size; | |
1578 | start = fixup_red_left(s, page_address(page)); | |
1579 | ||
1580 | /* First entry is used as the base of the freelist */ | |
1581 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1582 | freelist_count); | |
4d176711 | 1583 | cur = setup_object(s, page, cur); |
210e7a43 TG |
1584 | page->freelist = cur; |
1585 | ||
1586 | for (idx = 1; idx < page->objects; idx++) { | |
210e7a43 TG |
1587 | next = next_freelist_entry(s, page, &pos, start, page_limit, |
1588 | freelist_count); | |
4d176711 | 1589 | next = setup_object(s, page, next); |
210e7a43 TG |
1590 | set_freepointer(s, cur, next); |
1591 | cur = next; | |
1592 | } | |
210e7a43 TG |
1593 | set_freepointer(s, cur, NULL); |
1594 | ||
1595 | return true; | |
1596 | } | |
1597 | #else | |
1598 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1599 | { | |
1600 | return 0; | |
1601 | } | |
1602 | static inline void init_freelist_randomization(void) { } | |
1603 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1604 | { | |
1605 | return false; | |
1606 | } | |
1607 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1608 | ||
81819f0f CL |
1609 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1610 | { | |
06428780 | 1611 | struct page *page; |
834f3d11 | 1612 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1613 | gfp_t alloc_gfp; |
4d176711 | 1614 | void *start, *p, *next; |
588f8ba9 | 1615 | int idx, order; |
210e7a43 | 1616 | bool shuffle; |
81819f0f | 1617 | |
7e0528da CL |
1618 | flags &= gfp_allowed_mask; |
1619 | ||
d0164adc | 1620 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1621 | local_irq_enable(); |
1622 | ||
b7a49f0d | 1623 | flags |= s->allocflags; |
e12ba74d | 1624 | |
ba52270d PE |
1625 | /* |
1626 | * Let the initial higher-order allocation fail under memory pressure | |
1627 | * so we fall-back to the minimum order allocation. | |
1628 | */ | |
1629 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1630 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1631 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1632 | |
5dfb4175 | 1633 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1634 | if (unlikely(!page)) { |
1635 | oo = s->min; | |
80c3a998 | 1636 | alloc_gfp = flags; |
65c3376a CL |
1637 | /* |
1638 | * Allocation may have failed due to fragmentation. | |
1639 | * Try a lower order alloc if possible | |
1640 | */ | |
5dfb4175 | 1641 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1642 | if (unlikely(!page)) |
1643 | goto out; | |
1644 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1645 | } |
5a896d9e | 1646 | |
834f3d11 | 1647 | page->objects = oo_objects(oo); |
81819f0f | 1648 | |
1f458cbf | 1649 | order = compound_order(page); |
1b4f59e3 | 1650 | page->slab_cache = s; |
c03f94cc | 1651 | __SetPageSlab(page); |
2f064f34 | 1652 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1653 | SetPageSlabPfmemalloc(page); |
81819f0f | 1654 | |
a7101224 | 1655 | kasan_poison_slab(page); |
81819f0f | 1656 | |
a7101224 | 1657 | start = page_address(page); |
81819f0f | 1658 | |
a7101224 | 1659 | setup_page_debug(s, start, order); |
0316bec2 | 1660 | |
210e7a43 TG |
1661 | shuffle = shuffle_freelist(s, page); |
1662 | ||
1663 | if (!shuffle) { | |
4d176711 AK |
1664 | start = fixup_red_left(s, start); |
1665 | start = setup_object(s, page, start); | |
1666 | page->freelist = start; | |
18e50661 AK |
1667 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { |
1668 | next = p + s->size; | |
1669 | next = setup_object(s, page, next); | |
1670 | set_freepointer(s, p, next); | |
1671 | p = next; | |
1672 | } | |
1673 | set_freepointer(s, p, NULL); | |
81819f0f | 1674 | } |
81819f0f | 1675 | |
e6e82ea1 | 1676 | page->inuse = page->objects; |
8cb0a506 | 1677 | page->frozen = 1; |
588f8ba9 | 1678 | |
81819f0f | 1679 | out: |
d0164adc | 1680 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1681 | local_irq_disable(); |
1682 | if (!page) | |
1683 | return NULL; | |
1684 | ||
7779f212 | 1685 | mod_lruvec_page_state(page, |
588f8ba9 TG |
1686 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? |
1687 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1688 | 1 << oo_order(oo)); | |
1689 | ||
1690 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1691 | ||
81819f0f CL |
1692 | return page; |
1693 | } | |
1694 | ||
588f8ba9 TG |
1695 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1696 | { | |
1697 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
bacdcb34 | 1698 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
1699 | flags &= ~GFP_SLAB_BUG_MASK; |
1700 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
1701 | invalid_mask, &invalid_mask, flags, &flags); | |
65b9de75 | 1702 | dump_stack(); |
588f8ba9 TG |
1703 | } |
1704 | ||
1705 | return allocate_slab(s, | |
1706 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1707 | } | |
1708 | ||
81819f0f CL |
1709 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1710 | { | |
834f3d11 CL |
1711 | int order = compound_order(page); |
1712 | int pages = 1 << order; | |
81819f0f | 1713 | |
becfda68 | 1714 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { |
81819f0f CL |
1715 | void *p; |
1716 | ||
1717 | slab_pad_check(s, page); | |
224a88be CL |
1718 | for_each_object(p, s, page_address(page), |
1719 | page->objects) | |
f7cb1933 | 1720 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1721 | } |
1722 | ||
7779f212 | 1723 | mod_lruvec_page_state(page, |
81819f0f CL |
1724 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? |
1725 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1726 | -pages); |
81819f0f | 1727 | |
072bb0aa | 1728 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1729 | __ClearPageSlab(page); |
1f458cbf | 1730 | |
d4fc5069 | 1731 | page->mapping = NULL; |
1eb5ac64 NP |
1732 | if (current->reclaim_state) |
1733 | current->reclaim_state->reclaimed_slab += pages; | |
27ee57c9 VD |
1734 | memcg_uncharge_slab(page, order, s); |
1735 | __free_pages(page, order); | |
81819f0f CL |
1736 | } |
1737 | ||
1738 | static void rcu_free_slab(struct rcu_head *h) | |
1739 | { | |
bf68c214 | 1740 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 1741 | |
1b4f59e3 | 1742 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1743 | } |
1744 | ||
1745 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1746 | { | |
5f0d5a3a | 1747 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 1748 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
1749 | } else |
1750 | __free_slab(s, page); | |
1751 | } | |
1752 | ||
1753 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1754 | { | |
205ab99d | 1755 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1756 | free_slab(s, page); |
1757 | } | |
1758 | ||
1759 | /* | |
5cc6eee8 | 1760 | * Management of partially allocated slabs. |
81819f0f | 1761 | */ |
1e4dd946 SR |
1762 | static inline void |
1763 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1764 | { |
e95eed57 | 1765 | n->nr_partial++; |
136333d1 | 1766 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1767 | list_add_tail(&page->lru, &n->partial); |
1768 | else | |
1769 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1770 | } |
1771 | ||
1e4dd946 SR |
1772 | static inline void add_partial(struct kmem_cache_node *n, |
1773 | struct page *page, int tail) | |
62e346a8 | 1774 | { |
c65c1877 | 1775 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1776 | __add_partial(n, page, tail); |
1777 | } | |
c65c1877 | 1778 | |
1e4dd946 SR |
1779 | static inline void remove_partial(struct kmem_cache_node *n, |
1780 | struct page *page) | |
1781 | { | |
1782 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1783 | list_del(&page->lru); |
1784 | n->nr_partial--; | |
1e4dd946 SR |
1785 | } |
1786 | ||
81819f0f | 1787 | /* |
7ced3719 CL |
1788 | * Remove slab from the partial list, freeze it and |
1789 | * return the pointer to the freelist. | |
81819f0f | 1790 | * |
497b66f2 | 1791 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1792 | */ |
497b66f2 | 1793 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1794 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1795 | int mode, int *objects) |
81819f0f | 1796 | { |
2cfb7455 CL |
1797 | void *freelist; |
1798 | unsigned long counters; | |
1799 | struct page new; | |
1800 | ||
c65c1877 PZ |
1801 | lockdep_assert_held(&n->list_lock); |
1802 | ||
2cfb7455 CL |
1803 | /* |
1804 | * Zap the freelist and set the frozen bit. | |
1805 | * The old freelist is the list of objects for the | |
1806 | * per cpu allocation list. | |
1807 | */ | |
7ced3719 CL |
1808 | freelist = page->freelist; |
1809 | counters = page->counters; | |
1810 | new.counters = counters; | |
633b0764 | 1811 | *objects = new.objects - new.inuse; |
23910c50 | 1812 | if (mode) { |
7ced3719 | 1813 | new.inuse = page->objects; |
23910c50 PE |
1814 | new.freelist = NULL; |
1815 | } else { | |
1816 | new.freelist = freelist; | |
1817 | } | |
2cfb7455 | 1818 | |
a0132ac0 | 1819 | VM_BUG_ON(new.frozen); |
7ced3719 | 1820 | new.frozen = 1; |
2cfb7455 | 1821 | |
7ced3719 | 1822 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1823 | freelist, counters, |
02d7633f | 1824 | new.freelist, new.counters, |
7ced3719 | 1825 | "acquire_slab")) |
7ced3719 | 1826 | return NULL; |
2cfb7455 CL |
1827 | |
1828 | remove_partial(n, page); | |
7ced3719 | 1829 | WARN_ON(!freelist); |
49e22585 | 1830 | return freelist; |
81819f0f CL |
1831 | } |
1832 | ||
633b0764 | 1833 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1834 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1835 | |
81819f0f | 1836 | /* |
672bba3a | 1837 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1838 | */ |
8ba00bb6 JK |
1839 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1840 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1841 | { |
49e22585 CL |
1842 | struct page *page, *page2; |
1843 | void *object = NULL; | |
e5d9998f | 1844 | unsigned int available = 0; |
633b0764 | 1845 | int objects; |
81819f0f CL |
1846 | |
1847 | /* | |
1848 | * Racy check. If we mistakenly see no partial slabs then we | |
1849 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1850 | * partial slab and there is none available then get_partials() |
1851 | * will return NULL. | |
81819f0f CL |
1852 | */ |
1853 | if (!n || !n->nr_partial) | |
1854 | return NULL; | |
1855 | ||
1856 | spin_lock(&n->list_lock); | |
49e22585 | 1857 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1858 | void *t; |
49e22585 | 1859 | |
8ba00bb6 JK |
1860 | if (!pfmemalloc_match(page, flags)) |
1861 | continue; | |
1862 | ||
633b0764 | 1863 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1864 | if (!t) |
1865 | break; | |
1866 | ||
633b0764 | 1867 | available += objects; |
12d79634 | 1868 | if (!object) { |
49e22585 | 1869 | c->page = page; |
49e22585 | 1870 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1871 | object = t; |
49e22585 | 1872 | } else { |
633b0764 | 1873 | put_cpu_partial(s, page, 0); |
8028dcea | 1874 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1875 | } |
345c905d | 1876 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 1877 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
1878 | break; |
1879 | ||
497b66f2 | 1880 | } |
81819f0f | 1881 | spin_unlock(&n->list_lock); |
497b66f2 | 1882 | return object; |
81819f0f CL |
1883 | } |
1884 | ||
1885 | /* | |
672bba3a | 1886 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1887 | */ |
de3ec035 | 1888 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1889 | struct kmem_cache_cpu *c) |
81819f0f CL |
1890 | { |
1891 | #ifdef CONFIG_NUMA | |
1892 | struct zonelist *zonelist; | |
dd1a239f | 1893 | struct zoneref *z; |
54a6eb5c MG |
1894 | struct zone *zone; |
1895 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1896 | void *object; |
cc9a6c87 | 1897 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1898 | |
1899 | /* | |
672bba3a CL |
1900 | * The defrag ratio allows a configuration of the tradeoffs between |
1901 | * inter node defragmentation and node local allocations. A lower | |
1902 | * defrag_ratio increases the tendency to do local allocations | |
1903 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1904 | * |
672bba3a CL |
1905 | * If the defrag_ratio is set to 0 then kmalloc() always |
1906 | * returns node local objects. If the ratio is higher then kmalloc() | |
1907 | * may return off node objects because partial slabs are obtained | |
1908 | * from other nodes and filled up. | |
81819f0f | 1909 | * |
43efd3ea LP |
1910 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
1911 | * (which makes defrag_ratio = 1000) then every (well almost) | |
1912 | * allocation will first attempt to defrag slab caches on other nodes. | |
1913 | * This means scanning over all nodes to look for partial slabs which | |
1914 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 1915 | * with available objects. |
81819f0f | 1916 | */ |
9824601e CL |
1917 | if (!s->remote_node_defrag_ratio || |
1918 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1919 | return NULL; |
1920 | ||
cc9a6c87 | 1921 | do { |
d26914d1 | 1922 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1923 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1924 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1925 | struct kmem_cache_node *n; | |
1926 | ||
1927 | n = get_node(s, zone_to_nid(zone)); | |
1928 | ||
dee2f8aa | 1929 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1930 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1931 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1932 | if (object) { |
1933 | /* | |
d26914d1 MG |
1934 | * Don't check read_mems_allowed_retry() |
1935 | * here - if mems_allowed was updated in | |
1936 | * parallel, that was a harmless race | |
1937 | * between allocation and the cpuset | |
1938 | * update | |
cc9a6c87 | 1939 | */ |
cc9a6c87 MG |
1940 | return object; |
1941 | } | |
c0ff7453 | 1942 | } |
81819f0f | 1943 | } |
d26914d1 | 1944 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 1945 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
1946 | return NULL; |
1947 | } | |
1948 | ||
1949 | /* | |
1950 | * Get a partial page, lock it and return it. | |
1951 | */ | |
497b66f2 | 1952 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1953 | struct kmem_cache_cpu *c) |
81819f0f | 1954 | { |
497b66f2 | 1955 | void *object; |
a561ce00 JK |
1956 | int searchnode = node; |
1957 | ||
1958 | if (node == NUMA_NO_NODE) | |
1959 | searchnode = numa_mem_id(); | |
1960 | else if (!node_present_pages(node)) | |
1961 | searchnode = node_to_mem_node(node); | |
81819f0f | 1962 | |
8ba00bb6 | 1963 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1964 | if (object || node != NUMA_NO_NODE) |
1965 | return object; | |
81819f0f | 1966 | |
acd19fd1 | 1967 | return get_any_partial(s, flags, c); |
81819f0f CL |
1968 | } |
1969 | ||
8a5ec0ba CL |
1970 | #ifdef CONFIG_PREEMPT |
1971 | /* | |
1972 | * Calculate the next globally unique transaction for disambiguiation | |
1973 | * during cmpxchg. The transactions start with the cpu number and are then | |
1974 | * incremented by CONFIG_NR_CPUS. | |
1975 | */ | |
1976 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1977 | #else | |
1978 | /* | |
1979 | * No preemption supported therefore also no need to check for | |
1980 | * different cpus. | |
1981 | */ | |
1982 | #define TID_STEP 1 | |
1983 | #endif | |
1984 | ||
1985 | static inline unsigned long next_tid(unsigned long tid) | |
1986 | { | |
1987 | return tid + TID_STEP; | |
1988 | } | |
1989 | ||
1990 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1991 | { | |
1992 | return tid % TID_STEP; | |
1993 | } | |
1994 | ||
1995 | static inline unsigned long tid_to_event(unsigned long tid) | |
1996 | { | |
1997 | return tid / TID_STEP; | |
1998 | } | |
1999 | ||
2000 | static inline unsigned int init_tid(int cpu) | |
2001 | { | |
2002 | return cpu; | |
2003 | } | |
2004 | ||
2005 | static inline void note_cmpxchg_failure(const char *n, | |
2006 | const struct kmem_cache *s, unsigned long tid) | |
2007 | { | |
2008 | #ifdef SLUB_DEBUG_CMPXCHG | |
2009 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2010 | ||
f9f58285 | 2011 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
2012 | |
2013 | #ifdef CONFIG_PREEMPT | |
2014 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 2015 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2016 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2017 | else | |
2018 | #endif | |
2019 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2020 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2021 | tid_to_event(tid), tid_to_event(actual_tid)); |
2022 | else | |
f9f58285 | 2023 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2024 | actual_tid, tid, next_tid(tid)); |
2025 | #endif | |
4fdccdfb | 2026 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2027 | } |
2028 | ||
788e1aad | 2029 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2030 | { |
8a5ec0ba CL |
2031 | int cpu; |
2032 | ||
2033 | for_each_possible_cpu(cpu) | |
2034 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 2035 | } |
2cfb7455 | 2036 | |
81819f0f CL |
2037 | /* |
2038 | * Remove the cpu slab | |
2039 | */ | |
d0e0ac97 | 2040 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
d4ff6d35 | 2041 | void *freelist, struct kmem_cache_cpu *c) |
81819f0f | 2042 | { |
2cfb7455 | 2043 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
2044 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
2045 | int lock = 0; | |
2046 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 2047 | void *nextfree; |
136333d1 | 2048 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2049 | struct page new; |
2050 | struct page old; | |
2051 | ||
2052 | if (page->freelist) { | |
84e554e6 | 2053 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2054 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2055 | } |
2056 | ||
894b8788 | 2057 | /* |
2cfb7455 CL |
2058 | * Stage one: Free all available per cpu objects back |
2059 | * to the page freelist while it is still frozen. Leave the | |
2060 | * last one. | |
2061 | * | |
2062 | * There is no need to take the list->lock because the page | |
2063 | * is still frozen. | |
2064 | */ | |
2065 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2066 | void *prior; | |
2067 | unsigned long counters; | |
2068 | ||
2069 | do { | |
2070 | prior = page->freelist; | |
2071 | counters = page->counters; | |
2072 | set_freepointer(s, freelist, prior); | |
2073 | new.counters = counters; | |
2074 | new.inuse--; | |
a0132ac0 | 2075 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2076 | |
1d07171c | 2077 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2078 | prior, counters, |
2079 | freelist, new.counters, | |
2080 | "drain percpu freelist")); | |
2081 | ||
2082 | freelist = nextfree; | |
2083 | } | |
2084 | ||
894b8788 | 2085 | /* |
2cfb7455 CL |
2086 | * Stage two: Ensure that the page is unfrozen while the |
2087 | * list presence reflects the actual number of objects | |
2088 | * during unfreeze. | |
2089 | * | |
2090 | * We setup the list membership and then perform a cmpxchg | |
2091 | * with the count. If there is a mismatch then the page | |
2092 | * is not unfrozen but the page is on the wrong list. | |
2093 | * | |
2094 | * Then we restart the process which may have to remove | |
2095 | * the page from the list that we just put it on again | |
2096 | * because the number of objects in the slab may have | |
2097 | * changed. | |
894b8788 | 2098 | */ |
2cfb7455 | 2099 | redo: |
894b8788 | 2100 | |
2cfb7455 CL |
2101 | old.freelist = page->freelist; |
2102 | old.counters = page->counters; | |
a0132ac0 | 2103 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2104 | |
2cfb7455 CL |
2105 | /* Determine target state of the slab */ |
2106 | new.counters = old.counters; | |
2107 | if (freelist) { | |
2108 | new.inuse--; | |
2109 | set_freepointer(s, freelist, old.freelist); | |
2110 | new.freelist = freelist; | |
2111 | } else | |
2112 | new.freelist = old.freelist; | |
2113 | ||
2114 | new.frozen = 0; | |
2115 | ||
8a5b20ae | 2116 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2117 | m = M_FREE; |
2118 | else if (new.freelist) { | |
2119 | m = M_PARTIAL; | |
2120 | if (!lock) { | |
2121 | lock = 1; | |
2122 | /* | |
8bb4e7a2 | 2123 | * Taking the spinlock removes the possibility |
2cfb7455 CL |
2124 | * that acquire_slab() will see a slab page that |
2125 | * is frozen | |
2126 | */ | |
2127 | spin_lock(&n->list_lock); | |
2128 | } | |
2129 | } else { | |
2130 | m = M_FULL; | |
2131 | if (kmem_cache_debug(s) && !lock) { | |
2132 | lock = 1; | |
2133 | /* | |
2134 | * This also ensures that the scanning of full | |
2135 | * slabs from diagnostic functions will not see | |
2136 | * any frozen slabs. | |
2137 | */ | |
2138 | spin_lock(&n->list_lock); | |
2139 | } | |
2140 | } | |
2141 | ||
2142 | if (l != m) { | |
2cfb7455 | 2143 | if (l == M_PARTIAL) |
2cfb7455 | 2144 | remove_partial(n, page); |
2cfb7455 | 2145 | else if (l == M_FULL) |
c65c1877 | 2146 | remove_full(s, n, page); |
2cfb7455 | 2147 | |
88349a28 | 2148 | if (m == M_PARTIAL) |
2cfb7455 | 2149 | add_partial(n, page, tail); |
88349a28 | 2150 | else if (m == M_FULL) |
2cfb7455 | 2151 | add_full(s, n, page); |
2cfb7455 CL |
2152 | } |
2153 | ||
2154 | l = m; | |
1d07171c | 2155 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2156 | old.freelist, old.counters, |
2157 | new.freelist, new.counters, | |
2158 | "unfreezing slab")) | |
2159 | goto redo; | |
2160 | ||
2cfb7455 CL |
2161 | if (lock) |
2162 | spin_unlock(&n->list_lock); | |
2163 | ||
88349a28 WY |
2164 | if (m == M_PARTIAL) |
2165 | stat(s, tail); | |
2166 | else if (m == M_FULL) | |
2167 | stat(s, DEACTIVATE_FULL); | |
2168 | else if (m == M_FREE) { | |
2cfb7455 CL |
2169 | stat(s, DEACTIVATE_EMPTY); |
2170 | discard_slab(s, page); | |
2171 | stat(s, FREE_SLAB); | |
894b8788 | 2172 | } |
d4ff6d35 WY |
2173 | |
2174 | c->page = NULL; | |
2175 | c->freelist = NULL; | |
81819f0f CL |
2176 | } |
2177 | ||
d24ac77f JK |
2178 | /* |
2179 | * Unfreeze all the cpu partial slabs. | |
2180 | * | |
59a09917 CL |
2181 | * This function must be called with interrupts disabled |
2182 | * for the cpu using c (or some other guarantee must be there | |
2183 | * to guarantee no concurrent accesses). | |
d24ac77f | 2184 | */ |
59a09917 CL |
2185 | static void unfreeze_partials(struct kmem_cache *s, |
2186 | struct kmem_cache_cpu *c) | |
49e22585 | 2187 | { |
345c905d | 2188 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2189 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2190 | struct page *page, *discard_page = NULL; |
49e22585 CL |
2191 | |
2192 | while ((page = c->partial)) { | |
49e22585 CL |
2193 | struct page new; |
2194 | struct page old; | |
2195 | ||
2196 | c->partial = page->next; | |
43d77867 JK |
2197 | |
2198 | n2 = get_node(s, page_to_nid(page)); | |
2199 | if (n != n2) { | |
2200 | if (n) | |
2201 | spin_unlock(&n->list_lock); | |
2202 | ||
2203 | n = n2; | |
2204 | spin_lock(&n->list_lock); | |
2205 | } | |
49e22585 CL |
2206 | |
2207 | do { | |
2208 | ||
2209 | old.freelist = page->freelist; | |
2210 | old.counters = page->counters; | |
a0132ac0 | 2211 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2212 | |
2213 | new.counters = old.counters; | |
2214 | new.freelist = old.freelist; | |
2215 | ||
2216 | new.frozen = 0; | |
2217 | ||
d24ac77f | 2218 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2219 | old.freelist, old.counters, |
2220 | new.freelist, new.counters, | |
2221 | "unfreezing slab")); | |
2222 | ||
8a5b20ae | 2223 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2224 | page->next = discard_page; |
2225 | discard_page = page; | |
43d77867 JK |
2226 | } else { |
2227 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2228 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2229 | } |
2230 | } | |
2231 | ||
2232 | if (n) | |
2233 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2234 | |
2235 | while (discard_page) { | |
2236 | page = discard_page; | |
2237 | discard_page = discard_page->next; | |
2238 | ||
2239 | stat(s, DEACTIVATE_EMPTY); | |
2240 | discard_slab(s, page); | |
2241 | stat(s, FREE_SLAB); | |
2242 | } | |
6dfd1b65 | 2243 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2244 | } |
2245 | ||
2246 | /* | |
9234bae9 WY |
2247 | * Put a page that was just frozen (in __slab_free|get_partial_node) into a |
2248 | * partial page slot if available. | |
49e22585 CL |
2249 | * |
2250 | * If we did not find a slot then simply move all the partials to the | |
2251 | * per node partial list. | |
2252 | */ | |
633b0764 | 2253 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2254 | { |
345c905d | 2255 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2256 | struct page *oldpage; |
2257 | int pages; | |
2258 | int pobjects; | |
2259 | ||
d6e0b7fa | 2260 | preempt_disable(); |
49e22585 CL |
2261 | do { |
2262 | pages = 0; | |
2263 | pobjects = 0; | |
2264 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2265 | ||
2266 | if (oldpage) { | |
2267 | pobjects = oldpage->pobjects; | |
2268 | pages = oldpage->pages; | |
2269 | if (drain && pobjects > s->cpu_partial) { | |
2270 | unsigned long flags; | |
2271 | /* | |
2272 | * partial array is full. Move the existing | |
2273 | * set to the per node partial list. | |
2274 | */ | |
2275 | local_irq_save(flags); | |
59a09917 | 2276 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2277 | local_irq_restore(flags); |
e24fc410 | 2278 | oldpage = NULL; |
49e22585 CL |
2279 | pobjects = 0; |
2280 | pages = 0; | |
8028dcea | 2281 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2282 | } |
2283 | } | |
2284 | ||
2285 | pages++; | |
2286 | pobjects += page->objects - page->inuse; | |
2287 | ||
2288 | page->pages = pages; | |
2289 | page->pobjects = pobjects; | |
2290 | page->next = oldpage; | |
2291 | ||
d0e0ac97 CG |
2292 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2293 | != oldpage); | |
d6e0b7fa VD |
2294 | if (unlikely(!s->cpu_partial)) { |
2295 | unsigned long flags; | |
2296 | ||
2297 | local_irq_save(flags); | |
2298 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2299 | local_irq_restore(flags); | |
2300 | } | |
2301 | preempt_enable(); | |
6dfd1b65 | 2302 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2303 | } |
2304 | ||
dfb4f096 | 2305 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2306 | { |
84e554e6 | 2307 | stat(s, CPUSLAB_FLUSH); |
d4ff6d35 | 2308 | deactivate_slab(s, c->page, c->freelist, c); |
c17dda40 CL |
2309 | |
2310 | c->tid = next_tid(c->tid); | |
81819f0f CL |
2311 | } |
2312 | ||
2313 | /* | |
2314 | * Flush cpu slab. | |
6446faa2 | 2315 | * |
81819f0f CL |
2316 | * Called from IPI handler with interrupts disabled. |
2317 | */ | |
0c710013 | 2318 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2319 | { |
9dfc6e68 | 2320 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2321 | |
1265ef2d WY |
2322 | if (c->page) |
2323 | flush_slab(s, c); | |
49e22585 | 2324 | |
1265ef2d | 2325 | unfreeze_partials(s, c); |
81819f0f CL |
2326 | } |
2327 | ||
2328 | static void flush_cpu_slab(void *d) | |
2329 | { | |
2330 | struct kmem_cache *s = d; | |
81819f0f | 2331 | |
dfb4f096 | 2332 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2333 | } |
2334 | ||
a8364d55 GBY |
2335 | static bool has_cpu_slab(int cpu, void *info) |
2336 | { | |
2337 | struct kmem_cache *s = info; | |
2338 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2339 | ||
a93cf07b | 2340 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2341 | } |
2342 | ||
81819f0f CL |
2343 | static void flush_all(struct kmem_cache *s) |
2344 | { | |
a8364d55 | 2345 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2346 | } |
2347 | ||
a96a87bf SAS |
2348 | /* |
2349 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2350 | * necessary. | |
2351 | */ | |
2352 | static int slub_cpu_dead(unsigned int cpu) | |
2353 | { | |
2354 | struct kmem_cache *s; | |
2355 | unsigned long flags; | |
2356 | ||
2357 | mutex_lock(&slab_mutex); | |
2358 | list_for_each_entry(s, &slab_caches, list) { | |
2359 | local_irq_save(flags); | |
2360 | __flush_cpu_slab(s, cpu); | |
2361 | local_irq_restore(flags); | |
2362 | } | |
2363 | mutex_unlock(&slab_mutex); | |
2364 | return 0; | |
2365 | } | |
2366 | ||
dfb4f096 CL |
2367 | /* |
2368 | * Check if the objects in a per cpu structure fit numa | |
2369 | * locality expectations. | |
2370 | */ | |
57d437d2 | 2371 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2372 | { |
2373 | #ifdef CONFIG_NUMA | |
6159d0f5 | 2374 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2375 | return 0; |
2376 | #endif | |
2377 | return 1; | |
2378 | } | |
2379 | ||
9a02d699 | 2380 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2381 | static int count_free(struct page *page) |
2382 | { | |
2383 | return page->objects - page->inuse; | |
2384 | } | |
2385 | ||
9a02d699 DR |
2386 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2387 | { | |
2388 | return atomic_long_read(&n->total_objects); | |
2389 | } | |
2390 | #endif /* CONFIG_SLUB_DEBUG */ | |
2391 | ||
2392 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2393 | static unsigned long count_partial(struct kmem_cache_node *n, |
2394 | int (*get_count)(struct page *)) | |
2395 | { | |
2396 | unsigned long flags; | |
2397 | unsigned long x = 0; | |
2398 | struct page *page; | |
2399 | ||
2400 | spin_lock_irqsave(&n->list_lock, flags); | |
2401 | list_for_each_entry(page, &n->partial, lru) | |
2402 | x += get_count(page); | |
2403 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2404 | return x; | |
2405 | } | |
9a02d699 | 2406 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2407 | |
781b2ba6 PE |
2408 | static noinline void |
2409 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2410 | { | |
9a02d699 DR |
2411 | #ifdef CONFIG_SLUB_DEBUG |
2412 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2413 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2414 | int node; |
fa45dc25 | 2415 | struct kmem_cache_node *n; |
781b2ba6 | 2416 | |
9a02d699 DR |
2417 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2418 | return; | |
2419 | ||
5b3810e5 VB |
2420 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2421 | nid, gfpflags, &gfpflags); | |
19af27af | 2422 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2423 | s->name, s->object_size, s->size, oo_order(s->oo), |
2424 | oo_order(s->min)); | |
781b2ba6 | 2425 | |
3b0efdfa | 2426 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2427 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2428 | s->name); | |
fa5ec8a1 | 2429 | |
fa45dc25 | 2430 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2431 | unsigned long nr_slabs; |
2432 | unsigned long nr_objs; | |
2433 | unsigned long nr_free; | |
2434 | ||
26c02cf0 AB |
2435 | nr_free = count_partial(n, count_free); |
2436 | nr_slabs = node_nr_slabs(n); | |
2437 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2438 | |
f9f58285 | 2439 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2440 | node, nr_slabs, nr_objs, nr_free); |
2441 | } | |
9a02d699 | 2442 | #endif |
781b2ba6 PE |
2443 | } |
2444 | ||
497b66f2 CL |
2445 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2446 | int node, struct kmem_cache_cpu **pc) | |
2447 | { | |
6faa6833 | 2448 | void *freelist; |
188fd063 CL |
2449 | struct kmem_cache_cpu *c = *pc; |
2450 | struct page *page; | |
497b66f2 | 2451 | |
128227e7 MW |
2452 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2453 | ||
188fd063 | 2454 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2455 | |
188fd063 CL |
2456 | if (freelist) |
2457 | return freelist; | |
2458 | ||
2459 | page = new_slab(s, flags, node); | |
497b66f2 | 2460 | if (page) { |
7c8e0181 | 2461 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2462 | if (c->page) |
2463 | flush_slab(s, c); | |
2464 | ||
2465 | /* | |
2466 | * No other reference to the page yet so we can | |
2467 | * muck around with it freely without cmpxchg | |
2468 | */ | |
6faa6833 | 2469 | freelist = page->freelist; |
497b66f2 CL |
2470 | page->freelist = NULL; |
2471 | ||
2472 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2473 | c->page = page; |
2474 | *pc = c; | |
edde82b6 | 2475 | } |
497b66f2 | 2476 | |
6faa6833 | 2477 | return freelist; |
497b66f2 CL |
2478 | } |
2479 | ||
072bb0aa MG |
2480 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2481 | { | |
2482 | if (unlikely(PageSlabPfmemalloc(page))) | |
2483 | return gfp_pfmemalloc_allowed(gfpflags); | |
2484 | ||
2485 | return true; | |
2486 | } | |
2487 | ||
213eeb9f | 2488 | /* |
d0e0ac97 CG |
2489 | * Check the page->freelist of a page and either transfer the freelist to the |
2490 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2491 | * |
2492 | * The page is still frozen if the return value is not NULL. | |
2493 | * | |
2494 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2495 | * |
2496 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2497 | */ |
2498 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2499 | { | |
2500 | struct page new; | |
2501 | unsigned long counters; | |
2502 | void *freelist; | |
2503 | ||
2504 | do { | |
2505 | freelist = page->freelist; | |
2506 | counters = page->counters; | |
6faa6833 | 2507 | |
213eeb9f | 2508 | new.counters = counters; |
a0132ac0 | 2509 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2510 | |
2511 | new.inuse = page->objects; | |
2512 | new.frozen = freelist != NULL; | |
2513 | ||
d24ac77f | 2514 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2515 | freelist, counters, |
2516 | NULL, new.counters, | |
2517 | "get_freelist")); | |
2518 | ||
2519 | return freelist; | |
2520 | } | |
2521 | ||
81819f0f | 2522 | /* |
894b8788 CL |
2523 | * Slow path. The lockless freelist is empty or we need to perform |
2524 | * debugging duties. | |
2525 | * | |
894b8788 CL |
2526 | * Processing is still very fast if new objects have been freed to the |
2527 | * regular freelist. In that case we simply take over the regular freelist | |
2528 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2529 | * |
894b8788 CL |
2530 | * If that is not working then we fall back to the partial lists. We take the |
2531 | * first element of the freelist as the object to allocate now and move the | |
2532 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2533 | * |
894b8788 | 2534 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2535 | * we need to allocate a new slab. This is the slowest path since it involves |
2536 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2537 | * |
2538 | * Version of __slab_alloc to use when we know that interrupts are | |
2539 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2540 | */ |
a380a3c7 | 2541 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2542 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2543 | { |
6faa6833 | 2544 | void *freelist; |
f6e7def7 | 2545 | struct page *page; |
81819f0f | 2546 | |
f6e7def7 CL |
2547 | page = c->page; |
2548 | if (!page) | |
81819f0f | 2549 | goto new_slab; |
49e22585 | 2550 | redo: |
6faa6833 | 2551 | |
57d437d2 | 2552 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2553 | int searchnode = node; |
2554 | ||
2555 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2556 | searchnode = node_to_mem_node(node); | |
2557 | ||
2558 | if (unlikely(!node_match(page, searchnode))) { | |
2559 | stat(s, ALLOC_NODE_MISMATCH); | |
d4ff6d35 | 2560 | deactivate_slab(s, page, c->freelist, c); |
a561ce00 JK |
2561 | goto new_slab; |
2562 | } | |
fc59c053 | 2563 | } |
6446faa2 | 2564 | |
072bb0aa MG |
2565 | /* |
2566 | * By rights, we should be searching for a slab page that was | |
2567 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2568 | * information when the page leaves the per-cpu allocator | |
2569 | */ | |
2570 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
d4ff6d35 | 2571 | deactivate_slab(s, page, c->freelist, c); |
072bb0aa MG |
2572 | goto new_slab; |
2573 | } | |
2574 | ||
73736e03 | 2575 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2576 | freelist = c->freelist; |
2577 | if (freelist) | |
73736e03 | 2578 | goto load_freelist; |
03e404af | 2579 | |
f6e7def7 | 2580 | freelist = get_freelist(s, page); |
6446faa2 | 2581 | |
6faa6833 | 2582 | if (!freelist) { |
03e404af CL |
2583 | c->page = NULL; |
2584 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2585 | goto new_slab; |
03e404af | 2586 | } |
6446faa2 | 2587 | |
84e554e6 | 2588 | stat(s, ALLOC_REFILL); |
6446faa2 | 2589 | |
894b8788 | 2590 | load_freelist: |
507effea CL |
2591 | /* |
2592 | * freelist is pointing to the list of objects to be used. | |
2593 | * page is pointing to the page from which the objects are obtained. | |
2594 | * That page must be frozen for per cpu allocations to work. | |
2595 | */ | |
a0132ac0 | 2596 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2597 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2598 | c->tid = next_tid(c->tid); |
6faa6833 | 2599 | return freelist; |
81819f0f | 2600 | |
81819f0f | 2601 | new_slab: |
2cfb7455 | 2602 | |
a93cf07b WY |
2603 | if (slub_percpu_partial(c)) { |
2604 | page = c->page = slub_percpu_partial(c); | |
2605 | slub_set_percpu_partial(c, page); | |
49e22585 | 2606 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2607 | goto redo; |
81819f0f CL |
2608 | } |
2609 | ||
188fd063 | 2610 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2611 | |
f4697436 | 2612 | if (unlikely(!freelist)) { |
9a02d699 | 2613 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2614 | return NULL; |
81819f0f | 2615 | } |
2cfb7455 | 2616 | |
f6e7def7 | 2617 | page = c->page; |
5091b74a | 2618 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2619 | goto load_freelist; |
2cfb7455 | 2620 | |
497b66f2 | 2621 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2622 | if (kmem_cache_debug(s) && |
2623 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2624 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2625 | |
d4ff6d35 | 2626 | deactivate_slab(s, page, get_freepointer(s, freelist), c); |
6faa6833 | 2627 | return freelist; |
894b8788 CL |
2628 | } |
2629 | ||
a380a3c7 CL |
2630 | /* |
2631 | * Another one that disabled interrupt and compensates for possible | |
2632 | * cpu changes by refetching the per cpu area pointer. | |
2633 | */ | |
2634 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2635 | unsigned long addr, struct kmem_cache_cpu *c) | |
2636 | { | |
2637 | void *p; | |
2638 | unsigned long flags; | |
2639 | ||
2640 | local_irq_save(flags); | |
2641 | #ifdef CONFIG_PREEMPT | |
2642 | /* | |
2643 | * We may have been preempted and rescheduled on a different | |
2644 | * cpu before disabling interrupts. Need to reload cpu area | |
2645 | * pointer. | |
2646 | */ | |
2647 | c = this_cpu_ptr(s->cpu_slab); | |
2648 | #endif | |
2649 | ||
2650 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2651 | local_irq_restore(flags); | |
2652 | return p; | |
2653 | } | |
2654 | ||
894b8788 CL |
2655 | /* |
2656 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2657 | * have the fastpath folded into their functions. So no function call | |
2658 | * overhead for requests that can be satisfied on the fastpath. | |
2659 | * | |
2660 | * The fastpath works by first checking if the lockless freelist can be used. | |
2661 | * If not then __slab_alloc is called for slow processing. | |
2662 | * | |
2663 | * Otherwise we can simply pick the next object from the lockless free list. | |
2664 | */ | |
2b847c3c | 2665 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2666 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2667 | { |
03ec0ed5 | 2668 | void *object; |
dfb4f096 | 2669 | struct kmem_cache_cpu *c; |
57d437d2 | 2670 | struct page *page; |
8a5ec0ba | 2671 | unsigned long tid; |
1f84260c | 2672 | |
8135be5a VD |
2673 | s = slab_pre_alloc_hook(s, gfpflags); |
2674 | if (!s) | |
773ff60e | 2675 | return NULL; |
8a5ec0ba | 2676 | redo: |
8a5ec0ba CL |
2677 | /* |
2678 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2679 | * enabled. We may switch back and forth between cpus while | |
2680 | * reading from one cpu area. That does not matter as long | |
2681 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2682 | * |
9aabf810 JK |
2683 | * We should guarantee that tid and kmem_cache are retrieved on |
2684 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2685 | * to check if it is matched or not. | |
8a5ec0ba | 2686 | */ |
9aabf810 JK |
2687 | do { |
2688 | tid = this_cpu_read(s->cpu_slab->tid); | |
2689 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2690 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2691 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2692 | |
2693 | /* | |
2694 | * Irqless object alloc/free algorithm used here depends on sequence | |
2695 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2696 | * on c to guarantee that object and page associated with previous tid | |
2697 | * won't be used with current tid. If we fetch tid first, object and | |
2698 | * page could be one associated with next tid and our alloc/free | |
2699 | * request will be failed. In this case, we will retry. So, no problem. | |
2700 | */ | |
2701 | barrier(); | |
8a5ec0ba | 2702 | |
8a5ec0ba CL |
2703 | /* |
2704 | * The transaction ids are globally unique per cpu and per operation on | |
2705 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2706 | * occurs on the right processor and that there was no operation on the | |
2707 | * linked list in between. | |
2708 | */ | |
8a5ec0ba | 2709 | |
9dfc6e68 | 2710 | object = c->freelist; |
57d437d2 | 2711 | page = c->page; |
8eae1492 | 2712 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2713 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2714 | stat(s, ALLOC_SLOWPATH); |
2715 | } else { | |
0ad9500e ED |
2716 | void *next_object = get_freepointer_safe(s, object); |
2717 | ||
8a5ec0ba | 2718 | /* |
25985edc | 2719 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2720 | * operation and if we are on the right processor. |
2721 | * | |
d0e0ac97 CG |
2722 | * The cmpxchg does the following atomically (without lock |
2723 | * semantics!) | |
8a5ec0ba CL |
2724 | * 1. Relocate first pointer to the current per cpu area. |
2725 | * 2. Verify that tid and freelist have not been changed | |
2726 | * 3. If they were not changed replace tid and freelist | |
2727 | * | |
d0e0ac97 CG |
2728 | * Since this is without lock semantics the protection is only |
2729 | * against code executing on this cpu *not* from access by | |
2730 | * other cpus. | |
8a5ec0ba | 2731 | */ |
933393f5 | 2732 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2733 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2734 | object, tid, | |
0ad9500e | 2735 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2736 | |
2737 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2738 | goto redo; | |
2739 | } | |
0ad9500e | 2740 | prefetch_freepointer(s, next_object); |
84e554e6 | 2741 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2742 | } |
8a5ec0ba | 2743 | |
74e2134f | 2744 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2745 | memset(object, 0, s->object_size); |
d07dbea4 | 2746 | |
03ec0ed5 | 2747 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2748 | |
894b8788 | 2749 | return object; |
81819f0f CL |
2750 | } |
2751 | ||
2b847c3c EG |
2752 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2753 | gfp_t gfpflags, unsigned long addr) | |
2754 | { | |
2755 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2756 | } | |
2757 | ||
81819f0f CL |
2758 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2759 | { | |
2b847c3c | 2760 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2761 | |
d0e0ac97 CG |
2762 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2763 | s->size, gfpflags); | |
5b882be4 EGM |
2764 | |
2765 | return ret; | |
81819f0f CL |
2766 | } |
2767 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2768 | ||
0f24f128 | 2769 | #ifdef CONFIG_TRACING |
4a92379b RK |
2770 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2771 | { | |
2b847c3c | 2772 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2773 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0116523c | 2774 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2775 | return ret; |
2776 | } | |
2777 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2778 | #endif |
2779 | ||
81819f0f CL |
2780 | #ifdef CONFIG_NUMA |
2781 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2782 | { | |
2b847c3c | 2783 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2784 | |
ca2b84cb | 2785 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2786 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2787 | |
2788 | return ret; | |
81819f0f CL |
2789 | } |
2790 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2791 | |
0f24f128 | 2792 | #ifdef CONFIG_TRACING |
4a92379b | 2793 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2794 | gfp_t gfpflags, |
4a92379b | 2795 | int node, size_t size) |
5b882be4 | 2796 | { |
2b847c3c | 2797 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2798 | |
2799 | trace_kmalloc_node(_RET_IP_, ret, | |
2800 | size, s->size, gfpflags, node); | |
0316bec2 | 2801 | |
0116523c | 2802 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2803 | return ret; |
5b882be4 | 2804 | } |
4a92379b | 2805 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2806 | #endif |
6dfd1b65 | 2807 | #endif /* CONFIG_NUMA */ |
5b882be4 | 2808 | |
81819f0f | 2809 | /* |
94e4d712 | 2810 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2811 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2812 | * |
894b8788 CL |
2813 | * So we still attempt to reduce cache line usage. Just take the slab |
2814 | * lock and free the item. If there is no additional partial page | |
2815 | * handling required then we can return immediately. | |
81819f0f | 2816 | */ |
894b8788 | 2817 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2818 | void *head, void *tail, int cnt, |
2819 | unsigned long addr) | |
2820 | ||
81819f0f CL |
2821 | { |
2822 | void *prior; | |
2cfb7455 | 2823 | int was_frozen; |
2cfb7455 CL |
2824 | struct page new; |
2825 | unsigned long counters; | |
2826 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2827 | unsigned long uninitialized_var(flags); |
81819f0f | 2828 | |
8a5ec0ba | 2829 | stat(s, FREE_SLOWPATH); |
81819f0f | 2830 | |
19c7ff9e | 2831 | if (kmem_cache_debug(s) && |
282acb43 | 2832 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2833 | return; |
6446faa2 | 2834 | |
2cfb7455 | 2835 | do { |
837d678d JK |
2836 | if (unlikely(n)) { |
2837 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2838 | n = NULL; | |
2839 | } | |
2cfb7455 CL |
2840 | prior = page->freelist; |
2841 | counters = page->counters; | |
81084651 | 2842 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2843 | new.counters = counters; |
2844 | was_frozen = new.frozen; | |
81084651 | 2845 | new.inuse -= cnt; |
837d678d | 2846 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2847 | |
c65c1877 | 2848 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2849 | |
2850 | /* | |
d0e0ac97 CG |
2851 | * Slab was on no list before and will be |
2852 | * partially empty | |
2853 | * We can defer the list move and instead | |
2854 | * freeze it. | |
49e22585 CL |
2855 | */ |
2856 | new.frozen = 1; | |
2857 | ||
c65c1877 | 2858 | } else { /* Needs to be taken off a list */ |
49e22585 | 2859 | |
b455def2 | 2860 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2861 | /* |
2862 | * Speculatively acquire the list_lock. | |
2863 | * If the cmpxchg does not succeed then we may | |
2864 | * drop the list_lock without any processing. | |
2865 | * | |
2866 | * Otherwise the list_lock will synchronize with | |
2867 | * other processors updating the list of slabs. | |
2868 | */ | |
2869 | spin_lock_irqsave(&n->list_lock, flags); | |
2870 | ||
2871 | } | |
2cfb7455 | 2872 | } |
81819f0f | 2873 | |
2cfb7455 CL |
2874 | } while (!cmpxchg_double_slab(s, page, |
2875 | prior, counters, | |
81084651 | 2876 | head, new.counters, |
2cfb7455 | 2877 | "__slab_free")); |
81819f0f | 2878 | |
2cfb7455 | 2879 | if (likely(!n)) { |
49e22585 CL |
2880 | |
2881 | /* | |
2882 | * If we just froze the page then put it onto the | |
2883 | * per cpu partial list. | |
2884 | */ | |
8028dcea | 2885 | if (new.frozen && !was_frozen) { |
49e22585 | 2886 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2887 | stat(s, CPU_PARTIAL_FREE); |
2888 | } | |
49e22585 | 2889 | /* |
2cfb7455 CL |
2890 | * The list lock was not taken therefore no list |
2891 | * activity can be necessary. | |
2892 | */ | |
b455def2 L |
2893 | if (was_frozen) |
2894 | stat(s, FREE_FROZEN); | |
2895 | return; | |
2896 | } | |
81819f0f | 2897 | |
8a5b20ae | 2898 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2899 | goto slab_empty; |
2900 | ||
81819f0f | 2901 | /* |
837d678d JK |
2902 | * Objects left in the slab. If it was not on the partial list before |
2903 | * then add it. | |
81819f0f | 2904 | */ |
345c905d JK |
2905 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2906 | if (kmem_cache_debug(s)) | |
c65c1877 | 2907 | remove_full(s, n, page); |
837d678d JK |
2908 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2909 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2910 | } |
80f08c19 | 2911 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2912 | return; |
2913 | ||
2914 | slab_empty: | |
a973e9dd | 2915 | if (prior) { |
81819f0f | 2916 | /* |
6fbabb20 | 2917 | * Slab on the partial list. |
81819f0f | 2918 | */ |
5cc6eee8 | 2919 | remove_partial(n, page); |
84e554e6 | 2920 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2921 | } else { |
6fbabb20 | 2922 | /* Slab must be on the full list */ |
c65c1877 PZ |
2923 | remove_full(s, n, page); |
2924 | } | |
2cfb7455 | 2925 | |
80f08c19 | 2926 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2927 | stat(s, FREE_SLAB); |
81819f0f | 2928 | discard_slab(s, page); |
81819f0f CL |
2929 | } |
2930 | ||
894b8788 CL |
2931 | /* |
2932 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2933 | * can perform fastpath freeing without additional function calls. | |
2934 | * | |
2935 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2936 | * of this processor. This typically the case if we have just allocated | |
2937 | * the item before. | |
2938 | * | |
2939 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2940 | * with all sorts of special processing. | |
81084651 JDB |
2941 | * |
2942 | * Bulk free of a freelist with several objects (all pointing to the | |
2943 | * same page) possible by specifying head and tail ptr, plus objects | |
2944 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2945 | */ |
80a9201a AP |
2946 | static __always_inline void do_slab_free(struct kmem_cache *s, |
2947 | struct page *page, void *head, void *tail, | |
2948 | int cnt, unsigned long addr) | |
894b8788 | 2949 | { |
81084651 | 2950 | void *tail_obj = tail ? : head; |
dfb4f096 | 2951 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2952 | unsigned long tid; |
8a5ec0ba CL |
2953 | redo: |
2954 | /* | |
2955 | * Determine the currently cpus per cpu slab. | |
2956 | * The cpu may change afterward. However that does not matter since | |
2957 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2958 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2959 | */ |
9aabf810 JK |
2960 | do { |
2961 | tid = this_cpu_read(s->cpu_slab->tid); | |
2962 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2963 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2964 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2965 | |
9aabf810 JK |
2966 | /* Same with comment on barrier() in slab_alloc_node() */ |
2967 | barrier(); | |
c016b0bd | 2968 | |
442b06bc | 2969 | if (likely(page == c->page)) { |
81084651 | 2970 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2971 | |
933393f5 | 2972 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2973 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2974 | c->freelist, tid, | |
81084651 | 2975 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2976 | |
2977 | note_cmpxchg_failure("slab_free", s, tid); | |
2978 | goto redo; | |
2979 | } | |
84e554e6 | 2980 | stat(s, FREE_FASTPATH); |
894b8788 | 2981 | } else |
81084651 | 2982 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2983 | |
894b8788 CL |
2984 | } |
2985 | ||
80a9201a AP |
2986 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2987 | void *head, void *tail, int cnt, | |
2988 | unsigned long addr) | |
2989 | { | |
80a9201a | 2990 | /* |
c3895391 AK |
2991 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
2992 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 2993 | */ |
c3895391 AK |
2994 | if (slab_free_freelist_hook(s, &head, &tail)) |
2995 | do_slab_free(s, page, head, tail, cnt, addr); | |
80a9201a AP |
2996 | } |
2997 | ||
2bd926b4 | 2998 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
2999 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3000 | { | |
3001 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3002 | } | |
3003 | #endif | |
3004 | ||
81819f0f CL |
3005 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3006 | { | |
b9ce5ef4 GC |
3007 | s = cache_from_obj(s, x); |
3008 | if (!s) | |
79576102 | 3009 | return; |
81084651 | 3010 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 3011 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
3012 | } |
3013 | EXPORT_SYMBOL(kmem_cache_free); | |
3014 | ||
d0ecd894 | 3015 | struct detached_freelist { |
fbd02630 | 3016 | struct page *page; |
d0ecd894 JDB |
3017 | void *tail; |
3018 | void *freelist; | |
3019 | int cnt; | |
376bf125 | 3020 | struct kmem_cache *s; |
d0ecd894 | 3021 | }; |
fbd02630 | 3022 | |
d0ecd894 JDB |
3023 | /* |
3024 | * This function progressively scans the array with free objects (with | |
3025 | * a limited look ahead) and extract objects belonging to the same | |
3026 | * page. It builds a detached freelist directly within the given | |
3027 | * page/objects. This can happen without any need for | |
3028 | * synchronization, because the objects are owned by running process. | |
3029 | * The freelist is build up as a single linked list in the objects. | |
3030 | * The idea is, that this detached freelist can then be bulk | |
3031 | * transferred to the real freelist(s), but only requiring a single | |
3032 | * synchronization primitive. Look ahead in the array is limited due | |
3033 | * to performance reasons. | |
3034 | */ | |
376bf125 JDB |
3035 | static inline |
3036 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3037 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3038 | { |
3039 | size_t first_skipped_index = 0; | |
3040 | int lookahead = 3; | |
3041 | void *object; | |
ca257195 | 3042 | struct page *page; |
fbd02630 | 3043 | |
d0ecd894 JDB |
3044 | /* Always re-init detached_freelist */ |
3045 | df->page = NULL; | |
fbd02630 | 3046 | |
d0ecd894 JDB |
3047 | do { |
3048 | object = p[--size]; | |
ca257195 | 3049 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3050 | } while (!object && size); |
3eed034d | 3051 | |
d0ecd894 JDB |
3052 | if (!object) |
3053 | return 0; | |
fbd02630 | 3054 | |
ca257195 JDB |
3055 | page = virt_to_head_page(object); |
3056 | if (!s) { | |
3057 | /* Handle kalloc'ed objects */ | |
3058 | if (unlikely(!PageSlab(page))) { | |
3059 | BUG_ON(!PageCompound(page)); | |
3060 | kfree_hook(object); | |
4949148a | 3061 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3062 | p[size] = NULL; /* mark object processed */ |
3063 | return size; | |
3064 | } | |
3065 | /* Derive kmem_cache from object */ | |
3066 | df->s = page->slab_cache; | |
3067 | } else { | |
3068 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3069 | } | |
376bf125 | 3070 | |
d0ecd894 | 3071 | /* Start new detached freelist */ |
ca257195 | 3072 | df->page = page; |
376bf125 | 3073 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3074 | df->tail = object; |
3075 | df->freelist = object; | |
3076 | p[size] = NULL; /* mark object processed */ | |
3077 | df->cnt = 1; | |
3078 | ||
3079 | while (size) { | |
3080 | object = p[--size]; | |
3081 | if (!object) | |
3082 | continue; /* Skip processed objects */ | |
3083 | ||
3084 | /* df->page is always set at this point */ | |
3085 | if (df->page == virt_to_head_page(object)) { | |
3086 | /* Opportunity build freelist */ | |
376bf125 | 3087 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3088 | df->freelist = object; |
3089 | df->cnt++; | |
3090 | p[size] = NULL; /* mark object processed */ | |
3091 | ||
3092 | continue; | |
fbd02630 | 3093 | } |
d0ecd894 JDB |
3094 | |
3095 | /* Limit look ahead search */ | |
3096 | if (!--lookahead) | |
3097 | break; | |
3098 | ||
3099 | if (!first_skipped_index) | |
3100 | first_skipped_index = size + 1; | |
fbd02630 | 3101 | } |
d0ecd894 JDB |
3102 | |
3103 | return first_skipped_index; | |
3104 | } | |
3105 | ||
d0ecd894 | 3106 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3107 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3108 | { |
3109 | if (WARN_ON(!size)) | |
3110 | return; | |
3111 | ||
3112 | do { | |
3113 | struct detached_freelist df; | |
3114 | ||
3115 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3116 | if (!df.page) |
d0ecd894 JDB |
3117 | continue; |
3118 | ||
376bf125 | 3119 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3120 | } while (likely(size)); |
484748f0 CL |
3121 | } |
3122 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3123 | ||
994eb764 | 3124 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3125 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3126 | void **p) | |
484748f0 | 3127 | { |
994eb764 JDB |
3128 | struct kmem_cache_cpu *c; |
3129 | int i; | |
3130 | ||
03ec0ed5 JDB |
3131 | /* memcg and kmem_cache debug support */ |
3132 | s = slab_pre_alloc_hook(s, flags); | |
3133 | if (unlikely(!s)) | |
3134 | return false; | |
994eb764 JDB |
3135 | /* |
3136 | * Drain objects in the per cpu slab, while disabling local | |
3137 | * IRQs, which protects against PREEMPT and interrupts | |
3138 | * handlers invoking normal fastpath. | |
3139 | */ | |
3140 | local_irq_disable(); | |
3141 | c = this_cpu_ptr(s->cpu_slab); | |
3142 | ||
3143 | for (i = 0; i < size; i++) { | |
3144 | void *object = c->freelist; | |
3145 | ||
ebe909e0 | 3146 | if (unlikely(!object)) { |
ebe909e0 JDB |
3147 | /* |
3148 | * Invoking slow path likely have side-effect | |
3149 | * of re-populating per CPU c->freelist | |
3150 | */ | |
87098373 | 3151 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3152 | _RET_IP_, c); |
87098373 CL |
3153 | if (unlikely(!p[i])) |
3154 | goto error; | |
3155 | ||
ebe909e0 JDB |
3156 | c = this_cpu_ptr(s->cpu_slab); |
3157 | continue; /* goto for-loop */ | |
3158 | } | |
994eb764 JDB |
3159 | c->freelist = get_freepointer(s, object); |
3160 | p[i] = object; | |
3161 | } | |
3162 | c->tid = next_tid(c->tid); | |
3163 | local_irq_enable(); | |
3164 | ||
3165 | /* Clear memory outside IRQ disabled fastpath loop */ | |
3166 | if (unlikely(flags & __GFP_ZERO)) { | |
3167 | int j; | |
3168 | ||
3169 | for (j = 0; j < i; j++) | |
3170 | memset(p[j], 0, s->object_size); | |
3171 | } | |
3172 | ||
03ec0ed5 JDB |
3173 | /* memcg and kmem_cache debug support */ |
3174 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 3175 | return i; |
87098373 | 3176 | error: |
87098373 | 3177 | local_irq_enable(); |
03ec0ed5 JDB |
3178 | slab_post_alloc_hook(s, flags, i, p); |
3179 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 3180 | return 0; |
484748f0 CL |
3181 | } |
3182 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3183 | ||
3184 | ||
81819f0f | 3185 | /* |
672bba3a CL |
3186 | * Object placement in a slab is made very easy because we always start at |
3187 | * offset 0. If we tune the size of the object to the alignment then we can | |
3188 | * get the required alignment by putting one properly sized object after | |
3189 | * another. | |
81819f0f CL |
3190 | * |
3191 | * Notice that the allocation order determines the sizes of the per cpu | |
3192 | * caches. Each processor has always one slab available for allocations. | |
3193 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3194 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3195 | * locking overhead. |
81819f0f CL |
3196 | */ |
3197 | ||
3198 | /* | |
3199 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3200 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3201 | * and increases the number of allocations possible without having to | |
3202 | * take the list_lock. | |
3203 | */ | |
19af27af AD |
3204 | static unsigned int slub_min_order; |
3205 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3206 | static unsigned int slub_min_objects; | |
81819f0f | 3207 | |
81819f0f CL |
3208 | /* |
3209 | * Calculate the order of allocation given an slab object size. | |
3210 | * | |
672bba3a CL |
3211 | * The order of allocation has significant impact on performance and other |
3212 | * system components. Generally order 0 allocations should be preferred since | |
3213 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3214 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3215 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3216 | * would be wasted. |
3217 | * | |
3218 | * In order to reach satisfactory performance we must ensure that a minimum | |
3219 | * number of objects is in one slab. Otherwise we may generate too much | |
3220 | * activity on the partial lists which requires taking the list_lock. This is | |
3221 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3222 | * |
672bba3a CL |
3223 | * slub_max_order specifies the order where we begin to stop considering the |
3224 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3225 | * we try to keep the page order as low as possible. So we accept more waste | |
3226 | * of space in favor of a small page order. | |
81819f0f | 3227 | * |
672bba3a CL |
3228 | * Higher order allocations also allow the placement of more objects in a |
3229 | * slab and thereby reduce object handling overhead. If the user has | |
3230 | * requested a higher mininum order then we start with that one instead of | |
3231 | * the smallest order which will fit the object. | |
81819f0f | 3232 | */ |
19af27af AD |
3233 | static inline unsigned int slab_order(unsigned int size, |
3234 | unsigned int min_objects, unsigned int max_order, | |
9736d2a9 | 3235 | unsigned int fract_leftover) |
81819f0f | 3236 | { |
19af27af AD |
3237 | unsigned int min_order = slub_min_order; |
3238 | unsigned int order; | |
81819f0f | 3239 | |
9736d2a9 | 3240 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3241 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3242 | |
9736d2a9 | 3243 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3244 | order <= max_order; order++) { |
81819f0f | 3245 | |
19af27af AD |
3246 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3247 | unsigned int rem; | |
81819f0f | 3248 | |
9736d2a9 | 3249 | rem = slab_size % size; |
81819f0f | 3250 | |
5e6d444e | 3251 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3252 | break; |
81819f0f | 3253 | } |
672bba3a | 3254 | |
81819f0f CL |
3255 | return order; |
3256 | } | |
3257 | ||
9736d2a9 | 3258 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3259 | { |
19af27af AD |
3260 | unsigned int order; |
3261 | unsigned int min_objects; | |
3262 | unsigned int max_objects; | |
5e6d444e CL |
3263 | |
3264 | /* | |
3265 | * Attempt to find best configuration for a slab. This | |
3266 | * works by first attempting to generate a layout with | |
3267 | * the best configuration and backing off gradually. | |
3268 | * | |
422ff4d7 | 3269 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3270 | * we reduce the minimum objects required in a slab. |
3271 | */ | |
3272 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3273 | if (!min_objects) |
3274 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
9736d2a9 | 3275 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3276 | min_objects = min(min_objects, max_objects); |
3277 | ||
5e6d444e | 3278 | while (min_objects > 1) { |
19af27af AD |
3279 | unsigned int fraction; |
3280 | ||
c124f5b5 | 3281 | fraction = 16; |
5e6d444e CL |
3282 | while (fraction >= 4) { |
3283 | order = slab_order(size, min_objects, | |
9736d2a9 | 3284 | slub_max_order, fraction); |
5e6d444e CL |
3285 | if (order <= slub_max_order) |
3286 | return order; | |
3287 | fraction /= 2; | |
3288 | } | |
5086c389 | 3289 | min_objects--; |
5e6d444e CL |
3290 | } |
3291 | ||
3292 | /* | |
3293 | * We were unable to place multiple objects in a slab. Now | |
3294 | * lets see if we can place a single object there. | |
3295 | */ | |
9736d2a9 | 3296 | order = slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3297 | if (order <= slub_max_order) |
3298 | return order; | |
3299 | ||
3300 | /* | |
3301 | * Doh this slab cannot be placed using slub_max_order. | |
3302 | */ | |
9736d2a9 | 3303 | order = slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3304 | if (order < MAX_ORDER) |
5e6d444e CL |
3305 | return order; |
3306 | return -ENOSYS; | |
3307 | } | |
3308 | ||
5595cffc | 3309 | static void |
4053497d | 3310 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3311 | { |
3312 | n->nr_partial = 0; | |
81819f0f CL |
3313 | spin_lock_init(&n->list_lock); |
3314 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3315 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3316 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3317 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3318 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3319 | #endif |
81819f0f CL |
3320 | } |
3321 | ||
55136592 | 3322 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3323 | { |
6c182dc0 | 3324 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3325 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3326 | |
8a5ec0ba | 3327 | /* |
d4d84fef CM |
3328 | * Must align to double word boundary for the double cmpxchg |
3329 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3330 | */ |
d4d84fef CM |
3331 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3332 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3333 | |
3334 | if (!s->cpu_slab) | |
3335 | return 0; | |
3336 | ||
3337 | init_kmem_cache_cpus(s); | |
4c93c355 | 3338 | |
8a5ec0ba | 3339 | return 1; |
4c93c355 | 3340 | } |
4c93c355 | 3341 | |
51df1142 CL |
3342 | static struct kmem_cache *kmem_cache_node; |
3343 | ||
81819f0f CL |
3344 | /* |
3345 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3346 | * slab on the node for this slabcache. There are no concurrent accesses | |
3347 | * possible. | |
3348 | * | |
721ae22a ZYW |
3349 | * Note that this function only works on the kmem_cache_node |
3350 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3351 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3352 | */ |
55136592 | 3353 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3354 | { |
3355 | struct page *page; | |
3356 | struct kmem_cache_node *n; | |
3357 | ||
51df1142 | 3358 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3359 | |
51df1142 | 3360 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3361 | |
3362 | BUG_ON(!page); | |
a2f92ee7 | 3363 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3364 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3365 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3366 | } |
3367 | ||
81819f0f CL |
3368 | n = page->freelist; |
3369 | BUG_ON(!n); | |
8ab1372f | 3370 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3371 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3372 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3373 | #endif |
12b22386 | 3374 | n = kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
505f5dcb | 3375 | GFP_KERNEL); |
12b22386 AK |
3376 | page->freelist = get_freepointer(kmem_cache_node, n); |
3377 | page->inuse = 1; | |
3378 | page->frozen = 0; | |
3379 | kmem_cache_node->node[node] = n; | |
4053497d | 3380 | init_kmem_cache_node(n); |
51df1142 | 3381 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3382 | |
67b6c900 | 3383 | /* |
1e4dd946 SR |
3384 | * No locks need to be taken here as it has just been |
3385 | * initialized and there is no concurrent access. | |
67b6c900 | 3386 | */ |
1e4dd946 | 3387 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3388 | } |
3389 | ||
3390 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3391 | { | |
3392 | int node; | |
fa45dc25 | 3393 | struct kmem_cache_node *n; |
81819f0f | 3394 | |
fa45dc25 | 3395 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3396 | s->node[node] = NULL; |
ea37df54 | 3397 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3398 | } |
3399 | } | |
3400 | ||
52b4b950 DS |
3401 | void __kmem_cache_release(struct kmem_cache *s) |
3402 | { | |
210e7a43 | 3403 | cache_random_seq_destroy(s); |
52b4b950 DS |
3404 | free_percpu(s->cpu_slab); |
3405 | free_kmem_cache_nodes(s); | |
3406 | } | |
3407 | ||
55136592 | 3408 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3409 | { |
3410 | int node; | |
81819f0f | 3411 | |
f64dc58c | 3412 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3413 | struct kmem_cache_node *n; |
3414 | ||
73367bd8 | 3415 | if (slab_state == DOWN) { |
55136592 | 3416 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3417 | continue; |
3418 | } | |
51df1142 | 3419 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3420 | GFP_KERNEL, node); |
81819f0f | 3421 | |
73367bd8 AD |
3422 | if (!n) { |
3423 | free_kmem_cache_nodes(s); | |
3424 | return 0; | |
81819f0f | 3425 | } |
73367bd8 | 3426 | |
4053497d | 3427 | init_kmem_cache_node(n); |
ea37df54 | 3428 | s->node[node] = n; |
81819f0f CL |
3429 | } |
3430 | return 1; | |
3431 | } | |
81819f0f | 3432 | |
c0bdb232 | 3433 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3434 | { |
3435 | if (min < MIN_PARTIAL) | |
3436 | min = MIN_PARTIAL; | |
3437 | else if (min > MAX_PARTIAL) | |
3438 | min = MAX_PARTIAL; | |
3439 | s->min_partial = min; | |
3440 | } | |
3441 | ||
e6d0e1dc WY |
3442 | static void set_cpu_partial(struct kmem_cache *s) |
3443 | { | |
3444 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3445 | /* | |
3446 | * cpu_partial determined the maximum number of objects kept in the | |
3447 | * per cpu partial lists of a processor. | |
3448 | * | |
3449 | * Per cpu partial lists mainly contain slabs that just have one | |
3450 | * object freed. If they are used for allocation then they can be | |
3451 | * filled up again with minimal effort. The slab will never hit the | |
3452 | * per node partial lists and therefore no locking will be required. | |
3453 | * | |
3454 | * This setting also determines | |
3455 | * | |
3456 | * A) The number of objects from per cpu partial slabs dumped to the | |
3457 | * per node list when we reach the limit. | |
3458 | * B) The number of objects in cpu partial slabs to extract from the | |
3459 | * per node list when we run out of per cpu objects. We only fetch | |
3460 | * 50% to keep some capacity around for frees. | |
3461 | */ | |
3462 | if (!kmem_cache_has_cpu_partial(s)) | |
3463 | s->cpu_partial = 0; | |
3464 | else if (s->size >= PAGE_SIZE) | |
3465 | s->cpu_partial = 2; | |
3466 | else if (s->size >= 1024) | |
3467 | s->cpu_partial = 6; | |
3468 | else if (s->size >= 256) | |
3469 | s->cpu_partial = 13; | |
3470 | else | |
3471 | s->cpu_partial = 30; | |
3472 | #endif | |
3473 | } | |
3474 | ||
81819f0f CL |
3475 | /* |
3476 | * calculate_sizes() determines the order and the distribution of data within | |
3477 | * a slab object. | |
3478 | */ | |
06b285dc | 3479 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3480 | { |
d50112ed | 3481 | slab_flags_t flags = s->flags; |
be4a7988 | 3482 | unsigned int size = s->object_size; |
19af27af | 3483 | unsigned int order; |
81819f0f | 3484 | |
d8b42bf5 CL |
3485 | /* |
3486 | * Round up object size to the next word boundary. We can only | |
3487 | * place the free pointer at word boundaries and this determines | |
3488 | * the possible location of the free pointer. | |
3489 | */ | |
3490 | size = ALIGN(size, sizeof(void *)); | |
3491 | ||
3492 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3493 | /* |
3494 | * Determine if we can poison the object itself. If the user of | |
3495 | * the slab may touch the object after free or before allocation | |
3496 | * then we should never poison the object itself. | |
3497 | */ | |
5f0d5a3a | 3498 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3499 | !s->ctor) |
81819f0f CL |
3500 | s->flags |= __OBJECT_POISON; |
3501 | else | |
3502 | s->flags &= ~__OBJECT_POISON; | |
3503 | ||
81819f0f CL |
3504 | |
3505 | /* | |
672bba3a | 3506 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3507 | * end of the object and the free pointer. If not then add an |
672bba3a | 3508 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3509 | */ |
3b0efdfa | 3510 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3511 | size += sizeof(void *); |
41ecc55b | 3512 | #endif |
81819f0f CL |
3513 | |
3514 | /* | |
672bba3a CL |
3515 | * With that we have determined the number of bytes in actual use |
3516 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3517 | */ |
3518 | s->inuse = size; | |
3519 | ||
5f0d5a3a | 3520 | if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
c59def9f | 3521 | s->ctor)) { |
81819f0f CL |
3522 | /* |
3523 | * Relocate free pointer after the object if it is not | |
3524 | * permitted to overwrite the first word of the object on | |
3525 | * kmem_cache_free. | |
3526 | * | |
3527 | * This is the case if we do RCU, have a constructor or | |
3528 | * destructor or are poisoning the objects. | |
3529 | */ | |
3530 | s->offset = size; | |
3531 | size += sizeof(void *); | |
3532 | } | |
3533 | ||
c12b3c62 | 3534 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3535 | if (flags & SLAB_STORE_USER) |
3536 | /* | |
3537 | * Need to store information about allocs and frees after | |
3538 | * the object. | |
3539 | */ | |
3540 | size += 2 * sizeof(struct track); | |
80a9201a | 3541 | #endif |
81819f0f | 3542 | |
80a9201a AP |
3543 | kasan_cache_create(s, &size, &s->flags); |
3544 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3545 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3546 | /* |
3547 | * Add some empty padding so that we can catch | |
3548 | * overwrites from earlier objects rather than let | |
3549 | * tracking information or the free pointer be | |
0211a9c8 | 3550 | * corrupted if a user writes before the start |
81819f0f CL |
3551 | * of the object. |
3552 | */ | |
3553 | size += sizeof(void *); | |
d86bd1be JK |
3554 | |
3555 | s->red_left_pad = sizeof(void *); | |
3556 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3557 | size += s->red_left_pad; | |
3558 | } | |
41ecc55b | 3559 | #endif |
672bba3a | 3560 | |
81819f0f CL |
3561 | /* |
3562 | * SLUB stores one object immediately after another beginning from | |
3563 | * offset 0. In order to align the objects we have to simply size | |
3564 | * each object to conform to the alignment. | |
3565 | */ | |
45906855 | 3566 | size = ALIGN(size, s->align); |
81819f0f | 3567 | s->size = size; |
06b285dc CL |
3568 | if (forced_order >= 0) |
3569 | order = forced_order; | |
3570 | else | |
9736d2a9 | 3571 | order = calculate_order(size); |
81819f0f | 3572 | |
19af27af | 3573 | if ((int)order < 0) |
81819f0f CL |
3574 | return 0; |
3575 | ||
b7a49f0d | 3576 | s->allocflags = 0; |
834f3d11 | 3577 | if (order) |
b7a49f0d CL |
3578 | s->allocflags |= __GFP_COMP; |
3579 | ||
3580 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3581 | s->allocflags |= GFP_DMA; |
b7a49f0d | 3582 | |
6d6ea1e9 NB |
3583 | if (s->flags & SLAB_CACHE_DMA32) |
3584 | s->allocflags |= GFP_DMA32; | |
3585 | ||
b7a49f0d CL |
3586 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
3587 | s->allocflags |= __GFP_RECLAIMABLE; | |
3588 | ||
81819f0f CL |
3589 | /* |
3590 | * Determine the number of objects per slab | |
3591 | */ | |
9736d2a9 MW |
3592 | s->oo = oo_make(order, size); |
3593 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
3594 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3595 | s->max = s->oo; | |
81819f0f | 3596 | |
834f3d11 | 3597 | return !!oo_objects(s->oo); |
81819f0f CL |
3598 | } |
3599 | ||
d50112ed | 3600 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3601 | { |
8a13a4cc | 3602 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
2482ddec KC |
3603 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3604 | s->random = get_random_long(); | |
3605 | #endif | |
81819f0f | 3606 | |
06b285dc | 3607 | if (!calculate_sizes(s, -1)) |
81819f0f | 3608 | goto error; |
3de47213 DR |
3609 | if (disable_higher_order_debug) { |
3610 | /* | |
3611 | * Disable debugging flags that store metadata if the min slab | |
3612 | * order increased. | |
3613 | */ | |
3b0efdfa | 3614 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3615 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3616 | s->offset = 0; | |
3617 | if (!calculate_sizes(s, -1)) | |
3618 | goto error; | |
3619 | } | |
3620 | } | |
81819f0f | 3621 | |
2565409f HC |
3622 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3623 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3624 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3625 | /* Enable fast mode */ |
3626 | s->flags |= __CMPXCHG_DOUBLE; | |
3627 | #endif | |
3628 | ||
3b89d7d8 DR |
3629 | /* |
3630 | * The larger the object size is, the more pages we want on the partial | |
3631 | * list to avoid pounding the page allocator excessively. | |
3632 | */ | |
49e22585 CL |
3633 | set_min_partial(s, ilog2(s->size) / 2); |
3634 | ||
e6d0e1dc | 3635 | set_cpu_partial(s); |
49e22585 | 3636 | |
81819f0f | 3637 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3638 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3639 | #endif |
210e7a43 TG |
3640 | |
3641 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3642 | if (slab_state >= UP) { | |
3643 | if (init_cache_random_seq(s)) | |
3644 | goto error; | |
3645 | } | |
3646 | ||
55136592 | 3647 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3648 | goto error; |
81819f0f | 3649 | |
55136592 | 3650 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3651 | return 0; |
ff12059e | 3652 | |
4c93c355 | 3653 | free_kmem_cache_nodes(s); |
81819f0f CL |
3654 | error: |
3655 | if (flags & SLAB_PANIC) | |
44065b2e AD |
3656 | panic("Cannot create slab %s size=%u realsize=%u order=%u offset=%u flags=%lx\n", |
3657 | s->name, s->size, s->size, | |
4fd0b46e | 3658 | oo_order(s->oo), s->offset, (unsigned long)flags); |
278b1bb1 | 3659 | return -EINVAL; |
81819f0f | 3660 | } |
81819f0f | 3661 | |
33b12c38 CL |
3662 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3663 | const char *text) | |
3664 | { | |
3665 | #ifdef CONFIG_SLUB_DEBUG | |
3666 | void *addr = page_address(page); | |
3667 | void *p; | |
0684e652 | 3668 | unsigned long *map = bitmap_zalloc(page->objects, GFP_ATOMIC); |
bbd7d57b ED |
3669 | if (!map) |
3670 | return; | |
945cf2b6 | 3671 | slab_err(s, page, text, s->name); |
33b12c38 | 3672 | slab_lock(page); |
33b12c38 | 3673 | |
5f80b13a | 3674 | get_map(s, page, map); |
33b12c38 CL |
3675 | for_each_object(p, s, addr, page->objects) { |
3676 | ||
3677 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3678 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3679 | print_tracking(s, p); |
3680 | } | |
3681 | } | |
3682 | slab_unlock(page); | |
0684e652 | 3683 | bitmap_free(map); |
33b12c38 CL |
3684 | #endif |
3685 | } | |
3686 | ||
81819f0f | 3687 | /* |
599870b1 | 3688 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3689 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3690 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3691 | */ |
599870b1 | 3692 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3693 | { |
60398923 | 3694 | LIST_HEAD(discard); |
81819f0f CL |
3695 | struct page *page, *h; |
3696 | ||
52b4b950 DS |
3697 | BUG_ON(irqs_disabled()); |
3698 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3699 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3700 | if (!page->inuse) { |
52b4b950 | 3701 | remove_partial(n, page); |
60398923 | 3702 | list_add(&page->lru, &discard); |
33b12c38 CL |
3703 | } else { |
3704 | list_slab_objects(s, page, | |
52b4b950 | 3705 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3706 | } |
33b12c38 | 3707 | } |
52b4b950 | 3708 | spin_unlock_irq(&n->list_lock); |
60398923 CW |
3709 | |
3710 | list_for_each_entry_safe(page, h, &discard, lru) | |
3711 | discard_slab(s, page); | |
81819f0f CL |
3712 | } |
3713 | ||
f9e13c0a SB |
3714 | bool __kmem_cache_empty(struct kmem_cache *s) |
3715 | { | |
3716 | int node; | |
3717 | struct kmem_cache_node *n; | |
3718 | ||
3719 | for_each_kmem_cache_node(s, node, n) | |
3720 | if (n->nr_partial || slabs_node(s, node)) | |
3721 | return false; | |
3722 | return true; | |
3723 | } | |
3724 | ||
81819f0f | 3725 | /* |
672bba3a | 3726 | * Release all resources used by a slab cache. |
81819f0f | 3727 | */ |
52b4b950 | 3728 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3729 | { |
3730 | int node; | |
fa45dc25 | 3731 | struct kmem_cache_node *n; |
81819f0f CL |
3732 | |
3733 | flush_all(s); | |
81819f0f | 3734 | /* Attempt to free all objects */ |
fa45dc25 | 3735 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3736 | free_partial(s, n); |
3737 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3738 | return 1; |
3739 | } | |
bf5eb3de | 3740 | sysfs_slab_remove(s); |
81819f0f CL |
3741 | return 0; |
3742 | } | |
3743 | ||
81819f0f CL |
3744 | /******************************************************************** |
3745 | * Kmalloc subsystem | |
3746 | *******************************************************************/ | |
3747 | ||
81819f0f CL |
3748 | static int __init setup_slub_min_order(char *str) |
3749 | { | |
19af27af | 3750 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
3751 | |
3752 | return 1; | |
3753 | } | |
3754 | ||
3755 | __setup("slub_min_order=", setup_slub_min_order); | |
3756 | ||
3757 | static int __init setup_slub_max_order(char *str) | |
3758 | { | |
19af27af AD |
3759 | get_option(&str, (int *)&slub_max_order); |
3760 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
3761 | |
3762 | return 1; | |
3763 | } | |
3764 | ||
3765 | __setup("slub_max_order=", setup_slub_max_order); | |
3766 | ||
3767 | static int __init setup_slub_min_objects(char *str) | |
3768 | { | |
19af27af | 3769 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
3770 | |
3771 | return 1; | |
3772 | } | |
3773 | ||
3774 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3775 | ||
81819f0f CL |
3776 | void *__kmalloc(size_t size, gfp_t flags) |
3777 | { | |
aadb4bc4 | 3778 | struct kmem_cache *s; |
5b882be4 | 3779 | void *ret; |
81819f0f | 3780 | |
95a05b42 | 3781 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3782 | return kmalloc_large(size, flags); |
aadb4bc4 | 3783 | |
2c59dd65 | 3784 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3785 | |
3786 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3787 | return s; |
3788 | ||
2b847c3c | 3789 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3790 | |
ca2b84cb | 3791 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3792 | |
0116523c | 3793 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3794 | |
5b882be4 | 3795 | return ret; |
81819f0f CL |
3796 | } |
3797 | EXPORT_SYMBOL(__kmalloc); | |
3798 | ||
5d1f57e4 | 3799 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3800 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3801 | { | |
b1eeab67 | 3802 | struct page *page; |
e4f7c0b4 | 3803 | void *ptr = NULL; |
f619cfe1 | 3804 | |
75f296d9 | 3805 | flags |= __GFP_COMP; |
4949148a | 3806 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3807 | if (page) |
e4f7c0b4 CM |
3808 | ptr = page_address(page); |
3809 | ||
0116523c | 3810 | return kmalloc_large_node_hook(ptr, size, flags); |
f619cfe1 CL |
3811 | } |
3812 | ||
81819f0f CL |
3813 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3814 | { | |
aadb4bc4 | 3815 | struct kmem_cache *s; |
5b882be4 | 3816 | void *ret; |
81819f0f | 3817 | |
95a05b42 | 3818 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3819 | ret = kmalloc_large_node(size, flags, node); |
3820 | ||
ca2b84cb EGM |
3821 | trace_kmalloc_node(_RET_IP_, ret, |
3822 | size, PAGE_SIZE << get_order(size), | |
3823 | flags, node); | |
5b882be4 EGM |
3824 | |
3825 | return ret; | |
3826 | } | |
aadb4bc4 | 3827 | |
2c59dd65 | 3828 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3829 | |
3830 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3831 | return s; |
3832 | ||
2b847c3c | 3833 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3834 | |
ca2b84cb | 3835 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3836 | |
0116523c | 3837 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3838 | |
5b882be4 | 3839 | return ret; |
81819f0f CL |
3840 | } |
3841 | EXPORT_SYMBOL(__kmalloc_node); | |
6dfd1b65 | 3842 | #endif /* CONFIG_NUMA */ |
81819f0f | 3843 | |
ed18adc1 KC |
3844 | #ifdef CONFIG_HARDENED_USERCOPY |
3845 | /* | |
afcc90f8 KC |
3846 | * Rejects incorrectly sized objects and objects that are to be copied |
3847 | * to/from userspace but do not fall entirely within the containing slab | |
3848 | * cache's usercopy region. | |
ed18adc1 KC |
3849 | * |
3850 | * Returns NULL if check passes, otherwise const char * to name of cache | |
3851 | * to indicate an error. | |
3852 | */ | |
f4e6e289 KC |
3853 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
3854 | bool to_user) | |
ed18adc1 KC |
3855 | { |
3856 | struct kmem_cache *s; | |
44065b2e | 3857 | unsigned int offset; |
ed18adc1 KC |
3858 | size_t object_size; |
3859 | ||
96fedce2 AK |
3860 | ptr = kasan_reset_tag(ptr); |
3861 | ||
ed18adc1 KC |
3862 | /* Find object and usable object size. */ |
3863 | s = page->slab_cache; | |
ed18adc1 KC |
3864 | |
3865 | /* Reject impossible pointers. */ | |
3866 | if (ptr < page_address(page)) | |
f4e6e289 KC |
3867 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
3868 | to_user, 0, n); | |
ed18adc1 KC |
3869 | |
3870 | /* Find offset within object. */ | |
3871 | offset = (ptr - page_address(page)) % s->size; | |
3872 | ||
3873 | /* Adjust for redzone and reject if within the redzone. */ | |
3874 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) { | |
3875 | if (offset < s->red_left_pad) | |
f4e6e289 KC |
3876 | usercopy_abort("SLUB object in left red zone", |
3877 | s->name, to_user, offset, n); | |
ed18adc1 KC |
3878 | offset -= s->red_left_pad; |
3879 | } | |
3880 | ||
afcc90f8 KC |
3881 | /* Allow address range falling entirely within usercopy region. */ |
3882 | if (offset >= s->useroffset && | |
3883 | offset - s->useroffset <= s->usersize && | |
3884 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 3885 | return; |
ed18adc1 | 3886 | |
afcc90f8 KC |
3887 | /* |
3888 | * If the copy is still within the allocated object, produce | |
3889 | * a warning instead of rejecting the copy. This is intended | |
3890 | * to be a temporary method to find any missing usercopy | |
3891 | * whitelists. | |
3892 | */ | |
3893 | object_size = slab_ksize(s); | |
2d891fbc KC |
3894 | if (usercopy_fallback && |
3895 | offset <= object_size && n <= object_size - offset) { | |
afcc90f8 KC |
3896 | usercopy_warn("SLUB object", s->name, to_user, offset, n); |
3897 | return; | |
3898 | } | |
ed18adc1 | 3899 | |
f4e6e289 | 3900 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
3901 | } |
3902 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
3903 | ||
0316bec2 | 3904 | static size_t __ksize(const void *object) |
81819f0f | 3905 | { |
272c1d21 | 3906 | struct page *page; |
81819f0f | 3907 | |
ef8b4520 | 3908 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3909 | return 0; |
3910 | ||
294a80a8 | 3911 | page = virt_to_head_page(object); |
294a80a8 | 3912 | |
76994412 PE |
3913 | if (unlikely(!PageSlab(page))) { |
3914 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3915 | return PAGE_SIZE << compound_order(page); |
76994412 | 3916 | } |
81819f0f | 3917 | |
1b4f59e3 | 3918 | return slab_ksize(page->slab_cache); |
81819f0f | 3919 | } |
0316bec2 AR |
3920 | |
3921 | size_t ksize(const void *object) | |
3922 | { | |
3923 | size_t size = __ksize(object); | |
3924 | /* We assume that ksize callers could use whole allocated area, | |
4ebb31a4 AP |
3925 | * so we need to unpoison this area. |
3926 | */ | |
3927 | kasan_unpoison_shadow(object, size); | |
0316bec2 AR |
3928 | return size; |
3929 | } | |
b1aabecd | 3930 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3931 | |
3932 | void kfree(const void *x) | |
3933 | { | |
81819f0f | 3934 | struct page *page; |
5bb983b0 | 3935 | void *object = (void *)x; |
81819f0f | 3936 | |
2121db74 PE |
3937 | trace_kfree(_RET_IP_, x); |
3938 | ||
2408c550 | 3939 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3940 | return; |
3941 | ||
b49af68f | 3942 | page = virt_to_head_page(x); |
aadb4bc4 | 3943 | if (unlikely(!PageSlab(page))) { |
0937502a | 3944 | BUG_ON(!PageCompound(page)); |
47adccce | 3945 | kfree_hook(object); |
4949148a | 3946 | __free_pages(page, compound_order(page)); |
aadb4bc4 CL |
3947 | return; |
3948 | } | |
81084651 | 3949 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3950 | } |
3951 | EXPORT_SYMBOL(kfree); | |
3952 | ||
832f37f5 VD |
3953 | #define SHRINK_PROMOTE_MAX 32 |
3954 | ||
2086d26a | 3955 | /* |
832f37f5 VD |
3956 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3957 | * up most to the head of the partial lists. New allocations will then | |
3958 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3959 | * |
3960 | * The slabs with the least items are placed last. This results in them | |
3961 | * being allocated from last increasing the chance that the last objects | |
3962 | * are freed in them. | |
2086d26a | 3963 | */ |
c9fc5864 | 3964 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
3965 | { |
3966 | int node; | |
3967 | int i; | |
3968 | struct kmem_cache_node *n; | |
3969 | struct page *page; | |
3970 | struct page *t; | |
832f37f5 VD |
3971 | struct list_head discard; |
3972 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3973 | unsigned long flags; |
ce3712d7 | 3974 | int ret = 0; |
2086d26a | 3975 | |
2086d26a | 3976 | flush_all(s); |
fa45dc25 | 3977 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3978 | INIT_LIST_HEAD(&discard); |
3979 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3980 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3981 | |
3982 | spin_lock_irqsave(&n->list_lock, flags); | |
3983 | ||
3984 | /* | |
832f37f5 | 3985 | * Build lists of slabs to discard or promote. |
2086d26a | 3986 | * |
672bba3a CL |
3987 | * Note that concurrent frees may occur while we hold the |
3988 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3989 | */ |
3990 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3991 | int free = page->objects - page->inuse; |
3992 | ||
3993 | /* Do not reread page->inuse */ | |
3994 | barrier(); | |
3995 | ||
3996 | /* We do not keep full slabs on the list */ | |
3997 | BUG_ON(free <= 0); | |
3998 | ||
3999 | if (free == page->objects) { | |
4000 | list_move(&page->lru, &discard); | |
69cb8e6b | 4001 | n->nr_partial--; |
832f37f5 VD |
4002 | } else if (free <= SHRINK_PROMOTE_MAX) |
4003 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
4004 | } |
4005 | ||
2086d26a | 4006 | /* |
832f37f5 VD |
4007 | * Promote the slabs filled up most to the head of the |
4008 | * partial list. | |
2086d26a | 4009 | */ |
832f37f5 VD |
4010 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4011 | list_splice(promote + i, &n->partial); | |
2086d26a | 4012 | |
2086d26a | 4013 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4014 | |
4015 | /* Release empty slabs */ | |
832f37f5 | 4016 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 4017 | discard_slab(s, page); |
ce3712d7 VD |
4018 | |
4019 | if (slabs_node(s, node)) | |
4020 | ret = 1; | |
2086d26a CL |
4021 | } |
4022 | ||
ce3712d7 | 4023 | return ret; |
2086d26a | 4024 | } |
2086d26a | 4025 | |
c9fc5864 | 4026 | #ifdef CONFIG_MEMCG |
01fb58bc TH |
4027 | static void kmemcg_cache_deact_after_rcu(struct kmem_cache *s) |
4028 | { | |
50862ce7 TH |
4029 | /* |
4030 | * Called with all the locks held after a sched RCU grace period. | |
4031 | * Even if @s becomes empty after shrinking, we can't know that @s | |
4032 | * doesn't have allocations already in-flight and thus can't | |
4033 | * destroy @s until the associated memcg is released. | |
4034 | * | |
4035 | * However, let's remove the sysfs files for empty caches here. | |
4036 | * Each cache has a lot of interface files which aren't | |
4037 | * particularly useful for empty draining caches; otherwise, we can | |
4038 | * easily end up with millions of unnecessary sysfs files on | |
4039 | * systems which have a lot of memory and transient cgroups. | |
4040 | */ | |
4041 | if (!__kmem_cache_shrink(s)) | |
4042 | sysfs_slab_remove(s); | |
01fb58bc TH |
4043 | } |
4044 | ||
c9fc5864 TH |
4045 | void __kmemcg_cache_deactivate(struct kmem_cache *s) |
4046 | { | |
4047 | /* | |
4048 | * Disable empty slabs caching. Used to avoid pinning offline | |
4049 | * memory cgroups by kmem pages that can be freed. | |
4050 | */ | |
e6d0e1dc | 4051 | slub_set_cpu_partial(s, 0); |
c9fc5864 TH |
4052 | s->min_partial = 0; |
4053 | ||
4054 | /* | |
4055 | * s->cpu_partial is checked locklessly (see put_cpu_partial), so | |
01fb58bc | 4056 | * we have to make sure the change is visible before shrinking. |
c9fc5864 | 4057 | */ |
01fb58bc | 4058 | slab_deactivate_memcg_cache_rcu_sched(s, kmemcg_cache_deact_after_rcu); |
c9fc5864 | 4059 | } |
6dfd1b65 | 4060 | #endif /* CONFIG_MEMCG */ |
c9fc5864 | 4061 | |
b9049e23 YG |
4062 | static int slab_mem_going_offline_callback(void *arg) |
4063 | { | |
4064 | struct kmem_cache *s; | |
4065 | ||
18004c5d | 4066 | mutex_lock(&slab_mutex); |
b9049e23 | 4067 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4068 | __kmem_cache_shrink(s); |
18004c5d | 4069 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4070 | |
4071 | return 0; | |
4072 | } | |
4073 | ||
4074 | static void slab_mem_offline_callback(void *arg) | |
4075 | { | |
4076 | struct kmem_cache_node *n; | |
4077 | struct kmem_cache *s; | |
4078 | struct memory_notify *marg = arg; | |
4079 | int offline_node; | |
4080 | ||
b9d5ab25 | 4081 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4082 | |
4083 | /* | |
4084 | * If the node still has available memory. we need kmem_cache_node | |
4085 | * for it yet. | |
4086 | */ | |
4087 | if (offline_node < 0) | |
4088 | return; | |
4089 | ||
18004c5d | 4090 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4091 | list_for_each_entry(s, &slab_caches, list) { |
4092 | n = get_node(s, offline_node); | |
4093 | if (n) { | |
4094 | /* | |
4095 | * if n->nr_slabs > 0, slabs still exist on the node | |
4096 | * that is going down. We were unable to free them, | |
c9404c9c | 4097 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
4098 | * callback. So, we must fail. |
4099 | */ | |
0f389ec6 | 4100 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
4101 | |
4102 | s->node[offline_node] = NULL; | |
8de66a0c | 4103 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
4104 | } |
4105 | } | |
18004c5d | 4106 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4107 | } |
4108 | ||
4109 | static int slab_mem_going_online_callback(void *arg) | |
4110 | { | |
4111 | struct kmem_cache_node *n; | |
4112 | struct kmem_cache *s; | |
4113 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4114 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4115 | int ret = 0; |
4116 | ||
4117 | /* | |
4118 | * If the node's memory is already available, then kmem_cache_node is | |
4119 | * already created. Nothing to do. | |
4120 | */ | |
4121 | if (nid < 0) | |
4122 | return 0; | |
4123 | ||
4124 | /* | |
0121c619 | 4125 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4126 | * allocate a kmem_cache_node structure in order to bring the node |
4127 | * online. | |
4128 | */ | |
18004c5d | 4129 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4130 | list_for_each_entry(s, &slab_caches, list) { |
4131 | /* | |
4132 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4133 | * since memory is not yet available from the node that | |
4134 | * is brought up. | |
4135 | */ | |
8de66a0c | 4136 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4137 | if (!n) { |
4138 | ret = -ENOMEM; | |
4139 | goto out; | |
4140 | } | |
4053497d | 4141 | init_kmem_cache_node(n); |
b9049e23 YG |
4142 | s->node[nid] = n; |
4143 | } | |
4144 | out: | |
18004c5d | 4145 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4146 | return ret; |
4147 | } | |
4148 | ||
4149 | static int slab_memory_callback(struct notifier_block *self, | |
4150 | unsigned long action, void *arg) | |
4151 | { | |
4152 | int ret = 0; | |
4153 | ||
4154 | switch (action) { | |
4155 | case MEM_GOING_ONLINE: | |
4156 | ret = slab_mem_going_online_callback(arg); | |
4157 | break; | |
4158 | case MEM_GOING_OFFLINE: | |
4159 | ret = slab_mem_going_offline_callback(arg); | |
4160 | break; | |
4161 | case MEM_OFFLINE: | |
4162 | case MEM_CANCEL_ONLINE: | |
4163 | slab_mem_offline_callback(arg); | |
4164 | break; | |
4165 | case MEM_ONLINE: | |
4166 | case MEM_CANCEL_OFFLINE: | |
4167 | break; | |
4168 | } | |
dc19f9db KH |
4169 | if (ret) |
4170 | ret = notifier_from_errno(ret); | |
4171 | else | |
4172 | ret = NOTIFY_OK; | |
b9049e23 YG |
4173 | return ret; |
4174 | } | |
4175 | ||
3ac38faa AM |
4176 | static struct notifier_block slab_memory_callback_nb = { |
4177 | .notifier_call = slab_memory_callback, | |
4178 | .priority = SLAB_CALLBACK_PRI, | |
4179 | }; | |
b9049e23 | 4180 | |
81819f0f CL |
4181 | /******************************************************************** |
4182 | * Basic setup of slabs | |
4183 | *******************************************************************/ | |
4184 | ||
51df1142 CL |
4185 | /* |
4186 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4187 | * the page allocator. Allocate them properly then fix up the pointers |
4188 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4189 | */ |
4190 | ||
dffb4d60 | 4191 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4192 | { |
4193 | int node; | |
dffb4d60 | 4194 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4195 | struct kmem_cache_node *n; |
51df1142 | 4196 | |
dffb4d60 | 4197 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4198 | |
7d557b3c GC |
4199 | /* |
4200 | * This runs very early, and only the boot processor is supposed to be | |
4201 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4202 | * IPIs around. | |
4203 | */ | |
4204 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4205 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4206 | struct page *p; |
4207 | ||
fa45dc25 CL |
4208 | list_for_each_entry(p, &n->partial, lru) |
4209 | p->slab_cache = s; | |
51df1142 | 4210 | |
607bf324 | 4211 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
4212 | list_for_each_entry(p, &n->full, lru) |
4213 | p->slab_cache = s; | |
51df1142 | 4214 | #endif |
51df1142 | 4215 | } |
f7ce3190 | 4216 | slab_init_memcg_params(s); |
dffb4d60 | 4217 | list_add(&s->list, &slab_caches); |
510ded33 | 4218 | memcg_link_cache(s); |
dffb4d60 | 4219 | return s; |
51df1142 CL |
4220 | } |
4221 | ||
81819f0f CL |
4222 | void __init kmem_cache_init(void) |
4223 | { | |
dffb4d60 CL |
4224 | static __initdata struct kmem_cache boot_kmem_cache, |
4225 | boot_kmem_cache_node; | |
51df1142 | 4226 | |
fc8d8620 SG |
4227 | if (debug_guardpage_minorder()) |
4228 | slub_max_order = 0; | |
4229 | ||
dffb4d60 CL |
4230 | kmem_cache_node = &boot_kmem_cache_node; |
4231 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4232 | |
dffb4d60 | 4233 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4234 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4235 | |
3ac38faa | 4236 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4237 | |
4238 | /* Able to allocate the per node structures */ | |
4239 | slab_state = PARTIAL; | |
4240 | ||
dffb4d60 CL |
4241 | create_boot_cache(kmem_cache, "kmem_cache", |
4242 | offsetof(struct kmem_cache, node) + | |
4243 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4244 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4245 | |
dffb4d60 | 4246 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4247 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4248 | |
4249 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4250 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4251 | create_kmalloc_caches(0); |
81819f0f | 4252 | |
210e7a43 TG |
4253 | /* Setup random freelists for each cache */ |
4254 | init_freelist_randomization(); | |
4255 | ||
a96a87bf SAS |
4256 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4257 | slub_cpu_dead); | |
81819f0f | 4258 | |
b9726c26 | 4259 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 4260 | cache_line_size(), |
81819f0f CL |
4261 | slub_min_order, slub_max_order, slub_min_objects, |
4262 | nr_cpu_ids, nr_node_ids); | |
4263 | } | |
4264 | ||
7e85ee0c PE |
4265 | void __init kmem_cache_init_late(void) |
4266 | { | |
7e85ee0c PE |
4267 | } |
4268 | ||
2633d7a0 | 4269 | struct kmem_cache * |
f4957d5b | 4270 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4271 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4272 | { |
426589f5 | 4273 | struct kmem_cache *s, *c; |
81819f0f | 4274 | |
a44cb944 | 4275 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4276 | if (s) { |
4277 | s->refcount++; | |
84d0ddd6 | 4278 | |
81819f0f CL |
4279 | /* |
4280 | * Adjust the object sizes so that we clear | |
4281 | * the complete object on kzalloc. | |
4282 | */ | |
1b473f29 | 4283 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4284 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4285 | |
426589f5 | 4286 | for_each_memcg_cache(c, s) { |
84d0ddd6 | 4287 | c->object_size = s->object_size; |
52ee6d74 | 4288 | c->inuse = max(c->inuse, ALIGN(size, sizeof(void *))); |
84d0ddd6 VD |
4289 | } |
4290 | ||
7b8f3b66 | 4291 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4292 | s->refcount--; |
cbb79694 | 4293 | s = NULL; |
7b8f3b66 | 4294 | } |
a0e1d1be | 4295 | } |
6446faa2 | 4296 | |
cbb79694 CL |
4297 | return s; |
4298 | } | |
84c1cf62 | 4299 | |
d50112ed | 4300 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4301 | { |
aac3a166 PE |
4302 | int err; |
4303 | ||
4304 | err = kmem_cache_open(s, flags); | |
4305 | if (err) | |
4306 | return err; | |
20cea968 | 4307 | |
45530c44 CL |
4308 | /* Mutex is not taken during early boot */ |
4309 | if (slab_state <= UP) | |
4310 | return 0; | |
4311 | ||
107dab5c | 4312 | memcg_propagate_slab_attrs(s); |
aac3a166 | 4313 | err = sysfs_slab_add(s); |
aac3a166 | 4314 | if (err) |
52b4b950 | 4315 | __kmem_cache_release(s); |
20cea968 | 4316 | |
aac3a166 | 4317 | return err; |
81819f0f | 4318 | } |
81819f0f | 4319 | |
ce71e27c | 4320 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4321 | { |
aadb4bc4 | 4322 | struct kmem_cache *s; |
94b528d0 | 4323 | void *ret; |
aadb4bc4 | 4324 | |
95a05b42 | 4325 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4326 | return kmalloc_large(size, gfpflags); |
4327 | ||
2c59dd65 | 4328 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4329 | |
2408c550 | 4330 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4331 | return s; |
81819f0f | 4332 | |
2b847c3c | 4333 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4334 | |
25985edc | 4335 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4336 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4337 | |
4338 | return ret; | |
81819f0f CL |
4339 | } |
4340 | ||
5d1f57e4 | 4341 | #ifdef CONFIG_NUMA |
81819f0f | 4342 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4343 | int node, unsigned long caller) |
81819f0f | 4344 | { |
aadb4bc4 | 4345 | struct kmem_cache *s; |
94b528d0 | 4346 | void *ret; |
aadb4bc4 | 4347 | |
95a05b42 | 4348 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4349 | ret = kmalloc_large_node(size, gfpflags, node); |
4350 | ||
4351 | trace_kmalloc_node(caller, ret, | |
4352 | size, PAGE_SIZE << get_order(size), | |
4353 | gfpflags, node); | |
4354 | ||
4355 | return ret; | |
4356 | } | |
eada35ef | 4357 | |
2c59dd65 | 4358 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4359 | |
2408c550 | 4360 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4361 | return s; |
81819f0f | 4362 | |
2b847c3c | 4363 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4364 | |
25985edc | 4365 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4366 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4367 | |
4368 | return ret; | |
81819f0f | 4369 | } |
5d1f57e4 | 4370 | #endif |
81819f0f | 4371 | |
ab4d5ed5 | 4372 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4373 | static int count_inuse(struct page *page) |
4374 | { | |
4375 | return page->inuse; | |
4376 | } | |
4377 | ||
4378 | static int count_total(struct page *page) | |
4379 | { | |
4380 | return page->objects; | |
4381 | } | |
ab4d5ed5 | 4382 | #endif |
205ab99d | 4383 | |
ab4d5ed5 | 4384 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4385 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4386 | unsigned long *map) | |
53e15af0 CL |
4387 | { |
4388 | void *p; | |
a973e9dd | 4389 | void *addr = page_address(page); |
53e15af0 CL |
4390 | |
4391 | if (!check_slab(s, page) || | |
4392 | !on_freelist(s, page, NULL)) | |
4393 | return 0; | |
4394 | ||
4395 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4396 | bitmap_zero(map, page->objects); |
53e15af0 | 4397 | |
5f80b13a CL |
4398 | get_map(s, page, map); |
4399 | for_each_object(p, s, addr, page->objects) { | |
4400 | if (test_bit(slab_index(p, s, addr), map)) | |
4401 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4402 | return 0; | |
53e15af0 CL |
4403 | } |
4404 | ||
224a88be | 4405 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4406 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4407 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4408 | return 0; |
4409 | return 1; | |
4410 | } | |
4411 | ||
434e245d CL |
4412 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4413 | unsigned long *map) | |
53e15af0 | 4414 | { |
881db7fb CL |
4415 | slab_lock(page); |
4416 | validate_slab(s, page, map); | |
4417 | slab_unlock(page); | |
53e15af0 CL |
4418 | } |
4419 | ||
434e245d CL |
4420 | static int validate_slab_node(struct kmem_cache *s, |
4421 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4422 | { |
4423 | unsigned long count = 0; | |
4424 | struct page *page; | |
4425 | unsigned long flags; | |
4426 | ||
4427 | spin_lock_irqsave(&n->list_lock, flags); | |
4428 | ||
4429 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4430 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4431 | count++; |
4432 | } | |
4433 | if (count != n->nr_partial) | |
f9f58285 FF |
4434 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4435 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4436 | |
4437 | if (!(s->flags & SLAB_STORE_USER)) | |
4438 | goto out; | |
4439 | ||
4440 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4441 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4442 | count++; |
4443 | } | |
4444 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4445 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4446 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4447 | |
4448 | out: | |
4449 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4450 | return count; | |
4451 | } | |
4452 | ||
434e245d | 4453 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4454 | { |
4455 | int node; | |
4456 | unsigned long count = 0; | |
fa45dc25 | 4457 | struct kmem_cache_node *n; |
0684e652 | 4458 | unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL); |
434e245d CL |
4459 | |
4460 | if (!map) | |
4461 | return -ENOMEM; | |
53e15af0 CL |
4462 | |
4463 | flush_all(s); | |
fa45dc25 | 4464 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4465 | count += validate_slab_node(s, n, map); |
0684e652 | 4466 | bitmap_free(map); |
53e15af0 CL |
4467 | return count; |
4468 | } | |
88a420e4 | 4469 | /* |
672bba3a | 4470 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4471 | * and freed. |
4472 | */ | |
4473 | ||
4474 | struct location { | |
4475 | unsigned long count; | |
ce71e27c | 4476 | unsigned long addr; |
45edfa58 CL |
4477 | long long sum_time; |
4478 | long min_time; | |
4479 | long max_time; | |
4480 | long min_pid; | |
4481 | long max_pid; | |
174596a0 | 4482 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4483 | nodemask_t nodes; |
88a420e4 CL |
4484 | }; |
4485 | ||
4486 | struct loc_track { | |
4487 | unsigned long max; | |
4488 | unsigned long count; | |
4489 | struct location *loc; | |
4490 | }; | |
4491 | ||
4492 | static void free_loc_track(struct loc_track *t) | |
4493 | { | |
4494 | if (t->max) | |
4495 | free_pages((unsigned long)t->loc, | |
4496 | get_order(sizeof(struct location) * t->max)); | |
4497 | } | |
4498 | ||
68dff6a9 | 4499 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4500 | { |
4501 | struct location *l; | |
4502 | int order; | |
4503 | ||
88a420e4 CL |
4504 | order = get_order(sizeof(struct location) * max); |
4505 | ||
68dff6a9 | 4506 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4507 | if (!l) |
4508 | return 0; | |
4509 | ||
4510 | if (t->count) { | |
4511 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4512 | free_loc_track(t); | |
4513 | } | |
4514 | t->max = max; | |
4515 | t->loc = l; | |
4516 | return 1; | |
4517 | } | |
4518 | ||
4519 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4520 | const struct track *track) |
88a420e4 CL |
4521 | { |
4522 | long start, end, pos; | |
4523 | struct location *l; | |
ce71e27c | 4524 | unsigned long caddr; |
45edfa58 | 4525 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4526 | |
4527 | start = -1; | |
4528 | end = t->count; | |
4529 | ||
4530 | for ( ; ; ) { | |
4531 | pos = start + (end - start + 1) / 2; | |
4532 | ||
4533 | /* | |
4534 | * There is nothing at "end". If we end up there | |
4535 | * we need to add something to before end. | |
4536 | */ | |
4537 | if (pos == end) | |
4538 | break; | |
4539 | ||
4540 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4541 | if (track->addr == caddr) { |
4542 | ||
4543 | l = &t->loc[pos]; | |
4544 | l->count++; | |
4545 | if (track->when) { | |
4546 | l->sum_time += age; | |
4547 | if (age < l->min_time) | |
4548 | l->min_time = age; | |
4549 | if (age > l->max_time) | |
4550 | l->max_time = age; | |
4551 | ||
4552 | if (track->pid < l->min_pid) | |
4553 | l->min_pid = track->pid; | |
4554 | if (track->pid > l->max_pid) | |
4555 | l->max_pid = track->pid; | |
4556 | ||
174596a0 RR |
4557 | cpumask_set_cpu(track->cpu, |
4558 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4559 | } |
4560 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4561 | return 1; |
4562 | } | |
4563 | ||
45edfa58 | 4564 | if (track->addr < caddr) |
88a420e4 CL |
4565 | end = pos; |
4566 | else | |
4567 | start = pos; | |
4568 | } | |
4569 | ||
4570 | /* | |
672bba3a | 4571 | * Not found. Insert new tracking element. |
88a420e4 | 4572 | */ |
68dff6a9 | 4573 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4574 | return 0; |
4575 | ||
4576 | l = t->loc + pos; | |
4577 | if (pos < t->count) | |
4578 | memmove(l + 1, l, | |
4579 | (t->count - pos) * sizeof(struct location)); | |
4580 | t->count++; | |
4581 | l->count = 1; | |
45edfa58 CL |
4582 | l->addr = track->addr; |
4583 | l->sum_time = age; | |
4584 | l->min_time = age; | |
4585 | l->max_time = age; | |
4586 | l->min_pid = track->pid; | |
4587 | l->max_pid = track->pid; | |
174596a0 RR |
4588 | cpumask_clear(to_cpumask(l->cpus)); |
4589 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4590 | nodes_clear(l->nodes); |
4591 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4592 | return 1; |
4593 | } | |
4594 | ||
4595 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4596 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4597 | unsigned long *map) |
88a420e4 | 4598 | { |
a973e9dd | 4599 | void *addr = page_address(page); |
88a420e4 CL |
4600 | void *p; |
4601 | ||
39b26464 | 4602 | bitmap_zero(map, page->objects); |
5f80b13a | 4603 | get_map(s, page, map); |
88a420e4 | 4604 | |
224a88be | 4605 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4606 | if (!test_bit(slab_index(p, s, addr), map)) |
4607 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4608 | } |
4609 | ||
4610 | static int list_locations(struct kmem_cache *s, char *buf, | |
4611 | enum track_item alloc) | |
4612 | { | |
e374d483 | 4613 | int len = 0; |
88a420e4 | 4614 | unsigned long i; |
68dff6a9 | 4615 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4616 | int node; |
fa45dc25 | 4617 | struct kmem_cache_node *n; |
0684e652 | 4618 | unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL); |
88a420e4 | 4619 | |
bbd7d57b | 4620 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
0ee931c4 | 4621 | GFP_KERNEL)) { |
0684e652 | 4622 | bitmap_free(map); |
68dff6a9 | 4623 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4624 | } |
88a420e4 CL |
4625 | /* Push back cpu slabs */ |
4626 | flush_all(s); | |
4627 | ||
fa45dc25 | 4628 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4629 | unsigned long flags; |
4630 | struct page *page; | |
4631 | ||
9e86943b | 4632 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4633 | continue; |
4634 | ||
4635 | spin_lock_irqsave(&n->list_lock, flags); | |
4636 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4637 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4638 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4639 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4640 | spin_unlock_irqrestore(&n->list_lock, flags); |
4641 | } | |
4642 | ||
4643 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4644 | struct location *l = &t.loc[i]; |
88a420e4 | 4645 | |
9c246247 | 4646 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4647 | break; |
e374d483 | 4648 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4649 | |
4650 | if (l->addr) | |
62c70bce | 4651 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4652 | else |
e374d483 | 4653 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4654 | |
4655 | if (l->sum_time != l->min_time) { | |
e374d483 | 4656 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4657 | l->min_time, |
4658 | (long)div_u64(l->sum_time, l->count), | |
4659 | l->max_time); | |
45edfa58 | 4660 | } else |
e374d483 | 4661 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4662 | l->min_time); |
4663 | ||
4664 | if (l->min_pid != l->max_pid) | |
e374d483 | 4665 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4666 | l->min_pid, l->max_pid); |
4667 | else | |
e374d483 | 4668 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4669 | l->min_pid); |
4670 | ||
174596a0 RR |
4671 | if (num_online_cpus() > 1 && |
4672 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4673 | len < PAGE_SIZE - 60) |
4674 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4675 | " cpus=%*pbl", | |
4676 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4677 | |
62bc62a8 | 4678 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4679 | len < PAGE_SIZE - 60) |
4680 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4681 | " nodes=%*pbl", | |
4682 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4683 | |
e374d483 | 4684 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4685 | } |
4686 | ||
4687 | free_loc_track(&t); | |
0684e652 | 4688 | bitmap_free(map); |
88a420e4 | 4689 | if (!t.count) |
e374d483 HH |
4690 | len += sprintf(buf, "No data\n"); |
4691 | return len; | |
88a420e4 | 4692 | } |
6dfd1b65 | 4693 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 4694 | |
a5a84755 | 4695 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4696 | static void __init resiliency_test(void) |
a5a84755 CL |
4697 | { |
4698 | u8 *p; | |
cc252eae | 4699 | int type = KMALLOC_NORMAL; |
a5a84755 | 4700 | |
95a05b42 | 4701 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4702 | |
f9f58285 FF |
4703 | pr_err("SLUB resiliency testing\n"); |
4704 | pr_err("-----------------------\n"); | |
4705 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4706 | |
4707 | p = kzalloc(16, GFP_KERNEL); | |
4708 | p[16] = 0x12; | |
f9f58285 FF |
4709 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4710 | p + 16); | |
a5a84755 | 4711 | |
cc252eae | 4712 | validate_slab_cache(kmalloc_caches[type][4]); |
a5a84755 CL |
4713 | |
4714 | /* Hmmm... The next two are dangerous */ | |
4715 | p = kzalloc(32, GFP_KERNEL); | |
4716 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4717 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4718 | p); | |
4719 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 | 4720 | |
cc252eae | 4721 | validate_slab_cache(kmalloc_caches[type][5]); |
a5a84755 CL |
4722 | p = kzalloc(64, GFP_KERNEL); |
4723 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4724 | *p = 0x56; | |
f9f58285 FF |
4725 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4726 | p); | |
4727 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
cc252eae | 4728 | validate_slab_cache(kmalloc_caches[type][6]); |
a5a84755 | 4729 | |
f9f58285 | 4730 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4731 | p = kzalloc(128, GFP_KERNEL); |
4732 | kfree(p); | |
4733 | *p = 0x78; | |
f9f58285 | 4734 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
cc252eae | 4735 | validate_slab_cache(kmalloc_caches[type][7]); |
a5a84755 CL |
4736 | |
4737 | p = kzalloc(256, GFP_KERNEL); | |
4738 | kfree(p); | |
4739 | p[50] = 0x9a; | |
f9f58285 | 4740 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
cc252eae | 4741 | validate_slab_cache(kmalloc_caches[type][8]); |
a5a84755 CL |
4742 | |
4743 | p = kzalloc(512, GFP_KERNEL); | |
4744 | kfree(p); | |
4745 | p[512] = 0xab; | |
f9f58285 | 4746 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
cc252eae | 4747 | validate_slab_cache(kmalloc_caches[type][9]); |
a5a84755 CL |
4748 | } |
4749 | #else | |
4750 | #ifdef CONFIG_SYSFS | |
4751 | static void resiliency_test(void) {}; | |
4752 | #endif | |
6dfd1b65 | 4753 | #endif /* SLUB_RESILIENCY_TEST */ |
a5a84755 | 4754 | |
ab4d5ed5 | 4755 | #ifdef CONFIG_SYSFS |
81819f0f | 4756 | enum slab_stat_type { |
205ab99d CL |
4757 | SL_ALL, /* All slabs */ |
4758 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4759 | SL_CPU, /* Only slabs used for cpu caches */ | |
4760 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4761 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4762 | }; |
4763 | ||
205ab99d | 4764 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4765 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4766 | #define SO_CPU (1 << SL_CPU) | |
4767 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4768 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4769 | |
1663f26d TH |
4770 | #ifdef CONFIG_MEMCG |
4771 | static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON); | |
4772 | ||
4773 | static int __init setup_slub_memcg_sysfs(char *str) | |
4774 | { | |
4775 | int v; | |
4776 | ||
4777 | if (get_option(&str, &v) > 0) | |
4778 | memcg_sysfs_enabled = v; | |
4779 | ||
4780 | return 1; | |
4781 | } | |
4782 | ||
4783 | __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs); | |
4784 | #endif | |
4785 | ||
62e5c4b4 CG |
4786 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4787 | char *buf, unsigned long flags) | |
81819f0f CL |
4788 | { |
4789 | unsigned long total = 0; | |
81819f0f CL |
4790 | int node; |
4791 | int x; | |
4792 | unsigned long *nodes; | |
81819f0f | 4793 | |
6396bb22 | 4794 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
4795 | if (!nodes) |
4796 | return -ENOMEM; | |
81819f0f | 4797 | |
205ab99d CL |
4798 | if (flags & SO_CPU) { |
4799 | int cpu; | |
81819f0f | 4800 | |
205ab99d | 4801 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4802 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4803 | cpu); | |
ec3ab083 | 4804 | int node; |
49e22585 | 4805 | struct page *page; |
dfb4f096 | 4806 | |
4db0c3c2 | 4807 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4808 | if (!page) |
4809 | continue; | |
205ab99d | 4810 | |
ec3ab083 CL |
4811 | node = page_to_nid(page); |
4812 | if (flags & SO_TOTAL) | |
4813 | x = page->objects; | |
4814 | else if (flags & SO_OBJECTS) | |
4815 | x = page->inuse; | |
4816 | else | |
4817 | x = 1; | |
49e22585 | 4818 | |
ec3ab083 CL |
4819 | total += x; |
4820 | nodes[node] += x; | |
4821 | ||
a93cf07b | 4822 | page = slub_percpu_partial_read_once(c); |
49e22585 | 4823 | if (page) { |
8afb1474 LZ |
4824 | node = page_to_nid(page); |
4825 | if (flags & SO_TOTAL) | |
4826 | WARN_ON_ONCE(1); | |
4827 | else if (flags & SO_OBJECTS) | |
4828 | WARN_ON_ONCE(1); | |
4829 | else | |
4830 | x = page->pages; | |
bc6697d8 ED |
4831 | total += x; |
4832 | nodes[node] += x; | |
49e22585 | 4833 | } |
81819f0f CL |
4834 | } |
4835 | } | |
4836 | ||
bfc8c901 | 4837 | get_online_mems(); |
ab4d5ed5 | 4838 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4839 | if (flags & SO_ALL) { |
fa45dc25 CL |
4840 | struct kmem_cache_node *n; |
4841 | ||
4842 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4843 | |
d0e0ac97 CG |
4844 | if (flags & SO_TOTAL) |
4845 | x = atomic_long_read(&n->total_objects); | |
4846 | else if (flags & SO_OBJECTS) | |
4847 | x = atomic_long_read(&n->total_objects) - | |
4848 | count_partial(n, count_free); | |
81819f0f | 4849 | else |
205ab99d | 4850 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4851 | total += x; |
4852 | nodes[node] += x; | |
4853 | } | |
4854 | ||
ab4d5ed5 CL |
4855 | } else |
4856 | #endif | |
4857 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4858 | struct kmem_cache_node *n; |
81819f0f | 4859 | |
fa45dc25 | 4860 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4861 | if (flags & SO_TOTAL) |
4862 | x = count_partial(n, count_total); | |
4863 | else if (flags & SO_OBJECTS) | |
4864 | x = count_partial(n, count_inuse); | |
81819f0f | 4865 | else |
205ab99d | 4866 | x = n->nr_partial; |
81819f0f CL |
4867 | total += x; |
4868 | nodes[node] += x; | |
4869 | } | |
4870 | } | |
81819f0f CL |
4871 | x = sprintf(buf, "%lu", total); |
4872 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4873 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4874 | if (nodes[node]) |
4875 | x += sprintf(buf + x, " N%d=%lu", | |
4876 | node, nodes[node]); | |
4877 | #endif | |
bfc8c901 | 4878 | put_online_mems(); |
81819f0f CL |
4879 | kfree(nodes); |
4880 | return x + sprintf(buf + x, "\n"); | |
4881 | } | |
4882 | ||
ab4d5ed5 | 4883 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4884 | static int any_slab_objects(struct kmem_cache *s) |
4885 | { | |
4886 | int node; | |
fa45dc25 | 4887 | struct kmem_cache_node *n; |
81819f0f | 4888 | |
fa45dc25 | 4889 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4890 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4891 | return 1; |
fa45dc25 | 4892 | |
81819f0f CL |
4893 | return 0; |
4894 | } | |
ab4d5ed5 | 4895 | #endif |
81819f0f CL |
4896 | |
4897 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4898 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4899 | |
4900 | struct slab_attribute { | |
4901 | struct attribute attr; | |
4902 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4903 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4904 | }; | |
4905 | ||
4906 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4907 | static struct slab_attribute _name##_attr = \ |
4908 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4909 | |
4910 | #define SLAB_ATTR(_name) \ | |
4911 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4912 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4913 | |
81819f0f CL |
4914 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4915 | { | |
44065b2e | 4916 | return sprintf(buf, "%u\n", s->size); |
81819f0f CL |
4917 | } |
4918 | SLAB_ATTR_RO(slab_size); | |
4919 | ||
4920 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4921 | { | |
3a3791ec | 4922 | return sprintf(buf, "%u\n", s->align); |
81819f0f CL |
4923 | } |
4924 | SLAB_ATTR_RO(align); | |
4925 | ||
4926 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4927 | { | |
1b473f29 | 4928 | return sprintf(buf, "%u\n", s->object_size); |
81819f0f CL |
4929 | } |
4930 | SLAB_ATTR_RO(object_size); | |
4931 | ||
4932 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4933 | { | |
19af27af | 4934 | return sprintf(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
4935 | } |
4936 | SLAB_ATTR_RO(objs_per_slab); | |
4937 | ||
06b285dc CL |
4938 | static ssize_t order_store(struct kmem_cache *s, |
4939 | const char *buf, size_t length) | |
4940 | { | |
19af27af | 4941 | unsigned int order; |
0121c619 CL |
4942 | int err; |
4943 | ||
19af27af | 4944 | err = kstrtouint(buf, 10, &order); |
0121c619 CL |
4945 | if (err) |
4946 | return err; | |
06b285dc CL |
4947 | |
4948 | if (order > slub_max_order || order < slub_min_order) | |
4949 | return -EINVAL; | |
4950 | ||
4951 | calculate_sizes(s, order); | |
4952 | return length; | |
4953 | } | |
4954 | ||
81819f0f CL |
4955 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4956 | { | |
19af27af | 4957 | return sprintf(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 4958 | } |
06b285dc | 4959 | SLAB_ATTR(order); |
81819f0f | 4960 | |
73d342b1 DR |
4961 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4962 | { | |
4963 | return sprintf(buf, "%lu\n", s->min_partial); | |
4964 | } | |
4965 | ||
4966 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4967 | size_t length) | |
4968 | { | |
4969 | unsigned long min; | |
4970 | int err; | |
4971 | ||
3dbb95f7 | 4972 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4973 | if (err) |
4974 | return err; | |
4975 | ||
c0bdb232 | 4976 | set_min_partial(s, min); |
73d342b1 DR |
4977 | return length; |
4978 | } | |
4979 | SLAB_ATTR(min_partial); | |
4980 | ||
49e22585 CL |
4981 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4982 | { | |
e6d0e1dc | 4983 | return sprintf(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
4984 | } |
4985 | ||
4986 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4987 | size_t length) | |
4988 | { | |
e5d9998f | 4989 | unsigned int objects; |
49e22585 CL |
4990 | int err; |
4991 | ||
e5d9998f | 4992 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
4993 | if (err) |
4994 | return err; | |
345c905d | 4995 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4996 | return -EINVAL; |
49e22585 | 4997 | |
e6d0e1dc | 4998 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
4999 | flush_all(s); |
5000 | return length; | |
5001 | } | |
5002 | SLAB_ATTR(cpu_partial); | |
5003 | ||
81819f0f CL |
5004 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5005 | { | |
62c70bce JP |
5006 | if (!s->ctor) |
5007 | return 0; | |
5008 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
5009 | } |
5010 | SLAB_ATTR_RO(ctor); | |
5011 | ||
81819f0f CL |
5012 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5013 | { | |
4307c14f | 5014 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5015 | } |
5016 | SLAB_ATTR_RO(aliases); | |
5017 | ||
81819f0f CL |
5018 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5019 | { | |
d9acf4b7 | 5020 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5021 | } |
5022 | SLAB_ATTR_RO(partial); | |
5023 | ||
5024 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5025 | { | |
d9acf4b7 | 5026 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5027 | } |
5028 | SLAB_ATTR_RO(cpu_slabs); | |
5029 | ||
5030 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5031 | { | |
205ab99d | 5032 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5033 | } |
5034 | SLAB_ATTR_RO(objects); | |
5035 | ||
205ab99d CL |
5036 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5037 | { | |
5038 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5039 | } | |
5040 | SLAB_ATTR_RO(objects_partial); | |
5041 | ||
49e22585 CL |
5042 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5043 | { | |
5044 | int objects = 0; | |
5045 | int pages = 0; | |
5046 | int cpu; | |
5047 | int len; | |
5048 | ||
5049 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5050 | struct page *page; |
5051 | ||
5052 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5053 | |
5054 | if (page) { | |
5055 | pages += page->pages; | |
5056 | objects += page->pobjects; | |
5057 | } | |
5058 | } | |
5059 | ||
5060 | len = sprintf(buf, "%d(%d)", objects, pages); | |
5061 | ||
5062 | #ifdef CONFIG_SMP | |
5063 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5064 | struct page *page; |
5065 | ||
5066 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5067 | |
5068 | if (page && len < PAGE_SIZE - 20) | |
5069 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
5070 | page->pobjects, page->pages); | |
5071 | } | |
5072 | #endif | |
5073 | return len + sprintf(buf + len, "\n"); | |
5074 | } | |
5075 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5076 | ||
a5a84755 CL |
5077 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5078 | { | |
5079 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
5080 | } | |
5081 | ||
5082 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
5083 | const char *buf, size_t length) | |
5084 | { | |
5085 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
5086 | if (buf[0] == '1') | |
5087 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
5088 | return length; | |
5089 | } | |
5090 | SLAB_ATTR(reclaim_account); | |
5091 | ||
5092 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5093 | { | |
5094 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
5095 | } | |
5096 | SLAB_ATTR_RO(hwcache_align); | |
5097 | ||
5098 | #ifdef CONFIG_ZONE_DMA | |
5099 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5100 | { | |
5101 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
5102 | } | |
5103 | SLAB_ATTR_RO(cache_dma); | |
5104 | #endif | |
5105 | ||
8eb8284b DW |
5106 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5107 | { | |
7bbdb81e | 5108 | return sprintf(buf, "%u\n", s->usersize); |
8eb8284b DW |
5109 | } |
5110 | SLAB_ATTR_RO(usersize); | |
5111 | ||
a5a84755 CL |
5112 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5113 | { | |
5f0d5a3a | 5114 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5115 | } |
5116 | SLAB_ATTR_RO(destroy_by_rcu); | |
5117 | ||
ab4d5ed5 | 5118 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5119 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5120 | { | |
5121 | return show_slab_objects(s, buf, SO_ALL); | |
5122 | } | |
5123 | SLAB_ATTR_RO(slabs); | |
5124 | ||
205ab99d CL |
5125 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5126 | { | |
5127 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5128 | } | |
5129 | SLAB_ATTR_RO(total_objects); | |
5130 | ||
81819f0f CL |
5131 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5132 | { | |
becfda68 | 5133 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f CL |
5134 | } |
5135 | ||
5136 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
5137 | const char *buf, size_t length) | |
5138 | { | |
becfda68 | 5139 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; |
b789ef51 CL |
5140 | if (buf[0] == '1') { |
5141 | s->flags &= ~__CMPXCHG_DOUBLE; | |
becfda68 | 5142 | s->flags |= SLAB_CONSISTENCY_CHECKS; |
b789ef51 | 5143 | } |
81819f0f CL |
5144 | return length; |
5145 | } | |
5146 | SLAB_ATTR(sanity_checks); | |
5147 | ||
5148 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5149 | { | |
5150 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
5151 | } | |
5152 | ||
5153 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
5154 | size_t length) | |
5155 | { | |
c9e16131 CL |
5156 | /* |
5157 | * Tracing a merged cache is going to give confusing results | |
5158 | * as well as cause other issues like converting a mergeable | |
5159 | * cache into an umergeable one. | |
5160 | */ | |
5161 | if (s->refcount > 1) | |
5162 | return -EINVAL; | |
5163 | ||
81819f0f | 5164 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
5165 | if (buf[0] == '1') { |
5166 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5167 | s->flags |= SLAB_TRACE; |
b789ef51 | 5168 | } |
81819f0f CL |
5169 | return length; |
5170 | } | |
5171 | SLAB_ATTR(trace); | |
5172 | ||
81819f0f CL |
5173 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5174 | { | |
5175 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
5176 | } | |
5177 | ||
5178 | static ssize_t red_zone_store(struct kmem_cache *s, | |
5179 | const char *buf, size_t length) | |
5180 | { | |
5181 | if (any_slab_objects(s)) | |
5182 | return -EBUSY; | |
5183 | ||
5184 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 | 5185 | if (buf[0] == '1') { |
81819f0f | 5186 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 5187 | } |
06b285dc | 5188 | calculate_sizes(s, -1); |
81819f0f CL |
5189 | return length; |
5190 | } | |
5191 | SLAB_ATTR(red_zone); | |
5192 | ||
5193 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5194 | { | |
5195 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
5196 | } | |
5197 | ||
5198 | static ssize_t poison_store(struct kmem_cache *s, | |
5199 | const char *buf, size_t length) | |
5200 | { | |
5201 | if (any_slab_objects(s)) | |
5202 | return -EBUSY; | |
5203 | ||
5204 | s->flags &= ~SLAB_POISON; | |
b789ef51 | 5205 | if (buf[0] == '1') { |
81819f0f | 5206 | s->flags |= SLAB_POISON; |
b789ef51 | 5207 | } |
06b285dc | 5208 | calculate_sizes(s, -1); |
81819f0f CL |
5209 | return length; |
5210 | } | |
5211 | SLAB_ATTR(poison); | |
5212 | ||
5213 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5214 | { | |
5215 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
5216 | } | |
5217 | ||
5218 | static ssize_t store_user_store(struct kmem_cache *s, | |
5219 | const char *buf, size_t length) | |
5220 | { | |
5221 | if (any_slab_objects(s)) | |
5222 | return -EBUSY; | |
5223 | ||
5224 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
5225 | if (buf[0] == '1') { |
5226 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5227 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 5228 | } |
06b285dc | 5229 | calculate_sizes(s, -1); |
81819f0f CL |
5230 | return length; |
5231 | } | |
5232 | SLAB_ATTR(store_user); | |
5233 | ||
53e15af0 CL |
5234 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5235 | { | |
5236 | return 0; | |
5237 | } | |
5238 | ||
5239 | static ssize_t validate_store(struct kmem_cache *s, | |
5240 | const char *buf, size_t length) | |
5241 | { | |
434e245d CL |
5242 | int ret = -EINVAL; |
5243 | ||
5244 | if (buf[0] == '1') { | |
5245 | ret = validate_slab_cache(s); | |
5246 | if (ret >= 0) | |
5247 | ret = length; | |
5248 | } | |
5249 | return ret; | |
53e15af0 CL |
5250 | } |
5251 | SLAB_ATTR(validate); | |
a5a84755 CL |
5252 | |
5253 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5254 | { | |
5255 | if (!(s->flags & SLAB_STORE_USER)) | |
5256 | return -ENOSYS; | |
5257 | return list_locations(s, buf, TRACK_ALLOC); | |
5258 | } | |
5259 | SLAB_ATTR_RO(alloc_calls); | |
5260 | ||
5261 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5262 | { | |
5263 | if (!(s->flags & SLAB_STORE_USER)) | |
5264 | return -ENOSYS; | |
5265 | return list_locations(s, buf, TRACK_FREE); | |
5266 | } | |
5267 | SLAB_ATTR_RO(free_calls); | |
5268 | #endif /* CONFIG_SLUB_DEBUG */ | |
5269 | ||
5270 | #ifdef CONFIG_FAILSLAB | |
5271 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5272 | { | |
5273 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
5274 | } | |
5275 | ||
5276 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
5277 | size_t length) | |
5278 | { | |
c9e16131 CL |
5279 | if (s->refcount > 1) |
5280 | return -EINVAL; | |
5281 | ||
a5a84755 CL |
5282 | s->flags &= ~SLAB_FAILSLAB; |
5283 | if (buf[0] == '1') | |
5284 | s->flags |= SLAB_FAILSLAB; | |
5285 | return length; | |
5286 | } | |
5287 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 5288 | #endif |
53e15af0 | 5289 | |
2086d26a CL |
5290 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5291 | { | |
5292 | return 0; | |
5293 | } | |
5294 | ||
5295 | static ssize_t shrink_store(struct kmem_cache *s, | |
5296 | const char *buf, size_t length) | |
5297 | { | |
832f37f5 VD |
5298 | if (buf[0] == '1') |
5299 | kmem_cache_shrink(s); | |
5300 | else | |
2086d26a CL |
5301 | return -EINVAL; |
5302 | return length; | |
5303 | } | |
5304 | SLAB_ATTR(shrink); | |
5305 | ||
81819f0f | 5306 | #ifdef CONFIG_NUMA |
9824601e | 5307 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5308 | { |
eb7235eb | 5309 | return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5310 | } |
5311 | ||
9824601e | 5312 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5313 | const char *buf, size_t length) |
5314 | { | |
eb7235eb | 5315 | unsigned int ratio; |
0121c619 CL |
5316 | int err; |
5317 | ||
eb7235eb | 5318 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5319 | if (err) |
5320 | return err; | |
eb7235eb AD |
5321 | if (ratio > 100) |
5322 | return -ERANGE; | |
0121c619 | 5323 | |
eb7235eb | 5324 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5325 | |
81819f0f CL |
5326 | return length; |
5327 | } | |
9824601e | 5328 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5329 | #endif |
5330 | ||
8ff12cfc | 5331 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5332 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5333 | { | |
5334 | unsigned long sum = 0; | |
5335 | int cpu; | |
5336 | int len; | |
6da2ec56 | 5337 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5338 | |
5339 | if (!data) | |
5340 | return -ENOMEM; | |
5341 | ||
5342 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5343 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5344 | |
5345 | data[cpu] = x; | |
5346 | sum += x; | |
5347 | } | |
5348 | ||
5349 | len = sprintf(buf, "%lu", sum); | |
5350 | ||
50ef37b9 | 5351 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5352 | for_each_online_cpu(cpu) { |
5353 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5354 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5355 | } |
50ef37b9 | 5356 | #endif |
8ff12cfc CL |
5357 | kfree(data); |
5358 | return len + sprintf(buf + len, "\n"); | |
5359 | } | |
5360 | ||
78eb00cc DR |
5361 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5362 | { | |
5363 | int cpu; | |
5364 | ||
5365 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5366 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5367 | } |
5368 | ||
8ff12cfc CL |
5369 | #define STAT_ATTR(si, text) \ |
5370 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5371 | { \ | |
5372 | return show_stat(s, buf, si); \ | |
5373 | } \ | |
78eb00cc DR |
5374 | static ssize_t text##_store(struct kmem_cache *s, \ |
5375 | const char *buf, size_t length) \ | |
5376 | { \ | |
5377 | if (buf[0] != '0') \ | |
5378 | return -EINVAL; \ | |
5379 | clear_stat(s, si); \ | |
5380 | return length; \ | |
5381 | } \ | |
5382 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5383 | |
5384 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5385 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5386 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5387 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5388 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5389 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5390 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5391 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5392 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5393 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5394 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5395 | STAT_ATTR(FREE_SLAB, free_slab); |
5396 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5397 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5398 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5399 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5400 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5401 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5402 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5403 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5404 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5405 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5406 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5407 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5408 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5409 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 5410 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 5411 | |
06428780 | 5412 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5413 | &slab_size_attr.attr, |
5414 | &object_size_attr.attr, | |
5415 | &objs_per_slab_attr.attr, | |
5416 | &order_attr.attr, | |
73d342b1 | 5417 | &min_partial_attr.attr, |
49e22585 | 5418 | &cpu_partial_attr.attr, |
81819f0f | 5419 | &objects_attr.attr, |
205ab99d | 5420 | &objects_partial_attr.attr, |
81819f0f CL |
5421 | &partial_attr.attr, |
5422 | &cpu_slabs_attr.attr, | |
5423 | &ctor_attr.attr, | |
81819f0f CL |
5424 | &aliases_attr.attr, |
5425 | &align_attr.attr, | |
81819f0f CL |
5426 | &hwcache_align_attr.attr, |
5427 | &reclaim_account_attr.attr, | |
5428 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5429 | &shrink_attr.attr, |
49e22585 | 5430 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5431 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5432 | &total_objects_attr.attr, |
5433 | &slabs_attr.attr, | |
5434 | &sanity_checks_attr.attr, | |
5435 | &trace_attr.attr, | |
81819f0f CL |
5436 | &red_zone_attr.attr, |
5437 | &poison_attr.attr, | |
5438 | &store_user_attr.attr, | |
53e15af0 | 5439 | &validate_attr.attr, |
88a420e4 CL |
5440 | &alloc_calls_attr.attr, |
5441 | &free_calls_attr.attr, | |
ab4d5ed5 | 5442 | #endif |
81819f0f CL |
5443 | #ifdef CONFIG_ZONE_DMA |
5444 | &cache_dma_attr.attr, | |
5445 | #endif | |
5446 | #ifdef CONFIG_NUMA | |
9824601e | 5447 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5448 | #endif |
5449 | #ifdef CONFIG_SLUB_STATS | |
5450 | &alloc_fastpath_attr.attr, | |
5451 | &alloc_slowpath_attr.attr, | |
5452 | &free_fastpath_attr.attr, | |
5453 | &free_slowpath_attr.attr, | |
5454 | &free_frozen_attr.attr, | |
5455 | &free_add_partial_attr.attr, | |
5456 | &free_remove_partial_attr.attr, | |
5457 | &alloc_from_partial_attr.attr, | |
5458 | &alloc_slab_attr.attr, | |
5459 | &alloc_refill_attr.attr, | |
e36a2652 | 5460 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5461 | &free_slab_attr.attr, |
5462 | &cpuslab_flush_attr.attr, | |
5463 | &deactivate_full_attr.attr, | |
5464 | &deactivate_empty_attr.attr, | |
5465 | &deactivate_to_head_attr.attr, | |
5466 | &deactivate_to_tail_attr.attr, | |
5467 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5468 | &deactivate_bypass_attr.attr, |
65c3376a | 5469 | &order_fallback_attr.attr, |
b789ef51 CL |
5470 | &cmpxchg_double_fail_attr.attr, |
5471 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5472 | &cpu_partial_alloc_attr.attr, |
5473 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5474 | &cpu_partial_node_attr.attr, |
5475 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5476 | #endif |
4c13dd3b DM |
5477 | #ifdef CONFIG_FAILSLAB |
5478 | &failslab_attr.attr, | |
5479 | #endif | |
8eb8284b | 5480 | &usersize_attr.attr, |
4c13dd3b | 5481 | |
81819f0f CL |
5482 | NULL |
5483 | }; | |
5484 | ||
1fdaaa23 | 5485 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5486 | .attrs = slab_attrs, |
5487 | }; | |
5488 | ||
5489 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5490 | struct attribute *attr, | |
5491 | char *buf) | |
5492 | { | |
5493 | struct slab_attribute *attribute; | |
5494 | struct kmem_cache *s; | |
5495 | int err; | |
5496 | ||
5497 | attribute = to_slab_attr(attr); | |
5498 | s = to_slab(kobj); | |
5499 | ||
5500 | if (!attribute->show) | |
5501 | return -EIO; | |
5502 | ||
5503 | err = attribute->show(s, buf); | |
5504 | ||
5505 | return err; | |
5506 | } | |
5507 | ||
5508 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5509 | struct attribute *attr, | |
5510 | const char *buf, size_t len) | |
5511 | { | |
5512 | struct slab_attribute *attribute; | |
5513 | struct kmem_cache *s; | |
5514 | int err; | |
5515 | ||
5516 | attribute = to_slab_attr(attr); | |
5517 | s = to_slab(kobj); | |
5518 | ||
5519 | if (!attribute->store) | |
5520 | return -EIO; | |
5521 | ||
5522 | err = attribute->store(s, buf, len); | |
127424c8 | 5523 | #ifdef CONFIG_MEMCG |
107dab5c | 5524 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5525 | struct kmem_cache *c; |
81819f0f | 5526 | |
107dab5c GC |
5527 | mutex_lock(&slab_mutex); |
5528 | if (s->max_attr_size < len) | |
5529 | s->max_attr_size = len; | |
5530 | ||
ebe945c2 GC |
5531 | /* |
5532 | * This is a best effort propagation, so this function's return | |
5533 | * value will be determined by the parent cache only. This is | |
5534 | * basically because not all attributes will have a well | |
5535 | * defined semantics for rollbacks - most of the actions will | |
5536 | * have permanent effects. | |
5537 | * | |
5538 | * Returning the error value of any of the children that fail | |
5539 | * is not 100 % defined, in the sense that users seeing the | |
5540 | * error code won't be able to know anything about the state of | |
5541 | * the cache. | |
5542 | * | |
5543 | * Only returning the error code for the parent cache at least | |
5544 | * has well defined semantics. The cache being written to | |
5545 | * directly either failed or succeeded, in which case we loop | |
5546 | * through the descendants with best-effort propagation. | |
5547 | */ | |
426589f5 VD |
5548 | for_each_memcg_cache(c, s) |
5549 | attribute->store(c, buf, len); | |
107dab5c GC |
5550 | mutex_unlock(&slab_mutex); |
5551 | } | |
5552 | #endif | |
81819f0f CL |
5553 | return err; |
5554 | } | |
5555 | ||
107dab5c GC |
5556 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5557 | { | |
127424c8 | 5558 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5559 | int i; |
5560 | char *buffer = NULL; | |
93030d83 | 5561 | struct kmem_cache *root_cache; |
107dab5c | 5562 | |
93030d83 | 5563 | if (is_root_cache(s)) |
107dab5c GC |
5564 | return; |
5565 | ||
f7ce3190 | 5566 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5567 | |
107dab5c GC |
5568 | /* |
5569 | * This mean this cache had no attribute written. Therefore, no point | |
5570 | * in copying default values around | |
5571 | */ | |
93030d83 | 5572 | if (!root_cache->max_attr_size) |
107dab5c GC |
5573 | return; |
5574 | ||
5575 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5576 | char mbuf[64]; | |
5577 | char *buf; | |
5578 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
478fe303 | 5579 | ssize_t len; |
107dab5c GC |
5580 | |
5581 | if (!attr || !attr->store || !attr->show) | |
5582 | continue; | |
5583 | ||
5584 | /* | |
5585 | * It is really bad that we have to allocate here, so we will | |
5586 | * do it only as a fallback. If we actually allocate, though, | |
5587 | * we can just use the allocated buffer until the end. | |
5588 | * | |
5589 | * Most of the slub attributes will tend to be very small in | |
5590 | * size, but sysfs allows buffers up to a page, so they can | |
5591 | * theoretically happen. | |
5592 | */ | |
5593 | if (buffer) | |
5594 | buf = buffer; | |
93030d83 | 5595 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5596 | buf = mbuf; |
5597 | else { | |
5598 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5599 | if (WARN_ON(!buffer)) | |
5600 | continue; | |
5601 | buf = buffer; | |
5602 | } | |
5603 | ||
478fe303 TG |
5604 | len = attr->show(root_cache, buf); |
5605 | if (len > 0) | |
5606 | attr->store(s, buf, len); | |
107dab5c GC |
5607 | } |
5608 | ||
5609 | if (buffer) | |
5610 | free_page((unsigned long)buffer); | |
6dfd1b65 | 5611 | #endif /* CONFIG_MEMCG */ |
107dab5c GC |
5612 | } |
5613 | ||
41a21285 CL |
5614 | static void kmem_cache_release(struct kobject *k) |
5615 | { | |
5616 | slab_kmem_cache_release(to_slab(k)); | |
5617 | } | |
5618 | ||
52cf25d0 | 5619 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5620 | .show = slab_attr_show, |
5621 | .store = slab_attr_store, | |
5622 | }; | |
5623 | ||
5624 | static struct kobj_type slab_ktype = { | |
5625 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5626 | .release = kmem_cache_release, |
81819f0f CL |
5627 | }; |
5628 | ||
5629 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5630 | { | |
5631 | struct kobj_type *ktype = get_ktype(kobj); | |
5632 | ||
5633 | if (ktype == &slab_ktype) | |
5634 | return 1; | |
5635 | return 0; | |
5636 | } | |
5637 | ||
9cd43611 | 5638 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5639 | .filter = uevent_filter, |
5640 | }; | |
5641 | ||
27c3a314 | 5642 | static struct kset *slab_kset; |
81819f0f | 5643 | |
9a41707b VD |
5644 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5645 | { | |
127424c8 | 5646 | #ifdef CONFIG_MEMCG |
9a41707b | 5647 | if (!is_root_cache(s)) |
f7ce3190 | 5648 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5649 | #endif |
5650 | return slab_kset; | |
5651 | } | |
5652 | ||
81819f0f CL |
5653 | #define ID_STR_LENGTH 64 |
5654 | ||
5655 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5656 | * |
5657 | * Format :[flags-]size | |
81819f0f CL |
5658 | */ |
5659 | static char *create_unique_id(struct kmem_cache *s) | |
5660 | { | |
5661 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5662 | char *p = name; | |
5663 | ||
5664 | BUG_ON(!name); | |
5665 | ||
5666 | *p++ = ':'; | |
5667 | /* | |
5668 | * First flags affecting slabcache operations. We will only | |
5669 | * get here for aliasable slabs so we do not need to support | |
5670 | * too many flags. The flags here must cover all flags that | |
5671 | * are matched during merging to guarantee that the id is | |
5672 | * unique. | |
5673 | */ | |
5674 | if (s->flags & SLAB_CACHE_DMA) | |
5675 | *p++ = 'd'; | |
6d6ea1e9 NB |
5676 | if (s->flags & SLAB_CACHE_DMA32) |
5677 | *p++ = 'D'; | |
81819f0f CL |
5678 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5679 | *p++ = 'a'; | |
becfda68 | 5680 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5681 | *p++ = 'F'; |
230e9fc2 VD |
5682 | if (s->flags & SLAB_ACCOUNT) |
5683 | *p++ = 'A'; | |
81819f0f CL |
5684 | if (p != name + 1) |
5685 | *p++ = '-'; | |
44065b2e | 5686 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5687 | |
81819f0f CL |
5688 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5689 | return name; | |
5690 | } | |
5691 | ||
3b7b3140 TH |
5692 | static void sysfs_slab_remove_workfn(struct work_struct *work) |
5693 | { | |
5694 | struct kmem_cache *s = | |
5695 | container_of(work, struct kmem_cache, kobj_remove_work); | |
5696 | ||
5697 | if (!s->kobj.state_in_sysfs) | |
5698 | /* | |
5699 | * For a memcg cache, this may be called during | |
5700 | * deactivation and again on shutdown. Remove only once. | |
5701 | * A cache is never shut down before deactivation is | |
5702 | * complete, so no need to worry about synchronization. | |
5703 | */ | |
f6ba4880 | 5704 | goto out; |
3b7b3140 TH |
5705 | |
5706 | #ifdef CONFIG_MEMCG | |
5707 | kset_unregister(s->memcg_kset); | |
5708 | #endif | |
5709 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
f6ba4880 | 5710 | out: |
3b7b3140 TH |
5711 | kobject_put(&s->kobj); |
5712 | } | |
5713 | ||
81819f0f CL |
5714 | static int sysfs_slab_add(struct kmem_cache *s) |
5715 | { | |
5716 | int err; | |
5717 | const char *name; | |
1663f26d | 5718 | struct kset *kset = cache_kset(s); |
45530c44 | 5719 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5720 | |
3b7b3140 TH |
5721 | INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn); |
5722 | ||
1663f26d TH |
5723 | if (!kset) { |
5724 | kobject_init(&s->kobj, &slab_ktype); | |
5725 | return 0; | |
5726 | } | |
5727 | ||
11066386 MC |
5728 | if (!unmergeable && disable_higher_order_debug && |
5729 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5730 | unmergeable = 1; | |
5731 | ||
81819f0f CL |
5732 | if (unmergeable) { |
5733 | /* | |
5734 | * Slabcache can never be merged so we can use the name proper. | |
5735 | * This is typically the case for debug situations. In that | |
5736 | * case we can catch duplicate names easily. | |
5737 | */ | |
27c3a314 | 5738 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5739 | name = s->name; |
5740 | } else { | |
5741 | /* | |
5742 | * Create a unique name for the slab as a target | |
5743 | * for the symlinks. | |
5744 | */ | |
5745 | name = create_unique_id(s); | |
5746 | } | |
5747 | ||
1663f26d | 5748 | s->kobj.kset = kset; |
26e4f205 | 5749 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5750 | if (err) |
80da026a | 5751 | goto out; |
81819f0f CL |
5752 | |
5753 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5754 | if (err) |
5755 | goto out_del_kobj; | |
9a41707b | 5756 | |
127424c8 | 5757 | #ifdef CONFIG_MEMCG |
1663f26d | 5758 | if (is_root_cache(s) && memcg_sysfs_enabled) { |
9a41707b VD |
5759 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); |
5760 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5761 | err = -ENOMEM; |
5762 | goto out_del_kobj; | |
9a41707b VD |
5763 | } |
5764 | } | |
5765 | #endif | |
5766 | ||
81819f0f CL |
5767 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5768 | if (!unmergeable) { | |
5769 | /* Setup first alias */ | |
5770 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5771 | } |
54b6a731 DJ |
5772 | out: |
5773 | if (!unmergeable) | |
5774 | kfree(name); | |
5775 | return err; | |
5776 | out_del_kobj: | |
5777 | kobject_del(&s->kobj); | |
54b6a731 | 5778 | goto out; |
81819f0f CL |
5779 | } |
5780 | ||
bf5eb3de | 5781 | static void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5782 | { |
97d06609 | 5783 | if (slab_state < FULL) |
2bce6485 CL |
5784 | /* |
5785 | * Sysfs has not been setup yet so no need to remove the | |
5786 | * cache from sysfs. | |
5787 | */ | |
5788 | return; | |
5789 | ||
3b7b3140 TH |
5790 | kobject_get(&s->kobj); |
5791 | schedule_work(&s->kobj_remove_work); | |
bf5eb3de TH |
5792 | } |
5793 | ||
d50d82fa MP |
5794 | void sysfs_slab_unlink(struct kmem_cache *s) |
5795 | { | |
5796 | if (slab_state >= FULL) | |
5797 | kobject_del(&s->kobj); | |
5798 | } | |
5799 | ||
bf5eb3de TH |
5800 | void sysfs_slab_release(struct kmem_cache *s) |
5801 | { | |
5802 | if (slab_state >= FULL) | |
5803 | kobject_put(&s->kobj); | |
81819f0f CL |
5804 | } |
5805 | ||
5806 | /* | |
5807 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5808 | * available lest we lose that information. |
81819f0f CL |
5809 | */ |
5810 | struct saved_alias { | |
5811 | struct kmem_cache *s; | |
5812 | const char *name; | |
5813 | struct saved_alias *next; | |
5814 | }; | |
5815 | ||
5af328a5 | 5816 | static struct saved_alias *alias_list; |
81819f0f CL |
5817 | |
5818 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5819 | { | |
5820 | struct saved_alias *al; | |
5821 | ||
97d06609 | 5822 | if (slab_state == FULL) { |
81819f0f CL |
5823 | /* |
5824 | * If we have a leftover link then remove it. | |
5825 | */ | |
27c3a314 GKH |
5826 | sysfs_remove_link(&slab_kset->kobj, name); |
5827 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5828 | } |
5829 | ||
5830 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5831 | if (!al) | |
5832 | return -ENOMEM; | |
5833 | ||
5834 | al->s = s; | |
5835 | al->name = name; | |
5836 | al->next = alias_list; | |
5837 | alias_list = al; | |
5838 | return 0; | |
5839 | } | |
5840 | ||
5841 | static int __init slab_sysfs_init(void) | |
5842 | { | |
5b95a4ac | 5843 | struct kmem_cache *s; |
81819f0f CL |
5844 | int err; |
5845 | ||
18004c5d | 5846 | mutex_lock(&slab_mutex); |
2bce6485 | 5847 | |
0ff21e46 | 5848 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5849 | if (!slab_kset) { |
18004c5d | 5850 | mutex_unlock(&slab_mutex); |
f9f58285 | 5851 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5852 | return -ENOSYS; |
5853 | } | |
5854 | ||
97d06609 | 5855 | slab_state = FULL; |
26a7bd03 | 5856 | |
5b95a4ac | 5857 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5858 | err = sysfs_slab_add(s); |
5d540fb7 | 5859 | if (err) |
f9f58285 FF |
5860 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5861 | s->name); | |
26a7bd03 | 5862 | } |
81819f0f CL |
5863 | |
5864 | while (alias_list) { | |
5865 | struct saved_alias *al = alias_list; | |
5866 | ||
5867 | alias_list = alias_list->next; | |
5868 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5869 | if (err) |
f9f58285 FF |
5870 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5871 | al->name); | |
81819f0f CL |
5872 | kfree(al); |
5873 | } | |
5874 | ||
18004c5d | 5875 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5876 | resiliency_test(); |
5877 | return 0; | |
5878 | } | |
5879 | ||
5880 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5881 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5882 | |
5883 | /* | |
5884 | * The /proc/slabinfo ABI | |
5885 | */ | |
5b365771 | 5886 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 5887 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5888 | { |
57ed3eda | 5889 | unsigned long nr_slabs = 0; |
205ab99d CL |
5890 | unsigned long nr_objs = 0; |
5891 | unsigned long nr_free = 0; | |
57ed3eda | 5892 | int node; |
fa45dc25 | 5893 | struct kmem_cache_node *n; |
57ed3eda | 5894 | |
fa45dc25 | 5895 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5896 | nr_slabs += node_nr_slabs(n); |
5897 | nr_objs += node_nr_objs(n); | |
205ab99d | 5898 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5899 | } |
5900 | ||
0d7561c6 GC |
5901 | sinfo->active_objs = nr_objs - nr_free; |
5902 | sinfo->num_objs = nr_objs; | |
5903 | sinfo->active_slabs = nr_slabs; | |
5904 | sinfo->num_slabs = nr_slabs; | |
5905 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5906 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5907 | } |
5908 | ||
0d7561c6 | 5909 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5910 | { |
7b3c3a50 AD |
5911 | } |
5912 | ||
b7454ad3 GC |
5913 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5914 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5915 | { |
b7454ad3 | 5916 | return -EIO; |
7b3c3a50 | 5917 | } |
5b365771 | 5918 | #endif /* CONFIG_SLUB_DEBUG */ |