]>
Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
881db7fb CL |
5 | * The allocator synchronizes using per slab locks or atomic operatios |
6 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 7 | * |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 9 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
10 | */ |
11 | ||
12 | #include <linux/mm.h> | |
1eb5ac64 | 13 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
14 | #include <linux/module.h> |
15 | #include <linux/bit_spinlock.h> | |
16 | #include <linux/interrupt.h> | |
17 | #include <linux/bitops.h> | |
18 | #include <linux/slab.h> | |
97d06609 | 19 | #include "slab.h" |
7b3c3a50 | 20 | #include <linux/proc_fs.h> |
3ac38faa | 21 | #include <linux/notifier.h> |
81819f0f | 22 | #include <linux/seq_file.h> |
5a896d9e | 23 | #include <linux/kmemcheck.h> |
81819f0f CL |
24 | #include <linux/cpu.h> |
25 | #include <linux/cpuset.h> | |
26 | #include <linux/mempolicy.h> | |
27 | #include <linux/ctype.h> | |
3ac7fe5a | 28 | #include <linux/debugobjects.h> |
81819f0f | 29 | #include <linux/kallsyms.h> |
b9049e23 | 30 | #include <linux/memory.h> |
f8bd2258 | 31 | #include <linux/math64.h> |
773ff60e | 32 | #include <linux/fault-inject.h> |
bfa71457 | 33 | #include <linux/stacktrace.h> |
4de900b4 | 34 | #include <linux/prefetch.h> |
2633d7a0 | 35 | #include <linux/memcontrol.h> |
81819f0f | 36 | |
4a92379b RK |
37 | #include <trace/events/kmem.h> |
38 | ||
072bb0aa MG |
39 | #include "internal.h" |
40 | ||
81819f0f CL |
41 | /* |
42 | * Lock order: | |
18004c5d | 43 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
44 | * 2. node->list_lock |
45 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 46 | * |
18004c5d | 47 | * slab_mutex |
881db7fb | 48 | * |
18004c5d | 49 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
50 | * and to synchronize major metadata changes to slab cache structures. |
51 | * | |
52 | * The slab_lock is only used for debugging and on arches that do not | |
53 | * have the ability to do a cmpxchg_double. It only protects the second | |
54 | * double word in the page struct. Meaning | |
55 | * A. page->freelist -> List of object free in a page | |
56 | * B. page->counters -> Counters of objects | |
57 | * C. page->frozen -> frozen state | |
58 | * | |
59 | * If a slab is frozen then it is exempt from list management. It is not | |
60 | * on any list. The processor that froze the slab is the one who can | |
61 | * perform list operations on the page. Other processors may put objects | |
62 | * onto the freelist but the processor that froze the slab is the only | |
63 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
64 | * |
65 | * The list_lock protects the partial and full list on each node and | |
66 | * the partial slab counter. If taken then no new slabs may be added or | |
67 | * removed from the lists nor make the number of partial slabs be modified. | |
68 | * (Note that the total number of slabs is an atomic value that may be | |
69 | * modified without taking the list lock). | |
70 | * | |
71 | * The list_lock is a centralized lock and thus we avoid taking it as | |
72 | * much as possible. As long as SLUB does not have to handle partial | |
73 | * slabs, operations can continue without any centralized lock. F.e. | |
74 | * allocating a long series of objects that fill up slabs does not require | |
75 | * the list lock. | |
81819f0f CL |
76 | * Interrupts are disabled during allocation and deallocation in order to |
77 | * make the slab allocator safe to use in the context of an irq. In addition | |
78 | * interrupts are disabled to ensure that the processor does not change | |
79 | * while handling per_cpu slabs, due to kernel preemption. | |
80 | * | |
81 | * SLUB assigns one slab for allocation to each processor. | |
82 | * Allocations only occur from these slabs called cpu slabs. | |
83 | * | |
672bba3a CL |
84 | * Slabs with free elements are kept on a partial list and during regular |
85 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 86 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
87 | * We track full slabs for debugging purposes though because otherwise we |
88 | * cannot scan all objects. | |
81819f0f CL |
89 | * |
90 | * Slabs are freed when they become empty. Teardown and setup is | |
91 | * minimal so we rely on the page allocators per cpu caches for | |
92 | * fast frees and allocs. | |
93 | * | |
94 | * Overloading of page flags that are otherwise used for LRU management. | |
95 | * | |
4b6f0750 CL |
96 | * PageActive The slab is frozen and exempt from list processing. |
97 | * This means that the slab is dedicated to a purpose | |
98 | * such as satisfying allocations for a specific | |
99 | * processor. Objects may be freed in the slab while | |
100 | * it is frozen but slab_free will then skip the usual | |
101 | * list operations. It is up to the processor holding | |
102 | * the slab to integrate the slab into the slab lists | |
103 | * when the slab is no longer needed. | |
104 | * | |
105 | * One use of this flag is to mark slabs that are | |
106 | * used for allocations. Then such a slab becomes a cpu | |
107 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 108 | * freelist that allows lockless access to |
894b8788 CL |
109 | * free objects in addition to the regular freelist |
110 | * that requires the slab lock. | |
81819f0f CL |
111 | * |
112 | * PageError Slab requires special handling due to debug | |
113 | * options set. This moves slab handling out of | |
894b8788 | 114 | * the fast path and disables lockless freelists. |
81819f0f CL |
115 | */ |
116 | ||
af537b0a CL |
117 | static inline int kmem_cache_debug(struct kmem_cache *s) |
118 | { | |
5577bd8a | 119 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 120 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 121 | #else |
af537b0a | 122 | return 0; |
5577bd8a | 123 | #endif |
af537b0a | 124 | } |
5577bd8a | 125 | |
345c905d JK |
126 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
127 | { | |
128 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
129 | return !kmem_cache_debug(s); | |
130 | #else | |
131 | return false; | |
132 | #endif | |
133 | } | |
134 | ||
81819f0f CL |
135 | /* |
136 | * Issues still to be resolved: | |
137 | * | |
81819f0f CL |
138 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
139 | * | |
81819f0f CL |
140 | * - Variable sizing of the per node arrays |
141 | */ | |
142 | ||
143 | /* Enable to test recovery from slab corruption on boot */ | |
144 | #undef SLUB_RESILIENCY_TEST | |
145 | ||
b789ef51 CL |
146 | /* Enable to log cmpxchg failures */ |
147 | #undef SLUB_DEBUG_CMPXCHG | |
148 | ||
2086d26a CL |
149 | /* |
150 | * Mininum number of partial slabs. These will be left on the partial | |
151 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
152 | */ | |
76be8950 | 153 | #define MIN_PARTIAL 5 |
e95eed57 | 154 | |
2086d26a CL |
155 | /* |
156 | * Maximum number of desirable partial slabs. | |
157 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 158 | * sort the partial list by the number of objects in use. |
2086d26a CL |
159 | */ |
160 | #define MAX_PARTIAL 10 | |
161 | ||
81819f0f CL |
162 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
163 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 164 | |
fa5ec8a1 | 165 | /* |
3de47213 DR |
166 | * Debugging flags that require metadata to be stored in the slab. These get |
167 | * disabled when slub_debug=O is used and a cache's min order increases with | |
168 | * metadata. | |
fa5ec8a1 | 169 | */ |
3de47213 | 170 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 171 | |
210b5c06 CG |
172 | #define OO_SHIFT 16 |
173 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 174 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 175 | |
81819f0f | 176 | /* Internal SLUB flags */ |
f90ec390 | 177 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 178 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 179 | |
81819f0f CL |
180 | #ifdef CONFIG_SMP |
181 | static struct notifier_block slab_notifier; | |
182 | #endif | |
183 | ||
02cbc874 CL |
184 | /* |
185 | * Tracking user of a slab. | |
186 | */ | |
d6543e39 | 187 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 188 | struct track { |
ce71e27c | 189 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
190 | #ifdef CONFIG_STACKTRACE |
191 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
192 | #endif | |
02cbc874 CL |
193 | int cpu; /* Was running on cpu */ |
194 | int pid; /* Pid context */ | |
195 | unsigned long when; /* When did the operation occur */ | |
196 | }; | |
197 | ||
198 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
199 | ||
ab4d5ed5 | 200 | #ifdef CONFIG_SYSFS |
81819f0f CL |
201 | static int sysfs_slab_add(struct kmem_cache *); |
202 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 203 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
81819f0f | 204 | #else |
0c710013 CL |
205 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
206 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
207 | { return 0; } | |
107dab5c | 208 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
209 | #endif |
210 | ||
4fdccdfb | 211 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
212 | { |
213 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
214 | /* |
215 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
216 | * avoid this_cpu_add()'s irq-disable overhead. | |
217 | */ | |
218 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
219 | #endif |
220 | } | |
221 | ||
81819f0f CL |
222 | /******************************************************************** |
223 | * Core slab cache functions | |
224 | *******************************************************************/ | |
225 | ||
6446faa2 | 226 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
227 | static inline int check_valid_pointer(struct kmem_cache *s, |
228 | struct page *page, const void *object) | |
229 | { | |
230 | void *base; | |
231 | ||
a973e9dd | 232 | if (!object) |
02cbc874 CL |
233 | return 1; |
234 | ||
a973e9dd | 235 | base = page_address(page); |
39b26464 | 236 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
237 | (object - base) % s->size) { |
238 | return 0; | |
239 | } | |
240 | ||
241 | return 1; | |
242 | } | |
243 | ||
7656c72b CL |
244 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
245 | { | |
246 | return *(void **)(object + s->offset); | |
247 | } | |
248 | ||
0ad9500e ED |
249 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
250 | { | |
251 | prefetch(object + s->offset); | |
252 | } | |
253 | ||
1393d9a1 CL |
254 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
255 | { | |
256 | void *p; | |
257 | ||
258 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
259 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
260 | #else | |
261 | p = get_freepointer(s, object); | |
262 | #endif | |
263 | return p; | |
264 | } | |
265 | ||
7656c72b CL |
266 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
267 | { | |
268 | *(void **)(object + s->offset) = fp; | |
269 | } | |
270 | ||
271 | /* Loop over all objects in a slab */ | |
224a88be CL |
272 | #define for_each_object(__p, __s, __addr, __objects) \ |
273 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
274 | __p += (__s)->size) |
275 | ||
54266640 WY |
276 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
277 | for (__p = (__addr), __idx = 1; __idx <= __objects;\ | |
278 | __p += (__s)->size, __idx++) | |
279 | ||
7656c72b CL |
280 | /* Determine object index from a given position */ |
281 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
282 | { | |
283 | return (p - addr) / s->size; | |
284 | } | |
285 | ||
d71f606f MK |
286 | static inline size_t slab_ksize(const struct kmem_cache *s) |
287 | { | |
288 | #ifdef CONFIG_SLUB_DEBUG | |
289 | /* | |
290 | * Debugging requires use of the padding between object | |
291 | * and whatever may come after it. | |
292 | */ | |
293 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
3b0efdfa | 294 | return s->object_size; |
d71f606f MK |
295 | |
296 | #endif | |
297 | /* | |
298 | * If we have the need to store the freelist pointer | |
299 | * back there or track user information then we can | |
300 | * only use the space before that information. | |
301 | */ | |
302 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
303 | return s->inuse; | |
304 | /* | |
305 | * Else we can use all the padding etc for the allocation | |
306 | */ | |
307 | return s->size; | |
308 | } | |
309 | ||
ab9a0f19 LJ |
310 | static inline int order_objects(int order, unsigned long size, int reserved) |
311 | { | |
312 | return ((PAGE_SIZE << order) - reserved) / size; | |
313 | } | |
314 | ||
834f3d11 | 315 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 316 | unsigned long size, int reserved) |
834f3d11 CL |
317 | { |
318 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 319 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
320 | }; |
321 | ||
322 | return x; | |
323 | } | |
324 | ||
325 | static inline int oo_order(struct kmem_cache_order_objects x) | |
326 | { | |
210b5c06 | 327 | return x.x >> OO_SHIFT; |
834f3d11 CL |
328 | } |
329 | ||
330 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
331 | { | |
210b5c06 | 332 | return x.x & OO_MASK; |
834f3d11 CL |
333 | } |
334 | ||
881db7fb CL |
335 | /* |
336 | * Per slab locking using the pagelock | |
337 | */ | |
338 | static __always_inline void slab_lock(struct page *page) | |
339 | { | |
340 | bit_spin_lock(PG_locked, &page->flags); | |
341 | } | |
342 | ||
343 | static __always_inline void slab_unlock(struct page *page) | |
344 | { | |
345 | __bit_spin_unlock(PG_locked, &page->flags); | |
346 | } | |
347 | ||
a0320865 DH |
348 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
349 | { | |
350 | struct page tmp; | |
351 | tmp.counters = counters_new; | |
352 | /* | |
353 | * page->counters can cover frozen/inuse/objects as well | |
354 | * as page->_count. If we assign to ->counters directly | |
355 | * we run the risk of losing updates to page->_count, so | |
356 | * be careful and only assign to the fields we need. | |
357 | */ | |
358 | page->frozen = tmp.frozen; | |
359 | page->inuse = tmp.inuse; | |
360 | page->objects = tmp.objects; | |
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)) | |
376 | return 1; | |
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; |
a0320865 | 384 | set_page_slub_counters(page, counters_new); |
1d07171c CL |
385 | slab_unlock(page); |
386 | return 1; | |
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 | ||
398 | return 0; | |
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)) | |
412 | return 1; | |
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; |
a0320865 | 423 | set_page_slub_counters(page, counters_new); |
881db7fb | 424 | slab_unlock(page); |
1d07171c | 425 | local_irq_restore(flags); |
b789ef51 CL |
426 | return 1; |
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 | ||
439 | return 0; | |
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 | ||
41ecc55b CL |
458 | /* |
459 | * Debug settings: | |
460 | */ | |
f0630fff CL |
461 | #ifdef CONFIG_SLUB_DEBUG_ON |
462 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
463 | #else | |
41ecc55b | 464 | static int slub_debug; |
f0630fff | 465 | #endif |
41ecc55b CL |
466 | |
467 | static char *slub_debug_slabs; | |
fa5ec8a1 | 468 | static int disable_higher_order_debug; |
41ecc55b | 469 | |
81819f0f CL |
470 | /* |
471 | * Object debugging | |
472 | */ | |
473 | static void print_section(char *text, u8 *addr, unsigned int length) | |
474 | { | |
ffc79d28 SAS |
475 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
476 | length, 1); | |
81819f0f CL |
477 | } |
478 | ||
81819f0f CL |
479 | static struct track *get_track(struct kmem_cache *s, void *object, |
480 | enum track_item alloc) | |
481 | { | |
482 | struct track *p; | |
483 | ||
484 | if (s->offset) | |
485 | p = object + s->offset + sizeof(void *); | |
486 | else | |
487 | p = object + s->inuse; | |
488 | ||
489 | return p + alloc; | |
490 | } | |
491 | ||
492 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 493 | enum track_item alloc, unsigned long addr) |
81819f0f | 494 | { |
1a00df4a | 495 | struct track *p = get_track(s, object, alloc); |
81819f0f | 496 | |
81819f0f | 497 | if (addr) { |
d6543e39 BG |
498 | #ifdef CONFIG_STACKTRACE |
499 | struct stack_trace trace; | |
500 | int i; | |
501 | ||
502 | trace.nr_entries = 0; | |
503 | trace.max_entries = TRACK_ADDRS_COUNT; | |
504 | trace.entries = p->addrs; | |
505 | trace.skip = 3; | |
506 | save_stack_trace(&trace); | |
507 | ||
508 | /* See rant in lockdep.c */ | |
509 | if (trace.nr_entries != 0 && | |
510 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
511 | trace.nr_entries--; | |
512 | ||
513 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
514 | p->addrs[i] = 0; | |
515 | #endif | |
81819f0f CL |
516 | p->addr = addr; |
517 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 518 | p->pid = current->pid; |
81819f0f CL |
519 | p->when = jiffies; |
520 | } else | |
521 | memset(p, 0, sizeof(struct track)); | |
522 | } | |
523 | ||
81819f0f CL |
524 | static void init_tracking(struct kmem_cache *s, void *object) |
525 | { | |
24922684 CL |
526 | if (!(s->flags & SLAB_STORE_USER)) |
527 | return; | |
528 | ||
ce71e27c EGM |
529 | set_track(s, object, TRACK_FREE, 0UL); |
530 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
531 | } |
532 | ||
533 | static void print_track(const char *s, struct track *t) | |
534 | { | |
535 | if (!t->addr) | |
536 | return; | |
537 | ||
f9f58285 FF |
538 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
539 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
540 | #ifdef CONFIG_STACKTRACE |
541 | { | |
542 | int i; | |
543 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
544 | if (t->addrs[i]) | |
f9f58285 | 545 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
546 | else |
547 | break; | |
548 | } | |
549 | #endif | |
24922684 CL |
550 | } |
551 | ||
552 | static void print_tracking(struct kmem_cache *s, void *object) | |
553 | { | |
554 | if (!(s->flags & SLAB_STORE_USER)) | |
555 | return; | |
556 | ||
557 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
558 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
559 | } | |
560 | ||
561 | static void print_page_info(struct page *page) | |
562 | { | |
f9f58285 | 563 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 564 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
565 | |
566 | } | |
567 | ||
568 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
569 | { | |
ecc42fbe | 570 | struct va_format vaf; |
24922684 | 571 | va_list args; |
24922684 CL |
572 | |
573 | va_start(args, fmt); | |
ecc42fbe FF |
574 | vaf.fmt = fmt; |
575 | vaf.va = &args; | |
f9f58285 | 576 | pr_err("=============================================================================\n"); |
ecc42fbe | 577 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 578 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 579 | |
373d4d09 | 580 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 581 | va_end(args); |
81819f0f CL |
582 | } |
583 | ||
24922684 CL |
584 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
585 | { | |
ecc42fbe | 586 | struct va_format vaf; |
24922684 | 587 | va_list args; |
24922684 CL |
588 | |
589 | va_start(args, fmt); | |
ecc42fbe FF |
590 | vaf.fmt = fmt; |
591 | vaf.va = &args; | |
592 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 593 | va_end(args); |
24922684 CL |
594 | } |
595 | ||
596 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
597 | { |
598 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 599 | u8 *addr = page_address(page); |
24922684 CL |
600 | |
601 | print_tracking(s, p); | |
602 | ||
603 | print_page_info(page); | |
604 | ||
f9f58285 FF |
605 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
606 | p, p - addr, get_freepointer(s, p)); | |
24922684 CL |
607 | |
608 | if (p > addr + 16) | |
ffc79d28 | 609 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 610 | |
3b0efdfa | 611 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 612 | PAGE_SIZE)); |
81819f0f | 613 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
614 | print_section("Redzone ", p + s->object_size, |
615 | s->inuse - s->object_size); | |
81819f0f | 616 | |
81819f0f CL |
617 | if (s->offset) |
618 | off = s->offset + sizeof(void *); | |
619 | else | |
620 | off = s->inuse; | |
621 | ||
24922684 | 622 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 623 | off += 2 * sizeof(struct track); |
81819f0f CL |
624 | |
625 | if (off != s->size) | |
626 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 627 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
628 | |
629 | dump_stack(); | |
81819f0f CL |
630 | } |
631 | ||
632 | static void object_err(struct kmem_cache *s, struct page *page, | |
633 | u8 *object, char *reason) | |
634 | { | |
3dc50637 | 635 | slab_bug(s, "%s", reason); |
24922684 | 636 | print_trailer(s, page, object); |
81819f0f CL |
637 | } |
638 | ||
d0e0ac97 CG |
639 | static void slab_err(struct kmem_cache *s, struct page *page, |
640 | const char *fmt, ...) | |
81819f0f CL |
641 | { |
642 | va_list args; | |
643 | char buf[100]; | |
644 | ||
24922684 CL |
645 | va_start(args, fmt); |
646 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 647 | va_end(args); |
3dc50637 | 648 | slab_bug(s, "%s", buf); |
24922684 | 649 | print_page_info(page); |
81819f0f CL |
650 | dump_stack(); |
651 | } | |
652 | ||
f7cb1933 | 653 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
654 | { |
655 | u8 *p = object; | |
656 | ||
657 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
658 | memset(p, POISON_FREE, s->object_size - 1); |
659 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
660 | } |
661 | ||
662 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 663 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
664 | } |
665 | ||
24922684 CL |
666 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
667 | void *from, void *to) | |
668 | { | |
669 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
670 | memset(from, data, to - from); | |
671 | } | |
672 | ||
673 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
674 | u8 *object, char *what, | |
06428780 | 675 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
676 | { |
677 | u8 *fault; | |
678 | u8 *end; | |
679 | ||
79824820 | 680 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
681 | if (!fault) |
682 | return 1; | |
683 | ||
684 | end = start + bytes; | |
685 | while (end > fault && end[-1] == value) | |
686 | end--; | |
687 | ||
688 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 689 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
690 | fault, end - 1, fault[0], value); |
691 | print_trailer(s, page, object); | |
692 | ||
693 | restore_bytes(s, what, value, fault, end); | |
694 | return 0; | |
81819f0f CL |
695 | } |
696 | ||
81819f0f CL |
697 | /* |
698 | * Object layout: | |
699 | * | |
700 | * object address | |
701 | * Bytes of the object to be managed. | |
702 | * If the freepointer may overlay the object then the free | |
703 | * pointer is the first word of the object. | |
672bba3a | 704 | * |
81819f0f CL |
705 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
706 | * 0xa5 (POISON_END) | |
707 | * | |
3b0efdfa | 708 | * object + s->object_size |
81819f0f | 709 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 710 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 711 | * object_size == inuse. |
672bba3a | 712 | * |
81819f0f CL |
713 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
714 | * 0xcc (RED_ACTIVE) for objects in use. | |
715 | * | |
716 | * object + s->inuse | |
672bba3a CL |
717 | * Meta data starts here. |
718 | * | |
81819f0f CL |
719 | * A. Free pointer (if we cannot overwrite object on free) |
720 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 721 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 722 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
723 | * before the word boundary. |
724 | * | |
725 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
726 | * |
727 | * object + s->size | |
672bba3a | 728 | * Nothing is used beyond s->size. |
81819f0f | 729 | * |
3b0efdfa | 730 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 731 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
732 | * may be used with merged slabcaches. |
733 | */ | |
734 | ||
81819f0f CL |
735 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
736 | { | |
737 | unsigned long off = s->inuse; /* The end of info */ | |
738 | ||
739 | if (s->offset) | |
740 | /* Freepointer is placed after the object. */ | |
741 | off += sizeof(void *); | |
742 | ||
743 | if (s->flags & SLAB_STORE_USER) | |
744 | /* We also have user information there */ | |
745 | off += 2 * sizeof(struct track); | |
746 | ||
747 | if (s->size == off) | |
748 | return 1; | |
749 | ||
24922684 CL |
750 | return check_bytes_and_report(s, page, p, "Object padding", |
751 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
752 | } |
753 | ||
39b26464 | 754 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
755 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
756 | { | |
24922684 CL |
757 | u8 *start; |
758 | u8 *fault; | |
759 | u8 *end; | |
760 | int length; | |
761 | int remainder; | |
81819f0f CL |
762 | |
763 | if (!(s->flags & SLAB_POISON)) | |
764 | return 1; | |
765 | ||
a973e9dd | 766 | start = page_address(page); |
ab9a0f19 | 767 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
768 | end = start + length; |
769 | remainder = length % s->size; | |
81819f0f CL |
770 | if (!remainder) |
771 | return 1; | |
772 | ||
79824820 | 773 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
774 | if (!fault) |
775 | return 1; | |
776 | while (end > fault && end[-1] == POISON_INUSE) | |
777 | end--; | |
778 | ||
779 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 780 | print_section("Padding ", end - remainder, remainder); |
24922684 | 781 | |
8a3d271d | 782 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 783 | return 0; |
81819f0f CL |
784 | } |
785 | ||
786 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 787 | void *object, u8 val) |
81819f0f CL |
788 | { |
789 | u8 *p = object; | |
3b0efdfa | 790 | u8 *endobject = object + s->object_size; |
81819f0f CL |
791 | |
792 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 793 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 794 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 795 | return 0; |
81819f0f | 796 | } else { |
3b0efdfa | 797 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 798 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
799 | endobject, POISON_INUSE, |
800 | s->inuse - s->object_size); | |
3adbefee | 801 | } |
81819f0f CL |
802 | } |
803 | ||
804 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 805 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 806 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 807 | POISON_FREE, s->object_size - 1) || |
24922684 | 808 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 809 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 810 | return 0; |
81819f0f CL |
811 | /* |
812 | * check_pad_bytes cleans up on its own. | |
813 | */ | |
814 | check_pad_bytes(s, page, p); | |
815 | } | |
816 | ||
f7cb1933 | 817 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
818 | /* |
819 | * Object and freepointer overlap. Cannot check | |
820 | * freepointer while object is allocated. | |
821 | */ | |
822 | return 1; | |
823 | ||
824 | /* Check free pointer validity */ | |
825 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
826 | object_err(s, page, p, "Freepointer corrupt"); | |
827 | /* | |
9f6c708e | 828 | * No choice but to zap it and thus lose the remainder |
81819f0f | 829 | * of the free objects in this slab. May cause |
672bba3a | 830 | * another error because the object count is now wrong. |
81819f0f | 831 | */ |
a973e9dd | 832 | set_freepointer(s, p, NULL); |
81819f0f CL |
833 | return 0; |
834 | } | |
835 | return 1; | |
836 | } | |
837 | ||
838 | static int check_slab(struct kmem_cache *s, struct page *page) | |
839 | { | |
39b26464 CL |
840 | int maxobj; |
841 | ||
81819f0f CL |
842 | VM_BUG_ON(!irqs_disabled()); |
843 | ||
844 | if (!PageSlab(page)) { | |
24922684 | 845 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
846 | return 0; |
847 | } | |
39b26464 | 848 | |
ab9a0f19 | 849 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
850 | if (page->objects > maxobj) { |
851 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 852 | page->objects, maxobj); |
39b26464 CL |
853 | return 0; |
854 | } | |
855 | if (page->inuse > page->objects) { | |
24922684 | 856 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 857 | page->inuse, page->objects); |
81819f0f CL |
858 | return 0; |
859 | } | |
860 | /* Slab_pad_check fixes things up after itself */ | |
861 | slab_pad_check(s, page); | |
862 | return 1; | |
863 | } | |
864 | ||
865 | /* | |
672bba3a CL |
866 | * Determine if a certain object on a page is on the freelist. Must hold the |
867 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
868 | */ |
869 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
870 | { | |
871 | int nr = 0; | |
881db7fb | 872 | void *fp; |
81819f0f | 873 | void *object = NULL; |
f6edde9c | 874 | int max_objects; |
81819f0f | 875 | |
881db7fb | 876 | fp = page->freelist; |
39b26464 | 877 | while (fp && nr <= page->objects) { |
81819f0f CL |
878 | if (fp == search) |
879 | return 1; | |
880 | if (!check_valid_pointer(s, page, fp)) { | |
881 | if (object) { | |
882 | object_err(s, page, object, | |
883 | "Freechain corrupt"); | |
a973e9dd | 884 | set_freepointer(s, object, NULL); |
81819f0f | 885 | } else { |
24922684 | 886 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 887 | page->freelist = NULL; |
39b26464 | 888 | page->inuse = page->objects; |
24922684 | 889 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
890 | return 0; |
891 | } | |
892 | break; | |
893 | } | |
894 | object = fp; | |
895 | fp = get_freepointer(s, object); | |
896 | nr++; | |
897 | } | |
898 | ||
ab9a0f19 | 899 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
900 | if (max_objects > MAX_OBJS_PER_PAGE) |
901 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
902 | |
903 | if (page->objects != max_objects) { | |
904 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
905 | "should be %d", page->objects, max_objects); | |
906 | page->objects = max_objects; | |
907 | slab_fix(s, "Number of objects adjusted."); | |
908 | } | |
39b26464 | 909 | if (page->inuse != page->objects - nr) { |
70d71228 | 910 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
911 | "counted were %d", page->inuse, page->objects - nr); |
912 | page->inuse = page->objects - nr; | |
24922684 | 913 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
914 | } |
915 | return search == NULL; | |
916 | } | |
917 | ||
0121c619 CL |
918 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
919 | int alloc) | |
3ec09742 CL |
920 | { |
921 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 922 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
923 | s->name, |
924 | alloc ? "alloc" : "free", | |
925 | object, page->inuse, | |
926 | page->freelist); | |
927 | ||
928 | if (!alloc) | |
d0e0ac97 CG |
929 | print_section("Object ", (void *)object, |
930 | s->object_size); | |
3ec09742 CL |
931 | |
932 | dump_stack(); | |
933 | } | |
934 | } | |
935 | ||
643b1138 | 936 | /* |
672bba3a | 937 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 938 | */ |
5cc6eee8 CL |
939 | static void add_full(struct kmem_cache *s, |
940 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 941 | { |
5cc6eee8 CL |
942 | if (!(s->flags & SLAB_STORE_USER)) |
943 | return; | |
944 | ||
255d0884 | 945 | lockdep_assert_held(&n->list_lock); |
643b1138 | 946 | list_add(&page->lru, &n->full); |
643b1138 CL |
947 | } |
948 | ||
c65c1877 | 949 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 950 | { |
643b1138 CL |
951 | if (!(s->flags & SLAB_STORE_USER)) |
952 | return; | |
953 | ||
255d0884 | 954 | lockdep_assert_held(&n->list_lock); |
643b1138 | 955 | list_del(&page->lru); |
643b1138 CL |
956 | } |
957 | ||
0f389ec6 CL |
958 | /* Tracking of the number of slabs for debugging purposes */ |
959 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
960 | { | |
961 | struct kmem_cache_node *n = get_node(s, node); | |
962 | ||
963 | return atomic_long_read(&n->nr_slabs); | |
964 | } | |
965 | ||
26c02cf0 AB |
966 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
967 | { | |
968 | return atomic_long_read(&n->nr_slabs); | |
969 | } | |
970 | ||
205ab99d | 971 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
972 | { |
973 | struct kmem_cache_node *n = get_node(s, node); | |
974 | ||
975 | /* | |
976 | * May be called early in order to allocate a slab for the | |
977 | * kmem_cache_node structure. Solve the chicken-egg | |
978 | * dilemma by deferring the increment of the count during | |
979 | * bootstrap (see early_kmem_cache_node_alloc). | |
980 | */ | |
338b2642 | 981 | if (likely(n)) { |
0f389ec6 | 982 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
983 | atomic_long_add(objects, &n->total_objects); |
984 | } | |
0f389ec6 | 985 | } |
205ab99d | 986 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
987 | { |
988 | struct kmem_cache_node *n = get_node(s, node); | |
989 | ||
990 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 991 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
992 | } |
993 | ||
994 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
995 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
996 | void *object) | |
997 | { | |
998 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
999 | return; | |
1000 | ||
f7cb1933 | 1001 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1002 | init_tracking(s, object); |
1003 | } | |
1004 | ||
d0e0ac97 CG |
1005 | static noinline int alloc_debug_processing(struct kmem_cache *s, |
1006 | struct page *page, | |
ce71e27c | 1007 | void *object, unsigned long addr) |
81819f0f CL |
1008 | { |
1009 | if (!check_slab(s, page)) | |
1010 | goto bad; | |
1011 | ||
81819f0f CL |
1012 | if (!check_valid_pointer(s, page, object)) { |
1013 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1014 | goto bad; |
81819f0f CL |
1015 | } |
1016 | ||
f7cb1933 | 1017 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1018 | goto bad; |
81819f0f | 1019 | |
3ec09742 CL |
1020 | /* Success perform special debug activities for allocs */ |
1021 | if (s->flags & SLAB_STORE_USER) | |
1022 | set_track(s, object, TRACK_ALLOC, addr); | |
1023 | trace(s, page, object, 1); | |
f7cb1933 | 1024 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1025 | return 1; |
3ec09742 | 1026 | |
81819f0f CL |
1027 | bad: |
1028 | if (PageSlab(page)) { | |
1029 | /* | |
1030 | * If this is a slab page then lets do the best we can | |
1031 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1032 | * as used avoids touching the remaining objects. |
81819f0f | 1033 | */ |
24922684 | 1034 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1035 | page->inuse = page->objects; |
a973e9dd | 1036 | page->freelist = NULL; |
81819f0f CL |
1037 | } |
1038 | return 0; | |
1039 | } | |
1040 | ||
19c7ff9e CL |
1041 | static noinline struct kmem_cache_node *free_debug_processing( |
1042 | struct kmem_cache *s, struct page *page, void *object, | |
1043 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1044 | { |
19c7ff9e | 1045 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1046 | |
19c7ff9e | 1047 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1048 | slab_lock(page); |
1049 | ||
81819f0f CL |
1050 | if (!check_slab(s, page)) |
1051 | goto fail; | |
1052 | ||
1053 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1054 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1055 | goto fail; |
1056 | } | |
1057 | ||
1058 | if (on_freelist(s, page, object)) { | |
24922684 | 1059 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1060 | goto fail; |
1061 | } | |
1062 | ||
f7cb1933 | 1063 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1064 | goto out; |
81819f0f | 1065 | |
1b4f59e3 | 1066 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1067 | if (!PageSlab(page)) { |
70d71228 CL |
1068 | slab_err(s, page, "Attempt to free object(0x%p) " |
1069 | "outside of slab", object); | |
1b4f59e3 | 1070 | } else if (!page->slab_cache) { |
f9f58285 FF |
1071 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1072 | object); | |
70d71228 | 1073 | dump_stack(); |
06428780 | 1074 | } else |
24922684 CL |
1075 | object_err(s, page, object, |
1076 | "page slab pointer corrupt."); | |
81819f0f CL |
1077 | goto fail; |
1078 | } | |
3ec09742 | 1079 | |
3ec09742 CL |
1080 | if (s->flags & SLAB_STORE_USER) |
1081 | set_track(s, object, TRACK_FREE, addr); | |
1082 | trace(s, page, object, 0); | |
f7cb1933 | 1083 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1084 | out: |
881db7fb | 1085 | slab_unlock(page); |
19c7ff9e CL |
1086 | /* |
1087 | * Keep node_lock to preserve integrity | |
1088 | * until the object is actually freed | |
1089 | */ | |
1090 | return n; | |
3ec09742 | 1091 | |
81819f0f | 1092 | fail: |
19c7ff9e CL |
1093 | slab_unlock(page); |
1094 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1095 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1096 | return NULL; |
81819f0f CL |
1097 | } |
1098 | ||
41ecc55b CL |
1099 | static int __init setup_slub_debug(char *str) |
1100 | { | |
f0630fff CL |
1101 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1102 | if (*str++ != '=' || !*str) | |
1103 | /* | |
1104 | * No options specified. Switch on full debugging. | |
1105 | */ | |
1106 | goto out; | |
1107 | ||
1108 | if (*str == ',') | |
1109 | /* | |
1110 | * No options but restriction on slabs. This means full | |
1111 | * debugging for slabs matching a pattern. | |
1112 | */ | |
1113 | goto check_slabs; | |
1114 | ||
fa5ec8a1 DR |
1115 | if (tolower(*str) == 'o') { |
1116 | /* | |
1117 | * Avoid enabling debugging on caches if its minimum order | |
1118 | * would increase as a result. | |
1119 | */ | |
1120 | disable_higher_order_debug = 1; | |
1121 | goto out; | |
1122 | } | |
1123 | ||
f0630fff CL |
1124 | slub_debug = 0; |
1125 | if (*str == '-') | |
1126 | /* | |
1127 | * Switch off all debugging measures. | |
1128 | */ | |
1129 | goto out; | |
1130 | ||
1131 | /* | |
1132 | * Determine which debug features should be switched on | |
1133 | */ | |
06428780 | 1134 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1135 | switch (tolower(*str)) { |
1136 | case 'f': | |
1137 | slub_debug |= SLAB_DEBUG_FREE; | |
1138 | break; | |
1139 | case 'z': | |
1140 | slub_debug |= SLAB_RED_ZONE; | |
1141 | break; | |
1142 | case 'p': | |
1143 | slub_debug |= SLAB_POISON; | |
1144 | break; | |
1145 | case 'u': | |
1146 | slub_debug |= SLAB_STORE_USER; | |
1147 | break; | |
1148 | case 't': | |
1149 | slub_debug |= SLAB_TRACE; | |
1150 | break; | |
4c13dd3b DM |
1151 | case 'a': |
1152 | slub_debug |= SLAB_FAILSLAB; | |
1153 | break; | |
f0630fff | 1154 | default: |
f9f58285 FF |
1155 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1156 | *str); | |
f0630fff | 1157 | } |
41ecc55b CL |
1158 | } |
1159 | ||
f0630fff | 1160 | check_slabs: |
41ecc55b CL |
1161 | if (*str == ',') |
1162 | slub_debug_slabs = str + 1; | |
f0630fff | 1163 | out: |
41ecc55b CL |
1164 | return 1; |
1165 | } | |
1166 | ||
1167 | __setup("slub_debug", setup_slub_debug); | |
1168 | ||
423c929c | 1169 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1170 | unsigned long flags, const char *name, |
51cc5068 | 1171 | void (*ctor)(void *)) |
41ecc55b CL |
1172 | { |
1173 | /* | |
e153362a | 1174 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1175 | */ |
c6f58d9b CL |
1176 | if (slub_debug && (!slub_debug_slabs || (name && |
1177 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1178 | flags |= slub_debug; |
ba0268a8 CL |
1179 | |
1180 | return flags; | |
41ecc55b CL |
1181 | } |
1182 | #else | |
3ec09742 CL |
1183 | static inline void setup_object_debug(struct kmem_cache *s, |
1184 | struct page *page, void *object) {} | |
41ecc55b | 1185 | |
3ec09742 | 1186 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1187 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1188 | |
19c7ff9e CL |
1189 | static inline struct kmem_cache_node *free_debug_processing( |
1190 | struct kmem_cache *s, struct page *page, void *object, | |
1191 | unsigned long addr, unsigned long *flags) { return NULL; } | |
41ecc55b | 1192 | |
41ecc55b CL |
1193 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1194 | { return 1; } | |
1195 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1196 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1197 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1198 | struct page *page) {} | |
c65c1877 PZ |
1199 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1200 | struct page *page) {} | |
423c929c | 1201 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1202 | unsigned long flags, const char *name, |
51cc5068 | 1203 | void (*ctor)(void *)) |
ba0268a8 CL |
1204 | { |
1205 | return flags; | |
1206 | } | |
41ecc55b | 1207 | #define slub_debug 0 |
0f389ec6 | 1208 | |
fdaa45e9 IM |
1209 | #define disable_higher_order_debug 0 |
1210 | ||
0f389ec6 CL |
1211 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1212 | { return 0; } | |
26c02cf0 AB |
1213 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1214 | { return 0; } | |
205ab99d CL |
1215 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1216 | int objects) {} | |
1217 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1218 | int objects) {} | |
7d550c56 | 1219 | |
02e72cc6 AR |
1220 | #endif /* CONFIG_SLUB_DEBUG */ |
1221 | ||
1222 | /* | |
1223 | * Hooks for other subsystems that check memory allocations. In a typical | |
1224 | * production configuration these hooks all should produce no code at all. | |
1225 | */ | |
d56791b3 RB |
1226 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1227 | { | |
1228 | kmemleak_alloc(ptr, size, 1, flags); | |
1229 | } | |
1230 | ||
1231 | static inline void kfree_hook(const void *x) | |
1232 | { | |
1233 | kmemleak_free(x); | |
1234 | } | |
1235 | ||
7d550c56 | 1236 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) |
02e72cc6 AR |
1237 | { |
1238 | flags &= gfp_allowed_mask; | |
1239 | lockdep_trace_alloc(flags); | |
1240 | might_sleep_if(flags & __GFP_WAIT); | |
7d550c56 | 1241 | |
02e72cc6 AR |
1242 | return should_failslab(s->object_size, flags, s->flags); |
1243 | } | |
1244 | ||
1245 | static inline void slab_post_alloc_hook(struct kmem_cache *s, | |
1246 | gfp_t flags, void *object) | |
d56791b3 | 1247 | { |
02e72cc6 AR |
1248 | flags &= gfp_allowed_mask; |
1249 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); | |
1250 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); | |
d56791b3 | 1251 | } |
7d550c56 | 1252 | |
d56791b3 RB |
1253 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1254 | { | |
1255 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1256 | |
02e72cc6 AR |
1257 | /* |
1258 | * Trouble is that we may no longer disable interrupts in the fast path | |
1259 | * So in order to make the debug calls that expect irqs to be | |
1260 | * disabled we need to disable interrupts temporarily. | |
1261 | */ | |
1262 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1263 | { | |
1264 | unsigned long flags; | |
1265 | ||
1266 | local_irq_save(flags); | |
1267 | kmemcheck_slab_free(s, x, s->object_size); | |
1268 | debug_check_no_locks_freed(x, s->object_size); | |
1269 | local_irq_restore(flags); | |
1270 | } | |
1271 | #endif | |
1272 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1273 | debug_check_no_obj_freed(x, s->object_size); | |
1274 | } | |
205ab99d | 1275 | |
81819f0f CL |
1276 | /* |
1277 | * Slab allocation and freeing | |
1278 | */ | |
5dfb4175 VD |
1279 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1280 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1281 | { |
5dfb4175 | 1282 | struct page *page; |
65c3376a CL |
1283 | int order = oo_order(oo); |
1284 | ||
b1eeab67 VN |
1285 | flags |= __GFP_NOTRACK; |
1286 | ||
5dfb4175 VD |
1287 | if (memcg_charge_slab(s, flags, order)) |
1288 | return NULL; | |
1289 | ||
2154a336 | 1290 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1291 | page = alloc_pages(flags, order); |
65c3376a | 1292 | else |
5dfb4175 VD |
1293 | page = alloc_pages_exact_node(node, flags, order); |
1294 | ||
1295 | if (!page) | |
1296 | memcg_uncharge_slab(s, order); | |
1297 | ||
1298 | return page; | |
65c3376a CL |
1299 | } |
1300 | ||
81819f0f CL |
1301 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1302 | { | |
06428780 | 1303 | struct page *page; |
834f3d11 | 1304 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1305 | gfp_t alloc_gfp; |
81819f0f | 1306 | |
7e0528da CL |
1307 | flags &= gfp_allowed_mask; |
1308 | ||
1309 | if (flags & __GFP_WAIT) | |
1310 | local_irq_enable(); | |
1311 | ||
b7a49f0d | 1312 | flags |= s->allocflags; |
e12ba74d | 1313 | |
ba52270d PE |
1314 | /* |
1315 | * Let the initial higher-order allocation fail under memory pressure | |
1316 | * so we fall-back to the minimum order allocation. | |
1317 | */ | |
1318 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1319 | ||
5dfb4175 | 1320 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1321 | if (unlikely(!page)) { |
1322 | oo = s->min; | |
80c3a998 | 1323 | alloc_gfp = flags; |
65c3376a CL |
1324 | /* |
1325 | * Allocation may have failed due to fragmentation. | |
1326 | * Try a lower order alloc if possible | |
1327 | */ | |
5dfb4175 | 1328 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
81819f0f | 1329 | |
7e0528da CL |
1330 | if (page) |
1331 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1332 | } |
5a896d9e | 1333 | |
737b719e | 1334 | if (kmemcheck_enabled && page |
5086c389 | 1335 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1336 | int pages = 1 << oo_order(oo); |
1337 | ||
80c3a998 | 1338 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1339 | |
1340 | /* | |
1341 | * Objects from caches that have a constructor don't get | |
1342 | * cleared when they're allocated, so we need to do it here. | |
1343 | */ | |
1344 | if (s->ctor) | |
1345 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1346 | else | |
1347 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1348 | } |
1349 | ||
737b719e DR |
1350 | if (flags & __GFP_WAIT) |
1351 | local_irq_disable(); | |
1352 | if (!page) | |
1353 | return NULL; | |
1354 | ||
834f3d11 | 1355 | page->objects = oo_objects(oo); |
81819f0f CL |
1356 | mod_zone_page_state(page_zone(page), |
1357 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1358 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1359 | 1 << oo_order(oo)); |
81819f0f CL |
1360 | |
1361 | return page; | |
1362 | } | |
1363 | ||
1364 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1365 | void *object) | |
1366 | { | |
3ec09742 | 1367 | setup_object_debug(s, page, object); |
4f104934 | 1368 | if (unlikely(s->ctor)) |
51cc5068 | 1369 | s->ctor(object); |
81819f0f CL |
1370 | } |
1371 | ||
1372 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1373 | { | |
1374 | struct page *page; | |
81819f0f | 1375 | void *start; |
81819f0f | 1376 | void *p; |
1f458cbf | 1377 | int order; |
54266640 | 1378 | int idx; |
81819f0f | 1379 | |
c871ac4e AM |
1380 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
1381 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
1382 | BUG(); | |
1383 | } | |
81819f0f | 1384 | |
6cb06229 CL |
1385 | page = allocate_slab(s, |
1386 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1387 | if (!page) |
1388 | goto out; | |
1389 | ||
1f458cbf | 1390 | order = compound_order(page); |
205ab99d | 1391 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1b4f59e3 | 1392 | page->slab_cache = s; |
c03f94cc | 1393 | __SetPageSlab(page); |
072bb0aa MG |
1394 | if (page->pfmemalloc) |
1395 | SetPageSlabPfmemalloc(page); | |
81819f0f CL |
1396 | |
1397 | start = page_address(page); | |
81819f0f CL |
1398 | |
1399 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1400 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1401 | |
54266640 WY |
1402 | for_each_object_idx(p, idx, s, start, page->objects) { |
1403 | setup_object(s, page, p); | |
1404 | if (likely(idx < page->objects)) | |
1405 | set_freepointer(s, p, p + s->size); | |
1406 | else | |
1407 | set_freepointer(s, p, NULL); | |
81819f0f | 1408 | } |
81819f0f CL |
1409 | |
1410 | page->freelist = start; | |
e6e82ea1 | 1411 | page->inuse = page->objects; |
8cb0a506 | 1412 | page->frozen = 1; |
81819f0f | 1413 | out: |
81819f0f CL |
1414 | return page; |
1415 | } | |
1416 | ||
1417 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1418 | { | |
834f3d11 CL |
1419 | int order = compound_order(page); |
1420 | int pages = 1 << order; | |
81819f0f | 1421 | |
af537b0a | 1422 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1423 | void *p; |
1424 | ||
1425 | slab_pad_check(s, page); | |
224a88be CL |
1426 | for_each_object(p, s, page_address(page), |
1427 | page->objects) | |
f7cb1933 | 1428 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1429 | } |
1430 | ||
b1eeab67 | 1431 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1432 | |
81819f0f CL |
1433 | mod_zone_page_state(page_zone(page), |
1434 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1435 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1436 | -pages); |
81819f0f | 1437 | |
072bb0aa | 1438 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1439 | __ClearPageSlab(page); |
1f458cbf | 1440 | |
22b751c3 | 1441 | page_mapcount_reset(page); |
1eb5ac64 NP |
1442 | if (current->reclaim_state) |
1443 | current->reclaim_state->reclaimed_slab += pages; | |
5dfb4175 VD |
1444 | __free_pages(page, order); |
1445 | memcg_uncharge_slab(s, order); | |
81819f0f CL |
1446 | } |
1447 | ||
da9a638c LJ |
1448 | #define need_reserve_slab_rcu \ |
1449 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1450 | ||
81819f0f CL |
1451 | static void rcu_free_slab(struct rcu_head *h) |
1452 | { | |
1453 | struct page *page; | |
1454 | ||
da9a638c LJ |
1455 | if (need_reserve_slab_rcu) |
1456 | page = virt_to_head_page(h); | |
1457 | else | |
1458 | page = container_of((struct list_head *)h, struct page, lru); | |
1459 | ||
1b4f59e3 | 1460 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1461 | } |
1462 | ||
1463 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1464 | { | |
1465 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1466 | struct rcu_head *head; |
1467 | ||
1468 | if (need_reserve_slab_rcu) { | |
1469 | int order = compound_order(page); | |
1470 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1471 | ||
1472 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1473 | head = page_address(page) + offset; | |
1474 | } else { | |
1475 | /* | |
1476 | * RCU free overloads the RCU head over the LRU | |
1477 | */ | |
1478 | head = (void *)&page->lru; | |
1479 | } | |
81819f0f CL |
1480 | |
1481 | call_rcu(head, rcu_free_slab); | |
1482 | } else | |
1483 | __free_slab(s, page); | |
1484 | } | |
1485 | ||
1486 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1487 | { | |
205ab99d | 1488 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1489 | free_slab(s, page); |
1490 | } | |
1491 | ||
1492 | /* | |
5cc6eee8 | 1493 | * Management of partially allocated slabs. |
81819f0f | 1494 | */ |
1e4dd946 SR |
1495 | static inline void |
1496 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1497 | { |
e95eed57 | 1498 | n->nr_partial++; |
136333d1 | 1499 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1500 | list_add_tail(&page->lru, &n->partial); |
1501 | else | |
1502 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1503 | } |
1504 | ||
1e4dd946 SR |
1505 | static inline void add_partial(struct kmem_cache_node *n, |
1506 | struct page *page, int tail) | |
62e346a8 | 1507 | { |
c65c1877 | 1508 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1509 | __add_partial(n, page, tail); |
1510 | } | |
c65c1877 | 1511 | |
1e4dd946 SR |
1512 | static inline void |
1513 | __remove_partial(struct kmem_cache_node *n, struct page *page) | |
1514 | { | |
62e346a8 CL |
1515 | list_del(&page->lru); |
1516 | n->nr_partial--; | |
1517 | } | |
1518 | ||
1e4dd946 SR |
1519 | static inline void remove_partial(struct kmem_cache_node *n, |
1520 | struct page *page) | |
1521 | { | |
1522 | lockdep_assert_held(&n->list_lock); | |
1523 | __remove_partial(n, page); | |
1524 | } | |
1525 | ||
81819f0f | 1526 | /* |
7ced3719 CL |
1527 | * Remove slab from the partial list, freeze it and |
1528 | * return the pointer to the freelist. | |
81819f0f | 1529 | * |
497b66f2 | 1530 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1531 | */ |
497b66f2 | 1532 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1533 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1534 | int mode, int *objects) |
81819f0f | 1535 | { |
2cfb7455 CL |
1536 | void *freelist; |
1537 | unsigned long counters; | |
1538 | struct page new; | |
1539 | ||
c65c1877 PZ |
1540 | lockdep_assert_held(&n->list_lock); |
1541 | ||
2cfb7455 CL |
1542 | /* |
1543 | * Zap the freelist and set the frozen bit. | |
1544 | * The old freelist is the list of objects for the | |
1545 | * per cpu allocation list. | |
1546 | */ | |
7ced3719 CL |
1547 | freelist = page->freelist; |
1548 | counters = page->counters; | |
1549 | new.counters = counters; | |
633b0764 | 1550 | *objects = new.objects - new.inuse; |
23910c50 | 1551 | if (mode) { |
7ced3719 | 1552 | new.inuse = page->objects; |
23910c50 PE |
1553 | new.freelist = NULL; |
1554 | } else { | |
1555 | new.freelist = freelist; | |
1556 | } | |
2cfb7455 | 1557 | |
a0132ac0 | 1558 | VM_BUG_ON(new.frozen); |
7ced3719 | 1559 | new.frozen = 1; |
2cfb7455 | 1560 | |
7ced3719 | 1561 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1562 | freelist, counters, |
02d7633f | 1563 | new.freelist, new.counters, |
7ced3719 | 1564 | "acquire_slab")) |
7ced3719 | 1565 | return NULL; |
2cfb7455 CL |
1566 | |
1567 | remove_partial(n, page); | |
7ced3719 | 1568 | WARN_ON(!freelist); |
49e22585 | 1569 | return freelist; |
81819f0f CL |
1570 | } |
1571 | ||
633b0764 | 1572 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1573 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1574 | |
81819f0f | 1575 | /* |
672bba3a | 1576 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1577 | */ |
8ba00bb6 JK |
1578 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1579 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1580 | { |
49e22585 CL |
1581 | struct page *page, *page2; |
1582 | void *object = NULL; | |
633b0764 JK |
1583 | int available = 0; |
1584 | int objects; | |
81819f0f CL |
1585 | |
1586 | /* | |
1587 | * Racy check. If we mistakenly see no partial slabs then we | |
1588 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1589 | * partial slab and there is none available then get_partials() |
1590 | * will return NULL. | |
81819f0f CL |
1591 | */ |
1592 | if (!n || !n->nr_partial) | |
1593 | return NULL; | |
1594 | ||
1595 | spin_lock(&n->list_lock); | |
49e22585 | 1596 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1597 | void *t; |
49e22585 | 1598 | |
8ba00bb6 JK |
1599 | if (!pfmemalloc_match(page, flags)) |
1600 | continue; | |
1601 | ||
633b0764 | 1602 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1603 | if (!t) |
1604 | break; | |
1605 | ||
633b0764 | 1606 | available += objects; |
12d79634 | 1607 | if (!object) { |
49e22585 | 1608 | c->page = page; |
49e22585 | 1609 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1610 | object = t; |
49e22585 | 1611 | } else { |
633b0764 | 1612 | put_cpu_partial(s, page, 0); |
8028dcea | 1613 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1614 | } |
345c905d JK |
1615 | if (!kmem_cache_has_cpu_partial(s) |
1616 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1617 | break; |
1618 | ||
497b66f2 | 1619 | } |
81819f0f | 1620 | spin_unlock(&n->list_lock); |
497b66f2 | 1621 | return object; |
81819f0f CL |
1622 | } |
1623 | ||
1624 | /* | |
672bba3a | 1625 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1626 | */ |
de3ec035 | 1627 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1628 | struct kmem_cache_cpu *c) |
81819f0f CL |
1629 | { |
1630 | #ifdef CONFIG_NUMA | |
1631 | struct zonelist *zonelist; | |
dd1a239f | 1632 | struct zoneref *z; |
54a6eb5c MG |
1633 | struct zone *zone; |
1634 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1635 | void *object; |
cc9a6c87 | 1636 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1637 | |
1638 | /* | |
672bba3a CL |
1639 | * The defrag ratio allows a configuration of the tradeoffs between |
1640 | * inter node defragmentation and node local allocations. A lower | |
1641 | * defrag_ratio increases the tendency to do local allocations | |
1642 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1643 | * |
672bba3a CL |
1644 | * If the defrag_ratio is set to 0 then kmalloc() always |
1645 | * returns node local objects. If the ratio is higher then kmalloc() | |
1646 | * may return off node objects because partial slabs are obtained | |
1647 | * from other nodes and filled up. | |
81819f0f | 1648 | * |
6446faa2 | 1649 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1650 | * defrag_ratio = 1000) then every (well almost) allocation will |
1651 | * first attempt to defrag slab caches on other nodes. This means | |
1652 | * scanning over all nodes to look for partial slabs which may be | |
1653 | * expensive if we do it every time we are trying to find a slab | |
1654 | * with available objects. | |
81819f0f | 1655 | */ |
9824601e CL |
1656 | if (!s->remote_node_defrag_ratio || |
1657 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1658 | return NULL; |
1659 | ||
cc9a6c87 | 1660 | do { |
d26914d1 | 1661 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1662 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1663 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1664 | struct kmem_cache_node *n; | |
1665 | ||
1666 | n = get_node(s, zone_to_nid(zone)); | |
1667 | ||
1668 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && | |
1669 | n->nr_partial > s->min_partial) { | |
8ba00bb6 | 1670 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1671 | if (object) { |
1672 | /* | |
d26914d1 MG |
1673 | * Don't check read_mems_allowed_retry() |
1674 | * here - if mems_allowed was updated in | |
1675 | * parallel, that was a harmless race | |
1676 | * between allocation and the cpuset | |
1677 | * update | |
cc9a6c87 | 1678 | */ |
cc9a6c87 MG |
1679 | return object; |
1680 | } | |
c0ff7453 | 1681 | } |
81819f0f | 1682 | } |
d26914d1 | 1683 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1684 | #endif |
1685 | return NULL; | |
1686 | } | |
1687 | ||
1688 | /* | |
1689 | * Get a partial page, lock it and return it. | |
1690 | */ | |
497b66f2 | 1691 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1692 | struct kmem_cache_cpu *c) |
81819f0f | 1693 | { |
497b66f2 | 1694 | void *object; |
a561ce00 JK |
1695 | int searchnode = node; |
1696 | ||
1697 | if (node == NUMA_NO_NODE) | |
1698 | searchnode = numa_mem_id(); | |
1699 | else if (!node_present_pages(node)) | |
1700 | searchnode = node_to_mem_node(node); | |
81819f0f | 1701 | |
8ba00bb6 | 1702 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1703 | if (object || node != NUMA_NO_NODE) |
1704 | return object; | |
81819f0f | 1705 | |
acd19fd1 | 1706 | return get_any_partial(s, flags, c); |
81819f0f CL |
1707 | } |
1708 | ||
8a5ec0ba CL |
1709 | #ifdef CONFIG_PREEMPT |
1710 | /* | |
1711 | * Calculate the next globally unique transaction for disambiguiation | |
1712 | * during cmpxchg. The transactions start with the cpu number and are then | |
1713 | * incremented by CONFIG_NR_CPUS. | |
1714 | */ | |
1715 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1716 | #else | |
1717 | /* | |
1718 | * No preemption supported therefore also no need to check for | |
1719 | * different cpus. | |
1720 | */ | |
1721 | #define TID_STEP 1 | |
1722 | #endif | |
1723 | ||
1724 | static inline unsigned long next_tid(unsigned long tid) | |
1725 | { | |
1726 | return tid + TID_STEP; | |
1727 | } | |
1728 | ||
1729 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1730 | { | |
1731 | return tid % TID_STEP; | |
1732 | } | |
1733 | ||
1734 | static inline unsigned long tid_to_event(unsigned long tid) | |
1735 | { | |
1736 | return tid / TID_STEP; | |
1737 | } | |
1738 | ||
1739 | static inline unsigned int init_tid(int cpu) | |
1740 | { | |
1741 | return cpu; | |
1742 | } | |
1743 | ||
1744 | static inline void note_cmpxchg_failure(const char *n, | |
1745 | const struct kmem_cache *s, unsigned long tid) | |
1746 | { | |
1747 | #ifdef SLUB_DEBUG_CMPXCHG | |
1748 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1749 | ||
f9f58285 | 1750 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1751 | |
1752 | #ifdef CONFIG_PREEMPT | |
1753 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1754 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1755 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1756 | else | |
1757 | #endif | |
1758 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1759 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1760 | tid_to_event(tid), tid_to_event(actual_tid)); |
1761 | else | |
f9f58285 | 1762 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1763 | actual_tid, tid, next_tid(tid)); |
1764 | #endif | |
4fdccdfb | 1765 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1766 | } |
1767 | ||
788e1aad | 1768 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1769 | { |
8a5ec0ba CL |
1770 | int cpu; |
1771 | ||
1772 | for_each_possible_cpu(cpu) | |
1773 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1774 | } |
2cfb7455 | 1775 | |
81819f0f CL |
1776 | /* |
1777 | * Remove the cpu slab | |
1778 | */ | |
d0e0ac97 CG |
1779 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1780 | void *freelist) | |
81819f0f | 1781 | { |
2cfb7455 | 1782 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1783 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1784 | int lock = 0; | |
1785 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1786 | void *nextfree; |
136333d1 | 1787 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1788 | struct page new; |
1789 | struct page old; | |
1790 | ||
1791 | if (page->freelist) { | |
84e554e6 | 1792 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1793 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1794 | } |
1795 | ||
894b8788 | 1796 | /* |
2cfb7455 CL |
1797 | * Stage one: Free all available per cpu objects back |
1798 | * to the page freelist while it is still frozen. Leave the | |
1799 | * last one. | |
1800 | * | |
1801 | * There is no need to take the list->lock because the page | |
1802 | * is still frozen. | |
1803 | */ | |
1804 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1805 | void *prior; | |
1806 | unsigned long counters; | |
1807 | ||
1808 | do { | |
1809 | prior = page->freelist; | |
1810 | counters = page->counters; | |
1811 | set_freepointer(s, freelist, prior); | |
1812 | new.counters = counters; | |
1813 | new.inuse--; | |
a0132ac0 | 1814 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1815 | |
1d07171c | 1816 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1817 | prior, counters, |
1818 | freelist, new.counters, | |
1819 | "drain percpu freelist")); | |
1820 | ||
1821 | freelist = nextfree; | |
1822 | } | |
1823 | ||
894b8788 | 1824 | /* |
2cfb7455 CL |
1825 | * Stage two: Ensure that the page is unfrozen while the |
1826 | * list presence reflects the actual number of objects | |
1827 | * during unfreeze. | |
1828 | * | |
1829 | * We setup the list membership and then perform a cmpxchg | |
1830 | * with the count. If there is a mismatch then the page | |
1831 | * is not unfrozen but the page is on the wrong list. | |
1832 | * | |
1833 | * Then we restart the process which may have to remove | |
1834 | * the page from the list that we just put it on again | |
1835 | * because the number of objects in the slab may have | |
1836 | * changed. | |
894b8788 | 1837 | */ |
2cfb7455 | 1838 | redo: |
894b8788 | 1839 | |
2cfb7455 CL |
1840 | old.freelist = page->freelist; |
1841 | old.counters = page->counters; | |
a0132ac0 | 1842 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1843 | |
2cfb7455 CL |
1844 | /* Determine target state of the slab */ |
1845 | new.counters = old.counters; | |
1846 | if (freelist) { | |
1847 | new.inuse--; | |
1848 | set_freepointer(s, freelist, old.freelist); | |
1849 | new.freelist = freelist; | |
1850 | } else | |
1851 | new.freelist = old.freelist; | |
1852 | ||
1853 | new.frozen = 0; | |
1854 | ||
8a5b20ae | 1855 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
1856 | m = M_FREE; |
1857 | else if (new.freelist) { | |
1858 | m = M_PARTIAL; | |
1859 | if (!lock) { | |
1860 | lock = 1; | |
1861 | /* | |
1862 | * Taking the spinlock removes the possiblity | |
1863 | * that acquire_slab() will see a slab page that | |
1864 | * is frozen | |
1865 | */ | |
1866 | spin_lock(&n->list_lock); | |
1867 | } | |
1868 | } else { | |
1869 | m = M_FULL; | |
1870 | if (kmem_cache_debug(s) && !lock) { | |
1871 | lock = 1; | |
1872 | /* | |
1873 | * This also ensures that the scanning of full | |
1874 | * slabs from diagnostic functions will not see | |
1875 | * any frozen slabs. | |
1876 | */ | |
1877 | spin_lock(&n->list_lock); | |
1878 | } | |
1879 | } | |
1880 | ||
1881 | if (l != m) { | |
1882 | ||
1883 | if (l == M_PARTIAL) | |
1884 | ||
1885 | remove_partial(n, page); | |
1886 | ||
1887 | else if (l == M_FULL) | |
894b8788 | 1888 | |
c65c1877 | 1889 | remove_full(s, n, page); |
2cfb7455 CL |
1890 | |
1891 | if (m == M_PARTIAL) { | |
1892 | ||
1893 | add_partial(n, page, tail); | |
136333d1 | 1894 | stat(s, tail); |
2cfb7455 CL |
1895 | |
1896 | } else if (m == M_FULL) { | |
894b8788 | 1897 | |
2cfb7455 CL |
1898 | stat(s, DEACTIVATE_FULL); |
1899 | add_full(s, n, page); | |
1900 | ||
1901 | } | |
1902 | } | |
1903 | ||
1904 | l = m; | |
1d07171c | 1905 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1906 | old.freelist, old.counters, |
1907 | new.freelist, new.counters, | |
1908 | "unfreezing slab")) | |
1909 | goto redo; | |
1910 | ||
2cfb7455 CL |
1911 | if (lock) |
1912 | spin_unlock(&n->list_lock); | |
1913 | ||
1914 | if (m == M_FREE) { | |
1915 | stat(s, DEACTIVATE_EMPTY); | |
1916 | discard_slab(s, page); | |
1917 | stat(s, FREE_SLAB); | |
894b8788 | 1918 | } |
81819f0f CL |
1919 | } |
1920 | ||
d24ac77f JK |
1921 | /* |
1922 | * Unfreeze all the cpu partial slabs. | |
1923 | * | |
59a09917 CL |
1924 | * This function must be called with interrupts disabled |
1925 | * for the cpu using c (or some other guarantee must be there | |
1926 | * to guarantee no concurrent accesses). | |
d24ac77f | 1927 | */ |
59a09917 CL |
1928 | static void unfreeze_partials(struct kmem_cache *s, |
1929 | struct kmem_cache_cpu *c) | |
49e22585 | 1930 | { |
345c905d | 1931 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 1932 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1933 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1934 | |
1935 | while ((page = c->partial)) { | |
49e22585 CL |
1936 | struct page new; |
1937 | struct page old; | |
1938 | ||
1939 | c->partial = page->next; | |
43d77867 JK |
1940 | |
1941 | n2 = get_node(s, page_to_nid(page)); | |
1942 | if (n != n2) { | |
1943 | if (n) | |
1944 | spin_unlock(&n->list_lock); | |
1945 | ||
1946 | n = n2; | |
1947 | spin_lock(&n->list_lock); | |
1948 | } | |
49e22585 CL |
1949 | |
1950 | do { | |
1951 | ||
1952 | old.freelist = page->freelist; | |
1953 | old.counters = page->counters; | |
a0132ac0 | 1954 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
1955 | |
1956 | new.counters = old.counters; | |
1957 | new.freelist = old.freelist; | |
1958 | ||
1959 | new.frozen = 0; | |
1960 | ||
d24ac77f | 1961 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1962 | old.freelist, old.counters, |
1963 | new.freelist, new.counters, | |
1964 | "unfreezing slab")); | |
1965 | ||
8a5b20ae | 1966 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
1967 | page->next = discard_page; |
1968 | discard_page = page; | |
43d77867 JK |
1969 | } else { |
1970 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1971 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1972 | } |
1973 | } | |
1974 | ||
1975 | if (n) | |
1976 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
1977 | |
1978 | while (discard_page) { | |
1979 | page = discard_page; | |
1980 | discard_page = discard_page->next; | |
1981 | ||
1982 | stat(s, DEACTIVATE_EMPTY); | |
1983 | discard_slab(s, page); | |
1984 | stat(s, FREE_SLAB); | |
1985 | } | |
345c905d | 1986 | #endif |
49e22585 CL |
1987 | } |
1988 | ||
1989 | /* | |
1990 | * Put a page that was just frozen (in __slab_free) into a partial page | |
1991 | * slot if available. This is done without interrupts disabled and without | |
1992 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
1993 | * onto a random cpus partial slot. | |
1994 | * | |
1995 | * If we did not find a slot then simply move all the partials to the | |
1996 | * per node partial list. | |
1997 | */ | |
633b0764 | 1998 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 1999 | { |
345c905d | 2000 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2001 | struct page *oldpage; |
2002 | int pages; | |
2003 | int pobjects; | |
2004 | ||
2005 | do { | |
2006 | pages = 0; | |
2007 | pobjects = 0; | |
2008 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2009 | ||
2010 | if (oldpage) { | |
2011 | pobjects = oldpage->pobjects; | |
2012 | pages = oldpage->pages; | |
2013 | if (drain && pobjects > s->cpu_partial) { | |
2014 | unsigned long flags; | |
2015 | /* | |
2016 | * partial array is full. Move the existing | |
2017 | * set to the per node partial list. | |
2018 | */ | |
2019 | local_irq_save(flags); | |
59a09917 | 2020 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2021 | local_irq_restore(flags); |
e24fc410 | 2022 | oldpage = NULL; |
49e22585 CL |
2023 | pobjects = 0; |
2024 | pages = 0; | |
8028dcea | 2025 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2026 | } |
2027 | } | |
2028 | ||
2029 | pages++; | |
2030 | pobjects += page->objects - page->inuse; | |
2031 | ||
2032 | page->pages = pages; | |
2033 | page->pobjects = pobjects; | |
2034 | page->next = oldpage; | |
2035 | ||
d0e0ac97 CG |
2036 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2037 | != oldpage); | |
345c905d | 2038 | #endif |
49e22585 CL |
2039 | } |
2040 | ||
dfb4f096 | 2041 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2042 | { |
84e554e6 | 2043 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2044 | deactivate_slab(s, c->page, c->freelist); |
2045 | ||
2046 | c->tid = next_tid(c->tid); | |
2047 | c->page = NULL; | |
2048 | c->freelist = NULL; | |
81819f0f CL |
2049 | } |
2050 | ||
2051 | /* | |
2052 | * Flush cpu slab. | |
6446faa2 | 2053 | * |
81819f0f CL |
2054 | * Called from IPI handler with interrupts disabled. |
2055 | */ | |
0c710013 | 2056 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2057 | { |
9dfc6e68 | 2058 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2059 | |
49e22585 CL |
2060 | if (likely(c)) { |
2061 | if (c->page) | |
2062 | flush_slab(s, c); | |
2063 | ||
59a09917 | 2064 | unfreeze_partials(s, c); |
49e22585 | 2065 | } |
81819f0f CL |
2066 | } |
2067 | ||
2068 | static void flush_cpu_slab(void *d) | |
2069 | { | |
2070 | struct kmem_cache *s = d; | |
81819f0f | 2071 | |
dfb4f096 | 2072 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2073 | } |
2074 | ||
a8364d55 GBY |
2075 | static bool has_cpu_slab(int cpu, void *info) |
2076 | { | |
2077 | struct kmem_cache *s = info; | |
2078 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2079 | ||
02e1a9cd | 2080 | return c->page || c->partial; |
a8364d55 GBY |
2081 | } |
2082 | ||
81819f0f CL |
2083 | static void flush_all(struct kmem_cache *s) |
2084 | { | |
a8364d55 | 2085 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2086 | } |
2087 | ||
dfb4f096 CL |
2088 | /* |
2089 | * Check if the objects in a per cpu structure fit numa | |
2090 | * locality expectations. | |
2091 | */ | |
57d437d2 | 2092 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2093 | { |
2094 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2095 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2096 | return 0; |
2097 | #endif | |
2098 | return 1; | |
2099 | } | |
2100 | ||
9a02d699 | 2101 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2102 | static int count_free(struct page *page) |
2103 | { | |
2104 | return page->objects - page->inuse; | |
2105 | } | |
2106 | ||
9a02d699 DR |
2107 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2108 | { | |
2109 | return atomic_long_read(&n->total_objects); | |
2110 | } | |
2111 | #endif /* CONFIG_SLUB_DEBUG */ | |
2112 | ||
2113 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2114 | static unsigned long count_partial(struct kmem_cache_node *n, |
2115 | int (*get_count)(struct page *)) | |
2116 | { | |
2117 | unsigned long flags; | |
2118 | unsigned long x = 0; | |
2119 | struct page *page; | |
2120 | ||
2121 | spin_lock_irqsave(&n->list_lock, flags); | |
2122 | list_for_each_entry(page, &n->partial, lru) | |
2123 | x += get_count(page); | |
2124 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2125 | return x; | |
2126 | } | |
9a02d699 | 2127 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2128 | |
781b2ba6 PE |
2129 | static noinline void |
2130 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2131 | { | |
9a02d699 DR |
2132 | #ifdef CONFIG_SLUB_DEBUG |
2133 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2134 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2135 | int node; |
fa45dc25 | 2136 | struct kmem_cache_node *n; |
781b2ba6 | 2137 | |
9a02d699 DR |
2138 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2139 | return; | |
2140 | ||
f9f58285 | 2141 | pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", |
781b2ba6 | 2142 | nid, gfpflags); |
f9f58285 FF |
2143 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2144 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2145 | oo_order(s->min)); | |
781b2ba6 | 2146 | |
3b0efdfa | 2147 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2148 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2149 | s->name); | |
fa5ec8a1 | 2150 | |
fa45dc25 | 2151 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2152 | unsigned long nr_slabs; |
2153 | unsigned long nr_objs; | |
2154 | unsigned long nr_free; | |
2155 | ||
26c02cf0 AB |
2156 | nr_free = count_partial(n, count_free); |
2157 | nr_slabs = node_nr_slabs(n); | |
2158 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2159 | |
f9f58285 | 2160 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2161 | node, nr_slabs, nr_objs, nr_free); |
2162 | } | |
9a02d699 | 2163 | #endif |
781b2ba6 PE |
2164 | } |
2165 | ||
497b66f2 CL |
2166 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2167 | int node, struct kmem_cache_cpu **pc) | |
2168 | { | |
6faa6833 | 2169 | void *freelist; |
188fd063 CL |
2170 | struct kmem_cache_cpu *c = *pc; |
2171 | struct page *page; | |
497b66f2 | 2172 | |
188fd063 | 2173 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2174 | |
188fd063 CL |
2175 | if (freelist) |
2176 | return freelist; | |
2177 | ||
2178 | page = new_slab(s, flags, node); | |
497b66f2 | 2179 | if (page) { |
7c8e0181 | 2180 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2181 | if (c->page) |
2182 | flush_slab(s, c); | |
2183 | ||
2184 | /* | |
2185 | * No other reference to the page yet so we can | |
2186 | * muck around with it freely without cmpxchg | |
2187 | */ | |
6faa6833 | 2188 | freelist = page->freelist; |
497b66f2 CL |
2189 | page->freelist = NULL; |
2190 | ||
2191 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2192 | c->page = page; |
2193 | *pc = c; | |
2194 | } else | |
6faa6833 | 2195 | freelist = NULL; |
497b66f2 | 2196 | |
6faa6833 | 2197 | return freelist; |
497b66f2 CL |
2198 | } |
2199 | ||
072bb0aa MG |
2200 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2201 | { | |
2202 | if (unlikely(PageSlabPfmemalloc(page))) | |
2203 | return gfp_pfmemalloc_allowed(gfpflags); | |
2204 | ||
2205 | return true; | |
2206 | } | |
2207 | ||
213eeb9f | 2208 | /* |
d0e0ac97 CG |
2209 | * Check the page->freelist of a page and either transfer the freelist to the |
2210 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2211 | * |
2212 | * The page is still frozen if the return value is not NULL. | |
2213 | * | |
2214 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2215 | * |
2216 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2217 | */ |
2218 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2219 | { | |
2220 | struct page new; | |
2221 | unsigned long counters; | |
2222 | void *freelist; | |
2223 | ||
2224 | do { | |
2225 | freelist = page->freelist; | |
2226 | counters = page->counters; | |
6faa6833 | 2227 | |
213eeb9f | 2228 | new.counters = counters; |
a0132ac0 | 2229 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2230 | |
2231 | new.inuse = page->objects; | |
2232 | new.frozen = freelist != NULL; | |
2233 | ||
d24ac77f | 2234 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2235 | freelist, counters, |
2236 | NULL, new.counters, | |
2237 | "get_freelist")); | |
2238 | ||
2239 | return freelist; | |
2240 | } | |
2241 | ||
81819f0f | 2242 | /* |
894b8788 CL |
2243 | * Slow path. The lockless freelist is empty or we need to perform |
2244 | * debugging duties. | |
2245 | * | |
894b8788 CL |
2246 | * Processing is still very fast if new objects have been freed to the |
2247 | * regular freelist. In that case we simply take over the regular freelist | |
2248 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2249 | * |
894b8788 CL |
2250 | * If that is not working then we fall back to the partial lists. We take the |
2251 | * first element of the freelist as the object to allocate now and move the | |
2252 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2253 | * |
894b8788 | 2254 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2255 | * we need to allocate a new slab. This is the slowest path since it involves |
2256 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2257 | */ |
ce71e27c EGM |
2258 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2259 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2260 | { |
6faa6833 | 2261 | void *freelist; |
f6e7def7 | 2262 | struct page *page; |
8a5ec0ba CL |
2263 | unsigned long flags; |
2264 | ||
2265 | local_irq_save(flags); | |
2266 | #ifdef CONFIG_PREEMPT | |
2267 | /* | |
2268 | * We may have been preempted and rescheduled on a different | |
2269 | * cpu before disabling interrupts. Need to reload cpu area | |
2270 | * pointer. | |
2271 | */ | |
2272 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2273 | #endif |
81819f0f | 2274 | |
f6e7def7 CL |
2275 | page = c->page; |
2276 | if (!page) | |
81819f0f | 2277 | goto new_slab; |
49e22585 | 2278 | redo: |
6faa6833 | 2279 | |
57d437d2 | 2280 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2281 | int searchnode = node; |
2282 | ||
2283 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2284 | searchnode = node_to_mem_node(node); | |
2285 | ||
2286 | if (unlikely(!node_match(page, searchnode))) { | |
2287 | stat(s, ALLOC_NODE_MISMATCH); | |
2288 | deactivate_slab(s, page, c->freelist); | |
2289 | c->page = NULL; | |
2290 | c->freelist = NULL; | |
2291 | goto new_slab; | |
2292 | } | |
fc59c053 | 2293 | } |
6446faa2 | 2294 | |
072bb0aa MG |
2295 | /* |
2296 | * By rights, we should be searching for a slab page that was | |
2297 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2298 | * information when the page leaves the per-cpu allocator | |
2299 | */ | |
2300 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2301 | deactivate_slab(s, page, c->freelist); | |
2302 | c->page = NULL; | |
2303 | c->freelist = NULL; | |
2304 | goto new_slab; | |
2305 | } | |
2306 | ||
73736e03 | 2307 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2308 | freelist = c->freelist; |
2309 | if (freelist) | |
73736e03 | 2310 | goto load_freelist; |
03e404af | 2311 | |
f6e7def7 | 2312 | freelist = get_freelist(s, page); |
6446faa2 | 2313 | |
6faa6833 | 2314 | if (!freelist) { |
03e404af CL |
2315 | c->page = NULL; |
2316 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2317 | goto new_slab; |
03e404af | 2318 | } |
6446faa2 | 2319 | |
84e554e6 | 2320 | stat(s, ALLOC_REFILL); |
6446faa2 | 2321 | |
894b8788 | 2322 | load_freelist: |
507effea CL |
2323 | /* |
2324 | * freelist is pointing to the list of objects to be used. | |
2325 | * page is pointing to the page from which the objects are obtained. | |
2326 | * That page must be frozen for per cpu allocations to work. | |
2327 | */ | |
a0132ac0 | 2328 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2329 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2330 | c->tid = next_tid(c->tid); |
2331 | local_irq_restore(flags); | |
6faa6833 | 2332 | return freelist; |
81819f0f | 2333 | |
81819f0f | 2334 | new_slab: |
2cfb7455 | 2335 | |
49e22585 | 2336 | if (c->partial) { |
f6e7def7 CL |
2337 | page = c->page = c->partial; |
2338 | c->partial = page->next; | |
49e22585 CL |
2339 | stat(s, CPU_PARTIAL_ALLOC); |
2340 | c->freelist = NULL; | |
2341 | goto redo; | |
81819f0f CL |
2342 | } |
2343 | ||
188fd063 | 2344 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2345 | |
f4697436 | 2346 | if (unlikely(!freelist)) { |
9a02d699 | 2347 | slab_out_of_memory(s, gfpflags, node); |
f4697436 CL |
2348 | local_irq_restore(flags); |
2349 | return NULL; | |
81819f0f | 2350 | } |
2cfb7455 | 2351 | |
f6e7def7 | 2352 | page = c->page; |
5091b74a | 2353 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2354 | goto load_freelist; |
2cfb7455 | 2355 | |
497b66f2 | 2356 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2357 | if (kmem_cache_debug(s) && |
2358 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2359 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2360 | |
f6e7def7 | 2361 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2362 | c->page = NULL; |
2363 | c->freelist = NULL; | |
a71ae47a | 2364 | local_irq_restore(flags); |
6faa6833 | 2365 | return freelist; |
894b8788 CL |
2366 | } |
2367 | ||
2368 | /* | |
2369 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2370 | * have the fastpath folded into their functions. So no function call | |
2371 | * overhead for requests that can be satisfied on the fastpath. | |
2372 | * | |
2373 | * The fastpath works by first checking if the lockless freelist can be used. | |
2374 | * If not then __slab_alloc is called for slow processing. | |
2375 | * | |
2376 | * Otherwise we can simply pick the next object from the lockless free list. | |
2377 | */ | |
2b847c3c | 2378 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2379 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2380 | { |
894b8788 | 2381 | void **object; |
dfb4f096 | 2382 | struct kmem_cache_cpu *c; |
57d437d2 | 2383 | struct page *page; |
8a5ec0ba | 2384 | unsigned long tid; |
1f84260c | 2385 | |
c016b0bd | 2386 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2387 | return NULL; |
1f84260c | 2388 | |
d79923fa | 2389 | s = memcg_kmem_get_cache(s, gfpflags); |
8a5ec0ba | 2390 | redo: |
8a5ec0ba CL |
2391 | /* |
2392 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2393 | * enabled. We may switch back and forth between cpus while | |
2394 | * reading from one cpu area. That does not matter as long | |
2395 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b CL |
2396 | * |
2397 | * Preemption is disabled for the retrieval of the tid because that | |
2398 | * must occur from the current processor. We cannot allow rescheduling | |
2399 | * on a different processor between the determination of the pointer | |
2400 | * and the retrieval of the tid. | |
8a5ec0ba | 2401 | */ |
7cccd80b | 2402 | preempt_disable(); |
7c8e0181 | 2403 | c = this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2404 | |
8a5ec0ba CL |
2405 | /* |
2406 | * The transaction ids are globally unique per cpu and per operation on | |
2407 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2408 | * occurs on the right processor and that there was no operation on the | |
2409 | * linked list in between. | |
2410 | */ | |
2411 | tid = c->tid; | |
7cccd80b | 2412 | preempt_enable(); |
8a5ec0ba | 2413 | |
9dfc6e68 | 2414 | object = c->freelist; |
57d437d2 | 2415 | page = c->page; |
8eae1492 | 2416 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2417 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2418 | stat(s, ALLOC_SLOWPATH); |
2419 | } else { | |
0ad9500e ED |
2420 | void *next_object = get_freepointer_safe(s, object); |
2421 | ||
8a5ec0ba | 2422 | /* |
25985edc | 2423 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2424 | * operation and if we are on the right processor. |
2425 | * | |
d0e0ac97 CG |
2426 | * The cmpxchg does the following atomically (without lock |
2427 | * semantics!) | |
8a5ec0ba CL |
2428 | * 1. Relocate first pointer to the current per cpu area. |
2429 | * 2. Verify that tid and freelist have not been changed | |
2430 | * 3. If they were not changed replace tid and freelist | |
2431 | * | |
d0e0ac97 CG |
2432 | * Since this is without lock semantics the protection is only |
2433 | * against code executing on this cpu *not* from access by | |
2434 | * other cpus. | |
8a5ec0ba | 2435 | */ |
933393f5 | 2436 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2437 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2438 | object, tid, | |
0ad9500e | 2439 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2440 | |
2441 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2442 | goto redo; | |
2443 | } | |
0ad9500e | 2444 | prefetch_freepointer(s, next_object); |
84e554e6 | 2445 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2446 | } |
8a5ec0ba | 2447 | |
74e2134f | 2448 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2449 | memset(object, 0, s->object_size); |
d07dbea4 | 2450 | |
c016b0bd | 2451 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2452 | |
894b8788 | 2453 | return object; |
81819f0f CL |
2454 | } |
2455 | ||
2b847c3c EG |
2456 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2457 | gfp_t gfpflags, unsigned long addr) | |
2458 | { | |
2459 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2460 | } | |
2461 | ||
81819f0f CL |
2462 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2463 | { | |
2b847c3c | 2464 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2465 | |
d0e0ac97 CG |
2466 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2467 | s->size, gfpflags); | |
5b882be4 EGM |
2468 | |
2469 | return ret; | |
81819f0f CL |
2470 | } |
2471 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2472 | ||
0f24f128 | 2473 | #ifdef CONFIG_TRACING |
4a92379b RK |
2474 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2475 | { | |
2b847c3c | 2476 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b RK |
2477 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
2478 | return ret; | |
2479 | } | |
2480 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2481 | #endif |
2482 | ||
81819f0f CL |
2483 | #ifdef CONFIG_NUMA |
2484 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2485 | { | |
2b847c3c | 2486 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2487 | |
ca2b84cb | 2488 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2489 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2490 | |
2491 | return ret; | |
81819f0f CL |
2492 | } |
2493 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2494 | |
0f24f128 | 2495 | #ifdef CONFIG_TRACING |
4a92379b | 2496 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2497 | gfp_t gfpflags, |
4a92379b | 2498 | int node, size_t size) |
5b882be4 | 2499 | { |
2b847c3c | 2500 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2501 | |
2502 | trace_kmalloc_node(_RET_IP_, ret, | |
2503 | size, s->size, gfpflags, node); | |
2504 | return ret; | |
5b882be4 | 2505 | } |
4a92379b | 2506 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2507 | #endif |
5d1f57e4 | 2508 | #endif |
5b882be4 | 2509 | |
81819f0f | 2510 | /* |
894b8788 CL |
2511 | * Slow patch handling. This may still be called frequently since objects |
2512 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2513 | * |
894b8788 CL |
2514 | * So we still attempt to reduce cache line usage. Just take the slab |
2515 | * lock and free the item. If there is no additional partial page | |
2516 | * handling required then we can return immediately. | |
81819f0f | 2517 | */ |
894b8788 | 2518 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2519 | void *x, unsigned long addr) |
81819f0f CL |
2520 | { |
2521 | void *prior; | |
2522 | void **object = (void *)x; | |
2cfb7455 | 2523 | int was_frozen; |
2cfb7455 CL |
2524 | struct page new; |
2525 | unsigned long counters; | |
2526 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2527 | unsigned long uninitialized_var(flags); |
81819f0f | 2528 | |
8a5ec0ba | 2529 | stat(s, FREE_SLOWPATH); |
81819f0f | 2530 | |
19c7ff9e CL |
2531 | if (kmem_cache_debug(s) && |
2532 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2533 | return; |
6446faa2 | 2534 | |
2cfb7455 | 2535 | do { |
837d678d JK |
2536 | if (unlikely(n)) { |
2537 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2538 | n = NULL; | |
2539 | } | |
2cfb7455 CL |
2540 | prior = page->freelist; |
2541 | counters = page->counters; | |
2542 | set_freepointer(s, object, prior); | |
2543 | new.counters = counters; | |
2544 | was_frozen = new.frozen; | |
2545 | new.inuse--; | |
837d678d | 2546 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2547 | |
c65c1877 | 2548 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2549 | |
2550 | /* | |
d0e0ac97 CG |
2551 | * Slab was on no list before and will be |
2552 | * partially empty | |
2553 | * We can defer the list move and instead | |
2554 | * freeze it. | |
49e22585 CL |
2555 | */ |
2556 | new.frozen = 1; | |
2557 | ||
c65c1877 | 2558 | } else { /* Needs to be taken off a list */ |
49e22585 | 2559 | |
b455def2 | 2560 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2561 | /* |
2562 | * Speculatively acquire the list_lock. | |
2563 | * If the cmpxchg does not succeed then we may | |
2564 | * drop the list_lock without any processing. | |
2565 | * | |
2566 | * Otherwise the list_lock will synchronize with | |
2567 | * other processors updating the list of slabs. | |
2568 | */ | |
2569 | spin_lock_irqsave(&n->list_lock, flags); | |
2570 | ||
2571 | } | |
2cfb7455 | 2572 | } |
81819f0f | 2573 | |
2cfb7455 CL |
2574 | } while (!cmpxchg_double_slab(s, page, |
2575 | prior, counters, | |
2576 | object, new.counters, | |
2577 | "__slab_free")); | |
81819f0f | 2578 | |
2cfb7455 | 2579 | if (likely(!n)) { |
49e22585 CL |
2580 | |
2581 | /* | |
2582 | * If we just froze the page then put it onto the | |
2583 | * per cpu partial list. | |
2584 | */ | |
8028dcea | 2585 | if (new.frozen && !was_frozen) { |
49e22585 | 2586 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2587 | stat(s, CPU_PARTIAL_FREE); |
2588 | } | |
49e22585 | 2589 | /* |
2cfb7455 CL |
2590 | * The list lock was not taken therefore no list |
2591 | * activity can be necessary. | |
2592 | */ | |
b455def2 L |
2593 | if (was_frozen) |
2594 | stat(s, FREE_FROZEN); | |
2595 | return; | |
2596 | } | |
81819f0f | 2597 | |
8a5b20ae | 2598 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2599 | goto slab_empty; |
2600 | ||
81819f0f | 2601 | /* |
837d678d JK |
2602 | * Objects left in the slab. If it was not on the partial list before |
2603 | * then add it. | |
81819f0f | 2604 | */ |
345c905d JK |
2605 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2606 | if (kmem_cache_debug(s)) | |
c65c1877 | 2607 | remove_full(s, n, page); |
837d678d JK |
2608 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2609 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2610 | } |
80f08c19 | 2611 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2612 | return; |
2613 | ||
2614 | slab_empty: | |
a973e9dd | 2615 | if (prior) { |
81819f0f | 2616 | /* |
6fbabb20 | 2617 | * Slab on the partial list. |
81819f0f | 2618 | */ |
5cc6eee8 | 2619 | remove_partial(n, page); |
84e554e6 | 2620 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2621 | } else { |
6fbabb20 | 2622 | /* Slab must be on the full list */ |
c65c1877 PZ |
2623 | remove_full(s, n, page); |
2624 | } | |
2cfb7455 | 2625 | |
80f08c19 | 2626 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2627 | stat(s, FREE_SLAB); |
81819f0f | 2628 | discard_slab(s, page); |
81819f0f CL |
2629 | } |
2630 | ||
894b8788 CL |
2631 | /* |
2632 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2633 | * can perform fastpath freeing without additional function calls. | |
2634 | * | |
2635 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2636 | * of this processor. This typically the case if we have just allocated | |
2637 | * the item before. | |
2638 | * | |
2639 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2640 | * with all sorts of special processing. | |
2641 | */ | |
06428780 | 2642 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2643 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2644 | { |
2645 | void **object = (void *)x; | |
dfb4f096 | 2646 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2647 | unsigned long tid; |
1f84260c | 2648 | |
c016b0bd CL |
2649 | slab_free_hook(s, x); |
2650 | ||
8a5ec0ba CL |
2651 | redo: |
2652 | /* | |
2653 | * Determine the currently cpus per cpu slab. | |
2654 | * The cpu may change afterward. However that does not matter since | |
2655 | * data is retrieved via this pointer. If we are on the same cpu | |
2656 | * during the cmpxchg then the free will succedd. | |
2657 | */ | |
7cccd80b | 2658 | preempt_disable(); |
7c8e0181 | 2659 | c = this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2660 | |
8a5ec0ba | 2661 | tid = c->tid; |
7cccd80b | 2662 | preempt_enable(); |
c016b0bd | 2663 | |
442b06bc | 2664 | if (likely(page == c->page)) { |
ff12059e | 2665 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2666 | |
933393f5 | 2667 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2668 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2669 | c->freelist, tid, | |
2670 | object, next_tid(tid)))) { | |
2671 | ||
2672 | note_cmpxchg_failure("slab_free", s, tid); | |
2673 | goto redo; | |
2674 | } | |
84e554e6 | 2675 | stat(s, FREE_FASTPATH); |
894b8788 | 2676 | } else |
ff12059e | 2677 | __slab_free(s, page, x, addr); |
894b8788 | 2678 | |
894b8788 CL |
2679 | } |
2680 | ||
81819f0f CL |
2681 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2682 | { | |
b9ce5ef4 GC |
2683 | s = cache_from_obj(s, x); |
2684 | if (!s) | |
79576102 | 2685 | return; |
b9ce5ef4 | 2686 | slab_free(s, virt_to_head_page(x), x, _RET_IP_); |
ca2b84cb | 2687 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2688 | } |
2689 | EXPORT_SYMBOL(kmem_cache_free); | |
2690 | ||
81819f0f | 2691 | /* |
672bba3a CL |
2692 | * Object placement in a slab is made very easy because we always start at |
2693 | * offset 0. If we tune the size of the object to the alignment then we can | |
2694 | * get the required alignment by putting one properly sized object after | |
2695 | * another. | |
81819f0f CL |
2696 | * |
2697 | * Notice that the allocation order determines the sizes of the per cpu | |
2698 | * caches. Each processor has always one slab available for allocations. | |
2699 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2700 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2701 | * locking overhead. |
81819f0f CL |
2702 | */ |
2703 | ||
2704 | /* | |
2705 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2706 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2707 | * and increases the number of allocations possible without having to | |
2708 | * take the list_lock. | |
2709 | */ | |
2710 | static int slub_min_order; | |
114e9e89 | 2711 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2712 | static int slub_min_objects; |
81819f0f | 2713 | |
81819f0f CL |
2714 | /* |
2715 | * Calculate the order of allocation given an slab object size. | |
2716 | * | |
672bba3a CL |
2717 | * The order of allocation has significant impact on performance and other |
2718 | * system components. Generally order 0 allocations should be preferred since | |
2719 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2720 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2721 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2722 | * would be wasted. |
2723 | * | |
2724 | * In order to reach satisfactory performance we must ensure that a minimum | |
2725 | * number of objects is in one slab. Otherwise we may generate too much | |
2726 | * activity on the partial lists which requires taking the list_lock. This is | |
2727 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2728 | * |
672bba3a CL |
2729 | * slub_max_order specifies the order where we begin to stop considering the |
2730 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2731 | * we try to keep the page order as low as possible. So we accept more waste | |
2732 | * of space in favor of a small page order. | |
81819f0f | 2733 | * |
672bba3a CL |
2734 | * Higher order allocations also allow the placement of more objects in a |
2735 | * slab and thereby reduce object handling overhead. If the user has | |
2736 | * requested a higher mininum order then we start with that one instead of | |
2737 | * the smallest order which will fit the object. | |
81819f0f | 2738 | */ |
5e6d444e | 2739 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2740 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2741 | { |
2742 | int order; | |
2743 | int rem; | |
6300ea75 | 2744 | int min_order = slub_min_order; |
81819f0f | 2745 | |
ab9a0f19 | 2746 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2747 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2748 | |
6300ea75 | 2749 | for (order = max(min_order, |
5e6d444e CL |
2750 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2751 | order <= max_order; order++) { | |
81819f0f | 2752 | |
5e6d444e | 2753 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2754 | |
ab9a0f19 | 2755 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2756 | continue; |
2757 | ||
ab9a0f19 | 2758 | rem = (slab_size - reserved) % size; |
81819f0f | 2759 | |
5e6d444e | 2760 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2761 | break; |
2762 | ||
2763 | } | |
672bba3a | 2764 | |
81819f0f CL |
2765 | return order; |
2766 | } | |
2767 | ||
ab9a0f19 | 2768 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2769 | { |
2770 | int order; | |
2771 | int min_objects; | |
2772 | int fraction; | |
e8120ff1 | 2773 | int max_objects; |
5e6d444e CL |
2774 | |
2775 | /* | |
2776 | * Attempt to find best configuration for a slab. This | |
2777 | * works by first attempting to generate a layout with | |
2778 | * the best configuration and backing off gradually. | |
2779 | * | |
2780 | * First we reduce the acceptable waste in a slab. Then | |
2781 | * we reduce the minimum objects required in a slab. | |
2782 | */ | |
2783 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2784 | if (!min_objects) |
2785 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2786 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2787 | min_objects = min(min_objects, max_objects); |
2788 | ||
5e6d444e | 2789 | while (min_objects > 1) { |
c124f5b5 | 2790 | fraction = 16; |
5e6d444e CL |
2791 | while (fraction >= 4) { |
2792 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2793 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2794 | if (order <= slub_max_order) |
2795 | return order; | |
2796 | fraction /= 2; | |
2797 | } | |
5086c389 | 2798 | min_objects--; |
5e6d444e CL |
2799 | } |
2800 | ||
2801 | /* | |
2802 | * We were unable to place multiple objects in a slab. Now | |
2803 | * lets see if we can place a single object there. | |
2804 | */ | |
ab9a0f19 | 2805 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2806 | if (order <= slub_max_order) |
2807 | return order; | |
2808 | ||
2809 | /* | |
2810 | * Doh this slab cannot be placed using slub_max_order. | |
2811 | */ | |
ab9a0f19 | 2812 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2813 | if (order < MAX_ORDER) |
5e6d444e CL |
2814 | return order; |
2815 | return -ENOSYS; | |
2816 | } | |
2817 | ||
5595cffc | 2818 | static void |
4053497d | 2819 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2820 | { |
2821 | n->nr_partial = 0; | |
81819f0f CL |
2822 | spin_lock_init(&n->list_lock); |
2823 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2824 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2825 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2826 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2827 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2828 | #endif |
81819f0f CL |
2829 | } |
2830 | ||
55136592 | 2831 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2832 | { |
6c182dc0 | 2833 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 2834 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 2835 | |
8a5ec0ba | 2836 | /* |
d4d84fef CM |
2837 | * Must align to double word boundary for the double cmpxchg |
2838 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2839 | */ |
d4d84fef CM |
2840 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2841 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2842 | |
2843 | if (!s->cpu_slab) | |
2844 | return 0; | |
2845 | ||
2846 | init_kmem_cache_cpus(s); | |
4c93c355 | 2847 | |
8a5ec0ba | 2848 | return 1; |
4c93c355 | 2849 | } |
4c93c355 | 2850 | |
51df1142 CL |
2851 | static struct kmem_cache *kmem_cache_node; |
2852 | ||
81819f0f CL |
2853 | /* |
2854 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2855 | * slab on the node for this slabcache. There are no concurrent accesses | |
2856 | * possible. | |
2857 | * | |
721ae22a ZYW |
2858 | * Note that this function only works on the kmem_cache_node |
2859 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 2860 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 2861 | */ |
55136592 | 2862 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2863 | { |
2864 | struct page *page; | |
2865 | struct kmem_cache_node *n; | |
2866 | ||
51df1142 | 2867 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2868 | |
51df1142 | 2869 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2870 | |
2871 | BUG_ON(!page); | |
a2f92ee7 | 2872 | if (page_to_nid(page) != node) { |
f9f58285 FF |
2873 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
2874 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
2875 | } |
2876 | ||
81819f0f CL |
2877 | n = page->freelist; |
2878 | BUG_ON(!n); | |
51df1142 | 2879 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2880 | page->inuse = 1; |
8cb0a506 | 2881 | page->frozen = 0; |
51df1142 | 2882 | kmem_cache_node->node[node] = n; |
8ab1372f | 2883 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2884 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2885 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2886 | #endif |
4053497d | 2887 | init_kmem_cache_node(n); |
51df1142 | 2888 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2889 | |
67b6c900 | 2890 | /* |
1e4dd946 SR |
2891 | * No locks need to be taken here as it has just been |
2892 | * initialized and there is no concurrent access. | |
67b6c900 | 2893 | */ |
1e4dd946 | 2894 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2895 | } |
2896 | ||
2897 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2898 | { | |
2899 | int node; | |
fa45dc25 | 2900 | struct kmem_cache_node *n; |
81819f0f | 2901 | |
fa45dc25 CL |
2902 | for_each_kmem_cache_node(s, node, n) { |
2903 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
2904 | s->node[node] = NULL; |
2905 | } | |
2906 | } | |
2907 | ||
55136592 | 2908 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2909 | { |
2910 | int node; | |
81819f0f | 2911 | |
f64dc58c | 2912 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2913 | struct kmem_cache_node *n; |
2914 | ||
73367bd8 | 2915 | if (slab_state == DOWN) { |
55136592 | 2916 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2917 | continue; |
2918 | } | |
51df1142 | 2919 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2920 | GFP_KERNEL, node); |
81819f0f | 2921 | |
73367bd8 AD |
2922 | if (!n) { |
2923 | free_kmem_cache_nodes(s); | |
2924 | return 0; | |
81819f0f | 2925 | } |
73367bd8 | 2926 | |
81819f0f | 2927 | s->node[node] = n; |
4053497d | 2928 | init_kmem_cache_node(n); |
81819f0f CL |
2929 | } |
2930 | return 1; | |
2931 | } | |
81819f0f | 2932 | |
c0bdb232 | 2933 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2934 | { |
2935 | if (min < MIN_PARTIAL) | |
2936 | min = MIN_PARTIAL; | |
2937 | else if (min > MAX_PARTIAL) | |
2938 | min = MAX_PARTIAL; | |
2939 | s->min_partial = min; | |
2940 | } | |
2941 | ||
81819f0f CL |
2942 | /* |
2943 | * calculate_sizes() determines the order and the distribution of data within | |
2944 | * a slab object. | |
2945 | */ | |
06b285dc | 2946 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2947 | { |
2948 | unsigned long flags = s->flags; | |
3b0efdfa | 2949 | unsigned long size = s->object_size; |
834f3d11 | 2950 | int order; |
81819f0f | 2951 | |
d8b42bf5 CL |
2952 | /* |
2953 | * Round up object size to the next word boundary. We can only | |
2954 | * place the free pointer at word boundaries and this determines | |
2955 | * the possible location of the free pointer. | |
2956 | */ | |
2957 | size = ALIGN(size, sizeof(void *)); | |
2958 | ||
2959 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2960 | /* |
2961 | * Determine if we can poison the object itself. If the user of | |
2962 | * the slab may touch the object after free or before allocation | |
2963 | * then we should never poison the object itself. | |
2964 | */ | |
2965 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2966 | !s->ctor) |
81819f0f CL |
2967 | s->flags |= __OBJECT_POISON; |
2968 | else | |
2969 | s->flags &= ~__OBJECT_POISON; | |
2970 | ||
81819f0f CL |
2971 | |
2972 | /* | |
672bba3a | 2973 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2974 | * end of the object and the free pointer. If not then add an |
672bba3a | 2975 | * additional word to have some bytes to store Redzone information. |
81819f0f | 2976 | */ |
3b0efdfa | 2977 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 2978 | size += sizeof(void *); |
41ecc55b | 2979 | #endif |
81819f0f CL |
2980 | |
2981 | /* | |
672bba3a CL |
2982 | * With that we have determined the number of bytes in actual use |
2983 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2984 | */ |
2985 | s->inuse = size; | |
2986 | ||
2987 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2988 | s->ctor)) { |
81819f0f CL |
2989 | /* |
2990 | * Relocate free pointer after the object if it is not | |
2991 | * permitted to overwrite the first word of the object on | |
2992 | * kmem_cache_free. | |
2993 | * | |
2994 | * This is the case if we do RCU, have a constructor or | |
2995 | * destructor or are poisoning the objects. | |
2996 | */ | |
2997 | s->offset = size; | |
2998 | size += sizeof(void *); | |
2999 | } | |
3000 | ||
c12b3c62 | 3001 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3002 | if (flags & SLAB_STORE_USER) |
3003 | /* | |
3004 | * Need to store information about allocs and frees after | |
3005 | * the object. | |
3006 | */ | |
3007 | size += 2 * sizeof(struct track); | |
3008 | ||
be7b3fbc | 3009 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3010 | /* |
3011 | * Add some empty padding so that we can catch | |
3012 | * overwrites from earlier objects rather than let | |
3013 | * tracking information or the free pointer be | |
0211a9c8 | 3014 | * corrupted if a user writes before the start |
81819f0f CL |
3015 | * of the object. |
3016 | */ | |
3017 | size += sizeof(void *); | |
41ecc55b | 3018 | #endif |
672bba3a | 3019 | |
81819f0f CL |
3020 | /* |
3021 | * SLUB stores one object immediately after another beginning from | |
3022 | * offset 0. In order to align the objects we have to simply size | |
3023 | * each object to conform to the alignment. | |
3024 | */ | |
45906855 | 3025 | size = ALIGN(size, s->align); |
81819f0f | 3026 | s->size = size; |
06b285dc CL |
3027 | if (forced_order >= 0) |
3028 | order = forced_order; | |
3029 | else | |
ab9a0f19 | 3030 | order = calculate_order(size, s->reserved); |
81819f0f | 3031 | |
834f3d11 | 3032 | if (order < 0) |
81819f0f CL |
3033 | return 0; |
3034 | ||
b7a49f0d | 3035 | s->allocflags = 0; |
834f3d11 | 3036 | if (order) |
b7a49f0d CL |
3037 | s->allocflags |= __GFP_COMP; |
3038 | ||
3039 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3040 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3041 | |
3042 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3043 | s->allocflags |= __GFP_RECLAIMABLE; | |
3044 | ||
81819f0f CL |
3045 | /* |
3046 | * Determine the number of objects per slab | |
3047 | */ | |
ab9a0f19 LJ |
3048 | s->oo = oo_make(order, size, s->reserved); |
3049 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3050 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3051 | s->max = s->oo; | |
81819f0f | 3052 | |
834f3d11 | 3053 | return !!oo_objects(s->oo); |
81819f0f CL |
3054 | } |
3055 | ||
8a13a4cc | 3056 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3057 | { |
8a13a4cc | 3058 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3059 | s->reserved = 0; |
81819f0f | 3060 | |
da9a638c LJ |
3061 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3062 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3063 | |
06b285dc | 3064 | if (!calculate_sizes(s, -1)) |
81819f0f | 3065 | goto error; |
3de47213 DR |
3066 | if (disable_higher_order_debug) { |
3067 | /* | |
3068 | * Disable debugging flags that store metadata if the min slab | |
3069 | * order increased. | |
3070 | */ | |
3b0efdfa | 3071 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3072 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3073 | s->offset = 0; | |
3074 | if (!calculate_sizes(s, -1)) | |
3075 | goto error; | |
3076 | } | |
3077 | } | |
81819f0f | 3078 | |
2565409f HC |
3079 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3080 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3081 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3082 | /* Enable fast mode */ | |
3083 | s->flags |= __CMPXCHG_DOUBLE; | |
3084 | #endif | |
3085 | ||
3b89d7d8 DR |
3086 | /* |
3087 | * The larger the object size is, the more pages we want on the partial | |
3088 | * list to avoid pounding the page allocator excessively. | |
3089 | */ | |
49e22585 CL |
3090 | set_min_partial(s, ilog2(s->size) / 2); |
3091 | ||
3092 | /* | |
3093 | * cpu_partial determined the maximum number of objects kept in the | |
3094 | * per cpu partial lists of a processor. | |
3095 | * | |
3096 | * Per cpu partial lists mainly contain slabs that just have one | |
3097 | * object freed. If they are used for allocation then they can be | |
3098 | * filled up again with minimal effort. The slab will never hit the | |
3099 | * per node partial lists and therefore no locking will be required. | |
3100 | * | |
3101 | * This setting also determines | |
3102 | * | |
3103 | * A) The number of objects from per cpu partial slabs dumped to the | |
3104 | * per node list when we reach the limit. | |
9f264904 | 3105 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3106 | * per node list when we run out of per cpu objects. We only fetch |
3107 | * 50% to keep some capacity around for frees. | |
49e22585 | 3108 | */ |
345c905d | 3109 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3110 | s->cpu_partial = 0; |
3111 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3112 | s->cpu_partial = 2; |
3113 | else if (s->size >= 1024) | |
3114 | s->cpu_partial = 6; | |
3115 | else if (s->size >= 256) | |
3116 | s->cpu_partial = 13; | |
3117 | else | |
3118 | s->cpu_partial = 30; | |
3119 | ||
81819f0f | 3120 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3121 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3122 | #endif |
55136592 | 3123 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3124 | goto error; |
81819f0f | 3125 | |
55136592 | 3126 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3127 | return 0; |
ff12059e | 3128 | |
4c93c355 | 3129 | free_kmem_cache_nodes(s); |
81819f0f CL |
3130 | error: |
3131 | if (flags & SLAB_PANIC) | |
3132 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3133 | "order=%u offset=%u flags=%lx\n", | |
d0e0ac97 CG |
3134 | s->name, (unsigned long)s->size, s->size, |
3135 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3136 | return -EINVAL; |
81819f0f | 3137 | } |
81819f0f | 3138 | |
33b12c38 CL |
3139 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3140 | const char *text) | |
3141 | { | |
3142 | #ifdef CONFIG_SLUB_DEBUG | |
3143 | void *addr = page_address(page); | |
3144 | void *p; | |
a5dd5c11 NK |
3145 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3146 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3147 | if (!map) |
3148 | return; | |
945cf2b6 | 3149 | slab_err(s, page, text, s->name); |
33b12c38 | 3150 | slab_lock(page); |
33b12c38 | 3151 | |
5f80b13a | 3152 | get_map(s, page, map); |
33b12c38 CL |
3153 | for_each_object(p, s, addr, page->objects) { |
3154 | ||
3155 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3156 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3157 | print_tracking(s, p); |
3158 | } | |
3159 | } | |
3160 | slab_unlock(page); | |
bbd7d57b | 3161 | kfree(map); |
33b12c38 CL |
3162 | #endif |
3163 | } | |
3164 | ||
81819f0f | 3165 | /* |
599870b1 | 3166 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3167 | * This is called from kmem_cache_close(). We must be the last thread |
3168 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3169 | */ |
599870b1 | 3170 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3171 | { |
81819f0f CL |
3172 | struct page *page, *h; |
3173 | ||
33b12c38 | 3174 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3175 | if (!page->inuse) { |
1e4dd946 | 3176 | __remove_partial(n, page); |
81819f0f | 3177 | discard_slab(s, page); |
33b12c38 CL |
3178 | } else { |
3179 | list_slab_objects(s, page, | |
945cf2b6 | 3180 | "Objects remaining in %s on kmem_cache_close()"); |
599870b1 | 3181 | } |
33b12c38 | 3182 | } |
81819f0f CL |
3183 | } |
3184 | ||
3185 | /* | |
672bba3a | 3186 | * Release all resources used by a slab cache. |
81819f0f | 3187 | */ |
0c710013 | 3188 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3189 | { |
3190 | int node; | |
fa45dc25 | 3191 | struct kmem_cache_node *n; |
81819f0f CL |
3192 | |
3193 | flush_all(s); | |
81819f0f | 3194 | /* Attempt to free all objects */ |
fa45dc25 | 3195 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3196 | free_partial(s, n); |
3197 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3198 | return 1; |
3199 | } | |
945cf2b6 | 3200 | free_percpu(s->cpu_slab); |
81819f0f CL |
3201 | free_kmem_cache_nodes(s); |
3202 | return 0; | |
3203 | } | |
3204 | ||
945cf2b6 | 3205 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f | 3206 | { |
41a21285 | 3207 | return kmem_cache_close(s); |
81819f0f | 3208 | } |
81819f0f CL |
3209 | |
3210 | /******************************************************************** | |
3211 | * Kmalloc subsystem | |
3212 | *******************************************************************/ | |
3213 | ||
81819f0f CL |
3214 | static int __init setup_slub_min_order(char *str) |
3215 | { | |
06428780 | 3216 | get_option(&str, &slub_min_order); |
81819f0f CL |
3217 | |
3218 | return 1; | |
3219 | } | |
3220 | ||
3221 | __setup("slub_min_order=", setup_slub_min_order); | |
3222 | ||
3223 | static int __init setup_slub_max_order(char *str) | |
3224 | { | |
06428780 | 3225 | get_option(&str, &slub_max_order); |
818cf590 | 3226 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3227 | |
3228 | return 1; | |
3229 | } | |
3230 | ||
3231 | __setup("slub_max_order=", setup_slub_max_order); | |
3232 | ||
3233 | static int __init setup_slub_min_objects(char *str) | |
3234 | { | |
06428780 | 3235 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3236 | |
3237 | return 1; | |
3238 | } | |
3239 | ||
3240 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3241 | ||
81819f0f CL |
3242 | void *__kmalloc(size_t size, gfp_t flags) |
3243 | { | |
aadb4bc4 | 3244 | struct kmem_cache *s; |
5b882be4 | 3245 | void *ret; |
81819f0f | 3246 | |
95a05b42 | 3247 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3248 | return kmalloc_large(size, flags); |
aadb4bc4 | 3249 | |
2c59dd65 | 3250 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3251 | |
3252 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3253 | return s; |
3254 | ||
2b847c3c | 3255 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3256 | |
ca2b84cb | 3257 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3258 | |
3259 | return ret; | |
81819f0f CL |
3260 | } |
3261 | EXPORT_SYMBOL(__kmalloc); | |
3262 | ||
5d1f57e4 | 3263 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3264 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3265 | { | |
b1eeab67 | 3266 | struct page *page; |
e4f7c0b4 | 3267 | void *ptr = NULL; |
f619cfe1 | 3268 | |
52383431 VD |
3269 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3270 | page = alloc_kmem_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3271 | if (page) |
e4f7c0b4 CM |
3272 | ptr = page_address(page); |
3273 | ||
d56791b3 | 3274 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3275 | return ptr; |
f619cfe1 CL |
3276 | } |
3277 | ||
81819f0f CL |
3278 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3279 | { | |
aadb4bc4 | 3280 | struct kmem_cache *s; |
5b882be4 | 3281 | void *ret; |
81819f0f | 3282 | |
95a05b42 | 3283 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3284 | ret = kmalloc_large_node(size, flags, node); |
3285 | ||
ca2b84cb EGM |
3286 | trace_kmalloc_node(_RET_IP_, ret, |
3287 | size, PAGE_SIZE << get_order(size), | |
3288 | flags, node); | |
5b882be4 EGM |
3289 | |
3290 | return ret; | |
3291 | } | |
aadb4bc4 | 3292 | |
2c59dd65 | 3293 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3294 | |
3295 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3296 | return s; |
3297 | ||
2b847c3c | 3298 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3299 | |
ca2b84cb | 3300 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3301 | |
3302 | return ret; | |
81819f0f CL |
3303 | } |
3304 | EXPORT_SYMBOL(__kmalloc_node); | |
3305 | #endif | |
3306 | ||
3307 | size_t ksize(const void *object) | |
3308 | { | |
272c1d21 | 3309 | struct page *page; |
81819f0f | 3310 | |
ef8b4520 | 3311 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3312 | return 0; |
3313 | ||
294a80a8 | 3314 | page = virt_to_head_page(object); |
294a80a8 | 3315 | |
76994412 PE |
3316 | if (unlikely(!PageSlab(page))) { |
3317 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3318 | return PAGE_SIZE << compound_order(page); |
76994412 | 3319 | } |
81819f0f | 3320 | |
1b4f59e3 | 3321 | return slab_ksize(page->slab_cache); |
81819f0f | 3322 | } |
b1aabecd | 3323 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3324 | |
3325 | void kfree(const void *x) | |
3326 | { | |
81819f0f | 3327 | struct page *page; |
5bb983b0 | 3328 | void *object = (void *)x; |
81819f0f | 3329 | |
2121db74 PE |
3330 | trace_kfree(_RET_IP_, x); |
3331 | ||
2408c550 | 3332 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3333 | return; |
3334 | ||
b49af68f | 3335 | page = virt_to_head_page(x); |
aadb4bc4 | 3336 | if (unlikely(!PageSlab(page))) { |
0937502a | 3337 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3338 | kfree_hook(x); |
52383431 | 3339 | __free_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3340 | return; |
3341 | } | |
1b4f59e3 | 3342 | slab_free(page->slab_cache, page, object, _RET_IP_); |
81819f0f CL |
3343 | } |
3344 | EXPORT_SYMBOL(kfree); | |
3345 | ||
2086d26a | 3346 | /* |
672bba3a CL |
3347 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3348 | * the remaining slabs by the number of items in use. The slabs with the | |
3349 | * most items in use come first. New allocations will then fill those up | |
3350 | * and thus they can be removed from the partial lists. | |
3351 | * | |
3352 | * The slabs with the least items are placed last. This results in them | |
3353 | * being allocated from last increasing the chance that the last objects | |
3354 | * are freed in them. | |
2086d26a | 3355 | */ |
03afc0e2 | 3356 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
3357 | { |
3358 | int node; | |
3359 | int i; | |
3360 | struct kmem_cache_node *n; | |
3361 | struct page *page; | |
3362 | struct page *t; | |
205ab99d | 3363 | int objects = oo_objects(s->max); |
2086d26a | 3364 | struct list_head *slabs_by_inuse = |
834f3d11 | 3365 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3366 | unsigned long flags; |
3367 | ||
3368 | if (!slabs_by_inuse) | |
3369 | return -ENOMEM; | |
3370 | ||
3371 | flush_all(s); | |
fa45dc25 | 3372 | for_each_kmem_cache_node(s, node, n) { |
2086d26a CL |
3373 | if (!n->nr_partial) |
3374 | continue; | |
3375 | ||
834f3d11 | 3376 | for (i = 0; i < objects; i++) |
2086d26a CL |
3377 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3378 | ||
3379 | spin_lock_irqsave(&n->list_lock, flags); | |
3380 | ||
3381 | /* | |
672bba3a | 3382 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3383 | * |
672bba3a CL |
3384 | * Note that concurrent frees may occur while we hold the |
3385 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3386 | */ |
3387 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3388 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3389 | if (!page->inuse) | |
3390 | n->nr_partial--; | |
2086d26a CL |
3391 | } |
3392 | ||
2086d26a | 3393 | /* |
672bba3a CL |
3394 | * Rebuild the partial list with the slabs filled up most |
3395 | * first and the least used slabs at the end. | |
2086d26a | 3396 | */ |
69cb8e6b | 3397 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3398 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3399 | ||
2086d26a | 3400 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3401 | |
3402 | /* Release empty slabs */ | |
3403 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3404 | discard_slab(s, page); | |
2086d26a CL |
3405 | } |
3406 | ||
3407 | kfree(slabs_by_inuse); | |
3408 | return 0; | |
3409 | } | |
2086d26a | 3410 | |
b9049e23 YG |
3411 | static int slab_mem_going_offline_callback(void *arg) |
3412 | { | |
3413 | struct kmem_cache *s; | |
3414 | ||
18004c5d | 3415 | mutex_lock(&slab_mutex); |
b9049e23 | 3416 | list_for_each_entry(s, &slab_caches, list) |
03afc0e2 | 3417 | __kmem_cache_shrink(s); |
18004c5d | 3418 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3419 | |
3420 | return 0; | |
3421 | } | |
3422 | ||
3423 | static void slab_mem_offline_callback(void *arg) | |
3424 | { | |
3425 | struct kmem_cache_node *n; | |
3426 | struct kmem_cache *s; | |
3427 | struct memory_notify *marg = arg; | |
3428 | int offline_node; | |
3429 | ||
b9d5ab25 | 3430 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3431 | |
3432 | /* | |
3433 | * If the node still has available memory. we need kmem_cache_node | |
3434 | * for it yet. | |
3435 | */ | |
3436 | if (offline_node < 0) | |
3437 | return; | |
3438 | ||
18004c5d | 3439 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3440 | list_for_each_entry(s, &slab_caches, list) { |
3441 | n = get_node(s, offline_node); | |
3442 | if (n) { | |
3443 | /* | |
3444 | * if n->nr_slabs > 0, slabs still exist on the node | |
3445 | * that is going down. We were unable to free them, | |
c9404c9c | 3446 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3447 | * callback. So, we must fail. |
3448 | */ | |
0f389ec6 | 3449 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3450 | |
3451 | s->node[offline_node] = NULL; | |
8de66a0c | 3452 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3453 | } |
3454 | } | |
18004c5d | 3455 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3456 | } |
3457 | ||
3458 | static int slab_mem_going_online_callback(void *arg) | |
3459 | { | |
3460 | struct kmem_cache_node *n; | |
3461 | struct kmem_cache *s; | |
3462 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3463 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3464 | int ret = 0; |
3465 | ||
3466 | /* | |
3467 | * If the node's memory is already available, then kmem_cache_node is | |
3468 | * already created. Nothing to do. | |
3469 | */ | |
3470 | if (nid < 0) | |
3471 | return 0; | |
3472 | ||
3473 | /* | |
0121c619 | 3474 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3475 | * allocate a kmem_cache_node structure in order to bring the node |
3476 | * online. | |
3477 | */ | |
18004c5d | 3478 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3479 | list_for_each_entry(s, &slab_caches, list) { |
3480 | /* | |
3481 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3482 | * since memory is not yet available from the node that | |
3483 | * is brought up. | |
3484 | */ | |
8de66a0c | 3485 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3486 | if (!n) { |
3487 | ret = -ENOMEM; | |
3488 | goto out; | |
3489 | } | |
4053497d | 3490 | init_kmem_cache_node(n); |
b9049e23 YG |
3491 | s->node[nid] = n; |
3492 | } | |
3493 | out: | |
18004c5d | 3494 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3495 | return ret; |
3496 | } | |
3497 | ||
3498 | static int slab_memory_callback(struct notifier_block *self, | |
3499 | unsigned long action, void *arg) | |
3500 | { | |
3501 | int ret = 0; | |
3502 | ||
3503 | switch (action) { | |
3504 | case MEM_GOING_ONLINE: | |
3505 | ret = slab_mem_going_online_callback(arg); | |
3506 | break; | |
3507 | case MEM_GOING_OFFLINE: | |
3508 | ret = slab_mem_going_offline_callback(arg); | |
3509 | break; | |
3510 | case MEM_OFFLINE: | |
3511 | case MEM_CANCEL_ONLINE: | |
3512 | slab_mem_offline_callback(arg); | |
3513 | break; | |
3514 | case MEM_ONLINE: | |
3515 | case MEM_CANCEL_OFFLINE: | |
3516 | break; | |
3517 | } | |
dc19f9db KH |
3518 | if (ret) |
3519 | ret = notifier_from_errno(ret); | |
3520 | else | |
3521 | ret = NOTIFY_OK; | |
b9049e23 YG |
3522 | return ret; |
3523 | } | |
3524 | ||
3ac38faa AM |
3525 | static struct notifier_block slab_memory_callback_nb = { |
3526 | .notifier_call = slab_memory_callback, | |
3527 | .priority = SLAB_CALLBACK_PRI, | |
3528 | }; | |
b9049e23 | 3529 | |
81819f0f CL |
3530 | /******************************************************************** |
3531 | * Basic setup of slabs | |
3532 | *******************************************************************/ | |
3533 | ||
51df1142 CL |
3534 | /* |
3535 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3536 | * the page allocator. Allocate them properly then fix up the pointers |
3537 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3538 | */ |
3539 | ||
dffb4d60 | 3540 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3541 | { |
3542 | int node; | |
dffb4d60 | 3543 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 3544 | struct kmem_cache_node *n; |
51df1142 | 3545 | |
dffb4d60 | 3546 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3547 | |
7d557b3c GC |
3548 | /* |
3549 | * This runs very early, and only the boot processor is supposed to be | |
3550 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3551 | * IPIs around. | |
3552 | */ | |
3553 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 3554 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
3555 | struct page *p; |
3556 | ||
fa45dc25 CL |
3557 | list_for_each_entry(p, &n->partial, lru) |
3558 | p->slab_cache = s; | |
51df1142 | 3559 | |
607bf324 | 3560 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
3561 | list_for_each_entry(p, &n->full, lru) |
3562 | p->slab_cache = s; | |
51df1142 | 3563 | #endif |
51df1142 | 3564 | } |
dffb4d60 CL |
3565 | list_add(&s->list, &slab_caches); |
3566 | return s; | |
51df1142 CL |
3567 | } |
3568 | ||
81819f0f CL |
3569 | void __init kmem_cache_init(void) |
3570 | { | |
dffb4d60 CL |
3571 | static __initdata struct kmem_cache boot_kmem_cache, |
3572 | boot_kmem_cache_node; | |
51df1142 | 3573 | |
fc8d8620 SG |
3574 | if (debug_guardpage_minorder()) |
3575 | slub_max_order = 0; | |
3576 | ||
dffb4d60 CL |
3577 | kmem_cache_node = &boot_kmem_cache_node; |
3578 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3579 | |
dffb4d60 CL |
3580 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3581 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3582 | |
3ac38faa | 3583 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3584 | |
3585 | /* Able to allocate the per node structures */ | |
3586 | slab_state = PARTIAL; | |
3587 | ||
dffb4d60 CL |
3588 | create_boot_cache(kmem_cache, "kmem_cache", |
3589 | offsetof(struct kmem_cache, node) + | |
3590 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3591 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3592 | |
dffb4d60 | 3593 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3594 | |
51df1142 CL |
3595 | /* |
3596 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3597 | * kmem_cache_node is separately allocated so no need to | |
3598 | * update any list pointers. | |
3599 | */ | |
dffb4d60 | 3600 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3601 | |
3602 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f97d5f63 | 3603 | create_kmalloc_caches(0); |
81819f0f CL |
3604 | |
3605 | #ifdef CONFIG_SMP | |
3606 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3607 | #endif |
81819f0f | 3608 | |
f9f58285 | 3609 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3610 | cache_line_size(), |
81819f0f CL |
3611 | slub_min_order, slub_max_order, slub_min_objects, |
3612 | nr_cpu_ids, nr_node_ids); | |
3613 | } | |
3614 | ||
7e85ee0c PE |
3615 | void __init kmem_cache_init_late(void) |
3616 | { | |
7e85ee0c PE |
3617 | } |
3618 | ||
2633d7a0 | 3619 | struct kmem_cache * |
a44cb944 VD |
3620 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3621 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f CL |
3622 | { |
3623 | struct kmem_cache *s; | |
3624 | ||
a44cb944 | 3625 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3626 | if (s) { |
84d0ddd6 VD |
3627 | int i; |
3628 | struct kmem_cache *c; | |
3629 | ||
81819f0f | 3630 | s->refcount++; |
84d0ddd6 | 3631 | |
81819f0f CL |
3632 | /* |
3633 | * Adjust the object sizes so that we clear | |
3634 | * the complete object on kzalloc. | |
3635 | */ | |
3b0efdfa | 3636 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3637 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3638 | |
84d0ddd6 VD |
3639 | for_each_memcg_cache_index(i) { |
3640 | c = cache_from_memcg_idx(s, i); | |
3641 | if (!c) | |
3642 | continue; | |
3643 | c->object_size = s->object_size; | |
3644 | c->inuse = max_t(int, c->inuse, | |
3645 | ALIGN(size, sizeof(void *))); | |
3646 | } | |
3647 | ||
7b8f3b66 | 3648 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3649 | s->refcount--; |
cbb79694 | 3650 | s = NULL; |
7b8f3b66 | 3651 | } |
a0e1d1be | 3652 | } |
6446faa2 | 3653 | |
cbb79694 CL |
3654 | return s; |
3655 | } | |
84c1cf62 | 3656 | |
8a13a4cc | 3657 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3658 | { |
aac3a166 PE |
3659 | int err; |
3660 | ||
3661 | err = kmem_cache_open(s, flags); | |
3662 | if (err) | |
3663 | return err; | |
20cea968 | 3664 | |
45530c44 CL |
3665 | /* Mutex is not taken during early boot */ |
3666 | if (slab_state <= UP) | |
3667 | return 0; | |
3668 | ||
107dab5c | 3669 | memcg_propagate_slab_attrs(s); |
aac3a166 | 3670 | err = sysfs_slab_add(s); |
aac3a166 PE |
3671 | if (err) |
3672 | kmem_cache_close(s); | |
20cea968 | 3673 | |
aac3a166 | 3674 | return err; |
81819f0f | 3675 | } |
81819f0f | 3676 | |
81819f0f | 3677 | #ifdef CONFIG_SMP |
81819f0f | 3678 | /* |
672bba3a CL |
3679 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3680 | * necessary. | |
81819f0f | 3681 | */ |
0db0628d | 3682 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3683 | unsigned long action, void *hcpu) |
3684 | { | |
3685 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3686 | struct kmem_cache *s; |
3687 | unsigned long flags; | |
81819f0f CL |
3688 | |
3689 | switch (action) { | |
3690 | case CPU_UP_CANCELED: | |
8bb78442 | 3691 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3692 | case CPU_DEAD: |
8bb78442 | 3693 | case CPU_DEAD_FROZEN: |
18004c5d | 3694 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3695 | list_for_each_entry(s, &slab_caches, list) { |
3696 | local_irq_save(flags); | |
3697 | __flush_cpu_slab(s, cpu); | |
3698 | local_irq_restore(flags); | |
3699 | } | |
18004c5d | 3700 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3701 | break; |
3702 | default: | |
3703 | break; | |
3704 | } | |
3705 | return NOTIFY_OK; | |
3706 | } | |
3707 | ||
0db0628d | 3708 | static struct notifier_block slab_notifier = { |
3adbefee | 3709 | .notifier_call = slab_cpuup_callback |
06428780 | 3710 | }; |
81819f0f CL |
3711 | |
3712 | #endif | |
3713 | ||
ce71e27c | 3714 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3715 | { |
aadb4bc4 | 3716 | struct kmem_cache *s; |
94b528d0 | 3717 | void *ret; |
aadb4bc4 | 3718 | |
95a05b42 | 3719 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
3720 | return kmalloc_large(size, gfpflags); |
3721 | ||
2c59dd65 | 3722 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3723 | |
2408c550 | 3724 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3725 | return s; |
81819f0f | 3726 | |
2b847c3c | 3727 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 3728 | |
25985edc | 3729 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3730 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3731 | |
3732 | return ret; | |
81819f0f CL |
3733 | } |
3734 | ||
5d1f57e4 | 3735 | #ifdef CONFIG_NUMA |
81819f0f | 3736 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3737 | int node, unsigned long caller) |
81819f0f | 3738 | { |
aadb4bc4 | 3739 | struct kmem_cache *s; |
94b528d0 | 3740 | void *ret; |
aadb4bc4 | 3741 | |
95a05b42 | 3742 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
3743 | ret = kmalloc_large_node(size, gfpflags, node); |
3744 | ||
3745 | trace_kmalloc_node(caller, ret, | |
3746 | size, PAGE_SIZE << get_order(size), | |
3747 | gfpflags, node); | |
3748 | ||
3749 | return ret; | |
3750 | } | |
eada35ef | 3751 | |
2c59dd65 | 3752 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3753 | |
2408c550 | 3754 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3755 | return s; |
81819f0f | 3756 | |
2b847c3c | 3757 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 3758 | |
25985edc | 3759 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3760 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3761 | |
3762 | return ret; | |
81819f0f | 3763 | } |
5d1f57e4 | 3764 | #endif |
81819f0f | 3765 | |
ab4d5ed5 | 3766 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3767 | static int count_inuse(struct page *page) |
3768 | { | |
3769 | return page->inuse; | |
3770 | } | |
3771 | ||
3772 | static int count_total(struct page *page) | |
3773 | { | |
3774 | return page->objects; | |
3775 | } | |
ab4d5ed5 | 3776 | #endif |
205ab99d | 3777 | |
ab4d5ed5 | 3778 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3779 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3780 | unsigned long *map) | |
53e15af0 CL |
3781 | { |
3782 | void *p; | |
a973e9dd | 3783 | void *addr = page_address(page); |
53e15af0 CL |
3784 | |
3785 | if (!check_slab(s, page) || | |
3786 | !on_freelist(s, page, NULL)) | |
3787 | return 0; | |
3788 | ||
3789 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3790 | bitmap_zero(map, page->objects); |
53e15af0 | 3791 | |
5f80b13a CL |
3792 | get_map(s, page, map); |
3793 | for_each_object(p, s, addr, page->objects) { | |
3794 | if (test_bit(slab_index(p, s, addr), map)) | |
3795 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3796 | return 0; | |
53e15af0 CL |
3797 | } |
3798 | ||
224a88be | 3799 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3800 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3801 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3802 | return 0; |
3803 | return 1; | |
3804 | } | |
3805 | ||
434e245d CL |
3806 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3807 | unsigned long *map) | |
53e15af0 | 3808 | { |
881db7fb CL |
3809 | slab_lock(page); |
3810 | validate_slab(s, page, map); | |
3811 | slab_unlock(page); | |
53e15af0 CL |
3812 | } |
3813 | ||
434e245d CL |
3814 | static int validate_slab_node(struct kmem_cache *s, |
3815 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3816 | { |
3817 | unsigned long count = 0; | |
3818 | struct page *page; | |
3819 | unsigned long flags; | |
3820 | ||
3821 | spin_lock_irqsave(&n->list_lock, flags); | |
3822 | ||
3823 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3824 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3825 | count++; |
3826 | } | |
3827 | if (count != n->nr_partial) | |
f9f58285 FF |
3828 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
3829 | s->name, count, n->nr_partial); | |
53e15af0 CL |
3830 | |
3831 | if (!(s->flags & SLAB_STORE_USER)) | |
3832 | goto out; | |
3833 | ||
3834 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3835 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3836 | count++; |
3837 | } | |
3838 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
3839 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
3840 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
3841 | |
3842 | out: | |
3843 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3844 | return count; | |
3845 | } | |
3846 | ||
434e245d | 3847 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3848 | { |
3849 | int node; | |
3850 | unsigned long count = 0; | |
205ab99d | 3851 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 3852 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 3853 | struct kmem_cache_node *n; |
434e245d CL |
3854 | |
3855 | if (!map) | |
3856 | return -ENOMEM; | |
53e15af0 CL |
3857 | |
3858 | flush_all(s); | |
fa45dc25 | 3859 | for_each_kmem_cache_node(s, node, n) |
434e245d | 3860 | count += validate_slab_node(s, n, map); |
434e245d | 3861 | kfree(map); |
53e15af0 CL |
3862 | return count; |
3863 | } | |
88a420e4 | 3864 | /* |
672bba3a | 3865 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3866 | * and freed. |
3867 | */ | |
3868 | ||
3869 | struct location { | |
3870 | unsigned long count; | |
ce71e27c | 3871 | unsigned long addr; |
45edfa58 CL |
3872 | long long sum_time; |
3873 | long min_time; | |
3874 | long max_time; | |
3875 | long min_pid; | |
3876 | long max_pid; | |
174596a0 | 3877 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3878 | nodemask_t nodes; |
88a420e4 CL |
3879 | }; |
3880 | ||
3881 | struct loc_track { | |
3882 | unsigned long max; | |
3883 | unsigned long count; | |
3884 | struct location *loc; | |
3885 | }; | |
3886 | ||
3887 | static void free_loc_track(struct loc_track *t) | |
3888 | { | |
3889 | if (t->max) | |
3890 | free_pages((unsigned long)t->loc, | |
3891 | get_order(sizeof(struct location) * t->max)); | |
3892 | } | |
3893 | ||
68dff6a9 | 3894 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3895 | { |
3896 | struct location *l; | |
3897 | int order; | |
3898 | ||
88a420e4 CL |
3899 | order = get_order(sizeof(struct location) * max); |
3900 | ||
68dff6a9 | 3901 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3902 | if (!l) |
3903 | return 0; | |
3904 | ||
3905 | if (t->count) { | |
3906 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3907 | free_loc_track(t); | |
3908 | } | |
3909 | t->max = max; | |
3910 | t->loc = l; | |
3911 | return 1; | |
3912 | } | |
3913 | ||
3914 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3915 | const struct track *track) |
88a420e4 CL |
3916 | { |
3917 | long start, end, pos; | |
3918 | struct location *l; | |
ce71e27c | 3919 | unsigned long caddr; |
45edfa58 | 3920 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3921 | |
3922 | start = -1; | |
3923 | end = t->count; | |
3924 | ||
3925 | for ( ; ; ) { | |
3926 | pos = start + (end - start + 1) / 2; | |
3927 | ||
3928 | /* | |
3929 | * There is nothing at "end". If we end up there | |
3930 | * we need to add something to before end. | |
3931 | */ | |
3932 | if (pos == end) | |
3933 | break; | |
3934 | ||
3935 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3936 | if (track->addr == caddr) { |
3937 | ||
3938 | l = &t->loc[pos]; | |
3939 | l->count++; | |
3940 | if (track->when) { | |
3941 | l->sum_time += age; | |
3942 | if (age < l->min_time) | |
3943 | l->min_time = age; | |
3944 | if (age > l->max_time) | |
3945 | l->max_time = age; | |
3946 | ||
3947 | if (track->pid < l->min_pid) | |
3948 | l->min_pid = track->pid; | |
3949 | if (track->pid > l->max_pid) | |
3950 | l->max_pid = track->pid; | |
3951 | ||
174596a0 RR |
3952 | cpumask_set_cpu(track->cpu, |
3953 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3954 | } |
3955 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3956 | return 1; |
3957 | } | |
3958 | ||
45edfa58 | 3959 | if (track->addr < caddr) |
88a420e4 CL |
3960 | end = pos; |
3961 | else | |
3962 | start = pos; | |
3963 | } | |
3964 | ||
3965 | /* | |
672bba3a | 3966 | * Not found. Insert new tracking element. |
88a420e4 | 3967 | */ |
68dff6a9 | 3968 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3969 | return 0; |
3970 | ||
3971 | l = t->loc + pos; | |
3972 | if (pos < t->count) | |
3973 | memmove(l + 1, l, | |
3974 | (t->count - pos) * sizeof(struct location)); | |
3975 | t->count++; | |
3976 | l->count = 1; | |
45edfa58 CL |
3977 | l->addr = track->addr; |
3978 | l->sum_time = age; | |
3979 | l->min_time = age; | |
3980 | l->max_time = age; | |
3981 | l->min_pid = track->pid; | |
3982 | l->max_pid = track->pid; | |
174596a0 RR |
3983 | cpumask_clear(to_cpumask(l->cpus)); |
3984 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3985 | nodes_clear(l->nodes); |
3986 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3987 | return 1; |
3988 | } | |
3989 | ||
3990 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 3991 | struct page *page, enum track_item alloc, |
a5dd5c11 | 3992 | unsigned long *map) |
88a420e4 | 3993 | { |
a973e9dd | 3994 | void *addr = page_address(page); |
88a420e4 CL |
3995 | void *p; |
3996 | ||
39b26464 | 3997 | bitmap_zero(map, page->objects); |
5f80b13a | 3998 | get_map(s, page, map); |
88a420e4 | 3999 | |
224a88be | 4000 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4001 | if (!test_bit(slab_index(p, s, addr), map)) |
4002 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4003 | } |
4004 | ||
4005 | static int list_locations(struct kmem_cache *s, char *buf, | |
4006 | enum track_item alloc) | |
4007 | { | |
e374d483 | 4008 | int len = 0; |
88a420e4 | 4009 | unsigned long i; |
68dff6a9 | 4010 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4011 | int node; |
bbd7d57b ED |
4012 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4013 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4014 | struct kmem_cache_node *n; |
88a420e4 | 4015 | |
bbd7d57b ED |
4016 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4017 | GFP_TEMPORARY)) { | |
4018 | kfree(map); | |
68dff6a9 | 4019 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4020 | } |
88a420e4 CL |
4021 | /* Push back cpu slabs */ |
4022 | flush_all(s); | |
4023 | ||
fa45dc25 | 4024 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4025 | unsigned long flags; |
4026 | struct page *page; | |
4027 | ||
9e86943b | 4028 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4029 | continue; |
4030 | ||
4031 | spin_lock_irqsave(&n->list_lock, flags); | |
4032 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4033 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4034 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4035 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4036 | spin_unlock_irqrestore(&n->list_lock, flags); |
4037 | } | |
4038 | ||
4039 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4040 | struct location *l = &t.loc[i]; |
88a420e4 | 4041 | |
9c246247 | 4042 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4043 | break; |
e374d483 | 4044 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4045 | |
4046 | if (l->addr) | |
62c70bce | 4047 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4048 | else |
e374d483 | 4049 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4050 | |
4051 | if (l->sum_time != l->min_time) { | |
e374d483 | 4052 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4053 | l->min_time, |
4054 | (long)div_u64(l->sum_time, l->count), | |
4055 | l->max_time); | |
45edfa58 | 4056 | } else |
e374d483 | 4057 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4058 | l->min_time); |
4059 | ||
4060 | if (l->min_pid != l->max_pid) | |
e374d483 | 4061 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4062 | l->min_pid, l->max_pid); |
4063 | else | |
e374d483 | 4064 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4065 | l->min_pid); |
4066 | ||
174596a0 RR |
4067 | if (num_online_cpus() > 1 && |
4068 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4069 | len < PAGE_SIZE - 60) { |
4070 | len += sprintf(buf + len, " cpus="); | |
d0e0ac97 CG |
4071 | len += cpulist_scnprintf(buf + len, |
4072 | PAGE_SIZE - len - 50, | |
174596a0 | 4073 | to_cpumask(l->cpus)); |
45edfa58 CL |
4074 | } |
4075 | ||
62bc62a8 | 4076 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4077 | len < PAGE_SIZE - 60) { |
4078 | len += sprintf(buf + len, " nodes="); | |
d0e0ac97 CG |
4079 | len += nodelist_scnprintf(buf + len, |
4080 | PAGE_SIZE - len - 50, | |
4081 | l->nodes); | |
45edfa58 CL |
4082 | } |
4083 | ||
e374d483 | 4084 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4085 | } |
4086 | ||
4087 | free_loc_track(&t); | |
bbd7d57b | 4088 | kfree(map); |
88a420e4 | 4089 | if (!t.count) |
e374d483 HH |
4090 | len += sprintf(buf, "No data\n"); |
4091 | return len; | |
88a420e4 | 4092 | } |
ab4d5ed5 | 4093 | #endif |
88a420e4 | 4094 | |
a5a84755 | 4095 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4096 | static void __init resiliency_test(void) |
a5a84755 CL |
4097 | { |
4098 | u8 *p; | |
4099 | ||
95a05b42 | 4100 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4101 | |
f9f58285 FF |
4102 | pr_err("SLUB resiliency testing\n"); |
4103 | pr_err("-----------------------\n"); | |
4104 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4105 | |
4106 | p = kzalloc(16, GFP_KERNEL); | |
4107 | p[16] = 0x12; | |
f9f58285 FF |
4108 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4109 | p + 16); | |
a5a84755 CL |
4110 | |
4111 | validate_slab_cache(kmalloc_caches[4]); | |
4112 | ||
4113 | /* Hmmm... The next two are dangerous */ | |
4114 | p = kzalloc(32, GFP_KERNEL); | |
4115 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4116 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4117 | p); | |
4118 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4119 | |
4120 | validate_slab_cache(kmalloc_caches[5]); | |
4121 | p = kzalloc(64, GFP_KERNEL); | |
4122 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4123 | *p = 0x56; | |
f9f58285 FF |
4124 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4125 | p); | |
4126 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4127 | validate_slab_cache(kmalloc_caches[6]); |
4128 | ||
f9f58285 | 4129 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4130 | p = kzalloc(128, GFP_KERNEL); |
4131 | kfree(p); | |
4132 | *p = 0x78; | |
f9f58285 | 4133 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4134 | validate_slab_cache(kmalloc_caches[7]); |
4135 | ||
4136 | p = kzalloc(256, GFP_KERNEL); | |
4137 | kfree(p); | |
4138 | p[50] = 0x9a; | |
f9f58285 | 4139 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4140 | validate_slab_cache(kmalloc_caches[8]); |
4141 | ||
4142 | p = kzalloc(512, GFP_KERNEL); | |
4143 | kfree(p); | |
4144 | p[512] = 0xab; | |
f9f58285 | 4145 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4146 | validate_slab_cache(kmalloc_caches[9]); |
4147 | } | |
4148 | #else | |
4149 | #ifdef CONFIG_SYSFS | |
4150 | static void resiliency_test(void) {}; | |
4151 | #endif | |
4152 | #endif | |
4153 | ||
ab4d5ed5 | 4154 | #ifdef CONFIG_SYSFS |
81819f0f | 4155 | enum slab_stat_type { |
205ab99d CL |
4156 | SL_ALL, /* All slabs */ |
4157 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4158 | SL_CPU, /* Only slabs used for cpu caches */ | |
4159 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4160 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4161 | }; |
4162 | ||
205ab99d | 4163 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4164 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4165 | #define SO_CPU (1 << SL_CPU) | |
4166 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4167 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4168 | |
62e5c4b4 CG |
4169 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4170 | char *buf, unsigned long flags) | |
81819f0f CL |
4171 | { |
4172 | unsigned long total = 0; | |
81819f0f CL |
4173 | int node; |
4174 | int x; | |
4175 | unsigned long *nodes; | |
81819f0f | 4176 | |
e35e1a97 | 4177 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4178 | if (!nodes) |
4179 | return -ENOMEM; | |
81819f0f | 4180 | |
205ab99d CL |
4181 | if (flags & SO_CPU) { |
4182 | int cpu; | |
81819f0f | 4183 | |
205ab99d | 4184 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4185 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4186 | cpu); | |
ec3ab083 | 4187 | int node; |
49e22585 | 4188 | struct page *page; |
dfb4f096 | 4189 | |
bc6697d8 | 4190 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4191 | if (!page) |
4192 | continue; | |
205ab99d | 4193 | |
ec3ab083 CL |
4194 | node = page_to_nid(page); |
4195 | if (flags & SO_TOTAL) | |
4196 | x = page->objects; | |
4197 | else if (flags & SO_OBJECTS) | |
4198 | x = page->inuse; | |
4199 | else | |
4200 | x = 1; | |
49e22585 | 4201 | |
ec3ab083 CL |
4202 | total += x; |
4203 | nodes[node] += x; | |
4204 | ||
4205 | page = ACCESS_ONCE(c->partial); | |
49e22585 | 4206 | if (page) { |
8afb1474 LZ |
4207 | node = page_to_nid(page); |
4208 | if (flags & SO_TOTAL) | |
4209 | WARN_ON_ONCE(1); | |
4210 | else if (flags & SO_OBJECTS) | |
4211 | WARN_ON_ONCE(1); | |
4212 | else | |
4213 | x = page->pages; | |
bc6697d8 ED |
4214 | total += x; |
4215 | nodes[node] += x; | |
49e22585 | 4216 | } |
81819f0f CL |
4217 | } |
4218 | } | |
4219 | ||
bfc8c901 | 4220 | get_online_mems(); |
ab4d5ed5 | 4221 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4222 | if (flags & SO_ALL) { |
fa45dc25 CL |
4223 | struct kmem_cache_node *n; |
4224 | ||
4225 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4226 | |
d0e0ac97 CG |
4227 | if (flags & SO_TOTAL) |
4228 | x = atomic_long_read(&n->total_objects); | |
4229 | else if (flags & SO_OBJECTS) | |
4230 | x = atomic_long_read(&n->total_objects) - | |
4231 | count_partial(n, count_free); | |
81819f0f | 4232 | else |
205ab99d | 4233 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4234 | total += x; |
4235 | nodes[node] += x; | |
4236 | } | |
4237 | ||
ab4d5ed5 CL |
4238 | } else |
4239 | #endif | |
4240 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4241 | struct kmem_cache_node *n; |
81819f0f | 4242 | |
fa45dc25 | 4243 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4244 | if (flags & SO_TOTAL) |
4245 | x = count_partial(n, count_total); | |
4246 | else if (flags & SO_OBJECTS) | |
4247 | x = count_partial(n, count_inuse); | |
81819f0f | 4248 | else |
205ab99d | 4249 | x = n->nr_partial; |
81819f0f CL |
4250 | total += x; |
4251 | nodes[node] += x; | |
4252 | } | |
4253 | } | |
81819f0f CL |
4254 | x = sprintf(buf, "%lu", total); |
4255 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4256 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4257 | if (nodes[node]) |
4258 | x += sprintf(buf + x, " N%d=%lu", | |
4259 | node, nodes[node]); | |
4260 | #endif | |
bfc8c901 | 4261 | put_online_mems(); |
81819f0f CL |
4262 | kfree(nodes); |
4263 | return x + sprintf(buf + x, "\n"); | |
4264 | } | |
4265 | ||
ab4d5ed5 | 4266 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4267 | static int any_slab_objects(struct kmem_cache *s) |
4268 | { | |
4269 | int node; | |
fa45dc25 | 4270 | struct kmem_cache_node *n; |
81819f0f | 4271 | |
fa45dc25 | 4272 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4273 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4274 | return 1; |
fa45dc25 | 4275 | |
81819f0f CL |
4276 | return 0; |
4277 | } | |
ab4d5ed5 | 4278 | #endif |
81819f0f CL |
4279 | |
4280 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4281 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4282 | |
4283 | struct slab_attribute { | |
4284 | struct attribute attr; | |
4285 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4286 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4287 | }; | |
4288 | ||
4289 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4290 | static struct slab_attribute _name##_attr = \ |
4291 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4292 | |
4293 | #define SLAB_ATTR(_name) \ | |
4294 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4295 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4296 | |
81819f0f CL |
4297 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4298 | { | |
4299 | return sprintf(buf, "%d\n", s->size); | |
4300 | } | |
4301 | SLAB_ATTR_RO(slab_size); | |
4302 | ||
4303 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4304 | { | |
4305 | return sprintf(buf, "%d\n", s->align); | |
4306 | } | |
4307 | SLAB_ATTR_RO(align); | |
4308 | ||
4309 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4310 | { | |
3b0efdfa | 4311 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4312 | } |
4313 | SLAB_ATTR_RO(object_size); | |
4314 | ||
4315 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4316 | { | |
834f3d11 | 4317 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4318 | } |
4319 | SLAB_ATTR_RO(objs_per_slab); | |
4320 | ||
06b285dc CL |
4321 | static ssize_t order_store(struct kmem_cache *s, |
4322 | const char *buf, size_t length) | |
4323 | { | |
0121c619 CL |
4324 | unsigned long order; |
4325 | int err; | |
4326 | ||
3dbb95f7 | 4327 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4328 | if (err) |
4329 | return err; | |
06b285dc CL |
4330 | |
4331 | if (order > slub_max_order || order < slub_min_order) | |
4332 | return -EINVAL; | |
4333 | ||
4334 | calculate_sizes(s, order); | |
4335 | return length; | |
4336 | } | |
4337 | ||
81819f0f CL |
4338 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4339 | { | |
834f3d11 | 4340 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4341 | } |
06b285dc | 4342 | SLAB_ATTR(order); |
81819f0f | 4343 | |
73d342b1 DR |
4344 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4345 | { | |
4346 | return sprintf(buf, "%lu\n", s->min_partial); | |
4347 | } | |
4348 | ||
4349 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4350 | size_t length) | |
4351 | { | |
4352 | unsigned long min; | |
4353 | int err; | |
4354 | ||
3dbb95f7 | 4355 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4356 | if (err) |
4357 | return err; | |
4358 | ||
c0bdb232 | 4359 | set_min_partial(s, min); |
73d342b1 DR |
4360 | return length; |
4361 | } | |
4362 | SLAB_ATTR(min_partial); | |
4363 | ||
49e22585 CL |
4364 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4365 | { | |
4366 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4367 | } | |
4368 | ||
4369 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4370 | size_t length) | |
4371 | { | |
4372 | unsigned long objects; | |
4373 | int err; | |
4374 | ||
3dbb95f7 | 4375 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4376 | if (err) |
4377 | return err; | |
345c905d | 4378 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4379 | return -EINVAL; |
49e22585 CL |
4380 | |
4381 | s->cpu_partial = objects; | |
4382 | flush_all(s); | |
4383 | return length; | |
4384 | } | |
4385 | SLAB_ATTR(cpu_partial); | |
4386 | ||
81819f0f CL |
4387 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4388 | { | |
62c70bce JP |
4389 | if (!s->ctor) |
4390 | return 0; | |
4391 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4392 | } |
4393 | SLAB_ATTR_RO(ctor); | |
4394 | ||
81819f0f CL |
4395 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4396 | { | |
4307c14f | 4397 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4398 | } |
4399 | SLAB_ATTR_RO(aliases); | |
4400 | ||
81819f0f CL |
4401 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4402 | { | |
d9acf4b7 | 4403 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4404 | } |
4405 | SLAB_ATTR_RO(partial); | |
4406 | ||
4407 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4408 | { | |
d9acf4b7 | 4409 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4410 | } |
4411 | SLAB_ATTR_RO(cpu_slabs); | |
4412 | ||
4413 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4414 | { | |
205ab99d | 4415 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4416 | } |
4417 | SLAB_ATTR_RO(objects); | |
4418 | ||
205ab99d CL |
4419 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4420 | { | |
4421 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4422 | } | |
4423 | SLAB_ATTR_RO(objects_partial); | |
4424 | ||
49e22585 CL |
4425 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4426 | { | |
4427 | int objects = 0; | |
4428 | int pages = 0; | |
4429 | int cpu; | |
4430 | int len; | |
4431 | ||
4432 | for_each_online_cpu(cpu) { | |
4433 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4434 | ||
4435 | if (page) { | |
4436 | pages += page->pages; | |
4437 | objects += page->pobjects; | |
4438 | } | |
4439 | } | |
4440 | ||
4441 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4442 | ||
4443 | #ifdef CONFIG_SMP | |
4444 | for_each_online_cpu(cpu) { | |
4445 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4446 | ||
4447 | if (page && len < PAGE_SIZE - 20) | |
4448 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4449 | page->pobjects, page->pages); | |
4450 | } | |
4451 | #endif | |
4452 | return len + sprintf(buf + len, "\n"); | |
4453 | } | |
4454 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4455 | ||
a5a84755 CL |
4456 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4457 | { | |
4458 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4459 | } | |
4460 | ||
4461 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4462 | const char *buf, size_t length) | |
4463 | { | |
4464 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4465 | if (buf[0] == '1') | |
4466 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4467 | return length; | |
4468 | } | |
4469 | SLAB_ATTR(reclaim_account); | |
4470 | ||
4471 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4472 | { | |
4473 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4474 | } | |
4475 | SLAB_ATTR_RO(hwcache_align); | |
4476 | ||
4477 | #ifdef CONFIG_ZONE_DMA | |
4478 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4479 | { | |
4480 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4481 | } | |
4482 | SLAB_ATTR_RO(cache_dma); | |
4483 | #endif | |
4484 | ||
4485 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4486 | { | |
4487 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4488 | } | |
4489 | SLAB_ATTR_RO(destroy_by_rcu); | |
4490 | ||
ab9a0f19 LJ |
4491 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4492 | { | |
4493 | return sprintf(buf, "%d\n", s->reserved); | |
4494 | } | |
4495 | SLAB_ATTR_RO(reserved); | |
4496 | ||
ab4d5ed5 | 4497 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4498 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4499 | { | |
4500 | return show_slab_objects(s, buf, SO_ALL); | |
4501 | } | |
4502 | SLAB_ATTR_RO(slabs); | |
4503 | ||
205ab99d CL |
4504 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4505 | { | |
4506 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4507 | } | |
4508 | SLAB_ATTR_RO(total_objects); | |
4509 | ||
81819f0f CL |
4510 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4511 | { | |
4512 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4513 | } | |
4514 | ||
4515 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4516 | const char *buf, size_t length) | |
4517 | { | |
4518 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4519 | if (buf[0] == '1') { |
4520 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4521 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4522 | } |
81819f0f CL |
4523 | return length; |
4524 | } | |
4525 | SLAB_ATTR(sanity_checks); | |
4526 | ||
4527 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4528 | { | |
4529 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4530 | } | |
4531 | ||
4532 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4533 | size_t length) | |
4534 | { | |
c9e16131 CL |
4535 | /* |
4536 | * Tracing a merged cache is going to give confusing results | |
4537 | * as well as cause other issues like converting a mergeable | |
4538 | * cache into an umergeable one. | |
4539 | */ | |
4540 | if (s->refcount > 1) | |
4541 | return -EINVAL; | |
4542 | ||
81819f0f | 4543 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
4544 | if (buf[0] == '1') { |
4545 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4546 | s->flags |= SLAB_TRACE; |
b789ef51 | 4547 | } |
81819f0f CL |
4548 | return length; |
4549 | } | |
4550 | SLAB_ATTR(trace); | |
4551 | ||
81819f0f CL |
4552 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4553 | { | |
4554 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4555 | } | |
4556 | ||
4557 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4558 | const char *buf, size_t length) | |
4559 | { | |
4560 | if (any_slab_objects(s)) | |
4561 | return -EBUSY; | |
4562 | ||
4563 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4564 | if (buf[0] == '1') { |
4565 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4566 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4567 | } |
06b285dc | 4568 | calculate_sizes(s, -1); |
81819f0f CL |
4569 | return length; |
4570 | } | |
4571 | SLAB_ATTR(red_zone); | |
4572 | ||
4573 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4574 | { | |
4575 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4576 | } | |
4577 | ||
4578 | static ssize_t poison_store(struct kmem_cache *s, | |
4579 | const char *buf, size_t length) | |
4580 | { | |
4581 | if (any_slab_objects(s)) | |
4582 | return -EBUSY; | |
4583 | ||
4584 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4585 | if (buf[0] == '1') { |
4586 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4587 | s->flags |= SLAB_POISON; |
b789ef51 | 4588 | } |
06b285dc | 4589 | calculate_sizes(s, -1); |
81819f0f CL |
4590 | return length; |
4591 | } | |
4592 | SLAB_ATTR(poison); | |
4593 | ||
4594 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4595 | { | |
4596 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4597 | } | |
4598 | ||
4599 | static ssize_t store_user_store(struct kmem_cache *s, | |
4600 | const char *buf, size_t length) | |
4601 | { | |
4602 | if (any_slab_objects(s)) | |
4603 | return -EBUSY; | |
4604 | ||
4605 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4606 | if (buf[0] == '1') { |
4607 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4608 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4609 | } |
06b285dc | 4610 | calculate_sizes(s, -1); |
81819f0f CL |
4611 | return length; |
4612 | } | |
4613 | SLAB_ATTR(store_user); | |
4614 | ||
53e15af0 CL |
4615 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4616 | { | |
4617 | return 0; | |
4618 | } | |
4619 | ||
4620 | static ssize_t validate_store(struct kmem_cache *s, | |
4621 | const char *buf, size_t length) | |
4622 | { | |
434e245d CL |
4623 | int ret = -EINVAL; |
4624 | ||
4625 | if (buf[0] == '1') { | |
4626 | ret = validate_slab_cache(s); | |
4627 | if (ret >= 0) | |
4628 | ret = length; | |
4629 | } | |
4630 | return ret; | |
53e15af0 CL |
4631 | } |
4632 | SLAB_ATTR(validate); | |
a5a84755 CL |
4633 | |
4634 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4635 | { | |
4636 | if (!(s->flags & SLAB_STORE_USER)) | |
4637 | return -ENOSYS; | |
4638 | return list_locations(s, buf, TRACK_ALLOC); | |
4639 | } | |
4640 | SLAB_ATTR_RO(alloc_calls); | |
4641 | ||
4642 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4643 | { | |
4644 | if (!(s->flags & SLAB_STORE_USER)) | |
4645 | return -ENOSYS; | |
4646 | return list_locations(s, buf, TRACK_FREE); | |
4647 | } | |
4648 | SLAB_ATTR_RO(free_calls); | |
4649 | #endif /* CONFIG_SLUB_DEBUG */ | |
4650 | ||
4651 | #ifdef CONFIG_FAILSLAB | |
4652 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4653 | { | |
4654 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4655 | } | |
4656 | ||
4657 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4658 | size_t length) | |
4659 | { | |
c9e16131 CL |
4660 | if (s->refcount > 1) |
4661 | return -EINVAL; | |
4662 | ||
a5a84755 CL |
4663 | s->flags &= ~SLAB_FAILSLAB; |
4664 | if (buf[0] == '1') | |
4665 | s->flags |= SLAB_FAILSLAB; | |
4666 | return length; | |
4667 | } | |
4668 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4669 | #endif |
53e15af0 | 4670 | |
2086d26a CL |
4671 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4672 | { | |
4673 | return 0; | |
4674 | } | |
4675 | ||
4676 | static ssize_t shrink_store(struct kmem_cache *s, | |
4677 | const char *buf, size_t length) | |
4678 | { | |
4679 | if (buf[0] == '1') { | |
4680 | int rc = kmem_cache_shrink(s); | |
4681 | ||
4682 | if (rc) | |
4683 | return rc; | |
4684 | } else | |
4685 | return -EINVAL; | |
4686 | return length; | |
4687 | } | |
4688 | SLAB_ATTR(shrink); | |
4689 | ||
81819f0f | 4690 | #ifdef CONFIG_NUMA |
9824601e | 4691 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4692 | { |
9824601e | 4693 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4694 | } |
4695 | ||
9824601e | 4696 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4697 | const char *buf, size_t length) |
4698 | { | |
0121c619 CL |
4699 | unsigned long ratio; |
4700 | int err; | |
4701 | ||
3dbb95f7 | 4702 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
4703 | if (err) |
4704 | return err; | |
4705 | ||
e2cb96b7 | 4706 | if (ratio <= 100) |
0121c619 | 4707 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4708 | |
81819f0f CL |
4709 | return length; |
4710 | } | |
9824601e | 4711 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4712 | #endif |
4713 | ||
8ff12cfc | 4714 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4715 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4716 | { | |
4717 | unsigned long sum = 0; | |
4718 | int cpu; | |
4719 | int len; | |
4720 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4721 | ||
4722 | if (!data) | |
4723 | return -ENOMEM; | |
4724 | ||
4725 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4726 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4727 | |
4728 | data[cpu] = x; | |
4729 | sum += x; | |
4730 | } | |
4731 | ||
4732 | len = sprintf(buf, "%lu", sum); | |
4733 | ||
50ef37b9 | 4734 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4735 | for_each_online_cpu(cpu) { |
4736 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4737 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4738 | } |
50ef37b9 | 4739 | #endif |
8ff12cfc CL |
4740 | kfree(data); |
4741 | return len + sprintf(buf + len, "\n"); | |
4742 | } | |
4743 | ||
78eb00cc DR |
4744 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4745 | { | |
4746 | int cpu; | |
4747 | ||
4748 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4749 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4750 | } |
4751 | ||
8ff12cfc CL |
4752 | #define STAT_ATTR(si, text) \ |
4753 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4754 | { \ | |
4755 | return show_stat(s, buf, si); \ | |
4756 | } \ | |
78eb00cc DR |
4757 | static ssize_t text##_store(struct kmem_cache *s, \ |
4758 | const char *buf, size_t length) \ | |
4759 | { \ | |
4760 | if (buf[0] != '0') \ | |
4761 | return -EINVAL; \ | |
4762 | clear_stat(s, si); \ | |
4763 | return length; \ | |
4764 | } \ | |
4765 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4766 | |
4767 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4768 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4769 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4770 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4771 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4772 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4773 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4774 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4775 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4776 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4777 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4778 | STAT_ATTR(FREE_SLAB, free_slab); |
4779 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4780 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4781 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4782 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4783 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4784 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4785 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4786 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4787 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4788 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
4789 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
4790 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
4791 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
4792 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
4793 | #endif |
4794 | ||
06428780 | 4795 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4796 | &slab_size_attr.attr, |
4797 | &object_size_attr.attr, | |
4798 | &objs_per_slab_attr.attr, | |
4799 | &order_attr.attr, | |
73d342b1 | 4800 | &min_partial_attr.attr, |
49e22585 | 4801 | &cpu_partial_attr.attr, |
81819f0f | 4802 | &objects_attr.attr, |
205ab99d | 4803 | &objects_partial_attr.attr, |
81819f0f CL |
4804 | &partial_attr.attr, |
4805 | &cpu_slabs_attr.attr, | |
4806 | &ctor_attr.attr, | |
81819f0f CL |
4807 | &aliases_attr.attr, |
4808 | &align_attr.attr, | |
81819f0f CL |
4809 | &hwcache_align_attr.attr, |
4810 | &reclaim_account_attr.attr, | |
4811 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4812 | &shrink_attr.attr, |
ab9a0f19 | 4813 | &reserved_attr.attr, |
49e22585 | 4814 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 4815 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4816 | &total_objects_attr.attr, |
4817 | &slabs_attr.attr, | |
4818 | &sanity_checks_attr.attr, | |
4819 | &trace_attr.attr, | |
81819f0f CL |
4820 | &red_zone_attr.attr, |
4821 | &poison_attr.attr, | |
4822 | &store_user_attr.attr, | |
53e15af0 | 4823 | &validate_attr.attr, |
88a420e4 CL |
4824 | &alloc_calls_attr.attr, |
4825 | &free_calls_attr.attr, | |
ab4d5ed5 | 4826 | #endif |
81819f0f CL |
4827 | #ifdef CONFIG_ZONE_DMA |
4828 | &cache_dma_attr.attr, | |
4829 | #endif | |
4830 | #ifdef CONFIG_NUMA | |
9824601e | 4831 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4832 | #endif |
4833 | #ifdef CONFIG_SLUB_STATS | |
4834 | &alloc_fastpath_attr.attr, | |
4835 | &alloc_slowpath_attr.attr, | |
4836 | &free_fastpath_attr.attr, | |
4837 | &free_slowpath_attr.attr, | |
4838 | &free_frozen_attr.attr, | |
4839 | &free_add_partial_attr.attr, | |
4840 | &free_remove_partial_attr.attr, | |
4841 | &alloc_from_partial_attr.attr, | |
4842 | &alloc_slab_attr.attr, | |
4843 | &alloc_refill_attr.attr, | |
e36a2652 | 4844 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
4845 | &free_slab_attr.attr, |
4846 | &cpuslab_flush_attr.attr, | |
4847 | &deactivate_full_attr.attr, | |
4848 | &deactivate_empty_attr.attr, | |
4849 | &deactivate_to_head_attr.attr, | |
4850 | &deactivate_to_tail_attr.attr, | |
4851 | &deactivate_remote_frees_attr.attr, | |
03e404af | 4852 | &deactivate_bypass_attr.attr, |
65c3376a | 4853 | &order_fallback_attr.attr, |
b789ef51 CL |
4854 | &cmpxchg_double_fail_attr.attr, |
4855 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
4856 | &cpu_partial_alloc_attr.attr, |
4857 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
4858 | &cpu_partial_node_attr.attr, |
4859 | &cpu_partial_drain_attr.attr, | |
81819f0f | 4860 | #endif |
4c13dd3b DM |
4861 | #ifdef CONFIG_FAILSLAB |
4862 | &failslab_attr.attr, | |
4863 | #endif | |
4864 | ||
81819f0f CL |
4865 | NULL |
4866 | }; | |
4867 | ||
4868 | static struct attribute_group slab_attr_group = { | |
4869 | .attrs = slab_attrs, | |
4870 | }; | |
4871 | ||
4872 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4873 | struct attribute *attr, | |
4874 | char *buf) | |
4875 | { | |
4876 | struct slab_attribute *attribute; | |
4877 | struct kmem_cache *s; | |
4878 | int err; | |
4879 | ||
4880 | attribute = to_slab_attr(attr); | |
4881 | s = to_slab(kobj); | |
4882 | ||
4883 | if (!attribute->show) | |
4884 | return -EIO; | |
4885 | ||
4886 | err = attribute->show(s, buf); | |
4887 | ||
4888 | return err; | |
4889 | } | |
4890 | ||
4891 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4892 | struct attribute *attr, | |
4893 | const char *buf, size_t len) | |
4894 | { | |
4895 | struct slab_attribute *attribute; | |
4896 | struct kmem_cache *s; | |
4897 | int err; | |
4898 | ||
4899 | attribute = to_slab_attr(attr); | |
4900 | s = to_slab(kobj); | |
4901 | ||
4902 | if (!attribute->store) | |
4903 | return -EIO; | |
4904 | ||
4905 | err = attribute->store(s, buf, len); | |
107dab5c GC |
4906 | #ifdef CONFIG_MEMCG_KMEM |
4907 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { | |
4908 | int i; | |
81819f0f | 4909 | |
107dab5c GC |
4910 | mutex_lock(&slab_mutex); |
4911 | if (s->max_attr_size < len) | |
4912 | s->max_attr_size = len; | |
4913 | ||
ebe945c2 GC |
4914 | /* |
4915 | * This is a best effort propagation, so this function's return | |
4916 | * value will be determined by the parent cache only. This is | |
4917 | * basically because not all attributes will have a well | |
4918 | * defined semantics for rollbacks - most of the actions will | |
4919 | * have permanent effects. | |
4920 | * | |
4921 | * Returning the error value of any of the children that fail | |
4922 | * is not 100 % defined, in the sense that users seeing the | |
4923 | * error code won't be able to know anything about the state of | |
4924 | * the cache. | |
4925 | * | |
4926 | * Only returning the error code for the parent cache at least | |
4927 | * has well defined semantics. The cache being written to | |
4928 | * directly either failed or succeeded, in which case we loop | |
4929 | * through the descendants with best-effort propagation. | |
4930 | */ | |
107dab5c | 4931 | for_each_memcg_cache_index(i) { |
2ade4de8 | 4932 | struct kmem_cache *c = cache_from_memcg_idx(s, i); |
107dab5c GC |
4933 | if (c) |
4934 | attribute->store(c, buf, len); | |
4935 | } | |
4936 | mutex_unlock(&slab_mutex); | |
4937 | } | |
4938 | #endif | |
81819f0f CL |
4939 | return err; |
4940 | } | |
4941 | ||
107dab5c GC |
4942 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
4943 | { | |
4944 | #ifdef CONFIG_MEMCG_KMEM | |
4945 | int i; | |
4946 | char *buffer = NULL; | |
93030d83 | 4947 | struct kmem_cache *root_cache; |
107dab5c | 4948 | |
93030d83 | 4949 | if (is_root_cache(s)) |
107dab5c GC |
4950 | return; |
4951 | ||
93030d83 VD |
4952 | root_cache = s->memcg_params->root_cache; |
4953 | ||
107dab5c GC |
4954 | /* |
4955 | * This mean this cache had no attribute written. Therefore, no point | |
4956 | * in copying default values around | |
4957 | */ | |
93030d83 | 4958 | if (!root_cache->max_attr_size) |
107dab5c GC |
4959 | return; |
4960 | ||
4961 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
4962 | char mbuf[64]; | |
4963 | char *buf; | |
4964 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
4965 | ||
4966 | if (!attr || !attr->store || !attr->show) | |
4967 | continue; | |
4968 | ||
4969 | /* | |
4970 | * It is really bad that we have to allocate here, so we will | |
4971 | * do it only as a fallback. If we actually allocate, though, | |
4972 | * we can just use the allocated buffer until the end. | |
4973 | * | |
4974 | * Most of the slub attributes will tend to be very small in | |
4975 | * size, but sysfs allows buffers up to a page, so they can | |
4976 | * theoretically happen. | |
4977 | */ | |
4978 | if (buffer) | |
4979 | buf = buffer; | |
93030d83 | 4980 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
4981 | buf = mbuf; |
4982 | else { | |
4983 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
4984 | if (WARN_ON(!buffer)) | |
4985 | continue; | |
4986 | buf = buffer; | |
4987 | } | |
4988 | ||
93030d83 | 4989 | attr->show(root_cache, buf); |
107dab5c GC |
4990 | attr->store(s, buf, strlen(buf)); |
4991 | } | |
4992 | ||
4993 | if (buffer) | |
4994 | free_page((unsigned long)buffer); | |
4995 | #endif | |
4996 | } | |
4997 | ||
41a21285 CL |
4998 | static void kmem_cache_release(struct kobject *k) |
4999 | { | |
5000 | slab_kmem_cache_release(to_slab(k)); | |
5001 | } | |
5002 | ||
52cf25d0 | 5003 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5004 | .show = slab_attr_show, |
5005 | .store = slab_attr_store, | |
5006 | }; | |
5007 | ||
5008 | static struct kobj_type slab_ktype = { | |
5009 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5010 | .release = kmem_cache_release, |
81819f0f CL |
5011 | }; |
5012 | ||
5013 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5014 | { | |
5015 | struct kobj_type *ktype = get_ktype(kobj); | |
5016 | ||
5017 | if (ktype == &slab_ktype) | |
5018 | return 1; | |
5019 | return 0; | |
5020 | } | |
5021 | ||
9cd43611 | 5022 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5023 | .filter = uevent_filter, |
5024 | }; | |
5025 | ||
27c3a314 | 5026 | static struct kset *slab_kset; |
81819f0f | 5027 | |
9a41707b VD |
5028 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5029 | { | |
5030 | #ifdef CONFIG_MEMCG_KMEM | |
5031 | if (!is_root_cache(s)) | |
5032 | return s->memcg_params->root_cache->memcg_kset; | |
5033 | #endif | |
5034 | return slab_kset; | |
5035 | } | |
5036 | ||
81819f0f CL |
5037 | #define ID_STR_LENGTH 64 |
5038 | ||
5039 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5040 | * |
5041 | * Format :[flags-]size | |
81819f0f CL |
5042 | */ |
5043 | static char *create_unique_id(struct kmem_cache *s) | |
5044 | { | |
5045 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5046 | char *p = name; | |
5047 | ||
5048 | BUG_ON(!name); | |
5049 | ||
5050 | *p++ = ':'; | |
5051 | /* | |
5052 | * First flags affecting slabcache operations. We will only | |
5053 | * get here for aliasable slabs so we do not need to support | |
5054 | * too many flags. The flags here must cover all flags that | |
5055 | * are matched during merging to guarantee that the id is | |
5056 | * unique. | |
5057 | */ | |
5058 | if (s->flags & SLAB_CACHE_DMA) | |
5059 | *p++ = 'd'; | |
5060 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5061 | *p++ = 'a'; | |
5062 | if (s->flags & SLAB_DEBUG_FREE) | |
5063 | *p++ = 'F'; | |
5a896d9e VN |
5064 | if (!(s->flags & SLAB_NOTRACK)) |
5065 | *p++ = 't'; | |
81819f0f CL |
5066 | if (p != name + 1) |
5067 | *p++ = '-'; | |
5068 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5069 | |
81819f0f CL |
5070 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5071 | return name; | |
5072 | } | |
5073 | ||
5074 | static int sysfs_slab_add(struct kmem_cache *s) | |
5075 | { | |
5076 | int err; | |
5077 | const char *name; | |
45530c44 | 5078 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5079 | |
81819f0f CL |
5080 | if (unmergeable) { |
5081 | /* | |
5082 | * Slabcache can never be merged so we can use the name proper. | |
5083 | * This is typically the case for debug situations. In that | |
5084 | * case we can catch duplicate names easily. | |
5085 | */ | |
27c3a314 | 5086 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5087 | name = s->name; |
5088 | } else { | |
5089 | /* | |
5090 | * Create a unique name for the slab as a target | |
5091 | * for the symlinks. | |
5092 | */ | |
5093 | name = create_unique_id(s); | |
5094 | } | |
5095 | ||
9a41707b | 5096 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5097 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 DJ |
5098 | if (err) |
5099 | goto out_put_kobj; | |
81819f0f CL |
5100 | |
5101 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5102 | if (err) |
5103 | goto out_del_kobj; | |
9a41707b VD |
5104 | |
5105 | #ifdef CONFIG_MEMCG_KMEM | |
5106 | if (is_root_cache(s)) { | |
5107 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5108 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5109 | err = -ENOMEM; |
5110 | goto out_del_kobj; | |
9a41707b VD |
5111 | } |
5112 | } | |
5113 | #endif | |
5114 | ||
81819f0f CL |
5115 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5116 | if (!unmergeable) { | |
5117 | /* Setup first alias */ | |
5118 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5119 | } |
54b6a731 DJ |
5120 | out: |
5121 | if (!unmergeable) | |
5122 | kfree(name); | |
5123 | return err; | |
5124 | out_del_kobj: | |
5125 | kobject_del(&s->kobj); | |
5126 | out_put_kobj: | |
5127 | kobject_put(&s->kobj); | |
5128 | goto out; | |
81819f0f CL |
5129 | } |
5130 | ||
41a21285 | 5131 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5132 | { |
97d06609 | 5133 | if (slab_state < FULL) |
2bce6485 CL |
5134 | /* |
5135 | * Sysfs has not been setup yet so no need to remove the | |
5136 | * cache from sysfs. | |
5137 | */ | |
5138 | return; | |
5139 | ||
9a41707b VD |
5140 | #ifdef CONFIG_MEMCG_KMEM |
5141 | kset_unregister(s->memcg_kset); | |
5142 | #endif | |
81819f0f CL |
5143 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5144 | kobject_del(&s->kobj); | |
151c602f | 5145 | kobject_put(&s->kobj); |
81819f0f CL |
5146 | } |
5147 | ||
5148 | /* | |
5149 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5150 | * available lest we lose that information. |
81819f0f CL |
5151 | */ |
5152 | struct saved_alias { | |
5153 | struct kmem_cache *s; | |
5154 | const char *name; | |
5155 | struct saved_alias *next; | |
5156 | }; | |
5157 | ||
5af328a5 | 5158 | static struct saved_alias *alias_list; |
81819f0f CL |
5159 | |
5160 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5161 | { | |
5162 | struct saved_alias *al; | |
5163 | ||
97d06609 | 5164 | if (slab_state == FULL) { |
81819f0f CL |
5165 | /* |
5166 | * If we have a leftover link then remove it. | |
5167 | */ | |
27c3a314 GKH |
5168 | sysfs_remove_link(&slab_kset->kobj, name); |
5169 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5170 | } |
5171 | ||
5172 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5173 | if (!al) | |
5174 | return -ENOMEM; | |
5175 | ||
5176 | al->s = s; | |
5177 | al->name = name; | |
5178 | al->next = alias_list; | |
5179 | alias_list = al; | |
5180 | return 0; | |
5181 | } | |
5182 | ||
5183 | static int __init slab_sysfs_init(void) | |
5184 | { | |
5b95a4ac | 5185 | struct kmem_cache *s; |
81819f0f CL |
5186 | int err; |
5187 | ||
18004c5d | 5188 | mutex_lock(&slab_mutex); |
2bce6485 | 5189 | |
0ff21e46 | 5190 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5191 | if (!slab_kset) { |
18004c5d | 5192 | mutex_unlock(&slab_mutex); |
f9f58285 | 5193 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5194 | return -ENOSYS; |
5195 | } | |
5196 | ||
97d06609 | 5197 | slab_state = FULL; |
26a7bd03 | 5198 | |
5b95a4ac | 5199 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5200 | err = sysfs_slab_add(s); |
5d540fb7 | 5201 | if (err) |
f9f58285 FF |
5202 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5203 | s->name); | |
26a7bd03 | 5204 | } |
81819f0f CL |
5205 | |
5206 | while (alias_list) { | |
5207 | struct saved_alias *al = alias_list; | |
5208 | ||
5209 | alias_list = alias_list->next; | |
5210 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5211 | if (err) |
f9f58285 FF |
5212 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5213 | al->name); | |
81819f0f CL |
5214 | kfree(al); |
5215 | } | |
5216 | ||
18004c5d | 5217 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5218 | resiliency_test(); |
5219 | return 0; | |
5220 | } | |
5221 | ||
5222 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5223 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5224 | |
5225 | /* | |
5226 | * The /proc/slabinfo ABI | |
5227 | */ | |
158a9624 | 5228 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5229 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5230 | { |
57ed3eda | 5231 | unsigned long nr_slabs = 0; |
205ab99d CL |
5232 | unsigned long nr_objs = 0; |
5233 | unsigned long nr_free = 0; | |
57ed3eda | 5234 | int node; |
fa45dc25 | 5235 | struct kmem_cache_node *n; |
57ed3eda | 5236 | |
fa45dc25 | 5237 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5238 | nr_slabs += node_nr_slabs(n); |
5239 | nr_objs += node_nr_objs(n); | |
205ab99d | 5240 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5241 | } |
5242 | ||
0d7561c6 GC |
5243 | sinfo->active_objs = nr_objs - nr_free; |
5244 | sinfo->num_objs = nr_objs; | |
5245 | sinfo->active_slabs = nr_slabs; | |
5246 | sinfo->num_slabs = nr_slabs; | |
5247 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5248 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5249 | } |
5250 | ||
0d7561c6 | 5251 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5252 | { |
7b3c3a50 AD |
5253 | } |
5254 | ||
b7454ad3 GC |
5255 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5256 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5257 | { |
b7454ad3 | 5258 | return -EIO; |
7b3c3a50 | 5259 | } |
158a9624 | 5260 | #endif /* CONFIG_SLABINFO */ |