]>
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 BG |
554 | #ifdef CONFIG_STACKTRACE |
555 | struct stack_trace trace; | |
556 | int i; | |
557 | ||
558 | trace.nr_entries = 0; | |
559 | trace.max_entries = TRACK_ADDRS_COUNT; | |
560 | trace.entries = p->addrs; | |
561 | trace.skip = 3; | |
a79316c6 | 562 | metadata_access_enable(); |
d6543e39 | 563 | save_stack_trace(&trace); |
a79316c6 | 564 | metadata_access_disable(); |
d6543e39 BG |
565 | |
566 | /* See rant in lockdep.c */ | |
567 | if (trace.nr_entries != 0 && | |
568 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
569 | trace.nr_entries--; | |
570 | ||
571 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
572 | p->addrs[i] = 0; | |
573 | #endif | |
81819f0f CL |
574 | p->addr = addr; |
575 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 576 | p->pid = current->pid; |
81819f0f CL |
577 | p->when = jiffies; |
578 | } else | |
579 | memset(p, 0, sizeof(struct track)); | |
580 | } | |
581 | ||
81819f0f CL |
582 | static void init_tracking(struct kmem_cache *s, void *object) |
583 | { | |
24922684 CL |
584 | if (!(s->flags & SLAB_STORE_USER)) |
585 | return; | |
586 | ||
ce71e27c EGM |
587 | set_track(s, object, TRACK_FREE, 0UL); |
588 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
589 | } |
590 | ||
86609d33 | 591 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f CL |
592 | { |
593 | if (!t->addr) | |
594 | return; | |
595 | ||
f9f58285 | 596 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 597 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
d6543e39 BG |
598 | #ifdef CONFIG_STACKTRACE |
599 | { | |
600 | int i; | |
601 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
602 | if (t->addrs[i]) | |
f9f58285 | 603 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
604 | else |
605 | break; | |
606 | } | |
607 | #endif | |
24922684 CL |
608 | } |
609 | ||
610 | static void print_tracking(struct kmem_cache *s, void *object) | |
611 | { | |
86609d33 | 612 | unsigned long pr_time = jiffies; |
24922684 CL |
613 | if (!(s->flags & SLAB_STORE_USER)) |
614 | return; | |
615 | ||
86609d33 CP |
616 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
617 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
618 | } |
619 | ||
620 | static void print_page_info(struct page *page) | |
621 | { | |
f9f58285 | 622 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 623 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
624 | |
625 | } | |
626 | ||
627 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
628 | { | |
ecc42fbe | 629 | struct va_format vaf; |
24922684 | 630 | va_list args; |
24922684 CL |
631 | |
632 | va_start(args, fmt); | |
ecc42fbe FF |
633 | vaf.fmt = fmt; |
634 | vaf.va = &args; | |
f9f58285 | 635 | pr_err("=============================================================================\n"); |
ecc42fbe | 636 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 637 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 638 | |
373d4d09 | 639 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 640 | va_end(args); |
81819f0f CL |
641 | } |
642 | ||
24922684 CL |
643 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
644 | { | |
ecc42fbe | 645 | struct va_format vaf; |
24922684 | 646 | va_list args; |
24922684 CL |
647 | |
648 | va_start(args, fmt); | |
ecc42fbe FF |
649 | vaf.fmt = fmt; |
650 | vaf.va = &args; | |
651 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 652 | va_end(args); |
24922684 CL |
653 | } |
654 | ||
655 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
656 | { |
657 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 658 | u8 *addr = page_address(page); |
24922684 CL |
659 | |
660 | print_tracking(s, p); | |
661 | ||
662 | print_page_info(page); | |
663 | ||
f9f58285 FF |
664 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
665 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 666 | |
d86bd1be | 667 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e DT |
668 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
669 | s->red_left_pad); | |
d86bd1be | 670 | else if (p > addr + 16) |
aa2efd5e | 671 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 672 | |
aa2efd5e | 673 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 674 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 675 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e | 676 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 677 | s->inuse - s->object_size); |
81819f0f | 678 | |
81819f0f CL |
679 | if (s->offset) |
680 | off = s->offset + sizeof(void *); | |
681 | else | |
682 | off = s->inuse; | |
683 | ||
24922684 | 684 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 685 | off += 2 * sizeof(struct track); |
81819f0f | 686 | |
80a9201a AP |
687 | off += kasan_metadata_size(s); |
688 | ||
d86bd1be | 689 | if (off != size_from_object(s)) |
81819f0f | 690 | /* Beginning of the filler is the free pointer */ |
aa2efd5e DT |
691 | print_section(KERN_ERR, "Padding ", p + off, |
692 | size_from_object(s) - off); | |
24922684 CL |
693 | |
694 | dump_stack(); | |
81819f0f CL |
695 | } |
696 | ||
75c66def | 697 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
698 | u8 *object, char *reason) |
699 | { | |
3dc50637 | 700 | slab_bug(s, "%s", reason); |
24922684 | 701 | print_trailer(s, page, object); |
81819f0f CL |
702 | } |
703 | ||
a38965bf | 704 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, |
d0e0ac97 | 705 | const char *fmt, ...) |
81819f0f CL |
706 | { |
707 | va_list args; | |
708 | char buf[100]; | |
709 | ||
24922684 CL |
710 | va_start(args, fmt); |
711 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 712 | va_end(args); |
3dc50637 | 713 | slab_bug(s, "%s", buf); |
24922684 | 714 | print_page_info(page); |
81819f0f CL |
715 | dump_stack(); |
716 | } | |
717 | ||
f7cb1933 | 718 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
719 | { |
720 | u8 *p = object; | |
721 | ||
d86bd1be JK |
722 | if (s->flags & SLAB_RED_ZONE) |
723 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
724 | ||
81819f0f | 725 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
726 | memset(p, POISON_FREE, s->object_size - 1); |
727 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
728 | } |
729 | ||
730 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 731 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
732 | } |
733 | ||
24922684 CL |
734 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
735 | void *from, void *to) | |
736 | { | |
737 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
738 | memset(from, data, to - from); | |
739 | } | |
740 | ||
741 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
742 | u8 *object, char *what, | |
06428780 | 743 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
744 | { |
745 | u8 *fault; | |
746 | u8 *end; | |
747 | ||
a79316c6 | 748 | metadata_access_enable(); |
79824820 | 749 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 750 | metadata_access_disable(); |
24922684 CL |
751 | if (!fault) |
752 | return 1; | |
753 | ||
754 | end = start + bytes; | |
755 | while (end > fault && end[-1] == value) | |
756 | end--; | |
757 | ||
758 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 759 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
760 | fault, end - 1, fault[0], value); |
761 | print_trailer(s, page, object); | |
762 | ||
763 | restore_bytes(s, what, value, fault, end); | |
764 | return 0; | |
81819f0f CL |
765 | } |
766 | ||
81819f0f CL |
767 | /* |
768 | * Object layout: | |
769 | * | |
770 | * object address | |
771 | * Bytes of the object to be managed. | |
772 | * If the freepointer may overlay the object then the free | |
773 | * pointer is the first word of the object. | |
672bba3a | 774 | * |
81819f0f CL |
775 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
776 | * 0xa5 (POISON_END) | |
777 | * | |
3b0efdfa | 778 | * object + s->object_size |
81819f0f | 779 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 780 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 781 | * object_size == inuse. |
672bba3a | 782 | * |
81819f0f CL |
783 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
784 | * 0xcc (RED_ACTIVE) for objects in use. | |
785 | * | |
786 | * object + s->inuse | |
672bba3a CL |
787 | * Meta data starts here. |
788 | * | |
81819f0f CL |
789 | * A. Free pointer (if we cannot overwrite object on free) |
790 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 791 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 792 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
793 | * before the word boundary. |
794 | * | |
795 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
796 | * |
797 | * object + s->size | |
672bba3a | 798 | * Nothing is used beyond s->size. |
81819f0f | 799 | * |
3b0efdfa | 800 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 801 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
802 | * may be used with merged slabcaches. |
803 | */ | |
804 | ||
81819f0f CL |
805 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
806 | { | |
807 | unsigned long off = s->inuse; /* The end of info */ | |
808 | ||
809 | if (s->offset) | |
810 | /* Freepointer is placed after the object. */ | |
811 | off += sizeof(void *); | |
812 | ||
813 | if (s->flags & SLAB_STORE_USER) | |
814 | /* We also have user information there */ | |
815 | off += 2 * sizeof(struct track); | |
816 | ||
80a9201a AP |
817 | off += kasan_metadata_size(s); |
818 | ||
d86bd1be | 819 | if (size_from_object(s) == off) |
81819f0f CL |
820 | return 1; |
821 | ||
24922684 | 822 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 823 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
824 | } |
825 | ||
39b26464 | 826 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
827 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
828 | { | |
24922684 CL |
829 | u8 *start; |
830 | u8 *fault; | |
831 | u8 *end; | |
5d682681 | 832 | u8 *pad; |
24922684 CL |
833 | int length; |
834 | int remainder; | |
81819f0f CL |
835 | |
836 | if (!(s->flags & SLAB_POISON)) | |
837 | return 1; | |
838 | ||
a973e9dd | 839 | start = page_address(page); |
9736d2a9 | 840 | length = PAGE_SIZE << compound_order(page); |
39b26464 CL |
841 | end = start + length; |
842 | remainder = length % s->size; | |
81819f0f CL |
843 | if (!remainder) |
844 | return 1; | |
845 | ||
5d682681 | 846 | pad = end - remainder; |
a79316c6 | 847 | metadata_access_enable(); |
5d682681 | 848 | fault = memchr_inv(pad, POISON_INUSE, remainder); |
a79316c6 | 849 | metadata_access_disable(); |
24922684 CL |
850 | if (!fault) |
851 | return 1; | |
852 | while (end > fault && end[-1] == POISON_INUSE) | |
853 | end--; | |
854 | ||
855 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
5d682681 | 856 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 857 | |
5d682681 | 858 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
24922684 | 859 | return 0; |
81819f0f CL |
860 | } |
861 | ||
862 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 863 | void *object, u8 val) |
81819f0f CL |
864 | { |
865 | u8 *p = object; | |
3b0efdfa | 866 | u8 *endobject = object + s->object_size; |
81819f0f CL |
867 | |
868 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
869 | if (!check_bytes_and_report(s, page, object, "Redzone", |
870 | object - s->red_left_pad, val, s->red_left_pad)) | |
871 | return 0; | |
872 | ||
24922684 | 873 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 874 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 875 | return 0; |
81819f0f | 876 | } else { |
3b0efdfa | 877 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 878 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
879 | endobject, POISON_INUSE, |
880 | s->inuse - s->object_size); | |
3adbefee | 881 | } |
81819f0f CL |
882 | } |
883 | ||
884 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 885 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 886 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 887 | POISON_FREE, s->object_size - 1) || |
24922684 | 888 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 889 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 890 | return 0; |
81819f0f CL |
891 | /* |
892 | * check_pad_bytes cleans up on its own. | |
893 | */ | |
894 | check_pad_bytes(s, page, p); | |
895 | } | |
896 | ||
f7cb1933 | 897 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
898 | /* |
899 | * Object and freepointer overlap. Cannot check | |
900 | * freepointer while object is allocated. | |
901 | */ | |
902 | return 1; | |
903 | ||
904 | /* Check free pointer validity */ | |
905 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
906 | object_err(s, page, p, "Freepointer corrupt"); | |
907 | /* | |
9f6c708e | 908 | * No choice but to zap it and thus lose the remainder |
81819f0f | 909 | * of the free objects in this slab. May cause |
672bba3a | 910 | * another error because the object count is now wrong. |
81819f0f | 911 | */ |
a973e9dd | 912 | set_freepointer(s, p, NULL); |
81819f0f CL |
913 | return 0; |
914 | } | |
915 | return 1; | |
916 | } | |
917 | ||
918 | static int check_slab(struct kmem_cache *s, struct page *page) | |
919 | { | |
39b26464 CL |
920 | int maxobj; |
921 | ||
81819f0f CL |
922 | VM_BUG_ON(!irqs_disabled()); |
923 | ||
924 | if (!PageSlab(page)) { | |
24922684 | 925 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
926 | return 0; |
927 | } | |
39b26464 | 928 | |
9736d2a9 | 929 | maxobj = order_objects(compound_order(page), s->size); |
39b26464 CL |
930 | if (page->objects > maxobj) { |
931 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 932 | page->objects, maxobj); |
39b26464 CL |
933 | return 0; |
934 | } | |
935 | if (page->inuse > page->objects) { | |
24922684 | 936 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 937 | page->inuse, page->objects); |
81819f0f CL |
938 | return 0; |
939 | } | |
940 | /* Slab_pad_check fixes things up after itself */ | |
941 | slab_pad_check(s, page); | |
942 | return 1; | |
943 | } | |
944 | ||
945 | /* | |
672bba3a CL |
946 | * Determine if a certain object on a page is on the freelist. Must hold the |
947 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
948 | */ |
949 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
950 | { | |
951 | int nr = 0; | |
881db7fb | 952 | void *fp; |
81819f0f | 953 | void *object = NULL; |
f6edde9c | 954 | int max_objects; |
81819f0f | 955 | |
881db7fb | 956 | fp = page->freelist; |
39b26464 | 957 | while (fp && nr <= page->objects) { |
81819f0f CL |
958 | if (fp == search) |
959 | return 1; | |
960 | if (!check_valid_pointer(s, page, fp)) { | |
961 | if (object) { | |
962 | object_err(s, page, object, | |
963 | "Freechain corrupt"); | |
a973e9dd | 964 | set_freepointer(s, object, NULL); |
81819f0f | 965 | } else { |
24922684 | 966 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 967 | page->freelist = NULL; |
39b26464 | 968 | page->inuse = page->objects; |
24922684 | 969 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
970 | return 0; |
971 | } | |
972 | break; | |
973 | } | |
974 | object = fp; | |
975 | fp = get_freepointer(s, object); | |
976 | nr++; | |
977 | } | |
978 | ||
9736d2a9 | 979 | max_objects = order_objects(compound_order(page), s->size); |
210b5c06 CG |
980 | if (max_objects > MAX_OBJS_PER_PAGE) |
981 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
982 | |
983 | if (page->objects != max_objects) { | |
756a025f JP |
984 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
985 | page->objects, max_objects); | |
224a88be CL |
986 | page->objects = max_objects; |
987 | slab_fix(s, "Number of objects adjusted."); | |
988 | } | |
39b26464 | 989 | if (page->inuse != page->objects - nr) { |
756a025f JP |
990 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
991 | page->inuse, page->objects - nr); | |
39b26464 | 992 | page->inuse = page->objects - nr; |
24922684 | 993 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
994 | } |
995 | return search == NULL; | |
996 | } | |
997 | ||
0121c619 CL |
998 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
999 | int alloc) | |
3ec09742 CL |
1000 | { |
1001 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1002 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1003 | s->name, |
1004 | alloc ? "alloc" : "free", | |
1005 | object, page->inuse, | |
1006 | page->freelist); | |
1007 | ||
1008 | if (!alloc) | |
aa2efd5e | 1009 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1010 | s->object_size); |
3ec09742 CL |
1011 | |
1012 | dump_stack(); | |
1013 | } | |
1014 | } | |
1015 | ||
643b1138 | 1016 | /* |
672bba3a | 1017 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1018 | */ |
5cc6eee8 CL |
1019 | static void add_full(struct kmem_cache *s, |
1020 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1021 | { |
5cc6eee8 CL |
1022 | if (!(s->flags & SLAB_STORE_USER)) |
1023 | return; | |
1024 | ||
255d0884 | 1025 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1026 | list_add(&page->lru, &n->full); |
643b1138 CL |
1027 | } |
1028 | ||
c65c1877 | 1029 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1030 | { |
643b1138 CL |
1031 | if (!(s->flags & SLAB_STORE_USER)) |
1032 | return; | |
1033 | ||
255d0884 | 1034 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1035 | list_del(&page->lru); |
643b1138 CL |
1036 | } |
1037 | ||
0f389ec6 CL |
1038 | /* Tracking of the number of slabs for debugging purposes */ |
1039 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1040 | { | |
1041 | struct kmem_cache_node *n = get_node(s, node); | |
1042 | ||
1043 | return atomic_long_read(&n->nr_slabs); | |
1044 | } | |
1045 | ||
26c02cf0 AB |
1046 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1047 | { | |
1048 | return atomic_long_read(&n->nr_slabs); | |
1049 | } | |
1050 | ||
205ab99d | 1051 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1052 | { |
1053 | struct kmem_cache_node *n = get_node(s, node); | |
1054 | ||
1055 | /* | |
1056 | * May be called early in order to allocate a slab for the | |
1057 | * kmem_cache_node structure. Solve the chicken-egg | |
1058 | * dilemma by deferring the increment of the count during | |
1059 | * bootstrap (see early_kmem_cache_node_alloc). | |
1060 | */ | |
338b2642 | 1061 | if (likely(n)) { |
0f389ec6 | 1062 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1063 | atomic_long_add(objects, &n->total_objects); |
1064 | } | |
0f389ec6 | 1065 | } |
205ab99d | 1066 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1067 | { |
1068 | struct kmem_cache_node *n = get_node(s, node); | |
1069 | ||
1070 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1071 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1072 | } |
1073 | ||
1074 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1075 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1076 | void *object) | |
1077 | { | |
1078 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1079 | return; | |
1080 | ||
f7cb1933 | 1081 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1082 | init_tracking(s, object); |
1083 | } | |
1084 | ||
a7101224 AK |
1085 | static void setup_page_debug(struct kmem_cache *s, void *addr, int order) |
1086 | { | |
1087 | if (!(s->flags & SLAB_POISON)) | |
1088 | return; | |
1089 | ||
1090 | metadata_access_enable(); | |
1091 | memset(addr, POISON_INUSE, PAGE_SIZE << order); | |
1092 | metadata_access_disable(); | |
1093 | } | |
1094 | ||
becfda68 | 1095 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
d0e0ac97 | 1096 | struct page *page, |
ce71e27c | 1097 | void *object, unsigned long addr) |
81819f0f CL |
1098 | { |
1099 | if (!check_slab(s, page)) | |
becfda68 | 1100 | return 0; |
81819f0f | 1101 | |
81819f0f CL |
1102 | if (!check_valid_pointer(s, page, object)) { |
1103 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1104 | return 0; |
81819f0f CL |
1105 | } |
1106 | ||
f7cb1933 | 1107 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1108 | return 0; |
1109 | ||
1110 | return 1; | |
1111 | } | |
1112 | ||
1113 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1114 | struct page *page, | |
1115 | void *object, unsigned long addr) | |
1116 | { | |
1117 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1118 | if (!alloc_consistency_checks(s, page, object, addr)) | |
1119 | goto bad; | |
1120 | } | |
81819f0f | 1121 | |
3ec09742 CL |
1122 | /* Success perform special debug activities for allocs */ |
1123 | if (s->flags & SLAB_STORE_USER) | |
1124 | set_track(s, object, TRACK_ALLOC, addr); | |
1125 | trace(s, page, object, 1); | |
f7cb1933 | 1126 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1127 | return 1; |
3ec09742 | 1128 | |
81819f0f CL |
1129 | bad: |
1130 | if (PageSlab(page)) { | |
1131 | /* | |
1132 | * If this is a slab page then lets do the best we can | |
1133 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1134 | * as used avoids touching the remaining objects. |
81819f0f | 1135 | */ |
24922684 | 1136 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1137 | page->inuse = page->objects; |
a973e9dd | 1138 | page->freelist = NULL; |
81819f0f CL |
1139 | } |
1140 | return 0; | |
1141 | } | |
1142 | ||
becfda68 LA |
1143 | static inline int free_consistency_checks(struct kmem_cache *s, |
1144 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1145 | { |
81819f0f | 1146 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1147 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1148 | return 0; |
81819f0f CL |
1149 | } |
1150 | ||
1151 | if (on_freelist(s, page, object)) { | |
24922684 | 1152 | object_err(s, page, object, "Object already free"); |
becfda68 | 1153 | return 0; |
81819f0f CL |
1154 | } |
1155 | ||
f7cb1933 | 1156 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1157 | return 0; |
81819f0f | 1158 | |
1b4f59e3 | 1159 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1160 | if (!PageSlab(page)) { |
756a025f JP |
1161 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1162 | object); | |
1b4f59e3 | 1163 | } else if (!page->slab_cache) { |
f9f58285 FF |
1164 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1165 | object); | |
70d71228 | 1166 | dump_stack(); |
06428780 | 1167 | } else |
24922684 CL |
1168 | object_err(s, page, object, |
1169 | "page slab pointer corrupt."); | |
becfda68 LA |
1170 | return 0; |
1171 | } | |
1172 | return 1; | |
1173 | } | |
1174 | ||
1175 | /* Supports checking bulk free of a constructed freelist */ | |
1176 | static noinline int free_debug_processing( | |
1177 | struct kmem_cache *s, struct page *page, | |
1178 | void *head, void *tail, int bulk_cnt, | |
1179 | unsigned long addr) | |
1180 | { | |
1181 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1182 | void *object = head; | |
1183 | int cnt = 0; | |
1184 | unsigned long uninitialized_var(flags); | |
1185 | int ret = 0; | |
1186 | ||
1187 | spin_lock_irqsave(&n->list_lock, flags); | |
1188 | slab_lock(page); | |
1189 | ||
1190 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1191 | if (!check_slab(s, page)) | |
1192 | goto out; | |
1193 | } | |
1194 | ||
1195 | next_object: | |
1196 | cnt++; | |
1197 | ||
1198 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1199 | if (!free_consistency_checks(s, page, object, addr)) | |
1200 | goto out; | |
81819f0f | 1201 | } |
3ec09742 | 1202 | |
3ec09742 CL |
1203 | if (s->flags & SLAB_STORE_USER) |
1204 | set_track(s, object, TRACK_FREE, addr); | |
1205 | trace(s, page, object, 0); | |
81084651 | 1206 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1207 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1208 | |
1209 | /* Reached end of constructed freelist yet? */ | |
1210 | if (object != tail) { | |
1211 | object = get_freepointer(s, object); | |
1212 | goto next_object; | |
1213 | } | |
804aa132 LA |
1214 | ret = 1; |
1215 | ||
5c2e4bbb | 1216 | out: |
81084651 JDB |
1217 | if (cnt != bulk_cnt) |
1218 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1219 | bulk_cnt, cnt); | |
1220 | ||
881db7fb | 1221 | slab_unlock(page); |
282acb43 | 1222 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1223 | if (!ret) |
1224 | slab_fix(s, "Object at 0x%p not freed", object); | |
1225 | return ret; | |
81819f0f CL |
1226 | } |
1227 | ||
41ecc55b CL |
1228 | static int __init setup_slub_debug(char *str) |
1229 | { | |
f0630fff CL |
1230 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1231 | if (*str++ != '=' || !*str) | |
1232 | /* | |
1233 | * No options specified. Switch on full debugging. | |
1234 | */ | |
1235 | goto out; | |
1236 | ||
1237 | if (*str == ',') | |
1238 | /* | |
1239 | * No options but restriction on slabs. This means full | |
1240 | * debugging for slabs matching a pattern. | |
1241 | */ | |
1242 | goto check_slabs; | |
1243 | ||
1244 | slub_debug = 0; | |
1245 | if (*str == '-') | |
1246 | /* | |
1247 | * Switch off all debugging measures. | |
1248 | */ | |
1249 | goto out; | |
1250 | ||
1251 | /* | |
1252 | * Determine which debug features should be switched on | |
1253 | */ | |
06428780 | 1254 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1255 | switch (tolower(*str)) { |
1256 | case 'f': | |
becfda68 | 1257 | slub_debug |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1258 | break; |
1259 | case 'z': | |
1260 | slub_debug |= SLAB_RED_ZONE; | |
1261 | break; | |
1262 | case 'p': | |
1263 | slub_debug |= SLAB_POISON; | |
1264 | break; | |
1265 | case 'u': | |
1266 | slub_debug |= SLAB_STORE_USER; | |
1267 | break; | |
1268 | case 't': | |
1269 | slub_debug |= SLAB_TRACE; | |
1270 | break; | |
4c13dd3b DM |
1271 | case 'a': |
1272 | slub_debug |= SLAB_FAILSLAB; | |
1273 | break; | |
08303a73 CA |
1274 | case 'o': |
1275 | /* | |
1276 | * Avoid enabling debugging on caches if its minimum | |
1277 | * order would increase as a result. | |
1278 | */ | |
1279 | disable_higher_order_debug = 1; | |
1280 | break; | |
f0630fff | 1281 | default: |
f9f58285 FF |
1282 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1283 | *str); | |
f0630fff | 1284 | } |
41ecc55b CL |
1285 | } |
1286 | ||
f0630fff | 1287 | check_slabs: |
41ecc55b CL |
1288 | if (*str == ',') |
1289 | slub_debug_slabs = str + 1; | |
f0630fff | 1290 | out: |
41ecc55b CL |
1291 | return 1; |
1292 | } | |
1293 | ||
1294 | __setup("slub_debug", setup_slub_debug); | |
1295 | ||
c5fd3ca0 AT |
1296 | /* |
1297 | * kmem_cache_flags - apply debugging options to the cache | |
1298 | * @object_size: the size of an object without meta data | |
1299 | * @flags: flags to set | |
1300 | * @name: name of the cache | |
1301 | * @ctor: constructor function | |
1302 | * | |
1303 | * Debug option(s) are applied to @flags. In addition to the debug | |
1304 | * option(s), if a slab name (or multiple) is specified i.e. | |
1305 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1306 | * then only the select slabs will receive the debug option(s). | |
1307 | */ | |
0293d1fd | 1308 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1309 | slab_flags_t flags, const char *name, |
51cc5068 | 1310 | void (*ctor)(void *)) |
41ecc55b | 1311 | { |
c5fd3ca0 AT |
1312 | char *iter; |
1313 | size_t len; | |
1314 | ||
1315 | /* If slub_debug = 0, it folds into the if conditional. */ | |
1316 | if (!slub_debug_slabs) | |
1317 | return flags | slub_debug; | |
1318 | ||
1319 | len = strlen(name); | |
1320 | iter = slub_debug_slabs; | |
1321 | while (*iter) { | |
1322 | char *end, *glob; | |
1323 | size_t cmplen; | |
1324 | ||
1325 | end = strchr(iter, ','); | |
1326 | if (!end) | |
1327 | end = iter + strlen(iter); | |
1328 | ||
1329 | glob = strnchr(iter, end - iter, '*'); | |
1330 | if (glob) | |
1331 | cmplen = glob - iter; | |
1332 | else | |
1333 | cmplen = max_t(size_t, len, (end - iter)); | |
1334 | ||
1335 | if (!strncmp(name, iter, cmplen)) { | |
1336 | flags |= slub_debug; | |
1337 | break; | |
1338 | } | |
1339 | ||
1340 | if (!*end) | |
1341 | break; | |
1342 | iter = end + 1; | |
1343 | } | |
ba0268a8 CL |
1344 | |
1345 | return flags; | |
41ecc55b | 1346 | } |
b4a64718 | 1347 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1348 | static inline void setup_object_debug(struct kmem_cache *s, |
1349 | struct page *page, void *object) {} | |
a7101224 AK |
1350 | static inline void setup_page_debug(struct kmem_cache *s, |
1351 | void *addr, int order) {} | |
41ecc55b | 1352 | |
3ec09742 | 1353 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1354 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1355 | |
282acb43 | 1356 | static inline int free_debug_processing( |
81084651 JDB |
1357 | struct kmem_cache *s, struct page *page, |
1358 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1359 | unsigned long addr) { return 0; } |
41ecc55b | 1360 | |
41ecc55b CL |
1361 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1362 | { return 1; } | |
1363 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1364 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1365 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1366 | struct page *page) {} | |
c65c1877 PZ |
1367 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1368 | struct page *page) {} | |
0293d1fd | 1369 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1370 | slab_flags_t flags, const char *name, |
51cc5068 | 1371 | void (*ctor)(void *)) |
ba0268a8 CL |
1372 | { |
1373 | return flags; | |
1374 | } | |
41ecc55b | 1375 | #define slub_debug 0 |
0f389ec6 | 1376 | |
fdaa45e9 IM |
1377 | #define disable_higher_order_debug 0 |
1378 | ||
0f389ec6 CL |
1379 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1380 | { return 0; } | |
26c02cf0 AB |
1381 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1382 | { return 0; } | |
205ab99d CL |
1383 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1384 | int objects) {} | |
1385 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1386 | int objects) {} | |
7d550c56 | 1387 | |
02e72cc6 AR |
1388 | #endif /* CONFIG_SLUB_DEBUG */ |
1389 | ||
1390 | /* | |
1391 | * Hooks for other subsystems that check memory allocations. In a typical | |
1392 | * production configuration these hooks all should produce no code at all. | |
1393 | */ | |
0116523c | 1394 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
d56791b3 | 1395 | { |
53128245 | 1396 | ptr = kasan_kmalloc_large(ptr, size, flags); |
a2f77575 | 1397 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ |
d56791b3 | 1398 | kmemleak_alloc(ptr, size, 1, flags); |
53128245 | 1399 | return ptr; |
d56791b3 RB |
1400 | } |
1401 | ||
ee3ce779 | 1402 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1403 | { |
1404 | kmemleak_free(x); | |
ee3ce779 | 1405 | kasan_kfree_large(x, _RET_IP_); |
d56791b3 RB |
1406 | } |
1407 | ||
c3895391 | 1408 | static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x) |
d56791b3 RB |
1409 | { |
1410 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1411 | |
02e72cc6 AR |
1412 | /* |
1413 | * Trouble is that we may no longer disable interrupts in the fast path | |
1414 | * So in order to make the debug calls that expect irqs to be | |
1415 | * disabled we need to disable interrupts temporarily. | |
1416 | */ | |
4675ff05 | 1417 | #ifdef CONFIG_LOCKDEP |
02e72cc6 AR |
1418 | { |
1419 | unsigned long flags; | |
1420 | ||
1421 | local_irq_save(flags); | |
02e72cc6 AR |
1422 | debug_check_no_locks_freed(x, s->object_size); |
1423 | local_irq_restore(flags); | |
1424 | } | |
1425 | #endif | |
1426 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1427 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1428 | |
c3895391 AK |
1429 | /* KASAN might put x into memory quarantine, delaying its reuse */ |
1430 | return kasan_slab_free(s, x, _RET_IP_); | |
02e72cc6 | 1431 | } |
205ab99d | 1432 | |
c3895391 AK |
1433 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
1434 | void **head, void **tail) | |
81084651 JDB |
1435 | { |
1436 | /* | |
1437 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1438 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1439 | */ | |
4675ff05 | 1440 | #if defined(CONFIG_LOCKDEP) || \ |
81084651 JDB |
1441 | defined(CONFIG_DEBUG_KMEMLEAK) || \ |
1442 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1443 | defined(CONFIG_KASAN) | |
1444 | ||
c3895391 AK |
1445 | void *object; |
1446 | void *next = *head; | |
1447 | void *old_tail = *tail ? *tail : *head; | |
1448 | ||
1449 | /* Head and tail of the reconstructed freelist */ | |
1450 | *head = NULL; | |
1451 | *tail = NULL; | |
81084651 JDB |
1452 | |
1453 | do { | |
c3895391 AK |
1454 | object = next; |
1455 | next = get_freepointer(s, object); | |
1456 | /* If object's reuse doesn't have to be delayed */ | |
1457 | if (!slab_free_hook(s, object)) { | |
1458 | /* Move object to the new freelist */ | |
1459 | set_freepointer(s, object, *head); | |
1460 | *head = object; | |
1461 | if (!*tail) | |
1462 | *tail = object; | |
1463 | } | |
1464 | } while (object != old_tail); | |
1465 | ||
1466 | if (*head == *tail) | |
1467 | *tail = NULL; | |
1468 | ||
1469 | return *head != NULL; | |
1470 | #else | |
1471 | return true; | |
81084651 JDB |
1472 | #endif |
1473 | } | |
1474 | ||
4d176711 | 1475 | static void *setup_object(struct kmem_cache *s, struct page *page, |
588f8ba9 TG |
1476 | void *object) |
1477 | { | |
1478 | setup_object_debug(s, page, object); | |
4d176711 | 1479 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1480 | if (unlikely(s->ctor)) { |
1481 | kasan_unpoison_object_data(s, object); | |
1482 | s->ctor(object); | |
1483 | kasan_poison_object_data(s, object); | |
1484 | } | |
4d176711 | 1485 | return object; |
588f8ba9 TG |
1486 | } |
1487 | ||
81819f0f CL |
1488 | /* |
1489 | * Slab allocation and freeing | |
1490 | */ | |
5dfb4175 VD |
1491 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1492 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1493 | { |
5dfb4175 | 1494 | struct page *page; |
19af27af | 1495 | unsigned int order = oo_order(oo); |
65c3376a | 1496 | |
2154a336 | 1497 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1498 | page = alloc_pages(flags, order); |
65c3376a | 1499 | else |
96db800f | 1500 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1501 | |
f3ccb2c4 VD |
1502 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1503 | __free_pages(page, order); | |
1504 | page = NULL; | |
1505 | } | |
5dfb4175 VD |
1506 | |
1507 | return page; | |
65c3376a CL |
1508 | } |
1509 | ||
210e7a43 TG |
1510 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1511 | /* Pre-initialize the random sequence cache */ | |
1512 | static int init_cache_random_seq(struct kmem_cache *s) | |
1513 | { | |
19af27af | 1514 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1515 | int err; |
210e7a43 | 1516 | |
a810007a SR |
1517 | /* Bailout if already initialised */ |
1518 | if (s->random_seq) | |
1519 | return 0; | |
1520 | ||
210e7a43 TG |
1521 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1522 | if (err) { | |
1523 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1524 | s->name); | |
1525 | return err; | |
1526 | } | |
1527 | ||
1528 | /* Transform to an offset on the set of pages */ | |
1529 | if (s->random_seq) { | |
19af27af AD |
1530 | unsigned int i; |
1531 | ||
210e7a43 TG |
1532 | for (i = 0; i < count; i++) |
1533 | s->random_seq[i] *= s->size; | |
1534 | } | |
1535 | return 0; | |
1536 | } | |
1537 | ||
1538 | /* Initialize each random sequence freelist per cache */ | |
1539 | static void __init init_freelist_randomization(void) | |
1540 | { | |
1541 | struct kmem_cache *s; | |
1542 | ||
1543 | mutex_lock(&slab_mutex); | |
1544 | ||
1545 | list_for_each_entry(s, &slab_caches, list) | |
1546 | init_cache_random_seq(s); | |
1547 | ||
1548 | mutex_unlock(&slab_mutex); | |
1549 | } | |
1550 | ||
1551 | /* Get the next entry on the pre-computed freelist randomized */ | |
1552 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1553 | unsigned long *pos, void *start, | |
1554 | unsigned long page_limit, | |
1555 | unsigned long freelist_count) | |
1556 | { | |
1557 | unsigned int idx; | |
1558 | ||
1559 | /* | |
1560 | * If the target page allocation failed, the number of objects on the | |
1561 | * page might be smaller than the usual size defined by the cache. | |
1562 | */ | |
1563 | do { | |
1564 | idx = s->random_seq[*pos]; | |
1565 | *pos += 1; | |
1566 | if (*pos >= freelist_count) | |
1567 | *pos = 0; | |
1568 | } while (unlikely(idx >= page_limit)); | |
1569 | ||
1570 | return (char *)start + idx; | |
1571 | } | |
1572 | ||
1573 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1574 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1575 | { | |
1576 | void *start; | |
1577 | void *cur; | |
1578 | void *next; | |
1579 | unsigned long idx, pos, page_limit, freelist_count; | |
1580 | ||
1581 | if (page->objects < 2 || !s->random_seq) | |
1582 | return false; | |
1583 | ||
1584 | freelist_count = oo_objects(s->oo); | |
1585 | pos = get_random_int() % freelist_count; | |
1586 | ||
1587 | page_limit = page->objects * s->size; | |
1588 | start = fixup_red_left(s, page_address(page)); | |
1589 | ||
1590 | /* First entry is used as the base of the freelist */ | |
1591 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1592 | freelist_count); | |
4d176711 | 1593 | cur = setup_object(s, page, cur); |
210e7a43 TG |
1594 | page->freelist = cur; |
1595 | ||
1596 | for (idx = 1; idx < page->objects; idx++) { | |
210e7a43 TG |
1597 | next = next_freelist_entry(s, page, &pos, start, page_limit, |
1598 | freelist_count); | |
4d176711 | 1599 | next = setup_object(s, page, next); |
210e7a43 TG |
1600 | set_freepointer(s, cur, next); |
1601 | cur = next; | |
1602 | } | |
210e7a43 TG |
1603 | set_freepointer(s, cur, NULL); |
1604 | ||
1605 | return true; | |
1606 | } | |
1607 | #else | |
1608 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1609 | { | |
1610 | return 0; | |
1611 | } | |
1612 | static inline void init_freelist_randomization(void) { } | |
1613 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1614 | { | |
1615 | return false; | |
1616 | } | |
1617 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1618 | ||
81819f0f CL |
1619 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1620 | { | |
06428780 | 1621 | struct page *page; |
834f3d11 | 1622 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1623 | gfp_t alloc_gfp; |
4d176711 | 1624 | void *start, *p, *next; |
588f8ba9 | 1625 | int idx, order; |
210e7a43 | 1626 | bool shuffle; |
81819f0f | 1627 | |
7e0528da CL |
1628 | flags &= gfp_allowed_mask; |
1629 | ||
d0164adc | 1630 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1631 | local_irq_enable(); |
1632 | ||
b7a49f0d | 1633 | flags |= s->allocflags; |
e12ba74d | 1634 | |
ba52270d PE |
1635 | /* |
1636 | * Let the initial higher-order allocation fail under memory pressure | |
1637 | * so we fall-back to the minimum order allocation. | |
1638 | */ | |
1639 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1640 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1641 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1642 | |
5dfb4175 | 1643 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1644 | if (unlikely(!page)) { |
1645 | oo = s->min; | |
80c3a998 | 1646 | alloc_gfp = flags; |
65c3376a CL |
1647 | /* |
1648 | * Allocation may have failed due to fragmentation. | |
1649 | * Try a lower order alloc if possible | |
1650 | */ | |
5dfb4175 | 1651 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1652 | if (unlikely(!page)) |
1653 | goto out; | |
1654 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1655 | } |
5a896d9e | 1656 | |
834f3d11 | 1657 | page->objects = oo_objects(oo); |
81819f0f | 1658 | |
1f458cbf | 1659 | order = compound_order(page); |
1b4f59e3 | 1660 | page->slab_cache = s; |
c03f94cc | 1661 | __SetPageSlab(page); |
2f064f34 | 1662 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1663 | SetPageSlabPfmemalloc(page); |
81819f0f | 1664 | |
a7101224 | 1665 | kasan_poison_slab(page); |
81819f0f | 1666 | |
a7101224 | 1667 | start = page_address(page); |
81819f0f | 1668 | |
a7101224 | 1669 | setup_page_debug(s, start, order); |
0316bec2 | 1670 | |
210e7a43 TG |
1671 | shuffle = shuffle_freelist(s, page); |
1672 | ||
1673 | if (!shuffle) { | |
4d176711 AK |
1674 | start = fixup_red_left(s, start); |
1675 | start = setup_object(s, page, start); | |
1676 | page->freelist = start; | |
18e50661 AK |
1677 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { |
1678 | next = p + s->size; | |
1679 | next = setup_object(s, page, next); | |
1680 | set_freepointer(s, p, next); | |
1681 | p = next; | |
1682 | } | |
1683 | set_freepointer(s, p, NULL); | |
81819f0f | 1684 | } |
81819f0f | 1685 | |
e6e82ea1 | 1686 | page->inuse = page->objects; |
8cb0a506 | 1687 | page->frozen = 1; |
588f8ba9 | 1688 | |
81819f0f | 1689 | out: |
d0164adc | 1690 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1691 | local_irq_disable(); |
1692 | if (!page) | |
1693 | return NULL; | |
1694 | ||
7779f212 | 1695 | mod_lruvec_page_state(page, |
588f8ba9 TG |
1696 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? |
1697 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1698 | 1 << oo_order(oo)); | |
1699 | ||
1700 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1701 | ||
81819f0f CL |
1702 | return page; |
1703 | } | |
1704 | ||
588f8ba9 TG |
1705 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1706 | { | |
1707 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
bacdcb34 | 1708 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
1709 | flags &= ~GFP_SLAB_BUG_MASK; |
1710 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
1711 | invalid_mask, &invalid_mask, flags, &flags); | |
65b9de75 | 1712 | dump_stack(); |
588f8ba9 TG |
1713 | } |
1714 | ||
1715 | return allocate_slab(s, | |
1716 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1717 | } | |
1718 | ||
81819f0f CL |
1719 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1720 | { | |
834f3d11 CL |
1721 | int order = compound_order(page); |
1722 | int pages = 1 << order; | |
81819f0f | 1723 | |
becfda68 | 1724 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { |
81819f0f CL |
1725 | void *p; |
1726 | ||
1727 | slab_pad_check(s, page); | |
224a88be CL |
1728 | for_each_object(p, s, page_address(page), |
1729 | page->objects) | |
f7cb1933 | 1730 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1731 | } |
1732 | ||
7779f212 | 1733 | mod_lruvec_page_state(page, |
81819f0f CL |
1734 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? |
1735 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1736 | -pages); |
81819f0f | 1737 | |
072bb0aa | 1738 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1739 | __ClearPageSlab(page); |
1f458cbf | 1740 | |
d4fc5069 | 1741 | page->mapping = NULL; |
1eb5ac64 NP |
1742 | if (current->reclaim_state) |
1743 | current->reclaim_state->reclaimed_slab += pages; | |
27ee57c9 VD |
1744 | memcg_uncharge_slab(page, order, s); |
1745 | __free_pages(page, order); | |
81819f0f CL |
1746 | } |
1747 | ||
1748 | static void rcu_free_slab(struct rcu_head *h) | |
1749 | { | |
bf68c214 | 1750 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 1751 | |
1b4f59e3 | 1752 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1753 | } |
1754 | ||
1755 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1756 | { | |
5f0d5a3a | 1757 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 1758 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
1759 | } else |
1760 | __free_slab(s, page); | |
1761 | } | |
1762 | ||
1763 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1764 | { | |
205ab99d | 1765 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1766 | free_slab(s, page); |
1767 | } | |
1768 | ||
1769 | /* | |
5cc6eee8 | 1770 | * Management of partially allocated slabs. |
81819f0f | 1771 | */ |
1e4dd946 SR |
1772 | static inline void |
1773 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1774 | { |
e95eed57 | 1775 | n->nr_partial++; |
136333d1 | 1776 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1777 | list_add_tail(&page->lru, &n->partial); |
1778 | else | |
1779 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1780 | } |
1781 | ||
1e4dd946 SR |
1782 | static inline void add_partial(struct kmem_cache_node *n, |
1783 | struct page *page, int tail) | |
62e346a8 | 1784 | { |
c65c1877 | 1785 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1786 | __add_partial(n, page, tail); |
1787 | } | |
c65c1877 | 1788 | |
1e4dd946 SR |
1789 | static inline void remove_partial(struct kmem_cache_node *n, |
1790 | struct page *page) | |
1791 | { | |
1792 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1793 | list_del(&page->lru); |
1794 | n->nr_partial--; | |
1e4dd946 SR |
1795 | } |
1796 | ||
81819f0f | 1797 | /* |
7ced3719 CL |
1798 | * Remove slab from the partial list, freeze it and |
1799 | * return the pointer to the freelist. | |
81819f0f | 1800 | * |
497b66f2 | 1801 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1802 | */ |
497b66f2 | 1803 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1804 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1805 | int mode, int *objects) |
81819f0f | 1806 | { |
2cfb7455 CL |
1807 | void *freelist; |
1808 | unsigned long counters; | |
1809 | struct page new; | |
1810 | ||
c65c1877 PZ |
1811 | lockdep_assert_held(&n->list_lock); |
1812 | ||
2cfb7455 CL |
1813 | /* |
1814 | * Zap the freelist and set the frozen bit. | |
1815 | * The old freelist is the list of objects for the | |
1816 | * per cpu allocation list. | |
1817 | */ | |
7ced3719 CL |
1818 | freelist = page->freelist; |
1819 | counters = page->counters; | |
1820 | new.counters = counters; | |
633b0764 | 1821 | *objects = new.objects - new.inuse; |
23910c50 | 1822 | if (mode) { |
7ced3719 | 1823 | new.inuse = page->objects; |
23910c50 PE |
1824 | new.freelist = NULL; |
1825 | } else { | |
1826 | new.freelist = freelist; | |
1827 | } | |
2cfb7455 | 1828 | |
a0132ac0 | 1829 | VM_BUG_ON(new.frozen); |
7ced3719 | 1830 | new.frozen = 1; |
2cfb7455 | 1831 | |
7ced3719 | 1832 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1833 | freelist, counters, |
02d7633f | 1834 | new.freelist, new.counters, |
7ced3719 | 1835 | "acquire_slab")) |
7ced3719 | 1836 | return NULL; |
2cfb7455 CL |
1837 | |
1838 | remove_partial(n, page); | |
7ced3719 | 1839 | WARN_ON(!freelist); |
49e22585 | 1840 | return freelist; |
81819f0f CL |
1841 | } |
1842 | ||
633b0764 | 1843 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1844 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1845 | |
81819f0f | 1846 | /* |
672bba3a | 1847 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1848 | */ |
8ba00bb6 JK |
1849 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1850 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1851 | { |
49e22585 CL |
1852 | struct page *page, *page2; |
1853 | void *object = NULL; | |
e5d9998f | 1854 | unsigned int available = 0; |
633b0764 | 1855 | int objects; |
81819f0f CL |
1856 | |
1857 | /* | |
1858 | * Racy check. If we mistakenly see no partial slabs then we | |
1859 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1860 | * partial slab and there is none available then get_partials() |
1861 | * will return NULL. | |
81819f0f CL |
1862 | */ |
1863 | if (!n || !n->nr_partial) | |
1864 | return NULL; | |
1865 | ||
1866 | spin_lock(&n->list_lock); | |
49e22585 | 1867 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1868 | void *t; |
49e22585 | 1869 | |
8ba00bb6 JK |
1870 | if (!pfmemalloc_match(page, flags)) |
1871 | continue; | |
1872 | ||
633b0764 | 1873 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1874 | if (!t) |
1875 | break; | |
1876 | ||
633b0764 | 1877 | available += objects; |
12d79634 | 1878 | if (!object) { |
49e22585 | 1879 | c->page = page; |
49e22585 | 1880 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1881 | object = t; |
49e22585 | 1882 | } else { |
633b0764 | 1883 | put_cpu_partial(s, page, 0); |
8028dcea | 1884 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1885 | } |
345c905d | 1886 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 1887 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
1888 | break; |
1889 | ||
497b66f2 | 1890 | } |
81819f0f | 1891 | spin_unlock(&n->list_lock); |
497b66f2 | 1892 | return object; |
81819f0f CL |
1893 | } |
1894 | ||
1895 | /* | |
672bba3a | 1896 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1897 | */ |
de3ec035 | 1898 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1899 | struct kmem_cache_cpu *c) |
81819f0f CL |
1900 | { |
1901 | #ifdef CONFIG_NUMA | |
1902 | struct zonelist *zonelist; | |
dd1a239f | 1903 | struct zoneref *z; |
54a6eb5c MG |
1904 | struct zone *zone; |
1905 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1906 | void *object; |
cc9a6c87 | 1907 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1908 | |
1909 | /* | |
672bba3a CL |
1910 | * The defrag ratio allows a configuration of the tradeoffs between |
1911 | * inter node defragmentation and node local allocations. A lower | |
1912 | * defrag_ratio increases the tendency to do local allocations | |
1913 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1914 | * |
672bba3a CL |
1915 | * If the defrag_ratio is set to 0 then kmalloc() always |
1916 | * returns node local objects. If the ratio is higher then kmalloc() | |
1917 | * may return off node objects because partial slabs are obtained | |
1918 | * from other nodes and filled up. | |
81819f0f | 1919 | * |
43efd3ea LP |
1920 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
1921 | * (which makes defrag_ratio = 1000) then every (well almost) | |
1922 | * allocation will first attempt to defrag slab caches on other nodes. | |
1923 | * This means scanning over all nodes to look for partial slabs which | |
1924 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 1925 | * with available objects. |
81819f0f | 1926 | */ |
9824601e CL |
1927 | if (!s->remote_node_defrag_ratio || |
1928 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1929 | return NULL; |
1930 | ||
cc9a6c87 | 1931 | do { |
d26914d1 | 1932 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1933 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1934 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1935 | struct kmem_cache_node *n; | |
1936 | ||
1937 | n = get_node(s, zone_to_nid(zone)); | |
1938 | ||
dee2f8aa | 1939 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1940 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1941 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1942 | if (object) { |
1943 | /* | |
d26914d1 MG |
1944 | * Don't check read_mems_allowed_retry() |
1945 | * here - if mems_allowed was updated in | |
1946 | * parallel, that was a harmless race | |
1947 | * between allocation and the cpuset | |
1948 | * update | |
cc9a6c87 | 1949 | */ |
cc9a6c87 MG |
1950 | return object; |
1951 | } | |
c0ff7453 | 1952 | } |
81819f0f | 1953 | } |
d26914d1 | 1954 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1955 | #endif |
1956 | return NULL; | |
1957 | } | |
1958 | ||
1959 | /* | |
1960 | * Get a partial page, lock it and return it. | |
1961 | */ | |
497b66f2 | 1962 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1963 | struct kmem_cache_cpu *c) |
81819f0f | 1964 | { |
497b66f2 | 1965 | void *object; |
a561ce00 JK |
1966 | int searchnode = node; |
1967 | ||
1968 | if (node == NUMA_NO_NODE) | |
1969 | searchnode = numa_mem_id(); | |
1970 | else if (!node_present_pages(node)) | |
1971 | searchnode = node_to_mem_node(node); | |
81819f0f | 1972 | |
8ba00bb6 | 1973 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1974 | if (object || node != NUMA_NO_NODE) |
1975 | return object; | |
81819f0f | 1976 | |
acd19fd1 | 1977 | return get_any_partial(s, flags, c); |
81819f0f CL |
1978 | } |
1979 | ||
8a5ec0ba CL |
1980 | #ifdef CONFIG_PREEMPT |
1981 | /* | |
1982 | * Calculate the next globally unique transaction for disambiguiation | |
1983 | * during cmpxchg. The transactions start with the cpu number and are then | |
1984 | * incremented by CONFIG_NR_CPUS. | |
1985 | */ | |
1986 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1987 | #else | |
1988 | /* | |
1989 | * No preemption supported therefore also no need to check for | |
1990 | * different cpus. | |
1991 | */ | |
1992 | #define TID_STEP 1 | |
1993 | #endif | |
1994 | ||
1995 | static inline unsigned long next_tid(unsigned long tid) | |
1996 | { | |
1997 | return tid + TID_STEP; | |
1998 | } | |
1999 | ||
2000 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
2001 | { | |
2002 | return tid % TID_STEP; | |
2003 | } | |
2004 | ||
2005 | static inline unsigned long tid_to_event(unsigned long tid) | |
2006 | { | |
2007 | return tid / TID_STEP; | |
2008 | } | |
2009 | ||
2010 | static inline unsigned int init_tid(int cpu) | |
2011 | { | |
2012 | return cpu; | |
2013 | } | |
2014 | ||
2015 | static inline void note_cmpxchg_failure(const char *n, | |
2016 | const struct kmem_cache *s, unsigned long tid) | |
2017 | { | |
2018 | #ifdef SLUB_DEBUG_CMPXCHG | |
2019 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2020 | ||
f9f58285 | 2021 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
2022 | |
2023 | #ifdef CONFIG_PREEMPT | |
2024 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 2025 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2026 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2027 | else | |
2028 | #endif | |
2029 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2030 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2031 | tid_to_event(tid), tid_to_event(actual_tid)); |
2032 | else | |
f9f58285 | 2033 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2034 | actual_tid, tid, next_tid(tid)); |
2035 | #endif | |
4fdccdfb | 2036 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2037 | } |
2038 | ||
788e1aad | 2039 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2040 | { |
8a5ec0ba CL |
2041 | int cpu; |
2042 | ||
2043 | for_each_possible_cpu(cpu) | |
2044 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 2045 | } |
2cfb7455 | 2046 | |
81819f0f CL |
2047 | /* |
2048 | * Remove the cpu slab | |
2049 | */ | |
d0e0ac97 | 2050 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
d4ff6d35 | 2051 | void *freelist, struct kmem_cache_cpu *c) |
81819f0f | 2052 | { |
2cfb7455 | 2053 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
2054 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
2055 | int lock = 0; | |
2056 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 2057 | void *nextfree; |
136333d1 | 2058 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2059 | struct page new; |
2060 | struct page old; | |
2061 | ||
2062 | if (page->freelist) { | |
84e554e6 | 2063 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2064 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2065 | } |
2066 | ||
894b8788 | 2067 | /* |
2cfb7455 CL |
2068 | * Stage one: Free all available per cpu objects back |
2069 | * to the page freelist while it is still frozen. Leave the | |
2070 | * last one. | |
2071 | * | |
2072 | * There is no need to take the list->lock because the page | |
2073 | * is still frozen. | |
2074 | */ | |
2075 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2076 | void *prior; | |
2077 | unsigned long counters; | |
2078 | ||
2079 | do { | |
2080 | prior = page->freelist; | |
2081 | counters = page->counters; | |
2082 | set_freepointer(s, freelist, prior); | |
2083 | new.counters = counters; | |
2084 | new.inuse--; | |
a0132ac0 | 2085 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2086 | |
1d07171c | 2087 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2088 | prior, counters, |
2089 | freelist, new.counters, | |
2090 | "drain percpu freelist")); | |
2091 | ||
2092 | freelist = nextfree; | |
2093 | } | |
2094 | ||
894b8788 | 2095 | /* |
2cfb7455 CL |
2096 | * Stage two: Ensure that the page is unfrozen while the |
2097 | * list presence reflects the actual number of objects | |
2098 | * during unfreeze. | |
2099 | * | |
2100 | * We setup the list membership and then perform a cmpxchg | |
2101 | * with the count. If there is a mismatch then the page | |
2102 | * is not unfrozen but the page is on the wrong list. | |
2103 | * | |
2104 | * Then we restart the process which may have to remove | |
2105 | * the page from the list that we just put it on again | |
2106 | * because the number of objects in the slab may have | |
2107 | * changed. | |
894b8788 | 2108 | */ |
2cfb7455 | 2109 | redo: |
894b8788 | 2110 | |
2cfb7455 CL |
2111 | old.freelist = page->freelist; |
2112 | old.counters = page->counters; | |
a0132ac0 | 2113 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2114 | |
2cfb7455 CL |
2115 | /* Determine target state of the slab */ |
2116 | new.counters = old.counters; | |
2117 | if (freelist) { | |
2118 | new.inuse--; | |
2119 | set_freepointer(s, freelist, old.freelist); | |
2120 | new.freelist = freelist; | |
2121 | } else | |
2122 | new.freelist = old.freelist; | |
2123 | ||
2124 | new.frozen = 0; | |
2125 | ||
8a5b20ae | 2126 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2127 | m = M_FREE; |
2128 | else if (new.freelist) { | |
2129 | m = M_PARTIAL; | |
2130 | if (!lock) { | |
2131 | lock = 1; | |
2132 | /* | |
2133 | * Taking the spinlock removes the possiblity | |
2134 | * that acquire_slab() will see a slab page that | |
2135 | * is frozen | |
2136 | */ | |
2137 | spin_lock(&n->list_lock); | |
2138 | } | |
2139 | } else { | |
2140 | m = M_FULL; | |
2141 | if (kmem_cache_debug(s) && !lock) { | |
2142 | lock = 1; | |
2143 | /* | |
2144 | * This also ensures that the scanning of full | |
2145 | * slabs from diagnostic functions will not see | |
2146 | * any frozen slabs. | |
2147 | */ | |
2148 | spin_lock(&n->list_lock); | |
2149 | } | |
2150 | } | |
2151 | ||
2152 | if (l != m) { | |
2cfb7455 | 2153 | if (l == M_PARTIAL) |
2cfb7455 | 2154 | remove_partial(n, page); |
2cfb7455 | 2155 | else if (l == M_FULL) |
c65c1877 | 2156 | remove_full(s, n, page); |
2cfb7455 | 2157 | |
88349a28 | 2158 | if (m == M_PARTIAL) |
2cfb7455 | 2159 | add_partial(n, page, tail); |
88349a28 | 2160 | else if (m == M_FULL) |
2cfb7455 | 2161 | add_full(s, n, page); |
2cfb7455 CL |
2162 | } |
2163 | ||
2164 | l = m; | |
1d07171c | 2165 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2166 | old.freelist, old.counters, |
2167 | new.freelist, new.counters, | |
2168 | "unfreezing slab")) | |
2169 | goto redo; | |
2170 | ||
2cfb7455 CL |
2171 | if (lock) |
2172 | spin_unlock(&n->list_lock); | |
2173 | ||
88349a28 WY |
2174 | if (m == M_PARTIAL) |
2175 | stat(s, tail); | |
2176 | else if (m == M_FULL) | |
2177 | stat(s, DEACTIVATE_FULL); | |
2178 | else if (m == M_FREE) { | |
2cfb7455 CL |
2179 | stat(s, DEACTIVATE_EMPTY); |
2180 | discard_slab(s, page); | |
2181 | stat(s, FREE_SLAB); | |
894b8788 | 2182 | } |
d4ff6d35 WY |
2183 | |
2184 | c->page = NULL; | |
2185 | c->freelist = NULL; | |
81819f0f CL |
2186 | } |
2187 | ||
d24ac77f JK |
2188 | /* |
2189 | * Unfreeze all the cpu partial slabs. | |
2190 | * | |
59a09917 CL |
2191 | * This function must be called with interrupts disabled |
2192 | * for the cpu using c (or some other guarantee must be there | |
2193 | * to guarantee no concurrent accesses). | |
d24ac77f | 2194 | */ |
59a09917 CL |
2195 | static void unfreeze_partials(struct kmem_cache *s, |
2196 | struct kmem_cache_cpu *c) | |
49e22585 | 2197 | { |
345c905d | 2198 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2199 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2200 | struct page *page, *discard_page = NULL; |
49e22585 CL |
2201 | |
2202 | while ((page = c->partial)) { | |
49e22585 CL |
2203 | struct page new; |
2204 | struct page old; | |
2205 | ||
2206 | c->partial = page->next; | |
43d77867 JK |
2207 | |
2208 | n2 = get_node(s, page_to_nid(page)); | |
2209 | if (n != n2) { | |
2210 | if (n) | |
2211 | spin_unlock(&n->list_lock); | |
2212 | ||
2213 | n = n2; | |
2214 | spin_lock(&n->list_lock); | |
2215 | } | |
49e22585 CL |
2216 | |
2217 | do { | |
2218 | ||
2219 | old.freelist = page->freelist; | |
2220 | old.counters = page->counters; | |
a0132ac0 | 2221 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2222 | |
2223 | new.counters = old.counters; | |
2224 | new.freelist = old.freelist; | |
2225 | ||
2226 | new.frozen = 0; | |
2227 | ||
d24ac77f | 2228 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2229 | old.freelist, old.counters, |
2230 | new.freelist, new.counters, | |
2231 | "unfreezing slab")); | |
2232 | ||
8a5b20ae | 2233 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2234 | page->next = discard_page; |
2235 | discard_page = page; | |
43d77867 JK |
2236 | } else { |
2237 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2238 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2239 | } |
2240 | } | |
2241 | ||
2242 | if (n) | |
2243 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2244 | |
2245 | while (discard_page) { | |
2246 | page = discard_page; | |
2247 | discard_page = discard_page->next; | |
2248 | ||
2249 | stat(s, DEACTIVATE_EMPTY); | |
2250 | discard_slab(s, page); | |
2251 | stat(s, FREE_SLAB); | |
2252 | } | |
345c905d | 2253 | #endif |
49e22585 CL |
2254 | } |
2255 | ||
2256 | /* | |
2257 | * Put a page that was just frozen (in __slab_free) into a partial page | |
0d2d5d40 | 2258 | * slot if available. |
49e22585 CL |
2259 | * |
2260 | * If we did not find a slot then simply move all the partials to the | |
2261 | * per node partial list. | |
2262 | */ | |
633b0764 | 2263 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2264 | { |
345c905d | 2265 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2266 | struct page *oldpage; |
2267 | int pages; | |
2268 | int pobjects; | |
2269 | ||
d6e0b7fa | 2270 | preempt_disable(); |
49e22585 CL |
2271 | do { |
2272 | pages = 0; | |
2273 | pobjects = 0; | |
2274 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2275 | ||
2276 | if (oldpage) { | |
2277 | pobjects = oldpage->pobjects; | |
2278 | pages = oldpage->pages; | |
2279 | if (drain && pobjects > s->cpu_partial) { | |
2280 | unsigned long flags; | |
2281 | /* | |
2282 | * partial array is full. Move the existing | |
2283 | * set to the per node partial list. | |
2284 | */ | |
2285 | local_irq_save(flags); | |
59a09917 | 2286 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2287 | local_irq_restore(flags); |
e24fc410 | 2288 | oldpage = NULL; |
49e22585 CL |
2289 | pobjects = 0; |
2290 | pages = 0; | |
8028dcea | 2291 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2292 | } |
2293 | } | |
2294 | ||
2295 | pages++; | |
2296 | pobjects += page->objects - page->inuse; | |
2297 | ||
2298 | page->pages = pages; | |
2299 | page->pobjects = pobjects; | |
2300 | page->next = oldpage; | |
2301 | ||
d0e0ac97 CG |
2302 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2303 | != oldpage); | |
d6e0b7fa VD |
2304 | if (unlikely(!s->cpu_partial)) { |
2305 | unsigned long flags; | |
2306 | ||
2307 | local_irq_save(flags); | |
2308 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2309 | local_irq_restore(flags); | |
2310 | } | |
2311 | preempt_enable(); | |
345c905d | 2312 | #endif |
49e22585 CL |
2313 | } |
2314 | ||
dfb4f096 | 2315 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2316 | { |
84e554e6 | 2317 | stat(s, CPUSLAB_FLUSH); |
d4ff6d35 | 2318 | deactivate_slab(s, c->page, c->freelist, c); |
c17dda40 CL |
2319 | |
2320 | c->tid = next_tid(c->tid); | |
81819f0f CL |
2321 | } |
2322 | ||
2323 | /* | |
2324 | * Flush cpu slab. | |
6446faa2 | 2325 | * |
81819f0f CL |
2326 | * Called from IPI handler with interrupts disabled. |
2327 | */ | |
0c710013 | 2328 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2329 | { |
9dfc6e68 | 2330 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2331 | |
1265ef2d WY |
2332 | if (c->page) |
2333 | flush_slab(s, c); | |
49e22585 | 2334 | |
1265ef2d | 2335 | unfreeze_partials(s, c); |
81819f0f CL |
2336 | } |
2337 | ||
2338 | static void flush_cpu_slab(void *d) | |
2339 | { | |
2340 | struct kmem_cache *s = d; | |
81819f0f | 2341 | |
dfb4f096 | 2342 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2343 | } |
2344 | ||
a8364d55 GBY |
2345 | static bool has_cpu_slab(int cpu, void *info) |
2346 | { | |
2347 | struct kmem_cache *s = info; | |
2348 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2349 | ||
a93cf07b | 2350 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2351 | } |
2352 | ||
81819f0f CL |
2353 | static void flush_all(struct kmem_cache *s) |
2354 | { | |
a8364d55 | 2355 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2356 | } |
2357 | ||
a96a87bf SAS |
2358 | /* |
2359 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2360 | * necessary. | |
2361 | */ | |
2362 | static int slub_cpu_dead(unsigned int cpu) | |
2363 | { | |
2364 | struct kmem_cache *s; | |
2365 | unsigned long flags; | |
2366 | ||
2367 | mutex_lock(&slab_mutex); | |
2368 | list_for_each_entry(s, &slab_caches, list) { | |
2369 | local_irq_save(flags); | |
2370 | __flush_cpu_slab(s, cpu); | |
2371 | local_irq_restore(flags); | |
2372 | } | |
2373 | mutex_unlock(&slab_mutex); | |
2374 | return 0; | |
2375 | } | |
2376 | ||
dfb4f096 CL |
2377 | /* |
2378 | * Check if the objects in a per cpu structure fit numa | |
2379 | * locality expectations. | |
2380 | */ | |
57d437d2 | 2381 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2382 | { |
2383 | #ifdef CONFIG_NUMA | |
6159d0f5 | 2384 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2385 | return 0; |
2386 | #endif | |
2387 | return 1; | |
2388 | } | |
2389 | ||
9a02d699 | 2390 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2391 | static int count_free(struct page *page) |
2392 | { | |
2393 | return page->objects - page->inuse; | |
2394 | } | |
2395 | ||
9a02d699 DR |
2396 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2397 | { | |
2398 | return atomic_long_read(&n->total_objects); | |
2399 | } | |
2400 | #endif /* CONFIG_SLUB_DEBUG */ | |
2401 | ||
2402 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2403 | static unsigned long count_partial(struct kmem_cache_node *n, |
2404 | int (*get_count)(struct page *)) | |
2405 | { | |
2406 | unsigned long flags; | |
2407 | unsigned long x = 0; | |
2408 | struct page *page; | |
2409 | ||
2410 | spin_lock_irqsave(&n->list_lock, flags); | |
2411 | list_for_each_entry(page, &n->partial, lru) | |
2412 | x += get_count(page); | |
2413 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2414 | return x; | |
2415 | } | |
9a02d699 | 2416 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2417 | |
781b2ba6 PE |
2418 | static noinline void |
2419 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2420 | { | |
9a02d699 DR |
2421 | #ifdef CONFIG_SLUB_DEBUG |
2422 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2423 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2424 | int node; |
fa45dc25 | 2425 | struct kmem_cache_node *n; |
781b2ba6 | 2426 | |
9a02d699 DR |
2427 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2428 | return; | |
2429 | ||
5b3810e5 VB |
2430 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2431 | nid, gfpflags, &gfpflags); | |
19af27af | 2432 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2433 | s->name, s->object_size, s->size, oo_order(s->oo), |
2434 | oo_order(s->min)); | |
781b2ba6 | 2435 | |
3b0efdfa | 2436 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2437 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2438 | s->name); | |
fa5ec8a1 | 2439 | |
fa45dc25 | 2440 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2441 | unsigned long nr_slabs; |
2442 | unsigned long nr_objs; | |
2443 | unsigned long nr_free; | |
2444 | ||
26c02cf0 AB |
2445 | nr_free = count_partial(n, count_free); |
2446 | nr_slabs = node_nr_slabs(n); | |
2447 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2448 | |
f9f58285 | 2449 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2450 | node, nr_slabs, nr_objs, nr_free); |
2451 | } | |
9a02d699 | 2452 | #endif |
781b2ba6 PE |
2453 | } |
2454 | ||
497b66f2 CL |
2455 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2456 | int node, struct kmem_cache_cpu **pc) | |
2457 | { | |
6faa6833 | 2458 | void *freelist; |
188fd063 CL |
2459 | struct kmem_cache_cpu *c = *pc; |
2460 | struct page *page; | |
497b66f2 | 2461 | |
128227e7 MW |
2462 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2463 | ||
188fd063 | 2464 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2465 | |
188fd063 CL |
2466 | if (freelist) |
2467 | return freelist; | |
2468 | ||
2469 | page = new_slab(s, flags, node); | |
497b66f2 | 2470 | if (page) { |
7c8e0181 | 2471 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2472 | if (c->page) |
2473 | flush_slab(s, c); | |
2474 | ||
2475 | /* | |
2476 | * No other reference to the page yet so we can | |
2477 | * muck around with it freely without cmpxchg | |
2478 | */ | |
6faa6833 | 2479 | freelist = page->freelist; |
497b66f2 CL |
2480 | page->freelist = NULL; |
2481 | ||
2482 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2483 | c->page = page; |
2484 | *pc = c; | |
edde82b6 | 2485 | } |
497b66f2 | 2486 | |
6faa6833 | 2487 | return freelist; |
497b66f2 CL |
2488 | } |
2489 | ||
072bb0aa MG |
2490 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2491 | { | |
2492 | if (unlikely(PageSlabPfmemalloc(page))) | |
2493 | return gfp_pfmemalloc_allowed(gfpflags); | |
2494 | ||
2495 | return true; | |
2496 | } | |
2497 | ||
213eeb9f | 2498 | /* |
d0e0ac97 CG |
2499 | * Check the page->freelist of a page and either transfer the freelist to the |
2500 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2501 | * |
2502 | * The page is still frozen if the return value is not NULL. | |
2503 | * | |
2504 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2505 | * |
2506 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2507 | */ |
2508 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2509 | { | |
2510 | struct page new; | |
2511 | unsigned long counters; | |
2512 | void *freelist; | |
2513 | ||
2514 | do { | |
2515 | freelist = page->freelist; | |
2516 | counters = page->counters; | |
6faa6833 | 2517 | |
213eeb9f | 2518 | new.counters = counters; |
a0132ac0 | 2519 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2520 | |
2521 | new.inuse = page->objects; | |
2522 | new.frozen = freelist != NULL; | |
2523 | ||
d24ac77f | 2524 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2525 | freelist, counters, |
2526 | NULL, new.counters, | |
2527 | "get_freelist")); | |
2528 | ||
2529 | return freelist; | |
2530 | } | |
2531 | ||
81819f0f | 2532 | /* |
894b8788 CL |
2533 | * Slow path. The lockless freelist is empty or we need to perform |
2534 | * debugging duties. | |
2535 | * | |
894b8788 CL |
2536 | * Processing is still very fast if new objects have been freed to the |
2537 | * regular freelist. In that case we simply take over the regular freelist | |
2538 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2539 | * |
894b8788 CL |
2540 | * If that is not working then we fall back to the partial lists. We take the |
2541 | * first element of the freelist as the object to allocate now and move the | |
2542 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2543 | * |
894b8788 | 2544 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2545 | * we need to allocate a new slab. This is the slowest path since it involves |
2546 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2547 | * |
2548 | * Version of __slab_alloc to use when we know that interrupts are | |
2549 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2550 | */ |
a380a3c7 | 2551 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2552 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2553 | { |
6faa6833 | 2554 | void *freelist; |
f6e7def7 | 2555 | struct page *page; |
81819f0f | 2556 | |
f6e7def7 CL |
2557 | page = c->page; |
2558 | if (!page) | |
81819f0f | 2559 | goto new_slab; |
49e22585 | 2560 | redo: |
6faa6833 | 2561 | |
57d437d2 | 2562 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2563 | int searchnode = node; |
2564 | ||
2565 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2566 | searchnode = node_to_mem_node(node); | |
2567 | ||
2568 | if (unlikely(!node_match(page, searchnode))) { | |
2569 | stat(s, ALLOC_NODE_MISMATCH); | |
d4ff6d35 | 2570 | deactivate_slab(s, page, c->freelist, c); |
a561ce00 JK |
2571 | goto new_slab; |
2572 | } | |
fc59c053 | 2573 | } |
6446faa2 | 2574 | |
072bb0aa MG |
2575 | /* |
2576 | * By rights, we should be searching for a slab page that was | |
2577 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2578 | * information when the page leaves the per-cpu allocator | |
2579 | */ | |
2580 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
d4ff6d35 | 2581 | deactivate_slab(s, page, c->freelist, c); |
072bb0aa MG |
2582 | goto new_slab; |
2583 | } | |
2584 | ||
73736e03 | 2585 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2586 | freelist = c->freelist; |
2587 | if (freelist) | |
73736e03 | 2588 | goto load_freelist; |
03e404af | 2589 | |
f6e7def7 | 2590 | freelist = get_freelist(s, page); |
6446faa2 | 2591 | |
6faa6833 | 2592 | if (!freelist) { |
03e404af CL |
2593 | c->page = NULL; |
2594 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2595 | goto new_slab; |
03e404af | 2596 | } |
6446faa2 | 2597 | |
84e554e6 | 2598 | stat(s, ALLOC_REFILL); |
6446faa2 | 2599 | |
894b8788 | 2600 | load_freelist: |
507effea CL |
2601 | /* |
2602 | * freelist is pointing to the list of objects to be used. | |
2603 | * page is pointing to the page from which the objects are obtained. | |
2604 | * That page must be frozen for per cpu allocations to work. | |
2605 | */ | |
a0132ac0 | 2606 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2607 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2608 | c->tid = next_tid(c->tid); |
6faa6833 | 2609 | return freelist; |
81819f0f | 2610 | |
81819f0f | 2611 | new_slab: |
2cfb7455 | 2612 | |
a93cf07b WY |
2613 | if (slub_percpu_partial(c)) { |
2614 | page = c->page = slub_percpu_partial(c); | |
2615 | slub_set_percpu_partial(c, page); | |
49e22585 | 2616 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2617 | goto redo; |
81819f0f CL |
2618 | } |
2619 | ||
188fd063 | 2620 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2621 | |
f4697436 | 2622 | if (unlikely(!freelist)) { |
9a02d699 | 2623 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2624 | return NULL; |
81819f0f | 2625 | } |
2cfb7455 | 2626 | |
f6e7def7 | 2627 | page = c->page; |
5091b74a | 2628 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2629 | goto load_freelist; |
2cfb7455 | 2630 | |
497b66f2 | 2631 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2632 | if (kmem_cache_debug(s) && |
2633 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2634 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2635 | |
d4ff6d35 | 2636 | deactivate_slab(s, page, get_freepointer(s, freelist), c); |
6faa6833 | 2637 | return freelist; |
894b8788 CL |
2638 | } |
2639 | ||
a380a3c7 CL |
2640 | /* |
2641 | * Another one that disabled interrupt and compensates for possible | |
2642 | * cpu changes by refetching the per cpu area pointer. | |
2643 | */ | |
2644 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2645 | unsigned long addr, struct kmem_cache_cpu *c) | |
2646 | { | |
2647 | void *p; | |
2648 | unsigned long flags; | |
2649 | ||
2650 | local_irq_save(flags); | |
2651 | #ifdef CONFIG_PREEMPT | |
2652 | /* | |
2653 | * We may have been preempted and rescheduled on a different | |
2654 | * cpu before disabling interrupts. Need to reload cpu area | |
2655 | * pointer. | |
2656 | */ | |
2657 | c = this_cpu_ptr(s->cpu_slab); | |
2658 | #endif | |
2659 | ||
2660 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2661 | local_irq_restore(flags); | |
2662 | return p; | |
2663 | } | |
2664 | ||
894b8788 CL |
2665 | /* |
2666 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2667 | * have the fastpath folded into their functions. So no function call | |
2668 | * overhead for requests that can be satisfied on the fastpath. | |
2669 | * | |
2670 | * The fastpath works by first checking if the lockless freelist can be used. | |
2671 | * If not then __slab_alloc is called for slow processing. | |
2672 | * | |
2673 | * Otherwise we can simply pick the next object from the lockless free list. | |
2674 | */ | |
2b847c3c | 2675 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2676 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2677 | { |
03ec0ed5 | 2678 | void *object; |
dfb4f096 | 2679 | struct kmem_cache_cpu *c; |
57d437d2 | 2680 | struct page *page; |
8a5ec0ba | 2681 | unsigned long tid; |
1f84260c | 2682 | |
8135be5a VD |
2683 | s = slab_pre_alloc_hook(s, gfpflags); |
2684 | if (!s) | |
773ff60e | 2685 | return NULL; |
8a5ec0ba | 2686 | redo: |
8a5ec0ba CL |
2687 | /* |
2688 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2689 | * enabled. We may switch back and forth between cpus while | |
2690 | * reading from one cpu area. That does not matter as long | |
2691 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2692 | * |
9aabf810 JK |
2693 | * We should guarantee that tid and kmem_cache are retrieved on |
2694 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2695 | * to check if it is matched or not. | |
8a5ec0ba | 2696 | */ |
9aabf810 JK |
2697 | do { |
2698 | tid = this_cpu_read(s->cpu_slab->tid); | |
2699 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2700 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2701 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2702 | |
2703 | /* | |
2704 | * Irqless object alloc/free algorithm used here depends on sequence | |
2705 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2706 | * on c to guarantee that object and page associated with previous tid | |
2707 | * won't be used with current tid. If we fetch tid first, object and | |
2708 | * page could be one associated with next tid and our alloc/free | |
2709 | * request will be failed. In this case, we will retry. So, no problem. | |
2710 | */ | |
2711 | barrier(); | |
8a5ec0ba | 2712 | |
8a5ec0ba CL |
2713 | /* |
2714 | * The transaction ids are globally unique per cpu and per operation on | |
2715 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2716 | * occurs on the right processor and that there was no operation on the | |
2717 | * linked list in between. | |
2718 | */ | |
8a5ec0ba | 2719 | |
9dfc6e68 | 2720 | object = c->freelist; |
57d437d2 | 2721 | page = c->page; |
8eae1492 | 2722 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2723 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2724 | stat(s, ALLOC_SLOWPATH); |
2725 | } else { | |
0ad9500e ED |
2726 | void *next_object = get_freepointer_safe(s, object); |
2727 | ||
8a5ec0ba | 2728 | /* |
25985edc | 2729 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2730 | * operation and if we are on the right processor. |
2731 | * | |
d0e0ac97 CG |
2732 | * The cmpxchg does the following atomically (without lock |
2733 | * semantics!) | |
8a5ec0ba CL |
2734 | * 1. Relocate first pointer to the current per cpu area. |
2735 | * 2. Verify that tid and freelist have not been changed | |
2736 | * 3. If they were not changed replace tid and freelist | |
2737 | * | |
d0e0ac97 CG |
2738 | * Since this is without lock semantics the protection is only |
2739 | * against code executing on this cpu *not* from access by | |
2740 | * other cpus. | |
8a5ec0ba | 2741 | */ |
933393f5 | 2742 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2743 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2744 | object, tid, | |
0ad9500e | 2745 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2746 | |
2747 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2748 | goto redo; | |
2749 | } | |
0ad9500e | 2750 | prefetch_freepointer(s, next_object); |
84e554e6 | 2751 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2752 | } |
8a5ec0ba | 2753 | |
74e2134f | 2754 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2755 | memset(object, 0, s->object_size); |
d07dbea4 | 2756 | |
03ec0ed5 | 2757 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2758 | |
894b8788 | 2759 | return object; |
81819f0f CL |
2760 | } |
2761 | ||
2b847c3c EG |
2762 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2763 | gfp_t gfpflags, unsigned long addr) | |
2764 | { | |
2765 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2766 | } | |
2767 | ||
81819f0f CL |
2768 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2769 | { | |
2b847c3c | 2770 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2771 | |
d0e0ac97 CG |
2772 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2773 | s->size, gfpflags); | |
5b882be4 EGM |
2774 | |
2775 | return ret; | |
81819f0f CL |
2776 | } |
2777 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2778 | ||
0f24f128 | 2779 | #ifdef CONFIG_TRACING |
4a92379b RK |
2780 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2781 | { | |
2b847c3c | 2782 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2783 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0116523c | 2784 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2785 | return ret; |
2786 | } | |
2787 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2788 | #endif |
2789 | ||
81819f0f CL |
2790 | #ifdef CONFIG_NUMA |
2791 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2792 | { | |
2b847c3c | 2793 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2794 | |
ca2b84cb | 2795 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2796 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2797 | |
2798 | return ret; | |
81819f0f CL |
2799 | } |
2800 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2801 | |
0f24f128 | 2802 | #ifdef CONFIG_TRACING |
4a92379b | 2803 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2804 | gfp_t gfpflags, |
4a92379b | 2805 | int node, size_t size) |
5b882be4 | 2806 | { |
2b847c3c | 2807 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2808 | |
2809 | trace_kmalloc_node(_RET_IP_, ret, | |
2810 | size, s->size, gfpflags, node); | |
0316bec2 | 2811 | |
0116523c | 2812 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2813 | return ret; |
5b882be4 | 2814 | } |
4a92379b | 2815 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2816 | #endif |
5d1f57e4 | 2817 | #endif |
5b882be4 | 2818 | |
81819f0f | 2819 | /* |
94e4d712 | 2820 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2821 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2822 | * |
894b8788 CL |
2823 | * So we still attempt to reduce cache line usage. Just take the slab |
2824 | * lock and free the item. If there is no additional partial page | |
2825 | * handling required then we can return immediately. | |
81819f0f | 2826 | */ |
894b8788 | 2827 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2828 | void *head, void *tail, int cnt, |
2829 | unsigned long addr) | |
2830 | ||
81819f0f CL |
2831 | { |
2832 | void *prior; | |
2cfb7455 | 2833 | int was_frozen; |
2cfb7455 CL |
2834 | struct page new; |
2835 | unsigned long counters; | |
2836 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2837 | unsigned long uninitialized_var(flags); |
81819f0f | 2838 | |
8a5ec0ba | 2839 | stat(s, FREE_SLOWPATH); |
81819f0f | 2840 | |
19c7ff9e | 2841 | if (kmem_cache_debug(s) && |
282acb43 | 2842 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2843 | return; |
6446faa2 | 2844 | |
2cfb7455 | 2845 | do { |
837d678d JK |
2846 | if (unlikely(n)) { |
2847 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2848 | n = NULL; | |
2849 | } | |
2cfb7455 CL |
2850 | prior = page->freelist; |
2851 | counters = page->counters; | |
81084651 | 2852 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2853 | new.counters = counters; |
2854 | was_frozen = new.frozen; | |
81084651 | 2855 | new.inuse -= cnt; |
837d678d | 2856 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2857 | |
c65c1877 | 2858 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2859 | |
2860 | /* | |
d0e0ac97 CG |
2861 | * Slab was on no list before and will be |
2862 | * partially empty | |
2863 | * We can defer the list move and instead | |
2864 | * freeze it. | |
49e22585 CL |
2865 | */ |
2866 | new.frozen = 1; | |
2867 | ||
c65c1877 | 2868 | } else { /* Needs to be taken off a list */ |
49e22585 | 2869 | |
b455def2 | 2870 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2871 | /* |
2872 | * Speculatively acquire the list_lock. | |
2873 | * If the cmpxchg does not succeed then we may | |
2874 | * drop the list_lock without any processing. | |
2875 | * | |
2876 | * Otherwise the list_lock will synchronize with | |
2877 | * other processors updating the list of slabs. | |
2878 | */ | |
2879 | spin_lock_irqsave(&n->list_lock, flags); | |
2880 | ||
2881 | } | |
2cfb7455 | 2882 | } |
81819f0f | 2883 | |
2cfb7455 CL |
2884 | } while (!cmpxchg_double_slab(s, page, |
2885 | prior, counters, | |
81084651 | 2886 | head, new.counters, |
2cfb7455 | 2887 | "__slab_free")); |
81819f0f | 2888 | |
2cfb7455 | 2889 | if (likely(!n)) { |
49e22585 CL |
2890 | |
2891 | /* | |
2892 | * If we just froze the page then put it onto the | |
2893 | * per cpu partial list. | |
2894 | */ | |
8028dcea | 2895 | if (new.frozen && !was_frozen) { |
49e22585 | 2896 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2897 | stat(s, CPU_PARTIAL_FREE); |
2898 | } | |
49e22585 | 2899 | /* |
2cfb7455 CL |
2900 | * The list lock was not taken therefore no list |
2901 | * activity can be necessary. | |
2902 | */ | |
b455def2 L |
2903 | if (was_frozen) |
2904 | stat(s, FREE_FROZEN); | |
2905 | return; | |
2906 | } | |
81819f0f | 2907 | |
8a5b20ae | 2908 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2909 | goto slab_empty; |
2910 | ||
81819f0f | 2911 | /* |
837d678d JK |
2912 | * Objects left in the slab. If it was not on the partial list before |
2913 | * then add it. | |
81819f0f | 2914 | */ |
345c905d JK |
2915 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2916 | if (kmem_cache_debug(s)) | |
c65c1877 | 2917 | remove_full(s, n, page); |
837d678d JK |
2918 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2919 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2920 | } |
80f08c19 | 2921 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2922 | return; |
2923 | ||
2924 | slab_empty: | |
a973e9dd | 2925 | if (prior) { |
81819f0f | 2926 | /* |
6fbabb20 | 2927 | * Slab on the partial list. |
81819f0f | 2928 | */ |
5cc6eee8 | 2929 | remove_partial(n, page); |
84e554e6 | 2930 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2931 | } else { |
6fbabb20 | 2932 | /* Slab must be on the full list */ |
c65c1877 PZ |
2933 | remove_full(s, n, page); |
2934 | } | |
2cfb7455 | 2935 | |
80f08c19 | 2936 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2937 | stat(s, FREE_SLAB); |
81819f0f | 2938 | discard_slab(s, page); |
81819f0f CL |
2939 | } |
2940 | ||
894b8788 CL |
2941 | /* |
2942 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2943 | * can perform fastpath freeing without additional function calls. | |
2944 | * | |
2945 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2946 | * of this processor. This typically the case if we have just allocated | |
2947 | * the item before. | |
2948 | * | |
2949 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2950 | * with all sorts of special processing. | |
81084651 JDB |
2951 | * |
2952 | * Bulk free of a freelist with several objects (all pointing to the | |
2953 | * same page) possible by specifying head and tail ptr, plus objects | |
2954 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2955 | */ |
80a9201a AP |
2956 | static __always_inline void do_slab_free(struct kmem_cache *s, |
2957 | struct page *page, void *head, void *tail, | |
2958 | int cnt, unsigned long addr) | |
894b8788 | 2959 | { |
81084651 | 2960 | void *tail_obj = tail ? : head; |
dfb4f096 | 2961 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2962 | unsigned long tid; |
8a5ec0ba CL |
2963 | redo: |
2964 | /* | |
2965 | * Determine the currently cpus per cpu slab. | |
2966 | * The cpu may change afterward. However that does not matter since | |
2967 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2968 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2969 | */ |
9aabf810 JK |
2970 | do { |
2971 | tid = this_cpu_read(s->cpu_slab->tid); | |
2972 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2973 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2974 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2975 | |
9aabf810 JK |
2976 | /* Same with comment on barrier() in slab_alloc_node() */ |
2977 | barrier(); | |
c016b0bd | 2978 | |
442b06bc | 2979 | if (likely(page == c->page)) { |
81084651 | 2980 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2981 | |
933393f5 | 2982 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2983 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2984 | c->freelist, tid, | |
81084651 | 2985 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2986 | |
2987 | note_cmpxchg_failure("slab_free", s, tid); | |
2988 | goto redo; | |
2989 | } | |
84e554e6 | 2990 | stat(s, FREE_FASTPATH); |
894b8788 | 2991 | } else |
81084651 | 2992 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2993 | |
894b8788 CL |
2994 | } |
2995 | ||
80a9201a AP |
2996 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2997 | void *head, void *tail, int cnt, | |
2998 | unsigned long addr) | |
2999 | { | |
80a9201a | 3000 | /* |
c3895391 AK |
3001 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
3002 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 3003 | */ |
c3895391 AK |
3004 | if (slab_free_freelist_hook(s, &head, &tail)) |
3005 | do_slab_free(s, page, head, tail, cnt, addr); | |
80a9201a AP |
3006 | } |
3007 | ||
2bd926b4 | 3008 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
3009 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3010 | { | |
3011 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3012 | } | |
3013 | #endif | |
3014 | ||
81819f0f CL |
3015 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3016 | { | |
b9ce5ef4 GC |
3017 | s = cache_from_obj(s, x); |
3018 | if (!s) | |
79576102 | 3019 | return; |
81084651 | 3020 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 3021 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
3022 | } |
3023 | EXPORT_SYMBOL(kmem_cache_free); | |
3024 | ||
d0ecd894 | 3025 | struct detached_freelist { |
fbd02630 | 3026 | struct page *page; |
d0ecd894 JDB |
3027 | void *tail; |
3028 | void *freelist; | |
3029 | int cnt; | |
376bf125 | 3030 | struct kmem_cache *s; |
d0ecd894 | 3031 | }; |
fbd02630 | 3032 | |
d0ecd894 JDB |
3033 | /* |
3034 | * This function progressively scans the array with free objects (with | |
3035 | * a limited look ahead) and extract objects belonging to the same | |
3036 | * page. It builds a detached freelist directly within the given | |
3037 | * page/objects. This can happen without any need for | |
3038 | * synchronization, because the objects are owned by running process. | |
3039 | * The freelist is build up as a single linked list in the objects. | |
3040 | * The idea is, that this detached freelist can then be bulk | |
3041 | * transferred to the real freelist(s), but only requiring a single | |
3042 | * synchronization primitive. Look ahead in the array is limited due | |
3043 | * to performance reasons. | |
3044 | */ | |
376bf125 JDB |
3045 | static inline |
3046 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3047 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3048 | { |
3049 | size_t first_skipped_index = 0; | |
3050 | int lookahead = 3; | |
3051 | void *object; | |
ca257195 | 3052 | struct page *page; |
fbd02630 | 3053 | |
d0ecd894 JDB |
3054 | /* Always re-init detached_freelist */ |
3055 | df->page = NULL; | |
fbd02630 | 3056 | |
d0ecd894 JDB |
3057 | do { |
3058 | object = p[--size]; | |
ca257195 | 3059 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3060 | } while (!object && size); |
3eed034d | 3061 | |
d0ecd894 JDB |
3062 | if (!object) |
3063 | return 0; | |
fbd02630 | 3064 | |
ca257195 JDB |
3065 | page = virt_to_head_page(object); |
3066 | if (!s) { | |
3067 | /* Handle kalloc'ed objects */ | |
3068 | if (unlikely(!PageSlab(page))) { | |
3069 | BUG_ON(!PageCompound(page)); | |
3070 | kfree_hook(object); | |
4949148a | 3071 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3072 | p[size] = NULL; /* mark object processed */ |
3073 | return size; | |
3074 | } | |
3075 | /* Derive kmem_cache from object */ | |
3076 | df->s = page->slab_cache; | |
3077 | } else { | |
3078 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3079 | } | |
376bf125 | 3080 | |
d0ecd894 | 3081 | /* Start new detached freelist */ |
ca257195 | 3082 | df->page = page; |
376bf125 | 3083 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3084 | df->tail = object; |
3085 | df->freelist = object; | |
3086 | p[size] = NULL; /* mark object processed */ | |
3087 | df->cnt = 1; | |
3088 | ||
3089 | while (size) { | |
3090 | object = p[--size]; | |
3091 | if (!object) | |
3092 | continue; /* Skip processed objects */ | |
3093 | ||
3094 | /* df->page is always set at this point */ | |
3095 | if (df->page == virt_to_head_page(object)) { | |
3096 | /* Opportunity build freelist */ | |
376bf125 | 3097 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3098 | df->freelist = object; |
3099 | df->cnt++; | |
3100 | p[size] = NULL; /* mark object processed */ | |
3101 | ||
3102 | continue; | |
fbd02630 | 3103 | } |
d0ecd894 JDB |
3104 | |
3105 | /* Limit look ahead search */ | |
3106 | if (!--lookahead) | |
3107 | break; | |
3108 | ||
3109 | if (!first_skipped_index) | |
3110 | first_skipped_index = size + 1; | |
fbd02630 | 3111 | } |
d0ecd894 JDB |
3112 | |
3113 | return first_skipped_index; | |
3114 | } | |
3115 | ||
d0ecd894 | 3116 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3117 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3118 | { |
3119 | if (WARN_ON(!size)) | |
3120 | return; | |
3121 | ||
3122 | do { | |
3123 | struct detached_freelist df; | |
3124 | ||
3125 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3126 | if (!df.page) |
d0ecd894 JDB |
3127 | continue; |
3128 | ||
376bf125 | 3129 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3130 | } while (likely(size)); |
484748f0 CL |
3131 | } |
3132 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3133 | ||
994eb764 | 3134 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3135 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3136 | void **p) | |
484748f0 | 3137 | { |
994eb764 JDB |
3138 | struct kmem_cache_cpu *c; |
3139 | int i; | |
3140 | ||
03ec0ed5 JDB |
3141 | /* memcg and kmem_cache debug support */ |
3142 | s = slab_pre_alloc_hook(s, flags); | |
3143 | if (unlikely(!s)) | |
3144 | return false; | |
994eb764 JDB |
3145 | /* |
3146 | * Drain objects in the per cpu slab, while disabling local | |
3147 | * IRQs, which protects against PREEMPT and interrupts | |
3148 | * handlers invoking normal fastpath. | |
3149 | */ | |
3150 | local_irq_disable(); | |
3151 | c = this_cpu_ptr(s->cpu_slab); | |
3152 | ||
3153 | for (i = 0; i < size; i++) { | |
3154 | void *object = c->freelist; | |
3155 | ||
ebe909e0 | 3156 | if (unlikely(!object)) { |
ebe909e0 JDB |
3157 | /* |
3158 | * Invoking slow path likely have side-effect | |
3159 | * of re-populating per CPU c->freelist | |
3160 | */ | |
87098373 | 3161 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3162 | _RET_IP_, c); |
87098373 CL |
3163 | if (unlikely(!p[i])) |
3164 | goto error; | |
3165 | ||
ebe909e0 JDB |
3166 | c = this_cpu_ptr(s->cpu_slab); |
3167 | continue; /* goto for-loop */ | |
3168 | } | |
994eb764 JDB |
3169 | c->freelist = get_freepointer(s, object); |
3170 | p[i] = object; | |
3171 | } | |
3172 | c->tid = next_tid(c->tid); | |
3173 | local_irq_enable(); | |
3174 | ||
3175 | /* Clear memory outside IRQ disabled fastpath loop */ | |
3176 | if (unlikely(flags & __GFP_ZERO)) { | |
3177 | int j; | |
3178 | ||
3179 | for (j = 0; j < i; j++) | |
3180 | memset(p[j], 0, s->object_size); | |
3181 | } | |
3182 | ||
03ec0ed5 JDB |
3183 | /* memcg and kmem_cache debug support */ |
3184 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 3185 | return i; |
87098373 | 3186 | error: |
87098373 | 3187 | local_irq_enable(); |
03ec0ed5 JDB |
3188 | slab_post_alloc_hook(s, flags, i, p); |
3189 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 3190 | return 0; |
484748f0 CL |
3191 | } |
3192 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3193 | ||
3194 | ||
81819f0f | 3195 | /* |
672bba3a CL |
3196 | * Object placement in a slab is made very easy because we always start at |
3197 | * offset 0. If we tune the size of the object to the alignment then we can | |
3198 | * get the required alignment by putting one properly sized object after | |
3199 | * another. | |
81819f0f CL |
3200 | * |
3201 | * Notice that the allocation order determines the sizes of the per cpu | |
3202 | * caches. Each processor has always one slab available for allocations. | |
3203 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3204 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3205 | * locking overhead. |
81819f0f CL |
3206 | */ |
3207 | ||
3208 | /* | |
3209 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3210 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3211 | * and increases the number of allocations possible without having to | |
3212 | * take the list_lock. | |
3213 | */ | |
19af27af AD |
3214 | static unsigned int slub_min_order; |
3215 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3216 | static unsigned int slub_min_objects; | |
81819f0f | 3217 | |
81819f0f CL |
3218 | /* |
3219 | * Calculate the order of allocation given an slab object size. | |
3220 | * | |
672bba3a CL |
3221 | * The order of allocation has significant impact on performance and other |
3222 | * system components. Generally order 0 allocations should be preferred since | |
3223 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3224 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3225 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3226 | * would be wasted. |
3227 | * | |
3228 | * In order to reach satisfactory performance we must ensure that a minimum | |
3229 | * number of objects is in one slab. Otherwise we may generate too much | |
3230 | * activity on the partial lists which requires taking the list_lock. This is | |
3231 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3232 | * |
672bba3a CL |
3233 | * slub_max_order specifies the order where we begin to stop considering the |
3234 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3235 | * we try to keep the page order as low as possible. So we accept more waste | |
3236 | * of space in favor of a small page order. | |
81819f0f | 3237 | * |
672bba3a CL |
3238 | * Higher order allocations also allow the placement of more objects in a |
3239 | * slab and thereby reduce object handling overhead. If the user has | |
3240 | * requested a higher mininum order then we start with that one instead of | |
3241 | * the smallest order which will fit the object. | |
81819f0f | 3242 | */ |
19af27af AD |
3243 | static inline unsigned int slab_order(unsigned int size, |
3244 | unsigned int min_objects, unsigned int max_order, | |
9736d2a9 | 3245 | unsigned int fract_leftover) |
81819f0f | 3246 | { |
19af27af AD |
3247 | unsigned int min_order = slub_min_order; |
3248 | unsigned int order; | |
81819f0f | 3249 | |
9736d2a9 | 3250 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3251 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3252 | |
9736d2a9 | 3253 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3254 | order <= max_order; order++) { |
81819f0f | 3255 | |
19af27af AD |
3256 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3257 | unsigned int rem; | |
81819f0f | 3258 | |
9736d2a9 | 3259 | rem = slab_size % size; |
81819f0f | 3260 | |
5e6d444e | 3261 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3262 | break; |
81819f0f | 3263 | } |
672bba3a | 3264 | |
81819f0f CL |
3265 | return order; |
3266 | } | |
3267 | ||
9736d2a9 | 3268 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3269 | { |
19af27af AD |
3270 | unsigned int order; |
3271 | unsigned int min_objects; | |
3272 | unsigned int max_objects; | |
5e6d444e CL |
3273 | |
3274 | /* | |
3275 | * Attempt to find best configuration for a slab. This | |
3276 | * works by first attempting to generate a layout with | |
3277 | * the best configuration and backing off gradually. | |
3278 | * | |
422ff4d7 | 3279 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3280 | * we reduce the minimum objects required in a slab. |
3281 | */ | |
3282 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3283 | if (!min_objects) |
3284 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
9736d2a9 | 3285 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3286 | min_objects = min(min_objects, max_objects); |
3287 | ||
5e6d444e | 3288 | while (min_objects > 1) { |
19af27af AD |
3289 | unsigned int fraction; |
3290 | ||
c124f5b5 | 3291 | fraction = 16; |
5e6d444e CL |
3292 | while (fraction >= 4) { |
3293 | order = slab_order(size, min_objects, | |
9736d2a9 | 3294 | slub_max_order, fraction); |
5e6d444e CL |
3295 | if (order <= slub_max_order) |
3296 | return order; | |
3297 | fraction /= 2; | |
3298 | } | |
5086c389 | 3299 | min_objects--; |
5e6d444e CL |
3300 | } |
3301 | ||
3302 | /* | |
3303 | * We were unable to place multiple objects in a slab. Now | |
3304 | * lets see if we can place a single object there. | |
3305 | */ | |
9736d2a9 | 3306 | order = slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3307 | if (order <= slub_max_order) |
3308 | return order; | |
3309 | ||
3310 | /* | |
3311 | * Doh this slab cannot be placed using slub_max_order. | |
3312 | */ | |
9736d2a9 | 3313 | order = slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3314 | if (order < MAX_ORDER) |
5e6d444e CL |
3315 | return order; |
3316 | return -ENOSYS; | |
3317 | } | |
3318 | ||
5595cffc | 3319 | static void |
4053497d | 3320 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3321 | { |
3322 | n->nr_partial = 0; | |
81819f0f CL |
3323 | spin_lock_init(&n->list_lock); |
3324 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3325 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3326 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3327 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3328 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3329 | #endif |
81819f0f CL |
3330 | } |
3331 | ||
55136592 | 3332 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3333 | { |
6c182dc0 | 3334 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3335 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3336 | |
8a5ec0ba | 3337 | /* |
d4d84fef CM |
3338 | * Must align to double word boundary for the double cmpxchg |
3339 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3340 | */ |
d4d84fef CM |
3341 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3342 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3343 | |
3344 | if (!s->cpu_slab) | |
3345 | return 0; | |
3346 | ||
3347 | init_kmem_cache_cpus(s); | |
4c93c355 | 3348 | |
8a5ec0ba | 3349 | return 1; |
4c93c355 | 3350 | } |
4c93c355 | 3351 | |
51df1142 CL |
3352 | static struct kmem_cache *kmem_cache_node; |
3353 | ||
81819f0f CL |
3354 | /* |
3355 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3356 | * slab on the node for this slabcache. There are no concurrent accesses | |
3357 | * possible. | |
3358 | * | |
721ae22a ZYW |
3359 | * Note that this function only works on the kmem_cache_node |
3360 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3361 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3362 | */ |
55136592 | 3363 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3364 | { |
3365 | struct page *page; | |
3366 | struct kmem_cache_node *n; | |
3367 | ||
51df1142 | 3368 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3369 | |
51df1142 | 3370 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3371 | |
3372 | BUG_ON(!page); | |
a2f92ee7 | 3373 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3374 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3375 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3376 | } |
3377 | ||
81819f0f CL |
3378 | n = page->freelist; |
3379 | BUG_ON(!n); | |
8ab1372f | 3380 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3381 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3382 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3383 | #endif |
12b22386 | 3384 | n = kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
505f5dcb | 3385 | GFP_KERNEL); |
12b22386 AK |
3386 | page->freelist = get_freepointer(kmem_cache_node, n); |
3387 | page->inuse = 1; | |
3388 | page->frozen = 0; | |
3389 | kmem_cache_node->node[node] = n; | |
4053497d | 3390 | init_kmem_cache_node(n); |
51df1142 | 3391 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3392 | |
67b6c900 | 3393 | /* |
1e4dd946 SR |
3394 | * No locks need to be taken here as it has just been |
3395 | * initialized and there is no concurrent access. | |
67b6c900 | 3396 | */ |
1e4dd946 | 3397 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3398 | } |
3399 | ||
3400 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3401 | { | |
3402 | int node; | |
fa45dc25 | 3403 | struct kmem_cache_node *n; |
81819f0f | 3404 | |
fa45dc25 | 3405 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3406 | s->node[node] = NULL; |
ea37df54 | 3407 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3408 | } |
3409 | } | |
3410 | ||
52b4b950 DS |
3411 | void __kmem_cache_release(struct kmem_cache *s) |
3412 | { | |
210e7a43 | 3413 | cache_random_seq_destroy(s); |
52b4b950 DS |
3414 | free_percpu(s->cpu_slab); |
3415 | free_kmem_cache_nodes(s); | |
3416 | } | |
3417 | ||
55136592 | 3418 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3419 | { |
3420 | int node; | |
81819f0f | 3421 | |
f64dc58c | 3422 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3423 | struct kmem_cache_node *n; |
3424 | ||
73367bd8 | 3425 | if (slab_state == DOWN) { |
55136592 | 3426 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3427 | continue; |
3428 | } | |
51df1142 | 3429 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3430 | GFP_KERNEL, node); |
81819f0f | 3431 | |
73367bd8 AD |
3432 | if (!n) { |
3433 | free_kmem_cache_nodes(s); | |
3434 | return 0; | |
81819f0f | 3435 | } |
73367bd8 | 3436 | |
4053497d | 3437 | init_kmem_cache_node(n); |
ea37df54 | 3438 | s->node[node] = n; |
81819f0f CL |
3439 | } |
3440 | return 1; | |
3441 | } | |
81819f0f | 3442 | |
c0bdb232 | 3443 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3444 | { |
3445 | if (min < MIN_PARTIAL) | |
3446 | min = MIN_PARTIAL; | |
3447 | else if (min > MAX_PARTIAL) | |
3448 | min = MAX_PARTIAL; | |
3449 | s->min_partial = min; | |
3450 | } | |
3451 | ||
e6d0e1dc WY |
3452 | static void set_cpu_partial(struct kmem_cache *s) |
3453 | { | |
3454 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3455 | /* | |
3456 | * cpu_partial determined the maximum number of objects kept in the | |
3457 | * per cpu partial lists of a processor. | |
3458 | * | |
3459 | * Per cpu partial lists mainly contain slabs that just have one | |
3460 | * object freed. If they are used for allocation then they can be | |
3461 | * filled up again with minimal effort. The slab will never hit the | |
3462 | * per node partial lists and therefore no locking will be required. | |
3463 | * | |
3464 | * This setting also determines | |
3465 | * | |
3466 | * A) The number of objects from per cpu partial slabs dumped to the | |
3467 | * per node list when we reach the limit. | |
3468 | * B) The number of objects in cpu partial slabs to extract from the | |
3469 | * per node list when we run out of per cpu objects. We only fetch | |
3470 | * 50% to keep some capacity around for frees. | |
3471 | */ | |
3472 | if (!kmem_cache_has_cpu_partial(s)) | |
3473 | s->cpu_partial = 0; | |
3474 | else if (s->size >= PAGE_SIZE) | |
3475 | s->cpu_partial = 2; | |
3476 | else if (s->size >= 1024) | |
3477 | s->cpu_partial = 6; | |
3478 | else if (s->size >= 256) | |
3479 | s->cpu_partial = 13; | |
3480 | else | |
3481 | s->cpu_partial = 30; | |
3482 | #endif | |
3483 | } | |
3484 | ||
81819f0f CL |
3485 | /* |
3486 | * calculate_sizes() determines the order and the distribution of data within | |
3487 | * a slab object. | |
3488 | */ | |
06b285dc | 3489 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3490 | { |
d50112ed | 3491 | slab_flags_t flags = s->flags; |
be4a7988 | 3492 | unsigned int size = s->object_size; |
19af27af | 3493 | unsigned int order; |
81819f0f | 3494 | |
d8b42bf5 CL |
3495 | /* |
3496 | * Round up object size to the next word boundary. We can only | |
3497 | * place the free pointer at word boundaries and this determines | |
3498 | * the possible location of the free pointer. | |
3499 | */ | |
3500 | size = ALIGN(size, sizeof(void *)); | |
3501 | ||
3502 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3503 | /* |
3504 | * Determine if we can poison the object itself. If the user of | |
3505 | * the slab may touch the object after free or before allocation | |
3506 | * then we should never poison the object itself. | |
3507 | */ | |
5f0d5a3a | 3508 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3509 | !s->ctor) |
81819f0f CL |
3510 | s->flags |= __OBJECT_POISON; |
3511 | else | |
3512 | s->flags &= ~__OBJECT_POISON; | |
3513 | ||
81819f0f CL |
3514 | |
3515 | /* | |
672bba3a | 3516 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3517 | * end of the object and the free pointer. If not then add an |
672bba3a | 3518 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3519 | */ |
3b0efdfa | 3520 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3521 | size += sizeof(void *); |
41ecc55b | 3522 | #endif |
81819f0f CL |
3523 | |
3524 | /* | |
672bba3a CL |
3525 | * With that we have determined the number of bytes in actual use |
3526 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3527 | */ |
3528 | s->inuse = size; | |
3529 | ||
5f0d5a3a | 3530 | if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
c59def9f | 3531 | s->ctor)) { |
81819f0f CL |
3532 | /* |
3533 | * Relocate free pointer after the object if it is not | |
3534 | * permitted to overwrite the first word of the object on | |
3535 | * kmem_cache_free. | |
3536 | * | |
3537 | * This is the case if we do RCU, have a constructor or | |
3538 | * destructor or are poisoning the objects. | |
3539 | */ | |
3540 | s->offset = size; | |
3541 | size += sizeof(void *); | |
3542 | } | |
3543 | ||
c12b3c62 | 3544 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3545 | if (flags & SLAB_STORE_USER) |
3546 | /* | |
3547 | * Need to store information about allocs and frees after | |
3548 | * the object. | |
3549 | */ | |
3550 | size += 2 * sizeof(struct track); | |
80a9201a | 3551 | #endif |
81819f0f | 3552 | |
80a9201a AP |
3553 | kasan_cache_create(s, &size, &s->flags); |
3554 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3555 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3556 | /* |
3557 | * Add some empty padding so that we can catch | |
3558 | * overwrites from earlier objects rather than let | |
3559 | * tracking information or the free pointer be | |
0211a9c8 | 3560 | * corrupted if a user writes before the start |
81819f0f CL |
3561 | * of the object. |
3562 | */ | |
3563 | size += sizeof(void *); | |
d86bd1be JK |
3564 | |
3565 | s->red_left_pad = sizeof(void *); | |
3566 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3567 | size += s->red_left_pad; | |
3568 | } | |
41ecc55b | 3569 | #endif |
672bba3a | 3570 | |
81819f0f CL |
3571 | /* |
3572 | * SLUB stores one object immediately after another beginning from | |
3573 | * offset 0. In order to align the objects we have to simply size | |
3574 | * each object to conform to the alignment. | |
3575 | */ | |
45906855 | 3576 | size = ALIGN(size, s->align); |
81819f0f | 3577 | s->size = size; |
06b285dc CL |
3578 | if (forced_order >= 0) |
3579 | order = forced_order; | |
3580 | else | |
9736d2a9 | 3581 | order = calculate_order(size); |
81819f0f | 3582 | |
19af27af | 3583 | if ((int)order < 0) |
81819f0f CL |
3584 | return 0; |
3585 | ||
b7a49f0d | 3586 | s->allocflags = 0; |
834f3d11 | 3587 | if (order) |
b7a49f0d CL |
3588 | s->allocflags |= __GFP_COMP; |
3589 | ||
3590 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3591 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3592 | |
3593 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3594 | s->allocflags |= __GFP_RECLAIMABLE; | |
3595 | ||
81819f0f CL |
3596 | /* |
3597 | * Determine the number of objects per slab | |
3598 | */ | |
9736d2a9 MW |
3599 | s->oo = oo_make(order, size); |
3600 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
3601 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3602 | s->max = s->oo; | |
81819f0f | 3603 | |
834f3d11 | 3604 | return !!oo_objects(s->oo); |
81819f0f CL |
3605 | } |
3606 | ||
d50112ed | 3607 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3608 | { |
8a13a4cc | 3609 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
2482ddec KC |
3610 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3611 | s->random = get_random_long(); | |
3612 | #endif | |
81819f0f | 3613 | |
06b285dc | 3614 | if (!calculate_sizes(s, -1)) |
81819f0f | 3615 | goto error; |
3de47213 DR |
3616 | if (disable_higher_order_debug) { |
3617 | /* | |
3618 | * Disable debugging flags that store metadata if the min slab | |
3619 | * order increased. | |
3620 | */ | |
3b0efdfa | 3621 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3622 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3623 | s->offset = 0; | |
3624 | if (!calculate_sizes(s, -1)) | |
3625 | goto error; | |
3626 | } | |
3627 | } | |
81819f0f | 3628 | |
2565409f HC |
3629 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3630 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3631 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3632 | /* Enable fast mode */ |
3633 | s->flags |= __CMPXCHG_DOUBLE; | |
3634 | #endif | |
3635 | ||
3b89d7d8 DR |
3636 | /* |
3637 | * The larger the object size is, the more pages we want on the partial | |
3638 | * list to avoid pounding the page allocator excessively. | |
3639 | */ | |
49e22585 CL |
3640 | set_min_partial(s, ilog2(s->size) / 2); |
3641 | ||
e6d0e1dc | 3642 | set_cpu_partial(s); |
49e22585 | 3643 | |
81819f0f | 3644 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3645 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3646 | #endif |
210e7a43 TG |
3647 | |
3648 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3649 | if (slab_state >= UP) { | |
3650 | if (init_cache_random_seq(s)) | |
3651 | goto error; | |
3652 | } | |
3653 | ||
55136592 | 3654 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3655 | goto error; |
81819f0f | 3656 | |
55136592 | 3657 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3658 | return 0; |
ff12059e | 3659 | |
4c93c355 | 3660 | free_kmem_cache_nodes(s); |
81819f0f CL |
3661 | error: |
3662 | if (flags & SLAB_PANIC) | |
44065b2e AD |
3663 | panic("Cannot create slab %s size=%u realsize=%u order=%u offset=%u flags=%lx\n", |
3664 | s->name, s->size, s->size, | |
4fd0b46e | 3665 | oo_order(s->oo), s->offset, (unsigned long)flags); |
278b1bb1 | 3666 | return -EINVAL; |
81819f0f | 3667 | } |
81819f0f | 3668 | |
33b12c38 CL |
3669 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3670 | const char *text) | |
3671 | { | |
3672 | #ifdef CONFIG_SLUB_DEBUG | |
3673 | void *addr = page_address(page); | |
3674 | void *p; | |
0684e652 | 3675 | unsigned long *map = bitmap_zalloc(page->objects, GFP_ATOMIC); |
bbd7d57b ED |
3676 | if (!map) |
3677 | return; | |
945cf2b6 | 3678 | slab_err(s, page, text, s->name); |
33b12c38 | 3679 | slab_lock(page); |
33b12c38 | 3680 | |
5f80b13a | 3681 | get_map(s, page, map); |
33b12c38 CL |
3682 | for_each_object(p, s, addr, page->objects) { |
3683 | ||
3684 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3685 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3686 | print_tracking(s, p); |
3687 | } | |
3688 | } | |
3689 | slab_unlock(page); | |
0684e652 | 3690 | bitmap_free(map); |
33b12c38 CL |
3691 | #endif |
3692 | } | |
3693 | ||
81819f0f | 3694 | /* |
599870b1 | 3695 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3696 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3697 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3698 | */ |
599870b1 | 3699 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3700 | { |
60398923 | 3701 | LIST_HEAD(discard); |
81819f0f CL |
3702 | struct page *page, *h; |
3703 | ||
52b4b950 DS |
3704 | BUG_ON(irqs_disabled()); |
3705 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3706 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3707 | if (!page->inuse) { |
52b4b950 | 3708 | remove_partial(n, page); |
60398923 | 3709 | list_add(&page->lru, &discard); |
33b12c38 CL |
3710 | } else { |
3711 | list_slab_objects(s, page, | |
52b4b950 | 3712 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3713 | } |
33b12c38 | 3714 | } |
52b4b950 | 3715 | spin_unlock_irq(&n->list_lock); |
60398923 CW |
3716 | |
3717 | list_for_each_entry_safe(page, h, &discard, lru) | |
3718 | discard_slab(s, page); | |
81819f0f CL |
3719 | } |
3720 | ||
f9e13c0a SB |
3721 | bool __kmem_cache_empty(struct kmem_cache *s) |
3722 | { | |
3723 | int node; | |
3724 | struct kmem_cache_node *n; | |
3725 | ||
3726 | for_each_kmem_cache_node(s, node, n) | |
3727 | if (n->nr_partial || slabs_node(s, node)) | |
3728 | return false; | |
3729 | return true; | |
3730 | } | |
3731 | ||
81819f0f | 3732 | /* |
672bba3a | 3733 | * Release all resources used by a slab cache. |
81819f0f | 3734 | */ |
52b4b950 | 3735 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3736 | { |
3737 | int node; | |
fa45dc25 | 3738 | struct kmem_cache_node *n; |
81819f0f CL |
3739 | |
3740 | flush_all(s); | |
81819f0f | 3741 | /* Attempt to free all objects */ |
fa45dc25 | 3742 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3743 | free_partial(s, n); |
3744 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3745 | return 1; |
3746 | } | |
bf5eb3de | 3747 | sysfs_slab_remove(s); |
81819f0f CL |
3748 | return 0; |
3749 | } | |
3750 | ||
81819f0f CL |
3751 | /******************************************************************** |
3752 | * Kmalloc subsystem | |
3753 | *******************************************************************/ | |
3754 | ||
81819f0f CL |
3755 | static int __init setup_slub_min_order(char *str) |
3756 | { | |
19af27af | 3757 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
3758 | |
3759 | return 1; | |
3760 | } | |
3761 | ||
3762 | __setup("slub_min_order=", setup_slub_min_order); | |
3763 | ||
3764 | static int __init setup_slub_max_order(char *str) | |
3765 | { | |
19af27af AD |
3766 | get_option(&str, (int *)&slub_max_order); |
3767 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
3768 | |
3769 | return 1; | |
3770 | } | |
3771 | ||
3772 | __setup("slub_max_order=", setup_slub_max_order); | |
3773 | ||
3774 | static int __init setup_slub_min_objects(char *str) | |
3775 | { | |
19af27af | 3776 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
3777 | |
3778 | return 1; | |
3779 | } | |
3780 | ||
3781 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3782 | ||
81819f0f CL |
3783 | void *__kmalloc(size_t size, gfp_t flags) |
3784 | { | |
aadb4bc4 | 3785 | struct kmem_cache *s; |
5b882be4 | 3786 | void *ret; |
81819f0f | 3787 | |
95a05b42 | 3788 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3789 | return kmalloc_large(size, flags); |
aadb4bc4 | 3790 | |
2c59dd65 | 3791 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3792 | |
3793 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3794 | return s; |
3795 | ||
2b847c3c | 3796 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3797 | |
ca2b84cb | 3798 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3799 | |
0116523c | 3800 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3801 | |
5b882be4 | 3802 | return ret; |
81819f0f CL |
3803 | } |
3804 | EXPORT_SYMBOL(__kmalloc); | |
3805 | ||
5d1f57e4 | 3806 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3807 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3808 | { | |
b1eeab67 | 3809 | struct page *page; |
e4f7c0b4 | 3810 | void *ptr = NULL; |
f619cfe1 | 3811 | |
75f296d9 | 3812 | flags |= __GFP_COMP; |
4949148a | 3813 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3814 | if (page) |
e4f7c0b4 CM |
3815 | ptr = page_address(page); |
3816 | ||
0116523c | 3817 | return kmalloc_large_node_hook(ptr, size, flags); |
f619cfe1 CL |
3818 | } |
3819 | ||
81819f0f CL |
3820 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3821 | { | |
aadb4bc4 | 3822 | struct kmem_cache *s; |
5b882be4 | 3823 | void *ret; |
81819f0f | 3824 | |
95a05b42 | 3825 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3826 | ret = kmalloc_large_node(size, flags, node); |
3827 | ||
ca2b84cb EGM |
3828 | trace_kmalloc_node(_RET_IP_, ret, |
3829 | size, PAGE_SIZE << get_order(size), | |
3830 | flags, node); | |
5b882be4 EGM |
3831 | |
3832 | return ret; | |
3833 | } | |
aadb4bc4 | 3834 | |
2c59dd65 | 3835 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3836 | |
3837 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3838 | return s; |
3839 | ||
2b847c3c | 3840 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3841 | |
ca2b84cb | 3842 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3843 | |
0116523c | 3844 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3845 | |
5b882be4 | 3846 | return ret; |
81819f0f CL |
3847 | } |
3848 | EXPORT_SYMBOL(__kmalloc_node); | |
3849 | #endif | |
3850 | ||
ed18adc1 KC |
3851 | #ifdef CONFIG_HARDENED_USERCOPY |
3852 | /* | |
afcc90f8 KC |
3853 | * Rejects incorrectly sized objects and objects that are to be copied |
3854 | * to/from userspace but do not fall entirely within the containing slab | |
3855 | * cache's usercopy region. | |
ed18adc1 KC |
3856 | * |
3857 | * Returns NULL if check passes, otherwise const char * to name of cache | |
3858 | * to indicate an error. | |
3859 | */ | |
f4e6e289 KC |
3860 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
3861 | bool to_user) | |
ed18adc1 KC |
3862 | { |
3863 | struct kmem_cache *s; | |
44065b2e | 3864 | unsigned int offset; |
ed18adc1 KC |
3865 | size_t object_size; |
3866 | ||
96fedce2 AK |
3867 | ptr = kasan_reset_tag(ptr); |
3868 | ||
ed18adc1 KC |
3869 | /* Find object and usable object size. */ |
3870 | s = page->slab_cache; | |
ed18adc1 KC |
3871 | |
3872 | /* Reject impossible pointers. */ | |
3873 | if (ptr < page_address(page)) | |
f4e6e289 KC |
3874 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
3875 | to_user, 0, n); | |
ed18adc1 KC |
3876 | |
3877 | /* Find offset within object. */ | |
3878 | offset = (ptr - page_address(page)) % s->size; | |
3879 | ||
3880 | /* Adjust for redzone and reject if within the redzone. */ | |
3881 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) { | |
3882 | if (offset < s->red_left_pad) | |
f4e6e289 KC |
3883 | usercopy_abort("SLUB object in left red zone", |
3884 | s->name, to_user, offset, n); | |
ed18adc1 KC |
3885 | offset -= s->red_left_pad; |
3886 | } | |
3887 | ||
afcc90f8 KC |
3888 | /* Allow address range falling entirely within usercopy region. */ |
3889 | if (offset >= s->useroffset && | |
3890 | offset - s->useroffset <= s->usersize && | |
3891 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 3892 | return; |
ed18adc1 | 3893 | |
afcc90f8 KC |
3894 | /* |
3895 | * If the copy is still within the allocated object, produce | |
3896 | * a warning instead of rejecting the copy. This is intended | |
3897 | * to be a temporary method to find any missing usercopy | |
3898 | * whitelists. | |
3899 | */ | |
3900 | object_size = slab_ksize(s); | |
2d891fbc KC |
3901 | if (usercopy_fallback && |
3902 | offset <= object_size && n <= object_size - offset) { | |
afcc90f8 KC |
3903 | usercopy_warn("SLUB object", s->name, to_user, offset, n); |
3904 | return; | |
3905 | } | |
ed18adc1 | 3906 | |
f4e6e289 | 3907 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
3908 | } |
3909 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
3910 | ||
0316bec2 | 3911 | static size_t __ksize(const void *object) |
81819f0f | 3912 | { |
272c1d21 | 3913 | struct page *page; |
81819f0f | 3914 | |
ef8b4520 | 3915 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3916 | return 0; |
3917 | ||
294a80a8 | 3918 | page = virt_to_head_page(object); |
294a80a8 | 3919 | |
76994412 PE |
3920 | if (unlikely(!PageSlab(page))) { |
3921 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3922 | return PAGE_SIZE << compound_order(page); |
76994412 | 3923 | } |
81819f0f | 3924 | |
1b4f59e3 | 3925 | return slab_ksize(page->slab_cache); |
81819f0f | 3926 | } |
0316bec2 AR |
3927 | |
3928 | size_t ksize(const void *object) | |
3929 | { | |
3930 | size_t size = __ksize(object); | |
3931 | /* We assume that ksize callers could use whole allocated area, | |
4ebb31a4 AP |
3932 | * so we need to unpoison this area. |
3933 | */ | |
3934 | kasan_unpoison_shadow(object, size); | |
0316bec2 AR |
3935 | return size; |
3936 | } | |
b1aabecd | 3937 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3938 | |
3939 | void kfree(const void *x) | |
3940 | { | |
81819f0f | 3941 | struct page *page; |
5bb983b0 | 3942 | void *object = (void *)x; |
81819f0f | 3943 | |
2121db74 PE |
3944 | trace_kfree(_RET_IP_, x); |
3945 | ||
2408c550 | 3946 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3947 | return; |
3948 | ||
b49af68f | 3949 | page = virt_to_head_page(x); |
aadb4bc4 | 3950 | if (unlikely(!PageSlab(page))) { |
0937502a | 3951 | BUG_ON(!PageCompound(page)); |
47adccce | 3952 | kfree_hook(object); |
4949148a | 3953 | __free_pages(page, compound_order(page)); |
aadb4bc4 CL |
3954 | return; |
3955 | } | |
81084651 | 3956 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3957 | } |
3958 | EXPORT_SYMBOL(kfree); | |
3959 | ||
832f37f5 VD |
3960 | #define SHRINK_PROMOTE_MAX 32 |
3961 | ||
2086d26a | 3962 | /* |
832f37f5 VD |
3963 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3964 | * up most to the head of the partial lists. New allocations will then | |
3965 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3966 | * |
3967 | * The slabs with the least items are placed last. This results in them | |
3968 | * being allocated from last increasing the chance that the last objects | |
3969 | * are freed in them. | |
2086d26a | 3970 | */ |
c9fc5864 | 3971 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
3972 | { |
3973 | int node; | |
3974 | int i; | |
3975 | struct kmem_cache_node *n; | |
3976 | struct page *page; | |
3977 | struct page *t; | |
832f37f5 VD |
3978 | struct list_head discard; |
3979 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3980 | unsigned long flags; |
ce3712d7 | 3981 | int ret = 0; |
2086d26a | 3982 | |
2086d26a | 3983 | flush_all(s); |
fa45dc25 | 3984 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3985 | INIT_LIST_HEAD(&discard); |
3986 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3987 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3988 | |
3989 | spin_lock_irqsave(&n->list_lock, flags); | |
3990 | ||
3991 | /* | |
832f37f5 | 3992 | * Build lists of slabs to discard or promote. |
2086d26a | 3993 | * |
672bba3a CL |
3994 | * Note that concurrent frees may occur while we hold the |
3995 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3996 | */ |
3997 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3998 | int free = page->objects - page->inuse; |
3999 | ||
4000 | /* Do not reread page->inuse */ | |
4001 | barrier(); | |
4002 | ||
4003 | /* We do not keep full slabs on the list */ | |
4004 | BUG_ON(free <= 0); | |
4005 | ||
4006 | if (free == page->objects) { | |
4007 | list_move(&page->lru, &discard); | |
69cb8e6b | 4008 | n->nr_partial--; |
832f37f5 VD |
4009 | } else if (free <= SHRINK_PROMOTE_MAX) |
4010 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
4011 | } |
4012 | ||
2086d26a | 4013 | /* |
832f37f5 VD |
4014 | * Promote the slabs filled up most to the head of the |
4015 | * partial list. | |
2086d26a | 4016 | */ |
832f37f5 VD |
4017 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4018 | list_splice(promote + i, &n->partial); | |
2086d26a | 4019 | |
2086d26a | 4020 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4021 | |
4022 | /* Release empty slabs */ | |
832f37f5 | 4023 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 4024 | discard_slab(s, page); |
ce3712d7 VD |
4025 | |
4026 | if (slabs_node(s, node)) | |
4027 | ret = 1; | |
2086d26a CL |
4028 | } |
4029 | ||
ce3712d7 | 4030 | return ret; |
2086d26a | 4031 | } |
2086d26a | 4032 | |
c9fc5864 | 4033 | #ifdef CONFIG_MEMCG |
01fb58bc TH |
4034 | static void kmemcg_cache_deact_after_rcu(struct kmem_cache *s) |
4035 | { | |
50862ce7 TH |
4036 | /* |
4037 | * Called with all the locks held after a sched RCU grace period. | |
4038 | * Even if @s becomes empty after shrinking, we can't know that @s | |
4039 | * doesn't have allocations already in-flight and thus can't | |
4040 | * destroy @s until the associated memcg is released. | |
4041 | * | |
4042 | * However, let's remove the sysfs files for empty caches here. | |
4043 | * Each cache has a lot of interface files which aren't | |
4044 | * particularly useful for empty draining caches; otherwise, we can | |
4045 | * easily end up with millions of unnecessary sysfs files on | |
4046 | * systems which have a lot of memory and transient cgroups. | |
4047 | */ | |
4048 | if (!__kmem_cache_shrink(s)) | |
4049 | sysfs_slab_remove(s); | |
01fb58bc TH |
4050 | } |
4051 | ||
c9fc5864 TH |
4052 | void __kmemcg_cache_deactivate(struct kmem_cache *s) |
4053 | { | |
4054 | /* | |
4055 | * Disable empty slabs caching. Used to avoid pinning offline | |
4056 | * memory cgroups by kmem pages that can be freed. | |
4057 | */ | |
e6d0e1dc | 4058 | slub_set_cpu_partial(s, 0); |
c9fc5864 TH |
4059 | s->min_partial = 0; |
4060 | ||
4061 | /* | |
4062 | * s->cpu_partial is checked locklessly (see put_cpu_partial), so | |
01fb58bc | 4063 | * we have to make sure the change is visible before shrinking. |
c9fc5864 | 4064 | */ |
01fb58bc | 4065 | slab_deactivate_memcg_cache_rcu_sched(s, kmemcg_cache_deact_after_rcu); |
c9fc5864 TH |
4066 | } |
4067 | #endif | |
4068 | ||
b9049e23 YG |
4069 | static int slab_mem_going_offline_callback(void *arg) |
4070 | { | |
4071 | struct kmem_cache *s; | |
4072 | ||
18004c5d | 4073 | mutex_lock(&slab_mutex); |
b9049e23 | 4074 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4075 | __kmem_cache_shrink(s); |
18004c5d | 4076 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4077 | |
4078 | return 0; | |
4079 | } | |
4080 | ||
4081 | static void slab_mem_offline_callback(void *arg) | |
4082 | { | |
4083 | struct kmem_cache_node *n; | |
4084 | struct kmem_cache *s; | |
4085 | struct memory_notify *marg = arg; | |
4086 | int offline_node; | |
4087 | ||
b9d5ab25 | 4088 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4089 | |
4090 | /* | |
4091 | * If the node still has available memory. we need kmem_cache_node | |
4092 | * for it yet. | |
4093 | */ | |
4094 | if (offline_node < 0) | |
4095 | return; | |
4096 | ||
18004c5d | 4097 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4098 | list_for_each_entry(s, &slab_caches, list) { |
4099 | n = get_node(s, offline_node); | |
4100 | if (n) { | |
4101 | /* | |
4102 | * if n->nr_slabs > 0, slabs still exist on the node | |
4103 | * that is going down. We were unable to free them, | |
c9404c9c | 4104 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
4105 | * callback. So, we must fail. |
4106 | */ | |
0f389ec6 | 4107 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
4108 | |
4109 | s->node[offline_node] = NULL; | |
8de66a0c | 4110 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
4111 | } |
4112 | } | |
18004c5d | 4113 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4114 | } |
4115 | ||
4116 | static int slab_mem_going_online_callback(void *arg) | |
4117 | { | |
4118 | struct kmem_cache_node *n; | |
4119 | struct kmem_cache *s; | |
4120 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4121 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4122 | int ret = 0; |
4123 | ||
4124 | /* | |
4125 | * If the node's memory is already available, then kmem_cache_node is | |
4126 | * already created. Nothing to do. | |
4127 | */ | |
4128 | if (nid < 0) | |
4129 | return 0; | |
4130 | ||
4131 | /* | |
0121c619 | 4132 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4133 | * allocate a kmem_cache_node structure in order to bring the node |
4134 | * online. | |
4135 | */ | |
18004c5d | 4136 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4137 | list_for_each_entry(s, &slab_caches, list) { |
4138 | /* | |
4139 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4140 | * since memory is not yet available from the node that | |
4141 | * is brought up. | |
4142 | */ | |
8de66a0c | 4143 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4144 | if (!n) { |
4145 | ret = -ENOMEM; | |
4146 | goto out; | |
4147 | } | |
4053497d | 4148 | init_kmem_cache_node(n); |
b9049e23 YG |
4149 | s->node[nid] = n; |
4150 | } | |
4151 | out: | |
18004c5d | 4152 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4153 | return ret; |
4154 | } | |
4155 | ||
4156 | static int slab_memory_callback(struct notifier_block *self, | |
4157 | unsigned long action, void *arg) | |
4158 | { | |
4159 | int ret = 0; | |
4160 | ||
4161 | switch (action) { | |
4162 | case MEM_GOING_ONLINE: | |
4163 | ret = slab_mem_going_online_callback(arg); | |
4164 | break; | |
4165 | case MEM_GOING_OFFLINE: | |
4166 | ret = slab_mem_going_offline_callback(arg); | |
4167 | break; | |
4168 | case MEM_OFFLINE: | |
4169 | case MEM_CANCEL_ONLINE: | |
4170 | slab_mem_offline_callback(arg); | |
4171 | break; | |
4172 | case MEM_ONLINE: | |
4173 | case MEM_CANCEL_OFFLINE: | |
4174 | break; | |
4175 | } | |
dc19f9db KH |
4176 | if (ret) |
4177 | ret = notifier_from_errno(ret); | |
4178 | else | |
4179 | ret = NOTIFY_OK; | |
b9049e23 YG |
4180 | return ret; |
4181 | } | |
4182 | ||
3ac38faa AM |
4183 | static struct notifier_block slab_memory_callback_nb = { |
4184 | .notifier_call = slab_memory_callback, | |
4185 | .priority = SLAB_CALLBACK_PRI, | |
4186 | }; | |
b9049e23 | 4187 | |
81819f0f CL |
4188 | /******************************************************************** |
4189 | * Basic setup of slabs | |
4190 | *******************************************************************/ | |
4191 | ||
51df1142 CL |
4192 | /* |
4193 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4194 | * the page allocator. Allocate them properly then fix up the pointers |
4195 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4196 | */ |
4197 | ||
dffb4d60 | 4198 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4199 | { |
4200 | int node; | |
dffb4d60 | 4201 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4202 | struct kmem_cache_node *n; |
51df1142 | 4203 | |
dffb4d60 | 4204 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4205 | |
7d557b3c GC |
4206 | /* |
4207 | * This runs very early, and only the boot processor is supposed to be | |
4208 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4209 | * IPIs around. | |
4210 | */ | |
4211 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4212 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4213 | struct page *p; |
4214 | ||
fa45dc25 CL |
4215 | list_for_each_entry(p, &n->partial, lru) |
4216 | p->slab_cache = s; | |
51df1142 | 4217 | |
607bf324 | 4218 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
4219 | list_for_each_entry(p, &n->full, lru) |
4220 | p->slab_cache = s; | |
51df1142 | 4221 | #endif |
51df1142 | 4222 | } |
f7ce3190 | 4223 | slab_init_memcg_params(s); |
dffb4d60 | 4224 | list_add(&s->list, &slab_caches); |
510ded33 | 4225 | memcg_link_cache(s); |
dffb4d60 | 4226 | return s; |
51df1142 CL |
4227 | } |
4228 | ||
81819f0f CL |
4229 | void __init kmem_cache_init(void) |
4230 | { | |
dffb4d60 CL |
4231 | static __initdata struct kmem_cache boot_kmem_cache, |
4232 | boot_kmem_cache_node; | |
51df1142 | 4233 | |
fc8d8620 SG |
4234 | if (debug_guardpage_minorder()) |
4235 | slub_max_order = 0; | |
4236 | ||
dffb4d60 CL |
4237 | kmem_cache_node = &boot_kmem_cache_node; |
4238 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4239 | |
dffb4d60 | 4240 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4241 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4242 | |
3ac38faa | 4243 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4244 | |
4245 | /* Able to allocate the per node structures */ | |
4246 | slab_state = PARTIAL; | |
4247 | ||
dffb4d60 CL |
4248 | create_boot_cache(kmem_cache, "kmem_cache", |
4249 | offsetof(struct kmem_cache, node) + | |
4250 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4251 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4252 | |
dffb4d60 | 4253 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4254 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4255 | |
4256 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4257 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4258 | create_kmalloc_caches(0); |
81819f0f | 4259 | |
210e7a43 TG |
4260 | /* Setup random freelists for each cache */ |
4261 | init_freelist_randomization(); | |
4262 | ||
a96a87bf SAS |
4263 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4264 | slub_cpu_dead); | |
81819f0f | 4265 | |
19af27af | 4266 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%d\n", |
f97d5f63 | 4267 | cache_line_size(), |
81819f0f CL |
4268 | slub_min_order, slub_max_order, slub_min_objects, |
4269 | nr_cpu_ids, nr_node_ids); | |
4270 | } | |
4271 | ||
7e85ee0c PE |
4272 | void __init kmem_cache_init_late(void) |
4273 | { | |
7e85ee0c PE |
4274 | } |
4275 | ||
2633d7a0 | 4276 | struct kmem_cache * |
f4957d5b | 4277 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4278 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4279 | { |
426589f5 | 4280 | struct kmem_cache *s, *c; |
81819f0f | 4281 | |
a44cb944 | 4282 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4283 | if (s) { |
4284 | s->refcount++; | |
84d0ddd6 | 4285 | |
81819f0f CL |
4286 | /* |
4287 | * Adjust the object sizes so that we clear | |
4288 | * the complete object on kzalloc. | |
4289 | */ | |
1b473f29 | 4290 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4291 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4292 | |
426589f5 | 4293 | for_each_memcg_cache(c, s) { |
84d0ddd6 | 4294 | c->object_size = s->object_size; |
52ee6d74 | 4295 | c->inuse = max(c->inuse, ALIGN(size, sizeof(void *))); |
84d0ddd6 VD |
4296 | } |
4297 | ||
7b8f3b66 | 4298 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4299 | s->refcount--; |
cbb79694 | 4300 | s = NULL; |
7b8f3b66 | 4301 | } |
a0e1d1be | 4302 | } |
6446faa2 | 4303 | |
cbb79694 CL |
4304 | return s; |
4305 | } | |
84c1cf62 | 4306 | |
d50112ed | 4307 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4308 | { |
aac3a166 PE |
4309 | int err; |
4310 | ||
4311 | err = kmem_cache_open(s, flags); | |
4312 | if (err) | |
4313 | return err; | |
20cea968 | 4314 | |
45530c44 CL |
4315 | /* Mutex is not taken during early boot */ |
4316 | if (slab_state <= UP) | |
4317 | return 0; | |
4318 | ||
107dab5c | 4319 | memcg_propagate_slab_attrs(s); |
aac3a166 | 4320 | err = sysfs_slab_add(s); |
aac3a166 | 4321 | if (err) |
52b4b950 | 4322 | __kmem_cache_release(s); |
20cea968 | 4323 | |
aac3a166 | 4324 | return err; |
81819f0f | 4325 | } |
81819f0f | 4326 | |
ce71e27c | 4327 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4328 | { |
aadb4bc4 | 4329 | struct kmem_cache *s; |
94b528d0 | 4330 | void *ret; |
aadb4bc4 | 4331 | |
95a05b42 | 4332 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4333 | return kmalloc_large(size, gfpflags); |
4334 | ||
2c59dd65 | 4335 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4336 | |
2408c550 | 4337 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4338 | return s; |
81819f0f | 4339 | |
2b847c3c | 4340 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4341 | |
25985edc | 4342 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4343 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4344 | |
4345 | return ret; | |
81819f0f CL |
4346 | } |
4347 | ||
5d1f57e4 | 4348 | #ifdef CONFIG_NUMA |
81819f0f | 4349 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4350 | int node, unsigned long caller) |
81819f0f | 4351 | { |
aadb4bc4 | 4352 | struct kmem_cache *s; |
94b528d0 | 4353 | void *ret; |
aadb4bc4 | 4354 | |
95a05b42 | 4355 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4356 | ret = kmalloc_large_node(size, gfpflags, node); |
4357 | ||
4358 | trace_kmalloc_node(caller, ret, | |
4359 | size, PAGE_SIZE << get_order(size), | |
4360 | gfpflags, node); | |
4361 | ||
4362 | return ret; | |
4363 | } | |
eada35ef | 4364 | |
2c59dd65 | 4365 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4366 | |
2408c550 | 4367 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4368 | return s; |
81819f0f | 4369 | |
2b847c3c | 4370 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4371 | |
25985edc | 4372 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4373 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4374 | |
4375 | return ret; | |
81819f0f | 4376 | } |
5d1f57e4 | 4377 | #endif |
81819f0f | 4378 | |
ab4d5ed5 | 4379 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4380 | static int count_inuse(struct page *page) |
4381 | { | |
4382 | return page->inuse; | |
4383 | } | |
4384 | ||
4385 | static int count_total(struct page *page) | |
4386 | { | |
4387 | return page->objects; | |
4388 | } | |
ab4d5ed5 | 4389 | #endif |
205ab99d | 4390 | |
ab4d5ed5 | 4391 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4392 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4393 | unsigned long *map) | |
53e15af0 CL |
4394 | { |
4395 | void *p; | |
a973e9dd | 4396 | void *addr = page_address(page); |
53e15af0 CL |
4397 | |
4398 | if (!check_slab(s, page) || | |
4399 | !on_freelist(s, page, NULL)) | |
4400 | return 0; | |
4401 | ||
4402 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4403 | bitmap_zero(map, page->objects); |
53e15af0 | 4404 | |
5f80b13a CL |
4405 | get_map(s, page, map); |
4406 | for_each_object(p, s, addr, page->objects) { | |
4407 | if (test_bit(slab_index(p, s, addr), map)) | |
4408 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4409 | return 0; | |
53e15af0 CL |
4410 | } |
4411 | ||
224a88be | 4412 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4413 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4414 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4415 | return 0; |
4416 | return 1; | |
4417 | } | |
4418 | ||
434e245d CL |
4419 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4420 | unsigned long *map) | |
53e15af0 | 4421 | { |
881db7fb CL |
4422 | slab_lock(page); |
4423 | validate_slab(s, page, map); | |
4424 | slab_unlock(page); | |
53e15af0 CL |
4425 | } |
4426 | ||
434e245d CL |
4427 | static int validate_slab_node(struct kmem_cache *s, |
4428 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4429 | { |
4430 | unsigned long count = 0; | |
4431 | struct page *page; | |
4432 | unsigned long flags; | |
4433 | ||
4434 | spin_lock_irqsave(&n->list_lock, flags); | |
4435 | ||
4436 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4437 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4438 | count++; |
4439 | } | |
4440 | if (count != n->nr_partial) | |
f9f58285 FF |
4441 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4442 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4443 | |
4444 | if (!(s->flags & SLAB_STORE_USER)) | |
4445 | goto out; | |
4446 | ||
4447 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4448 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4449 | count++; |
4450 | } | |
4451 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4452 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4453 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4454 | |
4455 | out: | |
4456 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4457 | return count; | |
4458 | } | |
4459 | ||
434e245d | 4460 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4461 | { |
4462 | int node; | |
4463 | unsigned long count = 0; | |
fa45dc25 | 4464 | struct kmem_cache_node *n; |
0684e652 | 4465 | unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL); |
434e245d CL |
4466 | |
4467 | if (!map) | |
4468 | return -ENOMEM; | |
53e15af0 CL |
4469 | |
4470 | flush_all(s); | |
fa45dc25 | 4471 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4472 | count += validate_slab_node(s, n, map); |
0684e652 | 4473 | bitmap_free(map); |
53e15af0 CL |
4474 | return count; |
4475 | } | |
88a420e4 | 4476 | /* |
672bba3a | 4477 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4478 | * and freed. |
4479 | */ | |
4480 | ||
4481 | struct location { | |
4482 | unsigned long count; | |
ce71e27c | 4483 | unsigned long addr; |
45edfa58 CL |
4484 | long long sum_time; |
4485 | long min_time; | |
4486 | long max_time; | |
4487 | long min_pid; | |
4488 | long max_pid; | |
174596a0 | 4489 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4490 | nodemask_t nodes; |
88a420e4 CL |
4491 | }; |
4492 | ||
4493 | struct loc_track { | |
4494 | unsigned long max; | |
4495 | unsigned long count; | |
4496 | struct location *loc; | |
4497 | }; | |
4498 | ||
4499 | static void free_loc_track(struct loc_track *t) | |
4500 | { | |
4501 | if (t->max) | |
4502 | free_pages((unsigned long)t->loc, | |
4503 | get_order(sizeof(struct location) * t->max)); | |
4504 | } | |
4505 | ||
68dff6a9 | 4506 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4507 | { |
4508 | struct location *l; | |
4509 | int order; | |
4510 | ||
88a420e4 CL |
4511 | order = get_order(sizeof(struct location) * max); |
4512 | ||
68dff6a9 | 4513 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4514 | if (!l) |
4515 | return 0; | |
4516 | ||
4517 | if (t->count) { | |
4518 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4519 | free_loc_track(t); | |
4520 | } | |
4521 | t->max = max; | |
4522 | t->loc = l; | |
4523 | return 1; | |
4524 | } | |
4525 | ||
4526 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4527 | const struct track *track) |
88a420e4 CL |
4528 | { |
4529 | long start, end, pos; | |
4530 | struct location *l; | |
ce71e27c | 4531 | unsigned long caddr; |
45edfa58 | 4532 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4533 | |
4534 | start = -1; | |
4535 | end = t->count; | |
4536 | ||
4537 | for ( ; ; ) { | |
4538 | pos = start + (end - start + 1) / 2; | |
4539 | ||
4540 | /* | |
4541 | * There is nothing at "end". If we end up there | |
4542 | * we need to add something to before end. | |
4543 | */ | |
4544 | if (pos == end) | |
4545 | break; | |
4546 | ||
4547 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4548 | if (track->addr == caddr) { |
4549 | ||
4550 | l = &t->loc[pos]; | |
4551 | l->count++; | |
4552 | if (track->when) { | |
4553 | l->sum_time += age; | |
4554 | if (age < l->min_time) | |
4555 | l->min_time = age; | |
4556 | if (age > l->max_time) | |
4557 | l->max_time = age; | |
4558 | ||
4559 | if (track->pid < l->min_pid) | |
4560 | l->min_pid = track->pid; | |
4561 | if (track->pid > l->max_pid) | |
4562 | l->max_pid = track->pid; | |
4563 | ||
174596a0 RR |
4564 | cpumask_set_cpu(track->cpu, |
4565 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4566 | } |
4567 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4568 | return 1; |
4569 | } | |
4570 | ||
45edfa58 | 4571 | if (track->addr < caddr) |
88a420e4 CL |
4572 | end = pos; |
4573 | else | |
4574 | start = pos; | |
4575 | } | |
4576 | ||
4577 | /* | |
672bba3a | 4578 | * Not found. Insert new tracking element. |
88a420e4 | 4579 | */ |
68dff6a9 | 4580 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4581 | return 0; |
4582 | ||
4583 | l = t->loc + pos; | |
4584 | if (pos < t->count) | |
4585 | memmove(l + 1, l, | |
4586 | (t->count - pos) * sizeof(struct location)); | |
4587 | t->count++; | |
4588 | l->count = 1; | |
45edfa58 CL |
4589 | l->addr = track->addr; |
4590 | l->sum_time = age; | |
4591 | l->min_time = age; | |
4592 | l->max_time = age; | |
4593 | l->min_pid = track->pid; | |
4594 | l->max_pid = track->pid; | |
174596a0 RR |
4595 | cpumask_clear(to_cpumask(l->cpus)); |
4596 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4597 | nodes_clear(l->nodes); |
4598 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4599 | return 1; |
4600 | } | |
4601 | ||
4602 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4603 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4604 | unsigned long *map) |
88a420e4 | 4605 | { |
a973e9dd | 4606 | void *addr = page_address(page); |
88a420e4 CL |
4607 | void *p; |
4608 | ||
39b26464 | 4609 | bitmap_zero(map, page->objects); |
5f80b13a | 4610 | get_map(s, page, map); |
88a420e4 | 4611 | |
224a88be | 4612 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4613 | if (!test_bit(slab_index(p, s, addr), map)) |
4614 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4615 | } |
4616 | ||
4617 | static int list_locations(struct kmem_cache *s, char *buf, | |
4618 | enum track_item alloc) | |
4619 | { | |
e374d483 | 4620 | int len = 0; |
88a420e4 | 4621 | unsigned long i; |
68dff6a9 | 4622 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4623 | int node; |
fa45dc25 | 4624 | struct kmem_cache_node *n; |
0684e652 | 4625 | unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL); |
88a420e4 | 4626 | |
bbd7d57b | 4627 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
0ee931c4 | 4628 | GFP_KERNEL)) { |
0684e652 | 4629 | bitmap_free(map); |
68dff6a9 | 4630 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4631 | } |
88a420e4 CL |
4632 | /* Push back cpu slabs */ |
4633 | flush_all(s); | |
4634 | ||
fa45dc25 | 4635 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4636 | unsigned long flags; |
4637 | struct page *page; | |
4638 | ||
9e86943b | 4639 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4640 | continue; |
4641 | ||
4642 | spin_lock_irqsave(&n->list_lock, flags); | |
4643 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4644 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4645 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4646 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4647 | spin_unlock_irqrestore(&n->list_lock, flags); |
4648 | } | |
4649 | ||
4650 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4651 | struct location *l = &t.loc[i]; |
88a420e4 | 4652 | |
9c246247 | 4653 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4654 | break; |
e374d483 | 4655 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4656 | |
4657 | if (l->addr) | |
62c70bce | 4658 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4659 | else |
e374d483 | 4660 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4661 | |
4662 | if (l->sum_time != l->min_time) { | |
e374d483 | 4663 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4664 | l->min_time, |
4665 | (long)div_u64(l->sum_time, l->count), | |
4666 | l->max_time); | |
45edfa58 | 4667 | } else |
e374d483 | 4668 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4669 | l->min_time); |
4670 | ||
4671 | if (l->min_pid != l->max_pid) | |
e374d483 | 4672 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4673 | l->min_pid, l->max_pid); |
4674 | else | |
e374d483 | 4675 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4676 | l->min_pid); |
4677 | ||
174596a0 RR |
4678 | if (num_online_cpus() > 1 && |
4679 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4680 | len < PAGE_SIZE - 60) |
4681 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4682 | " cpus=%*pbl", | |
4683 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4684 | |
62bc62a8 | 4685 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4686 | len < PAGE_SIZE - 60) |
4687 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4688 | " nodes=%*pbl", | |
4689 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4690 | |
e374d483 | 4691 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4692 | } |
4693 | ||
4694 | free_loc_track(&t); | |
0684e652 | 4695 | bitmap_free(map); |
88a420e4 | 4696 | if (!t.count) |
e374d483 HH |
4697 | len += sprintf(buf, "No data\n"); |
4698 | return len; | |
88a420e4 | 4699 | } |
ab4d5ed5 | 4700 | #endif |
88a420e4 | 4701 | |
a5a84755 | 4702 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4703 | static void __init resiliency_test(void) |
a5a84755 CL |
4704 | { |
4705 | u8 *p; | |
cc252eae | 4706 | int type = KMALLOC_NORMAL; |
a5a84755 | 4707 | |
95a05b42 | 4708 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4709 | |
f9f58285 FF |
4710 | pr_err("SLUB resiliency testing\n"); |
4711 | pr_err("-----------------------\n"); | |
4712 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4713 | |
4714 | p = kzalloc(16, GFP_KERNEL); | |
4715 | p[16] = 0x12; | |
f9f58285 FF |
4716 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4717 | p + 16); | |
a5a84755 | 4718 | |
cc252eae | 4719 | validate_slab_cache(kmalloc_caches[type][4]); |
a5a84755 CL |
4720 | |
4721 | /* Hmmm... The next two are dangerous */ | |
4722 | p = kzalloc(32, GFP_KERNEL); | |
4723 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4724 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4725 | p); | |
4726 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 | 4727 | |
cc252eae | 4728 | validate_slab_cache(kmalloc_caches[type][5]); |
a5a84755 CL |
4729 | p = kzalloc(64, GFP_KERNEL); |
4730 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4731 | *p = 0x56; | |
f9f58285 FF |
4732 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4733 | p); | |
4734 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
cc252eae | 4735 | validate_slab_cache(kmalloc_caches[type][6]); |
a5a84755 | 4736 | |
f9f58285 | 4737 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4738 | p = kzalloc(128, GFP_KERNEL); |
4739 | kfree(p); | |
4740 | *p = 0x78; | |
f9f58285 | 4741 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
cc252eae | 4742 | validate_slab_cache(kmalloc_caches[type][7]); |
a5a84755 CL |
4743 | |
4744 | p = kzalloc(256, GFP_KERNEL); | |
4745 | kfree(p); | |
4746 | p[50] = 0x9a; | |
f9f58285 | 4747 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
cc252eae | 4748 | validate_slab_cache(kmalloc_caches[type][8]); |
a5a84755 CL |
4749 | |
4750 | p = kzalloc(512, GFP_KERNEL); | |
4751 | kfree(p); | |
4752 | p[512] = 0xab; | |
f9f58285 | 4753 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
cc252eae | 4754 | validate_slab_cache(kmalloc_caches[type][9]); |
a5a84755 CL |
4755 | } |
4756 | #else | |
4757 | #ifdef CONFIG_SYSFS | |
4758 | static void resiliency_test(void) {}; | |
4759 | #endif | |
4760 | #endif | |
4761 | ||
ab4d5ed5 | 4762 | #ifdef CONFIG_SYSFS |
81819f0f | 4763 | enum slab_stat_type { |
205ab99d CL |
4764 | SL_ALL, /* All slabs */ |
4765 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4766 | SL_CPU, /* Only slabs used for cpu caches */ | |
4767 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4768 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4769 | }; |
4770 | ||
205ab99d | 4771 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4772 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4773 | #define SO_CPU (1 << SL_CPU) | |
4774 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4775 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4776 | |
1663f26d TH |
4777 | #ifdef CONFIG_MEMCG |
4778 | static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON); | |
4779 | ||
4780 | static int __init setup_slub_memcg_sysfs(char *str) | |
4781 | { | |
4782 | int v; | |
4783 | ||
4784 | if (get_option(&str, &v) > 0) | |
4785 | memcg_sysfs_enabled = v; | |
4786 | ||
4787 | return 1; | |
4788 | } | |
4789 | ||
4790 | __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs); | |
4791 | #endif | |
4792 | ||
62e5c4b4 CG |
4793 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4794 | char *buf, unsigned long flags) | |
81819f0f CL |
4795 | { |
4796 | unsigned long total = 0; | |
81819f0f CL |
4797 | int node; |
4798 | int x; | |
4799 | unsigned long *nodes; | |
81819f0f | 4800 | |
6396bb22 | 4801 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
4802 | if (!nodes) |
4803 | return -ENOMEM; | |
81819f0f | 4804 | |
205ab99d CL |
4805 | if (flags & SO_CPU) { |
4806 | int cpu; | |
81819f0f | 4807 | |
205ab99d | 4808 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4809 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4810 | cpu); | |
ec3ab083 | 4811 | int node; |
49e22585 | 4812 | struct page *page; |
dfb4f096 | 4813 | |
4db0c3c2 | 4814 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4815 | if (!page) |
4816 | continue; | |
205ab99d | 4817 | |
ec3ab083 CL |
4818 | node = page_to_nid(page); |
4819 | if (flags & SO_TOTAL) | |
4820 | x = page->objects; | |
4821 | else if (flags & SO_OBJECTS) | |
4822 | x = page->inuse; | |
4823 | else | |
4824 | x = 1; | |
49e22585 | 4825 | |
ec3ab083 CL |
4826 | total += x; |
4827 | nodes[node] += x; | |
4828 | ||
a93cf07b | 4829 | page = slub_percpu_partial_read_once(c); |
49e22585 | 4830 | if (page) { |
8afb1474 LZ |
4831 | node = page_to_nid(page); |
4832 | if (flags & SO_TOTAL) | |
4833 | WARN_ON_ONCE(1); | |
4834 | else if (flags & SO_OBJECTS) | |
4835 | WARN_ON_ONCE(1); | |
4836 | else | |
4837 | x = page->pages; | |
bc6697d8 ED |
4838 | total += x; |
4839 | nodes[node] += x; | |
49e22585 | 4840 | } |
81819f0f CL |
4841 | } |
4842 | } | |
4843 | ||
bfc8c901 | 4844 | get_online_mems(); |
ab4d5ed5 | 4845 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4846 | if (flags & SO_ALL) { |
fa45dc25 CL |
4847 | struct kmem_cache_node *n; |
4848 | ||
4849 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4850 | |
d0e0ac97 CG |
4851 | if (flags & SO_TOTAL) |
4852 | x = atomic_long_read(&n->total_objects); | |
4853 | else if (flags & SO_OBJECTS) | |
4854 | x = atomic_long_read(&n->total_objects) - | |
4855 | count_partial(n, count_free); | |
81819f0f | 4856 | else |
205ab99d | 4857 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4858 | total += x; |
4859 | nodes[node] += x; | |
4860 | } | |
4861 | ||
ab4d5ed5 CL |
4862 | } else |
4863 | #endif | |
4864 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4865 | struct kmem_cache_node *n; |
81819f0f | 4866 | |
fa45dc25 | 4867 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4868 | if (flags & SO_TOTAL) |
4869 | x = count_partial(n, count_total); | |
4870 | else if (flags & SO_OBJECTS) | |
4871 | x = count_partial(n, count_inuse); | |
81819f0f | 4872 | else |
205ab99d | 4873 | x = n->nr_partial; |
81819f0f CL |
4874 | total += x; |
4875 | nodes[node] += x; | |
4876 | } | |
4877 | } | |
81819f0f CL |
4878 | x = sprintf(buf, "%lu", total); |
4879 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4880 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4881 | if (nodes[node]) |
4882 | x += sprintf(buf + x, " N%d=%lu", | |
4883 | node, nodes[node]); | |
4884 | #endif | |
bfc8c901 | 4885 | put_online_mems(); |
81819f0f CL |
4886 | kfree(nodes); |
4887 | return x + sprintf(buf + x, "\n"); | |
4888 | } | |
4889 | ||
ab4d5ed5 | 4890 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4891 | static int any_slab_objects(struct kmem_cache *s) |
4892 | { | |
4893 | int node; | |
fa45dc25 | 4894 | struct kmem_cache_node *n; |
81819f0f | 4895 | |
fa45dc25 | 4896 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4897 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4898 | return 1; |
fa45dc25 | 4899 | |
81819f0f CL |
4900 | return 0; |
4901 | } | |
ab4d5ed5 | 4902 | #endif |
81819f0f CL |
4903 | |
4904 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4905 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4906 | |
4907 | struct slab_attribute { | |
4908 | struct attribute attr; | |
4909 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4910 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4911 | }; | |
4912 | ||
4913 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4914 | static struct slab_attribute _name##_attr = \ |
4915 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4916 | |
4917 | #define SLAB_ATTR(_name) \ | |
4918 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4919 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4920 | |
81819f0f CL |
4921 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4922 | { | |
44065b2e | 4923 | return sprintf(buf, "%u\n", s->size); |
81819f0f CL |
4924 | } |
4925 | SLAB_ATTR_RO(slab_size); | |
4926 | ||
4927 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4928 | { | |
3a3791ec | 4929 | return sprintf(buf, "%u\n", s->align); |
81819f0f CL |
4930 | } |
4931 | SLAB_ATTR_RO(align); | |
4932 | ||
4933 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4934 | { | |
1b473f29 | 4935 | return sprintf(buf, "%u\n", s->object_size); |
81819f0f CL |
4936 | } |
4937 | SLAB_ATTR_RO(object_size); | |
4938 | ||
4939 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4940 | { | |
19af27af | 4941 | return sprintf(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
4942 | } |
4943 | SLAB_ATTR_RO(objs_per_slab); | |
4944 | ||
06b285dc CL |
4945 | static ssize_t order_store(struct kmem_cache *s, |
4946 | const char *buf, size_t length) | |
4947 | { | |
19af27af | 4948 | unsigned int order; |
0121c619 CL |
4949 | int err; |
4950 | ||
19af27af | 4951 | err = kstrtouint(buf, 10, &order); |
0121c619 CL |
4952 | if (err) |
4953 | return err; | |
06b285dc CL |
4954 | |
4955 | if (order > slub_max_order || order < slub_min_order) | |
4956 | return -EINVAL; | |
4957 | ||
4958 | calculate_sizes(s, order); | |
4959 | return length; | |
4960 | } | |
4961 | ||
81819f0f CL |
4962 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4963 | { | |
19af27af | 4964 | return sprintf(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 4965 | } |
06b285dc | 4966 | SLAB_ATTR(order); |
81819f0f | 4967 | |
73d342b1 DR |
4968 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4969 | { | |
4970 | return sprintf(buf, "%lu\n", s->min_partial); | |
4971 | } | |
4972 | ||
4973 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4974 | size_t length) | |
4975 | { | |
4976 | unsigned long min; | |
4977 | int err; | |
4978 | ||
3dbb95f7 | 4979 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4980 | if (err) |
4981 | return err; | |
4982 | ||
c0bdb232 | 4983 | set_min_partial(s, min); |
73d342b1 DR |
4984 | return length; |
4985 | } | |
4986 | SLAB_ATTR(min_partial); | |
4987 | ||
49e22585 CL |
4988 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4989 | { | |
e6d0e1dc | 4990 | return sprintf(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
4991 | } |
4992 | ||
4993 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4994 | size_t length) | |
4995 | { | |
e5d9998f | 4996 | unsigned int objects; |
49e22585 CL |
4997 | int err; |
4998 | ||
e5d9998f | 4999 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
5000 | if (err) |
5001 | return err; | |
345c905d | 5002 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 5003 | return -EINVAL; |
49e22585 | 5004 | |
e6d0e1dc | 5005 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
5006 | flush_all(s); |
5007 | return length; | |
5008 | } | |
5009 | SLAB_ATTR(cpu_partial); | |
5010 | ||
81819f0f CL |
5011 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5012 | { | |
62c70bce JP |
5013 | if (!s->ctor) |
5014 | return 0; | |
5015 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
5016 | } |
5017 | SLAB_ATTR_RO(ctor); | |
5018 | ||
81819f0f CL |
5019 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5020 | { | |
4307c14f | 5021 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5022 | } |
5023 | SLAB_ATTR_RO(aliases); | |
5024 | ||
81819f0f CL |
5025 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5026 | { | |
d9acf4b7 | 5027 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5028 | } |
5029 | SLAB_ATTR_RO(partial); | |
5030 | ||
5031 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5032 | { | |
d9acf4b7 | 5033 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5034 | } |
5035 | SLAB_ATTR_RO(cpu_slabs); | |
5036 | ||
5037 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5038 | { | |
205ab99d | 5039 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5040 | } |
5041 | SLAB_ATTR_RO(objects); | |
5042 | ||
205ab99d CL |
5043 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5044 | { | |
5045 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5046 | } | |
5047 | SLAB_ATTR_RO(objects_partial); | |
5048 | ||
49e22585 CL |
5049 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5050 | { | |
5051 | int objects = 0; | |
5052 | int pages = 0; | |
5053 | int cpu; | |
5054 | int len; | |
5055 | ||
5056 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5057 | struct page *page; |
5058 | ||
5059 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5060 | |
5061 | if (page) { | |
5062 | pages += page->pages; | |
5063 | objects += page->pobjects; | |
5064 | } | |
5065 | } | |
5066 | ||
5067 | len = sprintf(buf, "%d(%d)", objects, pages); | |
5068 | ||
5069 | #ifdef CONFIG_SMP | |
5070 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5071 | struct page *page; |
5072 | ||
5073 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5074 | |
5075 | if (page && len < PAGE_SIZE - 20) | |
5076 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
5077 | page->pobjects, page->pages); | |
5078 | } | |
5079 | #endif | |
5080 | return len + sprintf(buf + len, "\n"); | |
5081 | } | |
5082 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5083 | ||
a5a84755 CL |
5084 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5085 | { | |
5086 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
5087 | } | |
5088 | ||
5089 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
5090 | const char *buf, size_t length) | |
5091 | { | |
5092 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
5093 | if (buf[0] == '1') | |
5094 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
5095 | return length; | |
5096 | } | |
5097 | SLAB_ATTR(reclaim_account); | |
5098 | ||
5099 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5100 | { | |
5101 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
5102 | } | |
5103 | SLAB_ATTR_RO(hwcache_align); | |
5104 | ||
5105 | #ifdef CONFIG_ZONE_DMA | |
5106 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5107 | { | |
5108 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
5109 | } | |
5110 | SLAB_ATTR_RO(cache_dma); | |
5111 | #endif | |
5112 | ||
8eb8284b DW |
5113 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5114 | { | |
7bbdb81e | 5115 | return sprintf(buf, "%u\n", s->usersize); |
8eb8284b DW |
5116 | } |
5117 | SLAB_ATTR_RO(usersize); | |
5118 | ||
a5a84755 CL |
5119 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5120 | { | |
5f0d5a3a | 5121 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5122 | } |
5123 | SLAB_ATTR_RO(destroy_by_rcu); | |
5124 | ||
ab4d5ed5 | 5125 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5126 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5127 | { | |
5128 | return show_slab_objects(s, buf, SO_ALL); | |
5129 | } | |
5130 | SLAB_ATTR_RO(slabs); | |
5131 | ||
205ab99d CL |
5132 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5133 | { | |
5134 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5135 | } | |
5136 | SLAB_ATTR_RO(total_objects); | |
5137 | ||
81819f0f CL |
5138 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5139 | { | |
becfda68 | 5140 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f CL |
5141 | } |
5142 | ||
5143 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
5144 | const char *buf, size_t length) | |
5145 | { | |
becfda68 | 5146 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; |
b789ef51 CL |
5147 | if (buf[0] == '1') { |
5148 | s->flags &= ~__CMPXCHG_DOUBLE; | |
becfda68 | 5149 | s->flags |= SLAB_CONSISTENCY_CHECKS; |
b789ef51 | 5150 | } |
81819f0f CL |
5151 | return length; |
5152 | } | |
5153 | SLAB_ATTR(sanity_checks); | |
5154 | ||
5155 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5156 | { | |
5157 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
5158 | } | |
5159 | ||
5160 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
5161 | size_t length) | |
5162 | { | |
c9e16131 CL |
5163 | /* |
5164 | * Tracing a merged cache is going to give confusing results | |
5165 | * as well as cause other issues like converting a mergeable | |
5166 | * cache into an umergeable one. | |
5167 | */ | |
5168 | if (s->refcount > 1) | |
5169 | return -EINVAL; | |
5170 | ||
81819f0f | 5171 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
5172 | if (buf[0] == '1') { |
5173 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5174 | s->flags |= SLAB_TRACE; |
b789ef51 | 5175 | } |
81819f0f CL |
5176 | return length; |
5177 | } | |
5178 | SLAB_ATTR(trace); | |
5179 | ||
81819f0f CL |
5180 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5181 | { | |
5182 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
5183 | } | |
5184 | ||
5185 | static ssize_t red_zone_store(struct kmem_cache *s, | |
5186 | const char *buf, size_t length) | |
5187 | { | |
5188 | if (any_slab_objects(s)) | |
5189 | return -EBUSY; | |
5190 | ||
5191 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 | 5192 | if (buf[0] == '1') { |
81819f0f | 5193 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 5194 | } |
06b285dc | 5195 | calculate_sizes(s, -1); |
81819f0f CL |
5196 | return length; |
5197 | } | |
5198 | SLAB_ATTR(red_zone); | |
5199 | ||
5200 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5201 | { | |
5202 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
5203 | } | |
5204 | ||
5205 | static ssize_t poison_store(struct kmem_cache *s, | |
5206 | const char *buf, size_t length) | |
5207 | { | |
5208 | if (any_slab_objects(s)) | |
5209 | return -EBUSY; | |
5210 | ||
5211 | s->flags &= ~SLAB_POISON; | |
b789ef51 | 5212 | if (buf[0] == '1') { |
81819f0f | 5213 | s->flags |= SLAB_POISON; |
b789ef51 | 5214 | } |
06b285dc | 5215 | calculate_sizes(s, -1); |
81819f0f CL |
5216 | return length; |
5217 | } | |
5218 | SLAB_ATTR(poison); | |
5219 | ||
5220 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5221 | { | |
5222 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
5223 | } | |
5224 | ||
5225 | static ssize_t store_user_store(struct kmem_cache *s, | |
5226 | const char *buf, size_t length) | |
5227 | { | |
5228 | if (any_slab_objects(s)) | |
5229 | return -EBUSY; | |
5230 | ||
5231 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
5232 | if (buf[0] == '1') { |
5233 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5234 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 5235 | } |
06b285dc | 5236 | calculate_sizes(s, -1); |
81819f0f CL |
5237 | return length; |
5238 | } | |
5239 | SLAB_ATTR(store_user); | |
5240 | ||
53e15af0 CL |
5241 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5242 | { | |
5243 | return 0; | |
5244 | } | |
5245 | ||
5246 | static ssize_t validate_store(struct kmem_cache *s, | |
5247 | const char *buf, size_t length) | |
5248 | { | |
434e245d CL |
5249 | int ret = -EINVAL; |
5250 | ||
5251 | if (buf[0] == '1') { | |
5252 | ret = validate_slab_cache(s); | |
5253 | if (ret >= 0) | |
5254 | ret = length; | |
5255 | } | |
5256 | return ret; | |
53e15af0 CL |
5257 | } |
5258 | SLAB_ATTR(validate); | |
a5a84755 CL |
5259 | |
5260 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5261 | { | |
5262 | if (!(s->flags & SLAB_STORE_USER)) | |
5263 | return -ENOSYS; | |
5264 | return list_locations(s, buf, TRACK_ALLOC); | |
5265 | } | |
5266 | SLAB_ATTR_RO(alloc_calls); | |
5267 | ||
5268 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5269 | { | |
5270 | if (!(s->flags & SLAB_STORE_USER)) | |
5271 | return -ENOSYS; | |
5272 | return list_locations(s, buf, TRACK_FREE); | |
5273 | } | |
5274 | SLAB_ATTR_RO(free_calls); | |
5275 | #endif /* CONFIG_SLUB_DEBUG */ | |
5276 | ||
5277 | #ifdef CONFIG_FAILSLAB | |
5278 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5279 | { | |
5280 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
5281 | } | |
5282 | ||
5283 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
5284 | size_t length) | |
5285 | { | |
c9e16131 CL |
5286 | if (s->refcount > 1) |
5287 | return -EINVAL; | |
5288 | ||
a5a84755 CL |
5289 | s->flags &= ~SLAB_FAILSLAB; |
5290 | if (buf[0] == '1') | |
5291 | s->flags |= SLAB_FAILSLAB; | |
5292 | return length; | |
5293 | } | |
5294 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 5295 | #endif |
53e15af0 | 5296 | |
2086d26a CL |
5297 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5298 | { | |
5299 | return 0; | |
5300 | } | |
5301 | ||
5302 | static ssize_t shrink_store(struct kmem_cache *s, | |
5303 | const char *buf, size_t length) | |
5304 | { | |
832f37f5 VD |
5305 | if (buf[0] == '1') |
5306 | kmem_cache_shrink(s); | |
5307 | else | |
2086d26a CL |
5308 | return -EINVAL; |
5309 | return length; | |
5310 | } | |
5311 | SLAB_ATTR(shrink); | |
5312 | ||
81819f0f | 5313 | #ifdef CONFIG_NUMA |
9824601e | 5314 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5315 | { |
eb7235eb | 5316 | return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5317 | } |
5318 | ||
9824601e | 5319 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5320 | const char *buf, size_t length) |
5321 | { | |
eb7235eb | 5322 | unsigned int ratio; |
0121c619 CL |
5323 | int err; |
5324 | ||
eb7235eb | 5325 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5326 | if (err) |
5327 | return err; | |
eb7235eb AD |
5328 | if (ratio > 100) |
5329 | return -ERANGE; | |
0121c619 | 5330 | |
eb7235eb | 5331 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5332 | |
81819f0f CL |
5333 | return length; |
5334 | } | |
9824601e | 5335 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5336 | #endif |
5337 | ||
8ff12cfc | 5338 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5339 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5340 | { | |
5341 | unsigned long sum = 0; | |
5342 | int cpu; | |
5343 | int len; | |
6da2ec56 | 5344 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5345 | |
5346 | if (!data) | |
5347 | return -ENOMEM; | |
5348 | ||
5349 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5350 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5351 | |
5352 | data[cpu] = x; | |
5353 | sum += x; | |
5354 | } | |
5355 | ||
5356 | len = sprintf(buf, "%lu", sum); | |
5357 | ||
50ef37b9 | 5358 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5359 | for_each_online_cpu(cpu) { |
5360 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5361 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5362 | } |
50ef37b9 | 5363 | #endif |
8ff12cfc CL |
5364 | kfree(data); |
5365 | return len + sprintf(buf + len, "\n"); | |
5366 | } | |
5367 | ||
78eb00cc DR |
5368 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5369 | { | |
5370 | int cpu; | |
5371 | ||
5372 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5373 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5374 | } |
5375 | ||
8ff12cfc CL |
5376 | #define STAT_ATTR(si, text) \ |
5377 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5378 | { \ | |
5379 | return show_stat(s, buf, si); \ | |
5380 | } \ | |
78eb00cc DR |
5381 | static ssize_t text##_store(struct kmem_cache *s, \ |
5382 | const char *buf, size_t length) \ | |
5383 | { \ | |
5384 | if (buf[0] != '0') \ | |
5385 | return -EINVAL; \ | |
5386 | clear_stat(s, si); \ | |
5387 | return length; \ | |
5388 | } \ | |
5389 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5390 | |
5391 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5392 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5393 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5394 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5395 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5396 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5397 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5398 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5399 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5400 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5401 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5402 | STAT_ATTR(FREE_SLAB, free_slab); |
5403 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5404 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5405 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5406 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5407 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5408 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5409 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5410 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5411 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5412 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5413 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5414 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5415 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5416 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5417 | #endif |
5418 | ||
06428780 | 5419 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5420 | &slab_size_attr.attr, |
5421 | &object_size_attr.attr, | |
5422 | &objs_per_slab_attr.attr, | |
5423 | &order_attr.attr, | |
73d342b1 | 5424 | &min_partial_attr.attr, |
49e22585 | 5425 | &cpu_partial_attr.attr, |
81819f0f | 5426 | &objects_attr.attr, |
205ab99d | 5427 | &objects_partial_attr.attr, |
81819f0f CL |
5428 | &partial_attr.attr, |
5429 | &cpu_slabs_attr.attr, | |
5430 | &ctor_attr.attr, | |
81819f0f CL |
5431 | &aliases_attr.attr, |
5432 | &align_attr.attr, | |
81819f0f CL |
5433 | &hwcache_align_attr.attr, |
5434 | &reclaim_account_attr.attr, | |
5435 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5436 | &shrink_attr.attr, |
49e22585 | 5437 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5438 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5439 | &total_objects_attr.attr, |
5440 | &slabs_attr.attr, | |
5441 | &sanity_checks_attr.attr, | |
5442 | &trace_attr.attr, | |
81819f0f CL |
5443 | &red_zone_attr.attr, |
5444 | &poison_attr.attr, | |
5445 | &store_user_attr.attr, | |
53e15af0 | 5446 | &validate_attr.attr, |
88a420e4 CL |
5447 | &alloc_calls_attr.attr, |
5448 | &free_calls_attr.attr, | |
ab4d5ed5 | 5449 | #endif |
81819f0f CL |
5450 | #ifdef CONFIG_ZONE_DMA |
5451 | &cache_dma_attr.attr, | |
5452 | #endif | |
5453 | #ifdef CONFIG_NUMA | |
9824601e | 5454 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5455 | #endif |
5456 | #ifdef CONFIG_SLUB_STATS | |
5457 | &alloc_fastpath_attr.attr, | |
5458 | &alloc_slowpath_attr.attr, | |
5459 | &free_fastpath_attr.attr, | |
5460 | &free_slowpath_attr.attr, | |
5461 | &free_frozen_attr.attr, | |
5462 | &free_add_partial_attr.attr, | |
5463 | &free_remove_partial_attr.attr, | |
5464 | &alloc_from_partial_attr.attr, | |
5465 | &alloc_slab_attr.attr, | |
5466 | &alloc_refill_attr.attr, | |
e36a2652 | 5467 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5468 | &free_slab_attr.attr, |
5469 | &cpuslab_flush_attr.attr, | |
5470 | &deactivate_full_attr.attr, | |
5471 | &deactivate_empty_attr.attr, | |
5472 | &deactivate_to_head_attr.attr, | |
5473 | &deactivate_to_tail_attr.attr, | |
5474 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5475 | &deactivate_bypass_attr.attr, |
65c3376a | 5476 | &order_fallback_attr.attr, |
b789ef51 CL |
5477 | &cmpxchg_double_fail_attr.attr, |
5478 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5479 | &cpu_partial_alloc_attr.attr, |
5480 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5481 | &cpu_partial_node_attr.attr, |
5482 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5483 | #endif |
4c13dd3b DM |
5484 | #ifdef CONFIG_FAILSLAB |
5485 | &failslab_attr.attr, | |
5486 | #endif | |
8eb8284b | 5487 | &usersize_attr.attr, |
4c13dd3b | 5488 | |
81819f0f CL |
5489 | NULL |
5490 | }; | |
5491 | ||
1fdaaa23 | 5492 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5493 | .attrs = slab_attrs, |
5494 | }; | |
5495 | ||
5496 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5497 | struct attribute *attr, | |
5498 | char *buf) | |
5499 | { | |
5500 | struct slab_attribute *attribute; | |
5501 | struct kmem_cache *s; | |
5502 | int err; | |
5503 | ||
5504 | attribute = to_slab_attr(attr); | |
5505 | s = to_slab(kobj); | |
5506 | ||
5507 | if (!attribute->show) | |
5508 | return -EIO; | |
5509 | ||
5510 | err = attribute->show(s, buf); | |
5511 | ||
5512 | return err; | |
5513 | } | |
5514 | ||
5515 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5516 | struct attribute *attr, | |
5517 | const char *buf, size_t len) | |
5518 | { | |
5519 | struct slab_attribute *attribute; | |
5520 | struct kmem_cache *s; | |
5521 | int err; | |
5522 | ||
5523 | attribute = to_slab_attr(attr); | |
5524 | s = to_slab(kobj); | |
5525 | ||
5526 | if (!attribute->store) | |
5527 | return -EIO; | |
5528 | ||
5529 | err = attribute->store(s, buf, len); | |
127424c8 | 5530 | #ifdef CONFIG_MEMCG |
107dab5c | 5531 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5532 | struct kmem_cache *c; |
81819f0f | 5533 | |
107dab5c GC |
5534 | mutex_lock(&slab_mutex); |
5535 | if (s->max_attr_size < len) | |
5536 | s->max_attr_size = len; | |
5537 | ||
ebe945c2 GC |
5538 | /* |
5539 | * This is a best effort propagation, so this function's return | |
5540 | * value will be determined by the parent cache only. This is | |
5541 | * basically because not all attributes will have a well | |
5542 | * defined semantics for rollbacks - most of the actions will | |
5543 | * have permanent effects. | |
5544 | * | |
5545 | * Returning the error value of any of the children that fail | |
5546 | * is not 100 % defined, in the sense that users seeing the | |
5547 | * error code won't be able to know anything about the state of | |
5548 | * the cache. | |
5549 | * | |
5550 | * Only returning the error code for the parent cache at least | |
5551 | * has well defined semantics. The cache being written to | |
5552 | * directly either failed or succeeded, in which case we loop | |
5553 | * through the descendants with best-effort propagation. | |
5554 | */ | |
426589f5 VD |
5555 | for_each_memcg_cache(c, s) |
5556 | attribute->store(c, buf, len); | |
107dab5c GC |
5557 | mutex_unlock(&slab_mutex); |
5558 | } | |
5559 | #endif | |
81819f0f CL |
5560 | return err; |
5561 | } | |
5562 | ||
107dab5c GC |
5563 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5564 | { | |
127424c8 | 5565 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5566 | int i; |
5567 | char *buffer = NULL; | |
93030d83 | 5568 | struct kmem_cache *root_cache; |
107dab5c | 5569 | |
93030d83 | 5570 | if (is_root_cache(s)) |
107dab5c GC |
5571 | return; |
5572 | ||
f7ce3190 | 5573 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5574 | |
107dab5c GC |
5575 | /* |
5576 | * This mean this cache had no attribute written. Therefore, no point | |
5577 | * in copying default values around | |
5578 | */ | |
93030d83 | 5579 | if (!root_cache->max_attr_size) |
107dab5c GC |
5580 | return; |
5581 | ||
5582 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5583 | char mbuf[64]; | |
5584 | char *buf; | |
5585 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
478fe303 | 5586 | ssize_t len; |
107dab5c GC |
5587 | |
5588 | if (!attr || !attr->store || !attr->show) | |
5589 | continue; | |
5590 | ||
5591 | /* | |
5592 | * It is really bad that we have to allocate here, so we will | |
5593 | * do it only as a fallback. If we actually allocate, though, | |
5594 | * we can just use the allocated buffer until the end. | |
5595 | * | |
5596 | * Most of the slub attributes will tend to be very small in | |
5597 | * size, but sysfs allows buffers up to a page, so they can | |
5598 | * theoretically happen. | |
5599 | */ | |
5600 | if (buffer) | |
5601 | buf = buffer; | |
93030d83 | 5602 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5603 | buf = mbuf; |
5604 | else { | |
5605 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5606 | if (WARN_ON(!buffer)) | |
5607 | continue; | |
5608 | buf = buffer; | |
5609 | } | |
5610 | ||
478fe303 TG |
5611 | len = attr->show(root_cache, buf); |
5612 | if (len > 0) | |
5613 | attr->store(s, buf, len); | |
107dab5c GC |
5614 | } |
5615 | ||
5616 | if (buffer) | |
5617 | free_page((unsigned long)buffer); | |
5618 | #endif | |
5619 | } | |
5620 | ||
41a21285 CL |
5621 | static void kmem_cache_release(struct kobject *k) |
5622 | { | |
5623 | slab_kmem_cache_release(to_slab(k)); | |
5624 | } | |
5625 | ||
52cf25d0 | 5626 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5627 | .show = slab_attr_show, |
5628 | .store = slab_attr_store, | |
5629 | }; | |
5630 | ||
5631 | static struct kobj_type slab_ktype = { | |
5632 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5633 | .release = kmem_cache_release, |
81819f0f CL |
5634 | }; |
5635 | ||
5636 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5637 | { | |
5638 | struct kobj_type *ktype = get_ktype(kobj); | |
5639 | ||
5640 | if (ktype == &slab_ktype) | |
5641 | return 1; | |
5642 | return 0; | |
5643 | } | |
5644 | ||
9cd43611 | 5645 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5646 | .filter = uevent_filter, |
5647 | }; | |
5648 | ||
27c3a314 | 5649 | static struct kset *slab_kset; |
81819f0f | 5650 | |
9a41707b VD |
5651 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5652 | { | |
127424c8 | 5653 | #ifdef CONFIG_MEMCG |
9a41707b | 5654 | if (!is_root_cache(s)) |
f7ce3190 | 5655 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5656 | #endif |
5657 | return slab_kset; | |
5658 | } | |
5659 | ||
81819f0f CL |
5660 | #define ID_STR_LENGTH 64 |
5661 | ||
5662 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5663 | * |
5664 | * Format :[flags-]size | |
81819f0f CL |
5665 | */ |
5666 | static char *create_unique_id(struct kmem_cache *s) | |
5667 | { | |
5668 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5669 | char *p = name; | |
5670 | ||
5671 | BUG_ON(!name); | |
5672 | ||
5673 | *p++ = ':'; | |
5674 | /* | |
5675 | * First flags affecting slabcache operations. We will only | |
5676 | * get here for aliasable slabs so we do not need to support | |
5677 | * too many flags. The flags here must cover all flags that | |
5678 | * are matched during merging to guarantee that the id is | |
5679 | * unique. | |
5680 | */ | |
5681 | if (s->flags & SLAB_CACHE_DMA) | |
5682 | *p++ = 'd'; | |
5683 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5684 | *p++ = 'a'; | |
becfda68 | 5685 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5686 | *p++ = 'F'; |
230e9fc2 VD |
5687 | if (s->flags & SLAB_ACCOUNT) |
5688 | *p++ = 'A'; | |
81819f0f CL |
5689 | if (p != name + 1) |
5690 | *p++ = '-'; | |
44065b2e | 5691 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5692 | |
81819f0f CL |
5693 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5694 | return name; | |
5695 | } | |
5696 | ||
3b7b3140 TH |
5697 | static void sysfs_slab_remove_workfn(struct work_struct *work) |
5698 | { | |
5699 | struct kmem_cache *s = | |
5700 | container_of(work, struct kmem_cache, kobj_remove_work); | |
5701 | ||
5702 | if (!s->kobj.state_in_sysfs) | |
5703 | /* | |
5704 | * For a memcg cache, this may be called during | |
5705 | * deactivation and again on shutdown. Remove only once. | |
5706 | * A cache is never shut down before deactivation is | |
5707 | * complete, so no need to worry about synchronization. | |
5708 | */ | |
f6ba4880 | 5709 | goto out; |
3b7b3140 TH |
5710 | |
5711 | #ifdef CONFIG_MEMCG | |
5712 | kset_unregister(s->memcg_kset); | |
5713 | #endif | |
5714 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
f6ba4880 | 5715 | out: |
3b7b3140 TH |
5716 | kobject_put(&s->kobj); |
5717 | } | |
5718 | ||
81819f0f CL |
5719 | static int sysfs_slab_add(struct kmem_cache *s) |
5720 | { | |
5721 | int err; | |
5722 | const char *name; | |
1663f26d | 5723 | struct kset *kset = cache_kset(s); |
45530c44 | 5724 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5725 | |
3b7b3140 TH |
5726 | INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn); |
5727 | ||
1663f26d TH |
5728 | if (!kset) { |
5729 | kobject_init(&s->kobj, &slab_ktype); | |
5730 | return 0; | |
5731 | } | |
5732 | ||
11066386 MC |
5733 | if (!unmergeable && disable_higher_order_debug && |
5734 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5735 | unmergeable = 1; | |
5736 | ||
81819f0f CL |
5737 | if (unmergeable) { |
5738 | /* | |
5739 | * Slabcache can never be merged so we can use the name proper. | |
5740 | * This is typically the case for debug situations. In that | |
5741 | * case we can catch duplicate names easily. | |
5742 | */ | |
27c3a314 | 5743 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5744 | name = s->name; |
5745 | } else { | |
5746 | /* | |
5747 | * Create a unique name for the slab as a target | |
5748 | * for the symlinks. | |
5749 | */ | |
5750 | name = create_unique_id(s); | |
5751 | } | |
5752 | ||
1663f26d | 5753 | s->kobj.kset = kset; |
26e4f205 | 5754 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5755 | if (err) |
80da026a | 5756 | goto out; |
81819f0f CL |
5757 | |
5758 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5759 | if (err) |
5760 | goto out_del_kobj; | |
9a41707b | 5761 | |
127424c8 | 5762 | #ifdef CONFIG_MEMCG |
1663f26d | 5763 | if (is_root_cache(s) && memcg_sysfs_enabled) { |
9a41707b VD |
5764 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); |
5765 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5766 | err = -ENOMEM; |
5767 | goto out_del_kobj; | |
9a41707b VD |
5768 | } |
5769 | } | |
5770 | #endif | |
5771 | ||
81819f0f CL |
5772 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5773 | if (!unmergeable) { | |
5774 | /* Setup first alias */ | |
5775 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5776 | } |
54b6a731 DJ |
5777 | out: |
5778 | if (!unmergeable) | |
5779 | kfree(name); | |
5780 | return err; | |
5781 | out_del_kobj: | |
5782 | kobject_del(&s->kobj); | |
54b6a731 | 5783 | goto out; |
81819f0f CL |
5784 | } |
5785 | ||
bf5eb3de | 5786 | static void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5787 | { |
97d06609 | 5788 | if (slab_state < FULL) |
2bce6485 CL |
5789 | /* |
5790 | * Sysfs has not been setup yet so no need to remove the | |
5791 | * cache from sysfs. | |
5792 | */ | |
5793 | return; | |
5794 | ||
3b7b3140 TH |
5795 | kobject_get(&s->kobj); |
5796 | schedule_work(&s->kobj_remove_work); | |
bf5eb3de TH |
5797 | } |
5798 | ||
d50d82fa MP |
5799 | void sysfs_slab_unlink(struct kmem_cache *s) |
5800 | { | |
5801 | if (slab_state >= FULL) | |
5802 | kobject_del(&s->kobj); | |
5803 | } | |
5804 | ||
bf5eb3de TH |
5805 | void sysfs_slab_release(struct kmem_cache *s) |
5806 | { | |
5807 | if (slab_state >= FULL) | |
5808 | kobject_put(&s->kobj); | |
81819f0f CL |
5809 | } |
5810 | ||
5811 | /* | |
5812 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5813 | * available lest we lose that information. |
81819f0f CL |
5814 | */ |
5815 | struct saved_alias { | |
5816 | struct kmem_cache *s; | |
5817 | const char *name; | |
5818 | struct saved_alias *next; | |
5819 | }; | |
5820 | ||
5af328a5 | 5821 | static struct saved_alias *alias_list; |
81819f0f CL |
5822 | |
5823 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5824 | { | |
5825 | struct saved_alias *al; | |
5826 | ||
97d06609 | 5827 | if (slab_state == FULL) { |
81819f0f CL |
5828 | /* |
5829 | * If we have a leftover link then remove it. | |
5830 | */ | |
27c3a314 GKH |
5831 | sysfs_remove_link(&slab_kset->kobj, name); |
5832 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5833 | } |
5834 | ||
5835 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5836 | if (!al) | |
5837 | return -ENOMEM; | |
5838 | ||
5839 | al->s = s; | |
5840 | al->name = name; | |
5841 | al->next = alias_list; | |
5842 | alias_list = al; | |
5843 | return 0; | |
5844 | } | |
5845 | ||
5846 | static int __init slab_sysfs_init(void) | |
5847 | { | |
5b95a4ac | 5848 | struct kmem_cache *s; |
81819f0f CL |
5849 | int err; |
5850 | ||
18004c5d | 5851 | mutex_lock(&slab_mutex); |
2bce6485 | 5852 | |
0ff21e46 | 5853 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5854 | if (!slab_kset) { |
18004c5d | 5855 | mutex_unlock(&slab_mutex); |
f9f58285 | 5856 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5857 | return -ENOSYS; |
5858 | } | |
5859 | ||
97d06609 | 5860 | slab_state = FULL; |
26a7bd03 | 5861 | |
5b95a4ac | 5862 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5863 | err = sysfs_slab_add(s); |
5d540fb7 | 5864 | if (err) |
f9f58285 FF |
5865 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5866 | s->name); | |
26a7bd03 | 5867 | } |
81819f0f CL |
5868 | |
5869 | while (alias_list) { | |
5870 | struct saved_alias *al = alias_list; | |
5871 | ||
5872 | alias_list = alias_list->next; | |
5873 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5874 | if (err) |
f9f58285 FF |
5875 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5876 | al->name); | |
81819f0f CL |
5877 | kfree(al); |
5878 | } | |
5879 | ||
18004c5d | 5880 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5881 | resiliency_test(); |
5882 | return 0; | |
5883 | } | |
5884 | ||
5885 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5886 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5887 | |
5888 | /* | |
5889 | * The /proc/slabinfo ABI | |
5890 | */ | |
5b365771 | 5891 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 5892 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5893 | { |
57ed3eda | 5894 | unsigned long nr_slabs = 0; |
205ab99d CL |
5895 | unsigned long nr_objs = 0; |
5896 | unsigned long nr_free = 0; | |
57ed3eda | 5897 | int node; |
fa45dc25 | 5898 | struct kmem_cache_node *n; |
57ed3eda | 5899 | |
fa45dc25 | 5900 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5901 | nr_slabs += node_nr_slabs(n); |
5902 | nr_objs += node_nr_objs(n); | |
205ab99d | 5903 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5904 | } |
5905 | ||
0d7561c6 GC |
5906 | sinfo->active_objs = nr_objs - nr_free; |
5907 | sinfo->num_objs = nr_objs; | |
5908 | sinfo->active_slabs = nr_slabs; | |
5909 | sinfo->num_slabs = nr_slabs; | |
5910 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5911 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5912 | } |
5913 | ||
0d7561c6 | 5914 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5915 | { |
7b3c3a50 AD |
5916 | } |
5917 | ||
b7454ad3 GC |
5918 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5919 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5920 | { |
b7454ad3 | 5921 | return -EIO; |
7b3c3a50 | 5922 | } |
5b365771 | 5923 | #endif /* CONFIG_SLUB_DEBUG */ |